Switch "read_tree_recursive()" over to tree-walk functionality
[git/debian.git] / builtin-read-tree.c
blob5e513c85caee73d96a6efe862400e88ca43d717d
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(const char *name1, int dir1, const 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 const 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 struct tree *tree = lookup_tree(posns[i]->sha1);
165 any_dirs = 1;
166 parse_tree(tree);
167 subposns[i] = tree->entries;
168 posns[i] = posns[i]->next;
169 src[i + merge] = &df_conflict_entry;
170 continue;
173 if (!merge)
174 ce_stage = 0;
175 else if (i + 1 < head_idx)
176 ce_stage = 1;
177 else if (i + 1 > head_idx)
178 ce_stage = 3;
179 else
180 ce_stage = 2;
182 ce = xcalloc(1, ce_size);
183 ce->ce_mode = create_ce_mode(posns[i]->mode);
184 ce->ce_flags = create_ce_flags(baselen + pathlen,
185 ce_stage);
186 memcpy(ce->name, base, baselen);
187 memcpy(ce->name + baselen, first, pathlen + 1);
189 any_files = 1;
191 memcpy(ce->sha1, posns[i]->sha1, 20);
192 src[i + merge] = ce;
193 subposns[i] = &df_conflict_list;
194 posns[i] = posns[i]->next;
196 if (any_files) {
197 if (merge) {
198 int ret;
200 #if DBRT_DEBUG > 1
201 printf("%s:\n", first);
202 for (i = 0; i < src_size; i++) {
203 printf(" %d ", i);
204 if (src[i])
205 printf("%s\n", sha1_to_hex(src[i]->sha1));
206 else
207 printf("\n");
209 #endif
210 ret = fn(src);
212 #if DBRT_DEBUG > 1
213 printf("Added %d entries\n", ret);
214 #endif
215 *indpos += ret;
216 } else {
217 for (i = 0; i < src_size; i++) {
218 if (src[i]) {
219 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
224 if (any_dirs) {
225 char *newbase = xmalloc(baselen + 2 + pathlen);
226 memcpy(newbase, base, baselen);
227 memcpy(newbase + baselen, first, pathlen);
228 newbase[baselen + pathlen] = '/';
229 newbase[baselen + pathlen + 1] = '\0';
230 if (unpack_trees_rec(subposns, len, newbase, fn,
231 indpos))
232 return -1;
233 free(newbase);
235 free(subposns);
236 free(src);
237 } while (1);
240 static void reject_merge(struct cache_entry *ce)
242 die("Entry '%s' would be overwritten by merge. Cannot merge.",
243 ce->name);
246 /* Unlink the last component and attempt to remove leading
247 * directories, in case this unlink is the removal of the
248 * last entry in the directory -- empty directories are removed.
250 static void unlink_entry(char *name)
252 char *cp, *prev;
254 if (unlink(name))
255 return;
256 prev = NULL;
257 while (1) {
258 int status;
259 cp = strrchr(name, '/');
260 if (prev)
261 *prev = '/';
262 if (!cp)
263 break;
265 *cp = 0;
266 status = rmdir(name);
267 if (status) {
268 *cp = '/';
269 break;
271 prev = cp;
275 static void progress_interval(int signum)
277 progress_update = 1;
280 static void setup_progress_signal(void)
282 struct sigaction sa;
283 struct itimerval v;
285 memset(&sa, 0, sizeof(sa));
286 sa.sa_handler = progress_interval;
287 sigemptyset(&sa.sa_mask);
288 sa.sa_flags = SA_RESTART;
289 sigaction(SIGALRM, &sa, NULL);
291 v.it_interval.tv_sec = 1;
292 v.it_interval.tv_usec = 0;
293 v.it_value = v.it_interval;
294 setitimer(ITIMER_REAL, &v, NULL);
297 static void check_updates(struct cache_entry **src, int nr)
299 static struct checkout state = {
300 .base_dir = "",
301 .force = 1,
302 .quiet = 1,
303 .refresh_cache = 1,
305 unsigned short mask = htons(CE_UPDATE);
