Merge branch 'jc/cache-tree'
[git/dscho.git] / builtin-read-tree.c
blob716f7925142bca11cf7ab5438ce5908e55285463
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"
12 #include "cache-tree.h"
13 #include <sys/time.h>
14 #include <signal.h>
15 #include "builtin.h"
17 static int reset = 0;
18 static int merge = 0;
19 static int update = 0;
20 static int index_only = 0;
21 static int nontrivial_merge = 0;
22 static int trivial_merges_only = 0;
23 static int aggressive = 0;
24 static int verbose_update = 0;
25 static volatile int progress_update = 0;
27 static int head_idx = -1;
28 static int merge_size = 0;
30 static struct object_list *trees = NULL;
32 static struct cache_entry df_conflict_entry = {
35 static struct tree_entry_list df_conflict_list = {
36 .name = NULL,
37 .next = &df_conflict_list
40 typedef int (*merge_fn_t)(struct cache_entry **src);
42 static int entcmp(char *name1, int dir1, char *name2, int dir2)
44 int len1 = strlen(name1);
45 int len2 = strlen(name2);
46 int len = len1 < len2 ? len1 : len2;
47 int ret = memcmp(name1, name2, len);
48 unsigned char c1, c2;
49 if (ret)
50 return ret;
51 c1 = name1[len];
52 c2 = name2[len];
53 if (!c1 && dir1)
54 c1 = '/';
55 if (!c2 && dir2)
56 c2 = '/';
57 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
58 if (c1 && c2 && !ret)
59 ret = len1 - len2;
60 return ret;
63 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
64 const char *base, merge_fn_t fn, int *indpos)
66 int baselen = strlen(base);
67 int src_size = len + 1;
68 do {
69 int i;
70 char *first;
71 int firstdir = 0;
72 int pathlen;
73 unsigned ce_size;
74 struct tree_entry_list **subposns;
75 struct cache_entry **src;
76 int any_files = 0;
77 int any_dirs = 0;
78 char *cache_name;
79 int ce_stage;
81 /* Find the first name in the input. */
83 first = NULL;
84 cache_name = NULL;
86 /* Check the cache */
87 if (merge && *indpos < active_nr) {
88 /* This is a bit tricky: */
89 /* If the index has a subdirectory (with
90 * contents) as the first name, it'll get a
91 * filename like "foo/bar". But that's after
92 * "foo", so the entry in trees will get
93 * handled first, at which point we'll go into
94 * "foo", and deal with "bar" from the index,
95 * because the base will be "foo/". The only
96 * way we can actually have "foo/bar" first of
97 * all the things is if the trees don't
98 * contain "foo" at all, in which case we'll
99 * handle "foo/bar" without going into the
100 * directory, but that's fine (and will return
101 * an error anyway, with the added unknown
102 * file case.
105 cache_name = active_cache[*indpos]->name;
106 if (strlen(cache_name) > baselen &&
107 !memcmp(cache_name, base, baselen)) {
108 cache_name += baselen;
109 first = cache_name;
110 } else {
111 cache_name = NULL;
115 #if DBRT_DEBUG > 1
116 if (first)
117 printf("index %s\n", first);
118 #endif
119 for (i = 0; i < len; i++) {
120 if (!posns[i] || posns[i] == &df_conflict_list)
