Merge branch 'jc/dirwalk-n-cache-tree' into jc/cache-tree
[git/jrn.git] / read-tree.c
blob99e7c750444117fe07b25bd75d454f03da2134fe
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>
16 static int reset = 0;
17 static int merge = 0;
18 static int update = 0;
19 static int index_only = 0;
20 static int nontrivial_merge = 0;
21 static int trivial_merges_only = 0;
22 static int aggressive = 0;
23 static int verbose_update = 0;
24 static volatile int progress_update = 0;
26 static int head_idx = -1;
27 static int merge_size = 0;
29 static struct object_list *trees = NULL;
31 static struct cache_entry df_conflict_entry = {
34 static struct tree_entry_list df_conflict_list = {
35 .name = NULL,
36 .next = &df_conflict_list
39 typedef int (*merge_fn_t)(struct cache_entry **src);
41 static int entcmp(char *name1, int dir1, char *name2, int dir2)
43 int len1 = strlen(name1);
44 int len2 = strlen(name2);
45 int len = len1 < len2 ? len1 : len2;
46 int ret = memcmp(name1, name2, len);
47 unsigned char c1, c2;
48 if (ret)
49 return ret;
50 c1 = name1[len];
51 c2 = name2[len];
52 if (!c1 && dir1)
53 c1 = '/';
54 if (!c2 && dir2)
55 c2 = '/';
56 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
57 if (c1 && c2 && !ret)
58 ret = len1 - len2;
59 return ret;
62 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
63 const char *base, merge_fn_t fn, int *indpos)
65 int baselen = strlen(base);
66 int src_size = len + 1;
67 do {
68 int i;
69 char *first;
70 int firstdir = 0;
71 int pathlen;
72 unsigned ce_size;
73 struct tree_entry_list **subposns;
74 struct cache_entry **src;
75 int any_files = 0;
76 int any_dirs = 0;
77 char *cache_name;
78 int ce_stage;
80 /* Find the first name in the input. */
82 first = NULL;
83 cache_name = NULL;
85 /* Check the cache */
86 if (merge && *indpos < active_nr) {
87 /* This is a bit tricky: */
88 /* If the index has a subdirectory (with
89 * contents) as the first name, it'll get a
90 * filename like "foo/bar". But that's after
91 * "foo", so the entry in trees will get
92 * handled first, at which point we'll go into
93 * "foo", and deal with "bar" from the index,
94 * because the base will be "foo/". The only
95 * way we can actually have "foo/bar" first of
96 * all the things is if the trees don't
97 * contain "foo" at all, in which case we'll
98 * handle "foo/bar" without going into the
99 * directory, but that's fine (and will return
100 * an error anyway, with the added unknown
101 * file case.
104 cache_name = active_cache[*indpos]->name;
105 if (strlen(cache_name) > baselen &&
106 !memcmp(cache_name, base, baselen)) {
107 cache_name += baselen;
108 first = cache_name;
109 } else {
110 cache_name = NULL;
114 #if DBRT_DEBUG > 1
115 if (first)
116 printf("index %s\n", first);
117 #endif
118 for (i = 0; i < len; i++) {
119 if (!posns[i] || posns[i] == &df_conflict_list)
120 continue;
121 #if DBRT_DEBUG > 1
122 printf("%d %s\n", i + 1, posns[i]->name);
123 #endif
124 if (!first || entcmp(first, firstdir,
125 posns[i]->name,
126 posns[i]->directory) > 0) {
127 first = posns[i]->name;
128 firstdir = posns[i]->directory;
131 /* No name means we're done */
132 if (!first)
133 return 0;
135 pathlen = strlen(first);
136 ce_size = cache_entry_size(baselen + pathlen);
138 src = xcalloc(src_size, sizeof(struct cache_entry *));
140 subposns = xcalloc(len, sizeof(struct tree_list_entry *));
142 if (cache_name && !strcmp(cache_name, first)) {
143 any_files = 1;
144 src[0] = active_cache[*indpos];
145 remove_cache_entry_at(*indpos);
148 for (i = 0; i < len; i++) {
149 struct cache_entry *ce;
151 if (!posns[i] ||
152 (posns[i] != &df_conflict_list &&
153 strcmp(first, posns[i]->name))) {
154 continue;
157 if (posns[i] == &df_conflict_list) {
158 src[i + merge] = &df_conflict_entry;
159 continue;
162 if (posns[i]->directory) {
163 any_dirs = 1;
164 parse_tree(posns[i]->item.tree);
165 subposns[i] = posns[i]->item.tree->entries;
166 posns[i] = posns[i]->next;
167 src[i + merge] = &df_conflict_entry;
168 continue;
171 if (!merge)
172 ce_stage = 0;