306 unsigned last_percent = 200, cnt = 0, total = 0;
308 if (update && verbose_update) {
309 for (total = cnt = 0; cnt < nr; cnt++) {
310 struct cache_entry *ce = src[cnt];
311 if (!ce->ce_mode || ce->ce_flags & mask)
312 total++;
315 /* Don't bother doing this for very small updates */
316 if (total < 250)
317 total = 0;
319 if (total) {
320 fprintf(stderr, "Checking files out...\n");
321 setup_progress_signal();
322 progress_update = 1;
324 cnt = 0;
327 while (nr--) {
328 struct cache_entry *ce = *src++;
330 if (total) {
331 if (!ce->ce_mode || ce->ce_flags & mask) {
332 unsigned percent;
333 cnt++;
334 percent = (cnt * 100) / total;
335 if (percent != last_percent ||
336 progress_update) {
337 fprintf(stderr, "%4u%% (%u/%u) done\r",
338 percent, cnt, total);
339 last_percent = percent;
343 if (!ce->ce_mode) {
344 if (update)
345 unlink_entry(ce->name);
346 continue;
348 if (ce->ce_flags & mask) {
349 ce->ce_flags &= ~mask;
350 if (update)
351 checkout_entry(ce, &state, NULL);
354 if (total) {
355 signal(SIGALRM, SIG_IGN);
356 fputc('\n', stderr);
360 static int unpack_trees(merge_fn_t fn)
362 int indpos = 0;
363 unsigned len = object_list_length(trees);
364 struct tree_entry_list **posns;
365 int i;
366 struct object_list *posn = trees;
367 merge_size = len;
369 if (len) {
370 posns = xmalloc(len * sizeof(struct tree_entry_list *));
371 for (i = 0; i < len; i++) {
372 posns[i] = ((struct tree *) posn->item)->entries;
373 posn = posn->next;
375 if (unpack_trees_rec(posns, len, "", fn, &indpos))
376 return -1;
379 if (trivial_merges_only && nontrivial_merge)
380 die("Merge requires file-level merging");
382 check_updates(active_cache, active_nr);
383 return 0;
386 static int list_tree(unsigned char *sha1)
388 struct tree *tree = parse_tree_indirect(sha1);
389 if (!tree)
390 return -1;
391 object_list_append(&tree->object, &trees);
392 return 0;
395 static int same(struct cache_entry *a, struct cache_entry *b)
397 if (!!a != !!b)
398 return 0;
399 if (!a && !b)
400 return 1;
401 return a->ce_mode == b->ce_mode &&
402 !memcmp(a->sha1, b->sha1, 20);
407 * When a CE gets turned into an unmerged entry, we
408 * want it to be up-to-date
410 static void verify_uptodate(struct cache_entry *ce)
412 struct stat st;
414 if (index_only || reset)
415 return;
417 if (!lstat(ce->name, &st)) {
418 unsigned changed = ce_match_stat(ce, &st, 1);
419 if (!changed)
420 return;
421 errno = 0;
423 if (reset) {
424 ce->ce_flags |= htons(CE_UPDATE);
425 return;
427 if (errno == ENOENT)
428 return;
429 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
432 static void invalidate_ce_path(struct cache_entry *ce)
434 if (ce)
435 cache_tree_invalidate_path(active_cache_tree, ce->name);
439 * We do not want to remove or overwrite a working tree file that
440 * is not tracked.
442 static void verify_absent(const char *path, const char *action)
444 struct stat st;
446 if (index_only || reset || !update)
447 return;
448 if (!lstat(path, &st))
449 die("Untracked working tree file '%s' "
450 "would be %s by merge.", path, action);
453 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
455 merge->ce_flags |= htons(CE_UPDATE);
456 if (old) {
458 * See if we can re-use the old CE directly?
459 * That way we get the uptodate stat info.
461 * This also removes the UPDATE flag on
462 * a match.