121 continue;
122 #if DBRT_DEBUG > 1
123 printf("%d %s\n", i + 1, posns[i]->name);
124 #endif
125 if (!first || entcmp(first, firstdir,
126 posns[i]->name,
127 posns[i]->directory) > 0) {
128 first = posns[i]->name;
129 firstdir = posns[i]->directory;
132 /* No name means we're done */
133 if (!first)
134 return 0;
136 pathlen = strlen(first);
137 ce_size = cache_entry_size(baselen + pathlen);
139 src = xcalloc(src_size, sizeof(struct cache_entry *));
141 subposns = xcalloc(len, sizeof(struct tree_list_entry *));
143 if (cache_name && !strcmp(cache_name, first)) {
144 any_files = 1;
145 src[0] = active_cache[*indpos];
146 remove_cache_entry_at(*indpos);
149 for (i = 0; i < len; i++) {
150 struct cache_entry *ce;
152 if (!posns[i] ||
153 (posns[i] != &df_conflict_list &&
154 strcmp(first, posns[i]->name))) {
155 continue;
158 if (posns[i] == &df_conflict_list) {
159 src[i + merge] = &df_conflict_entry;
160 continue;
163 if (posns[i]->directory) {
164 any_dirs = 1;
165 parse_tree(posns[i]->item.tree);
166 subposns[i] = posns[i]->item.tree->entries;
167 posns[i] = posns[i]->next;
168 src[i + merge] = &df_conflict_entry;
169 continue;
172 if (!merge)
173 ce_stage = 0;
174 else if (i + 1 < head_idx)
175 ce_stage = 1;
176 else if (i + 1 > head_idx)
177 ce_stage = 3;
178 else
179 ce_stage = 2;
181 ce = xcalloc(1, ce_size);
182 ce->ce_mode = create_ce_mode(posns[i]->mode);
183 ce->ce_flags = create_ce_flags(baselen + pathlen,
184 ce_stage);
185 memcpy(ce->name, base, baselen);
186 memcpy(ce->name + baselen, first, pathlen + 1);
188 any_files = 1;
190 memcpy(ce->sha1, posns[i]->item.any->sha1, 20);
191 src[i + merge] = ce;
192 subposns[i] = &df_conflict_list;
193 posns[i] = posns[i]->next;
195 if (any_files) {
196 if (merge) {
197 int ret;
199 #if DBRT_DEBUG > 1
200 printf("%s:\n", first);
201 for (i = 0; i < src_size; i++) {
202 printf(" %d ", i);
203 if (src[i])
204 printf("%s\n", sha1_to_hex(src[i]->sha1));
205 else
206 printf("\n");
208 #endif
209 ret = fn(src);
211 #if DBRT_DEBUG > 1
212 printf("Added %d entries\n", ret);
213 #endif
214 *indpos += ret;
215 } else {
216 for (i = 0; i < src_size; i++) {
217 if (src[i]) {
218 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
223 if (any_dirs) {
224 char *newbase = xmalloc(baselen + 2 + pathlen);
225 memcpy(newbase, base, baselen);
226 memcpy(newbase + baselen, first, pathlen);
227 newbase[baselen + pathlen] = '/';
228 newbase[baselen + pathlen + 1] = '\0';
229 if (unpack_trees_rec(subposns, len, newbase, fn,
230 indpos))
231 return -1;
232 free(newbase);
234 free(subposns);
235 free(src);
236 } while (1);
239 static void reject_merge(struct cache_entry *ce)
241 die("Entry '%s' would be overwritten by merge. Cannot merge.",
242 ce->name);
245 /* Unlink the last component and attempt to remove leading
246 * directories, in case this unlink is the removal of the
247 * last entry in the directory -- empty directories are removed.