173 else if (i + 1 < head_idx)
174 ce_stage = 1;
175 else if (i + 1 > head_idx)
176 ce_stage = 3;
177 else
178 ce_stage = 2;
180 ce = xcalloc(1, ce_size);
181 ce->ce_mode = create_ce_mode(posns[i]->mode);
182 ce->ce_flags = create_ce_flags(baselen + pathlen,
183 ce_stage);
184 memcpy(ce->name, base, baselen);
185 memcpy(ce->name + baselen, first, pathlen + 1);
187 any_files = 1;
189 memcpy(ce->sha1, posns[i]->item.any->sha1, 20);
190 src[i + merge] = ce;
191 subposns[i] = &df_conflict_list;
192 posns[i] = posns[i]->next;
194 if (any_files) {
195 if (merge) {
196 int ret;
198 #if DBRT_DEBUG > 1
199 printf("%s:\n", first);
200 for (i = 0; i < src_size; i++) {
201 printf(" %d ", i);
202 if (src[i])
203 printf("%s\n", sha1_to_hex(src[i]->sha1));
204 else
205 printf("\n");
207 #endif
208 ret = fn(src);
210 #if DBRT_DEBUG > 1
211 printf("Added %d entries\n", ret);
212 #endif
213 *indpos += ret;
214 } else {
215 for (i = 0; i < src_size; i++) {
216 if (src[i]) {
217 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
222 if (any_dirs) {
223 char *newbase = xmalloc(baselen + 2 + pathlen);
224 memcpy(newbase, base, baselen);
225 memcpy(newbase + baselen, first, pathlen);
226 newbase[baselen + pathlen] = '/';
227 newbase[baselen + pathlen + 1] = '\0';
228 if (unpack_trees_rec(subposns, len, newbase, fn,
229 indpos))
230 return -1;
231 free(newbase);
233 free(subposns);
234 free(src);
235 } while (1);
238 static void reject_merge(struct cache_entry *ce)
240 die("Entry '%s' would be overwritten by merge. Cannot merge.",
241 ce->name);
244 /* Unlink the last component and attempt to remove leading
245 * directories, in case this unlink is the removal of the
246 * last entry in the directory -- empty directories are removed.
248 static void unlink_entry(char *name)
250 char *cp, *prev;
252 if (unlink(name))
253 return;
254 prev = NULL;
255 while (1) {
256 int status;
257 cp = strrchr(name, '/');
258 if (prev)
259 *prev = '/';
260 if (!cp)
261 break;
263 *cp = 0;
264 status = rmdir(name);
265 if (status) {
266 *cp = '/';
267 break;
269 prev = cp;
273 static void progress_interval(int signum)
275 progress_update = 1;
278 static void setup_progress_signal(void)
280 struct sigaction sa;
281 struct itimerval v;
283 memset(&sa, 0, sizeof(sa));
284 sa.sa_handler = progress_interval;
285 sigemptyset(&sa.sa_mask);
286 sa.sa_flags = SA_RESTART;
287 sigaction(SIGALRM, &sa, NULL);
289 v.it_interval.tv_sec = 1;
290 v.it_interval.tv_usec = 0;
291 v.it_value = v.it_interval;
292 setitimer(ITIMER_REAL, &v, NULL);
295 static void check_updates(struct cache_entry **src, int nr)
297 static struct checkout state = {
298 .base_dir = "",
299 .force = 1,
300 .quiet = 1,
301 .refresh_cache = 1,
303 unsigned short mask = htons(CE_UPDATE);
304 unsigned last_percent = 200, cnt = 0, total = 0;
306 if (update && verbose_update) {
307 for (total = cnt = 0; cnt < nr; cnt++) {
308 struct cache_entry *ce = src[cnt];
309 if (!ce->ce_mode || ce->ce_flags & mask)
310 total++;
313 /* Don't bother doing this for very small updates */
314 if (total < 250)
315 total = 0;
317 if (total) {
318 fprintf(stderr, "Checking files out...\n");
319 setup_progress_signal();
320 progress_update = 1;
322 cnt = 0;
325 while (nr--) {
326 struct cache_entry *ce = *src++;
328 if (total) {
329 if (!ce->ce_mode || ce->ce_flags & mask) {
330 unsigned percent;
331 cnt++;
332 percent = (cnt * 100) / total;
333 if (percent != last_percent ||
334 progress_update) {
335 fprintf(stderr, "%4u%% (%u/%u) done\r",
336 percent, cnt, total);
337 last_percent = percent;
341 if (!ce->ce_mode) {
342 if (update)
343 unlink_entry(ce->name);
344 continue;
346 if (ce->ce_flags & mask) {
347 ce->ce_flags &= ~mask;
348 if (update)
349 checkout_entry(ce, &state, NULL);
352 if (total) {
353 signal(SIGALRM, SIG_IGN);
354 fputc('\n', stderr);
358 static int unpack_trees(merge_fn_t fn)
360 int indpos = 0;
361 unsigned len = object_list_length(trees);
362 struct tree_entry_list **posns;
363 int i;
364 struct object_list *posn = trees;
365 merge_size = len;
367 if (len) {