464 if (same(old, merge)) {
465 *merge = *old;
466 } else {
467 verify_uptodate(old);
468 invalidate_ce_path(old);
471 else {
472 verify_absent(merge->name, "overwritten");
473 invalidate_ce_path(merge);
476 merge->ce_flags &= ~htons(CE_STAGEMASK);
477 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
478 return 1;
481 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
483 if (old)
484 verify_uptodate(old);
485 else
486 verify_absent(ce->name, "removed");
487 ce->ce_mode = 0;
488 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
489 invalidate_ce_path(ce);
490 return 1;
493 static int keep_entry(struct cache_entry *ce)
495 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
496 return 1;
499 #if DBRT_DEBUG
500 static void show_stage_entry(FILE *o,
501 const char *label, const struct cache_entry *ce)
503 if (!ce)
504 fprintf(o, "%s (missing)\n", label);
505 else
506 fprintf(o, "%s%06o %s %d\t%s\n",
507 label,
508 ntohl(ce->ce_mode),
509 sha1_to_hex(ce->sha1),
510 ce_stage(ce),
511 ce->name);
513 #endif
515 static int threeway_merge(struct cache_entry **stages)
517 struct cache_entry *index;
518 struct cache_entry *head;
519 struct cache_entry *remote = stages[head_idx + 1];
520 int count;
521 int head_match = 0;
522 int remote_match = 0;
523 const char *path = NULL;
525 int df_conflict_head = 0;
526 int df_conflict_remote = 0;
528 int any_anc_missing = 0;
529 int no_anc_exists = 1;
530 int i;
532 for (i = 1; i < head_idx; i++) {
533 if (!stages[i])
534 any_anc_missing = 1;
535 else {
536 if (!path)
537 path = stages[i]->name;
538 no_anc_exists = 0;
542 index = stages[0];
543 head = stages[head_idx];
545 if (head == &df_conflict_entry) {
546 df_conflict_head = 1;
547 head = NULL;
550 if (remote == &df_conflict_entry) {
551 df_conflict_remote = 1;
552 remote = NULL;
555 if (!path && index)
556 path = index->name;
557 if (!path && head)
558 path = head->name;
559 if (!path && remote)
560 path = remote->name;
562 /* First, if there's a #16 situation, note that to prevent #13
563 * and #14.
565 if (!same(remote, head)) {
566 for (i = 1; i < head_idx; i++) {
567 if (same(stages[i], head)) {
568 head_match = i;
570 if (same(stages[i], remote)) {
571 remote_match = i;
576 /* We start with cases where the index is allowed to match
577 * something other than the head: #14(ALT) and #2ALT, where it
578 * is permitted to match the result instead.
580 /* #14, #14ALT, #2ALT */
581 if (remote && !df_conflict_head && head_match && !remote_match) {
582 if (index && !same(index, remote) && !same(index, head))
583 reject_merge(index);
584 return merged_entry(remote, index);
587 * If we have an entry in the index cache, then we want to
588 * make sure that it matches head.
590 if (index && !same(index, head)) {
591 reject_merge(index);
594 if (head) {
595 /* #5ALT, #15 */
596 if (same(head, remote))
597 return merged_entry(head, index);
598 /* #13, #3ALT */
599 if (!df_conflict_remote && remote_match && !head_match)
600 return merged_entry(head, index);
603 /* #1 */
604 if (!head && !remote && any_anc_missing)
605 return 0;
607 /* Under the new "aggressive" rule, we resolve mostly trivial
608 * cases that we historically had git-merge-one-file resolve.
610 if (aggressive) {
611 int head_deleted = !head && !df_conflict_head;
612 int remote_deleted = !remote && !df_conflict_remote;
614 * Deleted in both.
615 * Deleted in one and unchanged in the other.
617 if ((head_deleted && remote_deleted) ||
618 (head_deleted && remote && remote_match) ||
619 (remote_deleted && head && head_match)) {
620 if (index)
621 return deleted_entry(index, index);
622 else if (path)
623 verify_absent(path, "removed");
624 return 0;
627 * Added in both, identically.