249 static void unlink_entry(char *name)
251 char *cp, *prev;
253 if (unlink(name))
254 return;
255 prev = NULL;
256 while (1) {
257 int status;
258 cp = strrchr(name, '/');
259 if (prev)
260 *prev = '/';
261 if (!cp)
262 break;
264 *cp = 0;
265 status = rmdir(name);
266 if (status) {
267 *cp = '/';
268 break;
270 prev = cp;
274 static void progress_interval(int signum)
276 progress_update = 1;
279 static void setup_progress_signal(void)
281 struct sigaction sa;
282 struct itimerval v;
284 memset(&sa, 0, sizeof(sa));
285 sa.sa_handler = progress_interval;
286 sigemptyset(&sa.sa_mask);
287 sa.sa_flags = SA_RESTART;
288 sigaction(SIGALRM, &sa, NULL);
290 v.it_interval.tv_sec = 1;
291 v.it_interval.tv_usec = 0;
292 v.it_value = v.it_interval;
293 setitimer(ITIMER_REAL, &v, NULL);
296 static void check_updates(struct cache_entry **src, int nr)
298 static struct checkout state = {
299 .base_dir = "",
300 .force = 1,
301 .quiet = 1,
302 .refresh_cache = 1,
304 unsigned short mask = htons(CE_UPDATE);
305 unsigned last_percent = 200, cnt = 0, total = 0;
307 if (update && verbose_update) {
308 for (total = cnt = 0; cnt < nr; cnt++) {
309 struct cache_entry *ce = src[cnt];
310 if (!ce->ce_mode || ce->ce_flags & mask)
311 total++;
314 /* Don't bother doing this for very small updates */
315 if (total < 250)
316 total = 0;
318 if (total) {
319 fprintf(stderr, "Checking files out...\n");
320 setup_progress_signal();
321 progress_update = 1;
323 cnt = 0;
326 while (nr--) {
327 struct cache_entry *ce = *src++;
329 if (total) {
330 if (!ce->ce_mode || ce->ce_flags & mask) {
331 unsigned percent;
332 cnt++;
333 percent = (cnt * 100) / total;
334 if (percent != last_percent ||
335 progress_update) {
336 fprintf(stderr, "%4u%% (%u/%u) done\r",
337 percent, cnt, total);
338 last_percent = percent;
342 if (!ce->ce_mode) {
343 if (update)
344 unlink_entry(ce->name);
345 continue;
347 if (ce->ce_flags & mask) {
348 ce->ce_flags &= ~mask;
349 if (update)
350 checkout_entry(ce, &state, NULL);
353 if (total) {
354 signal(SIGALRM, SIG_IGN);
355 fputc('\n', stderr);
359 static int unpack_trees(merge_fn_t fn)
361 int indpos = 0;
362 unsigned len = object_list_length(trees);
363 struct tree_entry_list **posns;
364 int i;
365 struct object_list *posn = trees;
366 merge_size = len;
368 if (len) {
369 posns = xmalloc(len * sizeof(struct tree_entry_list *));
370 for (i = 0; i < len; i++) {
371 posns[i] = ((struct tree *) posn->item)->entries;
372 posn = posn->next;
374 if (unpack_trees_rec(posns, len, "", fn, &indpos))
375 return -1;
378 if (trivial_merges_only && nontrivial_merge)
379 die("Merge requires file-level merging");
381 check_updates(active_cache, active_nr);
382 return 0;
385 static int list_tree(unsigned char *sha1)
387 struct tree *tree = parse_tree_indirect(sha1);
388 if (!tree)
389 return -1;
390 object_list_append(&tree->object, &trees);
391 return 0;
394 static int same(struct cache_entry *a, struct cache_entry *b)
396 if (!!a != !!b)
397 return 0;
398 if (!a && !b)
399 return 1;
400 return a->ce_mode == b->ce_mode &&
401 !memcmp(a->sha1, b->sha1, 20);
406 * When a CE gets turned into an unmerged entry, we
407 * want it to be up-to-date
409 static void verify_uptodate(struct cache_entry *ce)
411 struct stat st;
413 if (index_only || reset)
414 return;
416 if (!lstat(ce->name, &st)) {
417 unsigned changed = ce_match_stat(ce, &st, 1);
418 if (!changed)
419 return;
420 errno = 0;
422 if (reset) {
423 ce->ce_flags |= htons(CE_UPDATE);
424 return;
426 if (errno == ENOENT)
427 return;
428 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
431 static void invalidate_ce_path(struct cache_entry *ce)
433 if (ce)
434 cache_tree_invalidate_path(active_cache_tree, ce->name);
438 * We do not want to remove or overwrite a working tree file that
439 * is not tracked.
441 static void verify_absent(const char *path, const char *action)
443 struct stat st;
445 if (index_only || reset || !update)
446 return;
447 if (!lstat(path, &st))
448 die("Untracked working tree file '%s' "
449 "would be %s by merge.", path, action);
452 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
454 merge->ce_flags |= htons(CE_UPDATE);
455 if (old) {
457 * See if we can re-use the old CE directly?
458 * That way we get the uptodate stat info.
460 * This also removes the UPDATE flag on
461 * a match.