368 posns = xmalloc(len * sizeof(struct tree_entry_list *));
369 for (i = 0; i < len; i++) {
370 posns[i] = ((struct tree *) posn->item)->entries;
371 posn = posn->next;
373 if (unpack_trees_rec(posns, len, "", fn, &indpos))
374 return -1;
377 if (trivial_merges_only && nontrivial_merge)
378 die("Merge requires file-level merging");
380 check_updates(active_cache, active_nr);
381 return 0;
384 static int list_tree(unsigned char *sha1)
386 struct tree *tree = parse_tree_indirect(sha1);
387 if (!tree)
388 return -1;
389 object_list_append(&tree->object, &trees);
390 return 0;
393 static int same(struct cache_entry *a, struct cache_entry *b)
395 if (!!a != !!b)
396 return 0;
397 if (!a && !b)
398 return 1;
399 return a->ce_mode == b->ce_mode &&
400 !memcmp(a->sha1, b->sha1, 20);
405 * When a CE gets turned into an unmerged entry, we
406 * want it to be up-to-date
408 static void verify_uptodate(struct cache_entry *ce)
410 struct stat st;
412 if (index_only || reset)
413 return;
415 if (!lstat(ce->name, &st)) {
416 unsigned changed = ce_match_stat(ce, &st, 1);
417 if (!changed)
418 return;
419 errno = 0;
421 if (reset) {
422 ce->ce_flags |= htons(CE_UPDATE);
423 return;
425 if (errno == ENOENT)
426 return;
427 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
430 static void invalidate_ce_path(struct cache_entry *ce)
432 if (ce)
433 cache_tree_invalidate_path(active_cache_tree, ce->name);
437 * We do not want to remove or overwrite a working tree file that
438 * is not tracked.
440 static void verify_absent(const char *path, const char *action)
442 struct stat st;
444 if (index_only || reset || !update)
445 return;
446 if (!lstat(path, &st))
447 die("Untracked working tree file '%s' "
448 "would be %s by merge.", path, action);
451 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
453 merge->ce_flags |= htons(CE_UPDATE);
454 if (old) {
456 * See if we can re-use the old CE directly?
457 * That way we get the uptodate stat info.
459 * This also removes the UPDATE flag on
460 * a match.
462 if (same(old, merge)) {
463 *merge = *old;
464 } else {
465 verify_uptodate(old);
466 invalidate_ce_path(old);
469 else {
470 verify_absent(merge->name, "overwritten");
471 invalidate_ce_path(merge);
474 merge->ce_flags &= ~htons(CE_STAGEMASK);
475 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
476 return 1;
479 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
481 if (old)
482 verify_uptodate(old);
483 else
484 verify_absent(ce->name, "removed");
485 ce->ce_mode = 0;
486 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
487 invalidate_ce_path(ce);
488 return 1;
491 static int keep_entry(struct cache_entry *ce)
493 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
494 return 1;
497 #if DBRT_DEBUG
498 static void show_stage_entry(FILE *o,
499 const char *label, const struct cache_entry *ce)
501 if (!ce)
502 fprintf(o, "%s (missing)\n", label);
503 else
504 fprintf(o, "%s%06o %s %d\t%s\n",
505 label,
506 ntohl(ce->ce_mode),
507 sha1_to_hex(ce->sha1),
508 ce_stage(ce),
509 ce->name);
511 #endif
513 static int threeway_merge(struct cache_entry **stages)
515 struct cache_entry *index;
516 struct cache_entry *head;
517 struct cache_entry *remote = stages[head_idx + 1];
518 int count;
519 int head_match = 0;
520 int remote_match = 0;
521 const char *path = NULL;
523 int df_conflict_head = 0;
524 int df_conflict_remote = 0;
526 int any_anc_missing = 0;
527 int no_anc_exists = 1;
528 int i;
530 for (i = 1; i < head_idx; i++) {
531 if (!stages[i])
532 any_anc_missing = 1;
533 else {
534 if (!path)
535 path = stages[i]->name;
536 no_anc_exists = 0;
540 index = stages[0];
541 head = stages[head_idx];
543 if (head == &df_conflict_entry) {
544 df_conflict_head = 1;
545 head = NULL;
548 if (remote == &df_conflict_entry) {
549 df_conflict_remote = 1;
550 remote = NULL;
553 if (!path && index)
554 path = index->name;
555 if (!path && head)
556 path = head->name;
557 if (!path && remote)
558 path = remote->name;
560 /* First, if there's a #16 situation, note that to prevent #13
561 * and #14.