629 if (no_anc_exists && head && remote && same(head, remote))
630 return merged_entry(head, index);
634 /* Below are "no merge" cases, which require that the index be
635 * up-to-date to avoid the files getting overwritten with
636 * conflict resolution files.
638 if (index) {
639 verify_uptodate(index);
641 else if (path)
642 verify_absent(path, "overwritten");
644 nontrivial_merge = 1;
646 /* #2, #3, #4, #6, #7, #9, #11. */
647 count = 0;
648 if (!head_match || !remote_match) {
649 for (i = 1; i < head_idx; i++) {
650 if (stages[i]) {
651 keep_entry(stages[i]);
652 count++;
653 break;
657 #if DBRT_DEBUG
658 else {
659 fprintf(stderr, "read-tree: warning #16 detected\n");
660 show_stage_entry(stderr, "head ", stages[head_match]);
661 show_stage_entry(stderr, "remote ", stages[remote_match]);
663 #endif
664 if (head) { count += keep_entry(head); }
665 if (remote) { count += keep_entry(remote); }
666 return count;
670 * Two-way merge.
672 * The rule is to "carry forward" what is in the index without losing
673 * information across a "fast forward", favoring a successful merge
674 * over a merge failure when it makes sense. For details of the
675 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
678 static int twoway_merge(struct cache_entry **src)
680 struct cache_entry *current = src[0];
681 struct cache_entry *oldtree = src[1], *newtree = src[2];
683 if (merge_size != 2)
684 return error("Cannot do a twoway merge of %d trees",
685 merge_size);
687 if (current) {
688 if ((!oldtree && !newtree) || /* 4 and 5 */
689 (!oldtree && newtree &&
690 same(current, newtree)) || /* 6 and 7 */
691 (oldtree && newtree &&
692 same(oldtree, newtree)) || /* 14 and 15 */
693 (oldtree && newtree &&
694 !same(oldtree, newtree) && /* 18 and 19*/
695 same(current, newtree))) {
696 return keep_entry(current);
698 else if (oldtree && !newtree && same(current, oldtree)) {
699 /* 10 or 11 */
700 return deleted_entry(oldtree, current);
702 else if (oldtree && newtree &&
703 same(current, oldtree) && !same(current, newtree)) {
704 /* 20 or 21 */
705 return merged_entry(newtree, current);
707 else {
708 /* all other failures */
709 if (oldtree)
710 reject_merge(oldtree);
711 if (current)
712 reject_merge(current);
713 if (newtree)
714 reject_merge(newtree);
715 return -1;
718 else if (newtree)
719 return merged_entry(newtree, current);
720 else
721 return deleted_entry(oldtree, current);
725 * One-way merge.
727 * The rule is:
728 * - take the stat information from stage0, take the data from stage1
730 static int oneway_merge(struct cache_entry **src)
732 struct cache_entry *old = src[0];
733 struct cache_entry *a = src[1];
735 if (merge_size != 1)
736 return error("Cannot do a oneway merge of %d trees",
737 merge_size);
739 if (!a)
740 return deleted_entry(old, old);
741 if (old && same(old, a)) {
742 if (reset) {
743 struct stat st;
744 if (lstat(old->name, &st) ||
745 ce_match_stat(old, &st, 1))
746 old->ce_flags |= htons(CE_UPDATE);
748 return keep_entry(old);
750 return merged_entry(a, old);
753 static int read_cache_unmerged(void)
755 int i, deleted;
756 struct cache_entry **dst;
758 read_cache();
759 dst = active_cache;
760 deleted = 0;
761 for (i = 0; i < active_nr; i++) {
762 struct cache_entry *ce = active_cache[i];
763 if (ce_stage(ce)) {
764 deleted++;
765 invalidate_ce_path(ce);
766 continue;
768 if (deleted)
769 *dst = ce;
770 dst++;
772 active_nr -= deleted;
773 return deleted;
776 static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
778 struct tree_entry_list *ent;
779 int cnt;
781 memcpy(it->sha1, tree->object.sha1, 20);