463 if (same(old, merge)) {
464 *merge = *old;
465 } else {
466 verify_uptodate(old);
467 invalidate_ce_path(old);
470 else {
471 verify_absent(merge->name, "overwritten");
472 invalidate_ce_path(merge);
475 merge->ce_flags &= ~htons(CE_STAGEMASK);
476 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
477 return 1;
480 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
482 if (old)
483 verify_uptodate(old);
484 else
485 verify_absent(ce->name, "removed");
486 ce->ce_mode = 0;
487 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
488 invalidate_ce_path(ce);
489 return 1;
492 static int keep_entry(struct cache_entry *ce)
494 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
495 return 1;
498 #if DBRT_DEBUG
499 static void show_stage_entry(FILE *o,
500 const char *label, const struct cache_entry *ce)
502 if (!ce)
503 fprintf(o, "%s (missing)\n", label);
504 else
505 fprintf(o, "%s%06o %s %d\t%s\n",
506 label,
507 ntohl(ce->ce_mode),
508 sha1_to_hex(ce->sha1),
509 ce_stage(ce),
510 ce->name);
512 #endif
514 static int threeway_merge(struct cache_entry **stages)
516 struct cache_entry *index;
517 struct cache_entry *head;
518 struct cache_entry *remote = stages[head_idx + 1];
519 int count;
520 int head_match = 0;
521 int remote_match = 0;
522 const char *path = NULL;
524 int df_conflict_head = 0;
525 int df_conflict_remote = 0;
527 int any_anc_missing = 0;
528 int no_anc_exists = 1;
529 int i;
531 for (i = 1; i < head_idx; i++) {
532 if (!stages[i])
533 any_anc_missing = 1;
534 else {
535 if (!path)
536 path = stages[i]->name;
537 no_anc_exists = 0;
541 index = stages[0];
542 head = stages[head_idx];
544 if (head == &df_conflict_entry) {
545 df_conflict_head = 1;
546 head = NULL;
549 if (remote == &df_conflict_entry) {
550 df_conflict_remote = 1;
551 remote = NULL;
554 if (!path && index)
555 path = index->name;
556 if (!path && head)
557 path = head->name;
558 if (!path && remote)
559 path = remote->name;
561 /* First, if there's a #16 situation, note that to prevent #13
562 * and #14.
564 if (!same(remote, head)) {
565 for (i = 1; i < head_idx; i++) {
566 if (same(stages[i], head)) {
567 head_match = i;
569 if (same(stages[i], remote)) {
570 remote_match = i;
575 /* We start with cases where the index is allowed to match
576 * something other than the head: #14(ALT) and #2ALT, where it
577 * is permitted to match the result instead.
579 /* #14, #14ALT, #2ALT */
580 if (remote && !df_conflict_head && head_match && !remote_match) {
581 if (index && !same(index, remote) && !same(index, head))
582 reject_merge(index);
583 return merged_entry(remote, index);
586 * If we have an entry in the index cache, then we want to
587 * make sure that it matches head.
589 if (index && !same(index, head)) {
590 reject_merge(index);
593 if (head) {
594 /* #5ALT, #15 */
595 if (same(head, remote))
596 return merged_entry(head, index);
597 /* #13, #3ALT */
598 if (!df_conflict_remote && remote_match && !head_match)
599 return merged_entry(head, index);
602 /* #1 */
603 if (!head && !remote && any_anc_missing)
604 return 0;
606 /* Under the new "aggressive" rule, we resolve mostly trivial
607 * cases that we historically had git-merge-one-file resolve.
609 if (aggressive) {
610 int head_deleted = !head && !df_conflict_head;
611 int remote_deleted = !remote && !df_conflict_remote;
613 * Deleted in both.
614 * Deleted in one and unchanged in the other.
616 if ((head_deleted && remote_deleted) ||
617 (head_deleted && remote && remote_match) ||
618 (remote_deleted && head && head_match)) {
619 if (index)
620 return deleted_entry(index, index);
621 else if (path)
622 verify_absent(path, "removed");
623 return 0;
626 * Added in both, identically.