563 if (!same(remote, head)) {
564 for (i = 1; i < head_idx; i++) {
565 if (same(stages[i], head)) {
566 head_match = i;
568 if (same(stages[i], remote)) {
569 remote_match = i;
574 /* We start with cases where the index is allowed to match
575 * something other than the head: #14(ALT) and #2ALT, where it
576 * is permitted to match the result instead.
578 /* #14, #14ALT, #2ALT */
579 if (remote && !df_conflict_head && head_match && !remote_match) {
580 if (index && !same(index, remote) && !same(index, head))
581 reject_merge(index);
582 return merged_entry(remote, index);
585 * If we have an entry in the index cache, then we want to
586 * make sure that it matches head.
588 if (index && !same(index, head)) {
589 reject_merge(index);
592 if (head) {
593 /* #5ALT, #15 */
594 if (same(head, remote))
595 return merged_entry(head, index);
596 /* #13, #3ALT */
597 if (!df_conflict_remote && remote_match && !head_match)
598 return merged_entry(head, index);
601 /* #1 */
602 if (!head && !remote && any_anc_missing)
603 return 0;
605 /* Under the new "aggressive" rule, we resolve mostly trivial
606 * cases that we historically had git-merge-one-file resolve.
608 if (aggressive) {
609 int head_deleted = !head && !df_conflict_head;
610 int remote_deleted = !remote && !df_conflict_remote;
612 * Deleted in both.
613 * Deleted in one and unchanged in the other.
615 if ((head_deleted && remote_deleted) ||
616 (head_deleted && remote && remote_match) ||
617 (remote_deleted && head && head_match)) {
618 if (index)
619 return deleted_entry(index, index);
620 else if (path)
621 verify_absent(path, "removed");
622 return 0;
625 * Added in both, identically.
627 if (no_anc_exists && head && remote && same(head, remote))
628 return merged_entry(head, index);
632 /* Below are "no merge" cases, which require that the index be
633 * up-to-date to avoid the files getting overwritten with
634 * conflict resolution files.
636 if (index) {
637 verify_uptodate(index);
639 else if (path)
640 verify_absent(path, "overwritten");
642 nontrivial_merge = 1;
644 /* #2, #3, #4, #6, #7, #9, #11. */
645 count = 0;
646 if (!head_match || !remote_match) {
647 for (i = 1; i < head_idx; i++) {
648 if (stages[i]) {
649 keep_entry(stages[i]);
650 count++;
651 break;
655 #if DBRT_DEBUG
656 else {
657 fprintf(stderr, "read-tree: warning #16 detected\n");
658 show_stage_entry(stderr, "head ", stages[head_match]);
659 show_stage_entry(stderr, "remote ", stages[remote_match]);
661 #endif
662 if (head) { count += keep_entry(head); }
663 if (remote) { count += keep_entry(remote); }
664 return count;
668 * Two-way merge.
670 * The rule is to "carry forward" what is in the index without losing
671 * information across a "fast forward", favoring a successful merge
672 * over a merge failure when it makes sense. For details of the
673 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
676 static int twoway_merge(struct cache_entry **src)
678 struct cache_entry *current = src[0];
679 struct cache_entry *oldtree = src[1], *newtree = src[2];
681 if (merge_size != 2)
682 return error("Cannot do a twoway merge of %d trees",
683 merge_size);
685 if (current) {
686 if ((!oldtree && !newtree) || /* 4 and 5 */
687 (!oldtree && newtree &&
688 same(current, newtree)) || /* 6 and 7 */
689 (oldtree && newtree &&
690 same(oldtree, newtree)) || /* 14 and 15 */
691 (oldtree && newtree &&
692 !same(oldtree, newtree) && /* 18 and 19*/
693 same(current, newtree))) {
694 return keep_entry(current);
696 else if (oldtree && !newtree && same(current, oldtree)) {
697 /* 10 or 11 */
698 return deleted_entry(oldtree, current);
700 else if (oldtree && newtree &&
701 same(current, oldtree) && !same(current, newtree)) {
702 /* 20 or 21 */
703 return merged_entry(newtree, current);
705 else {
706 /* all other failures */
707 if (oldtree)
708 reject_merge(oldtree);
709 if (current)
710 reject_merge(current);
711 if (newtree)
712 reject_merge(newtree);
713 return -1;
716 else if (newtree)
717 return merged_entry(newtree, current);
718 else
719 return deleted_entry(oldtree, current);
723 * One-way merge.