782 for (cnt = 0, ent = tree->entries; ent; ent = ent->next) {
783 if (!ent->directory)
784 cnt++;
785 else {
786 struct cache_tree_sub *sub;
787 struct tree *subtree = lookup_tree(ent->sha1);
788 if (!subtree->object.parsed)
789 parse_tree(subtree);
790 sub = cache_tree_sub(it, ent->name);
791 sub->cache_tree = cache_tree();
792 prime_cache_tree_rec(sub->cache_tree, subtree);
793 cnt += sub->cache_tree->entry_count;
796 it->entry_count = cnt;
799 static void prime_cache_tree(void)
801 struct tree *tree = (struct tree *)trees->item;
802 if (!tree)
803 return;
804 active_cache_tree = cache_tree();
805 prime_cache_tree_rec(active_cache_tree, tree);
809 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
811 static struct cache_file cache_file;
813 int cmd_read_tree(int argc, const char **argv, char **envp)
815 int i, newfd, stage = 0;
816 unsigned char sha1[20];
817 merge_fn_t fn = NULL;
819 setup_git_directory();
820 git_config(git_default_config);
822 newfd = hold_index_file_for_update(&cache_file, get_index_file());
823 if (newfd < 0)
824 die("unable to create new cachefile");
826 git_config(git_default_config);
828 merge = 0;
829 reset = 0;
830 for (i = 1; i < argc; i++) {
831 const char *arg = argv[i];
833 /* "-u" means "update", meaning that a merge will update
834 * the working tree.
836 if (!strcmp(arg, "-u")) {
837 update = 1;
838 continue;
841 if (!strcmp(arg, "-v")) {
842 verbose_update = 1;
843 continue;
846 /* "-i" means "index only", meaning that a merge will
847 * not even look at the working tree.
849 if (!strcmp(arg, "-i")) {
850 index_only = 1;
851 continue;
854 /* This differs from "-m" in that we'll silently ignore unmerged entries */
855 if (!strcmp(arg, "--reset")) {
856 if (stage || merge)
857 usage(read_tree_usage);
858 reset = 1;
859 merge = 1;
860 stage = 1;
861 read_cache_unmerged();
862 continue;
865 if (!strcmp(arg, "--trivial")) {
866 trivial_merges_only = 1;
867 continue;
870 if (!strcmp(arg, "--aggressive")) {
871 aggressive = 1;
872 continue;
875 /* "-m" stands for "merge", meaning we start in stage 1 */
876 if (!strcmp(arg, "-m")) {
877 if (stage || merge)
878 usage(read_tree_usage);
879 if (read_cache_unmerged())
880 die("you need to resolve your current index first");
881 stage = 1;
882 merge = 1;
883 continue;
886 /* using -u and -i at the same time makes no sense */
887 if (1 < index_only + update)
888 usage(read_tree_usage);
890 if (get_sha1(arg, sha1))
891 die("Not a valid object name %s", arg);
892 if (list_tree(sha1) < 0)
893 die("failed to unpack tree object %s", arg);
894 stage++;
896 if ((update||index_only) && !merge)
897 usage(read_tree_usage);
899 if (merge) {
900 if (stage < 2)
901 die("just how do you expect me to merge %d trees?", stage-1);
902 switch (stage - 1) {
903 case 1:
904 fn = oneway_merge;
905 break;
906 case 2:
907 fn = twoway_merge;
908 break;
909 case 3:
910 default:
911 fn = threeway_merge;
912 cache_tree_free(&active_cache_tree);
913 break;
916 if (stage - 1 >= 3)
917 head_idx = stage - 2;
918 else
919 head_idx = 1;
922 unpack_trees(fn);
925 * When reading only one tree (either the most basic form,
926 * "-m ent" or "--reset ent" form), we can obtain a fully
927 * valid cache-tree because the index must match exactly
928 * what came from the tree.
930 if (trees && trees->item && (!merge || (stage == 2))) {
931 cache_tree_free(&active_cache_tree);
932 prime_cache_tree();
935 if (write_cache(newfd, active_cache, active_nr) ||
936 commit_index_file(&cache_file))
937 die("unable to write new index file");
938 return 0;