628 if (no_anc_exists && head && remote && same(head, remote))
629 return merged_entry(head, index);
633 /* Below are "no merge" cases, which require that the index be
634 * up-to-date to avoid the files getting overwritten with
635 * conflict resolution files.
637 if (index) {
638 verify_uptodate(index);
640 else if (path)
641 verify_absent(path, "overwritten");
643 nontrivial_merge = 1;
645 /* #2, #3, #4, #6, #7, #9, #11. */
646 count = 0;
647 if (!head_match || !remote_match) {
648 for (i = 1; i < head_idx; i++) {
649 if (stages[i]) {
650 keep_entry(stages[i]);
651 count++;
652 break;
656 #if DBRT_DEBUG
657 else {
658 fprintf(stderr, "read-tree: warning #16 detected\n");
659 show_stage_entry(stderr, "head ", stages[head_match]);
660 show_stage_entry(stderr, "remote ", stages[remote_match]);
662 #endif
663 if (head) { count += keep_entry(head); }
664 if (remote) { count += keep_entry(remote); }
665 return count;
669 * Two-way merge.
671 * The rule is to "carry forward" what is in the index without losing
672 * information across a "fast forward", favoring a successful merge
673 * over a merge failure when it makes sense. For details of the
674 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
677 static int twoway_merge(struct cache_entry **src)
679 struct cache_entry *current = src[0];
680 struct cache_entry *oldtree = src[1], *newtree = src[2];
682 if (merge_size != 2)
683 return error("Cannot do a twoway merge of %d trees",
684 merge_size);
686 if (current) {
687 if ((!oldtree && !newtree) || /* 4 and 5 */
688 (!oldtree && newtree &&
689 same(current, newtree)) || /* 6 and 7 */
690 (oldtree && newtree &&
691 same(oldtree, newtree)) || /* 14 and 15 */
692 (oldtree && newtree &&
693 !same(oldtree, newtree) && /* 18 and 19*/
694 same(current, newtree))) {
695 return keep_entry(current);
697 else if (oldtree && !newtree && same(current, oldtree)) {
698 /* 10 or 11 */
699 return deleted_entry(oldtree, current);
701 else if (oldtree && newtree &&
702 same(current, oldtree) && !same(current, newtree)) {
703 /* 20 or 21 */
704 return merged_entry(newtree, current);
706 else {
707 /* all other failures */
708 if (oldtree)
709 reject_merge(oldtree);
710 if (current)
711 reject_merge(current);
712 if (newtree)
713 reject_merge(newtree);
714 return -1;
717 else if (newtree)
718 return merged_entry(newtree, current);
719 else
720 return deleted_entry(oldtree, current);
724 * One-way merge.
726 * The rule is:
727 * - take the stat information from stage0, take the data from stage1
729 static int oneway_merge(struct cache_entry **src)
731 struct cache_entry *old = src[0];
732 struct cache_entry *a = src[1];
734 if (merge_size != 1)
735 return error("Cannot do a oneway merge of %d trees",
736 merge_size);
738 if (!a)
739 return deleted_entry(old, old);
740 if (old && same(old, a)) {
741 if (reset) {
742 struct stat st;
743 if (lstat(old->name, &st) ||
744 ce_match_stat(old, &st, 1))
745 old->ce_flags |= htons(CE_UPDATE);
747 return keep_entry(old);
749 return merged_entry(a, old);
752 static int read_cache_unmerged(void)
754 int i, deleted;
755 struct cache_entry **dst;
757 read_cache();
758 dst = active_cache;
759 deleted = 0;
760 for (i = 0; i < active_nr; i++) {
761 struct cache_entry *ce = active_cache[i];
762 if (ce_stage(ce)) {
763 deleted++;
764 invalidate_ce_path(ce);
765 continue;
767 if (deleted)
768 *dst = ce;
769 dst++;
771 active_nr -= deleted;
772 return deleted;
775 static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
777 struct tree_entry_list *ent;
778 int cnt;
780 memcpy(it->sha1, tree->object.sha1, 20);
781 for (cnt = 0, ent = tree->entries; ent; ent = ent->next) {
782 if (!ent->directory)
783 cnt++;
784 else {
785 struct cache_tree_sub *sub;
786 struct tree *subtree = (struct tree *)ent->item.tree;
787 if (!subtree->object.parsed)
788 parse_tree(subtree);
789 sub = cache_tree_sub(it, ent->name);
790 sub->cache_tree = cache_tree();
791 prime_cache_tree_rec(sub->cache_tree, subtree);
792 cnt += sub->cache_tree->entry_count;
795 it->entry_count = cnt;
798 static void prime_cache_tree(void)
800 struct tree *tree = (struct tree *)trees->item;
801 if (!tree)
802 return;
803 active_cache_tree = cache_tree();
804 prime_cache_tree_rec(active_cache_tree, tree);
808 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
810 static struct cache_file cache_file;
812 int cmd_read_tree(int argc, const char **argv, char **envp)
814 int i, newfd, stage = 0;
815 unsigned char sha1[20];
816 merge_fn_t fn = NULL;
818 setup_git_directory();
819 git_config(git_default_config);
821 newfd = hold_index_file_for_update(&cache_file, get_index_file());
822 if (newfd < 0)
823 die("unable to create new cachefile");
825 git_config(git_default_config);
827 merge = 0;
828 reset = 0;
829 for (i = 1; i < argc; i++) {
830 const char *arg = argv[i];
832 /* "-u" means "update", meaning that a merge will update
833 * the working tree.
835 if (!strcmp(arg, "-u")) {
836 update = 1;
837 continue;
840 if (!strcmp(arg, "-v")) {
841 verbose_update = 1;
842 continue;
845 /* "-i" means "index only", meaning that a merge will
846 * not even look at the working tree.
848 if (!strcmp(arg, "-i")) {
849 index_only = 1;
850 continue;
853 /* This differs from "-m" in that we'll silently ignore unmerged entries */
854 if (!strcmp(arg, "--reset")) {
855 if (stage || merge)
856 usage(read_tree_usage);
857 reset = 1;
858 merge = 1;
859 stage = 1;
860 read_cache_unmerged();
861 continue;
864 if (!strcmp(arg, "--trivial")) {
865 trivial_merges_only = 1;
866 continue;
869 if (!strcmp(arg, "--aggressive")) {
870 aggressive = 1;
871 continue;
874 /* "-m" stands for "merge", meaning we start in stage 1 */
875 if (!strcmp(arg, "-m")) {
876 if (stage || merge)
877 usage(read_tree_usage);
878 if (read_cache_unmerged())
879 die("you need to resolve your current index first");
880 stage = 1;
881 merge = 1;
882 continue;
885 /* using -u and -i at the same time makes no sense */
886 if (1 < index_only + update)
887 usage(read_tree_usage);
889 if (get_sha1(arg, sha1))
890 die("Not a valid object name %s", arg);
891 if (list_tree(sha1) < 0)
892 die("failed to unpack tree object %s", arg);
893 stage++;
895 if ((update||index_only) && !merge)
896 usage(read_tree_usage);
898 if (merge) {
899 if (stage < 2)
900 die("just how do you expect me to merge %d trees?", stage-1);
901 switch (stage - 1) {
902 case 1:
903 fn = oneway_merge;
904 break;
905 case 2:
906 fn = twoway_merge;
907 break;
908 case 3:
909 default:
910 fn = threeway_merge;
911 cache_tree_free(&active_cache_tree);
912 break;
915 if (stage - 1 >= 3)
916 head_idx = stage - 2;
917 else
918 head_idx = 1;
921 unpack_trees(fn);
924 * When reading only one tree (either the most basic form,
925 * "-m ent" or "--reset ent" form), we can obtain a fully
926 * valid cache-tree because the index must match exactly
927 * what came from the tree.
929 if (trees && trees->item && (!merge || (stage == 2))) {
930 cache_tree_free(&active_cache_tree);
931 prime_cache_tree();
934 if (write_cache(newfd, active_cache, active_nr) ||
935 commit_index_file(&cache_file))
936 die("unable to write new index file");
937 return 0;