725 * The rule is:
726 * - take the stat information from stage0, take the data from stage1
728 static int oneway_merge(struct cache_entry **src)
730 struct cache_entry *old = src[0];
731 struct cache_entry *a = src[1];
733 if (merge_size != 1)
734 return error("Cannot do a oneway merge of %d trees",
735 merge_size);
737 if (!a)
738 return deleted_entry(old, old);
739 if (old && same(old, a)) {
740 if (reset) {
741 struct stat st;
742 if (lstat(old->name, &st) ||
743 ce_match_stat(old, &st, 1))
744 old->ce_flags |= htons(CE_UPDATE);
746 return keep_entry(old);
748 return merged_entry(a, old);
751 static int read_cache_unmerged(void)
753 int i, deleted;
754 struct cache_entry **dst;
756 read_cache();
757 dst = active_cache;
758 deleted = 0;
759 for (i = 0; i < active_nr; i++) {
760 struct cache_entry *ce = active_cache[i];
761 if (ce_stage(ce)) {
762 deleted++;
763 invalidate_ce_path(ce);
764 continue;
766 if (deleted)
767 *dst = ce;
768 dst++;
770 active_nr -= deleted;
771 return deleted;
774 static void prime_cache_tree_rec(struct cache_tree *it, struct tree *tree)
776 struct tree_entry_list *ent;
777 int cnt;
779 memcpy(it->sha1, tree->object.sha1, 20);
780 for (cnt = 0, ent = tree->entries; ent; ent = ent->next) {
781 if (!ent->directory)
782 cnt++;
783 else {
784 struct cache_tree_sub *sub;
785 struct tree *subtree = (struct tree *)ent->item.tree;
786 if (!subtree->object.parsed)
787 parse_tree(subtree);
788 sub = cache_tree_sub(it, ent->name);
789 sub->cache_tree = cache_tree();
790 prime_cache_tree_rec(sub->cache_tree, subtree);
791 cnt += sub->cache_tree->entry_count;
794 it->entry_count = cnt;
797 static void prime_cache_tree(void)
799 struct tree *tree = (struct tree *)trees->item;
800 if (!tree)
801 return;
802 active_cache_tree = cache_tree();
803 prime_cache_tree_rec(active_cache_tree, tree);
807 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
809 static struct cache_file cache_file;
811 int main(int argc, char **argv)
813 int i, newfd, stage = 0;
814 unsigned char sha1[20];
815 merge_fn_t fn = NULL;
817 setup_git_directory();
818 git_config(git_default_config);
820 newfd = hold_index_file_for_update(&cache_file, get_index_file());
821 if (newfd < 0)
822 die("unable to create new cachefile");
824 git_config(git_default_config);
826 merge = 0;
827 reset = 0;
828 for (i = 1; i < argc; i++) {
829 const char *arg = argv[i];
831 /* "-u" means "update", meaning that a merge will update
832 * the working tree.
834 if (!strcmp(arg, "-u")) {
835 update = 1;
836 continue;
839 if (!strcmp(arg, "-v")) {
840 verbose_update = 1;
841 continue;
844 /* "-i" means "index only", meaning that a merge will
845 * not even look at the working tree.
847 if (!strcmp(arg, "-i")) {
848 index_only = 1;
849 continue;
852 /* This differs from "-m" in that we'll silently ignore unmerged entries */
853 if (!strcmp(arg, "--reset")) {
854 if (stage || merge)
855 usage(read_tree_usage);
856 reset = 1;
857 merge = 1;
858 stage = 1;
859 read_cache_unmerged();
860 continue;
863 if (!strcmp(arg, "--trivial")) {
864 trivial_merges_only = 1;
865 continue;
868 if (!strcmp(arg, "--aggressive")) {
869 aggressive = 1;
870 continue;
873 /* "-m" stands for "merge", meaning we start in stage 1 */
874 if (!strcmp(arg, "-m")) {
875 if (stage || merge)
876 usage(read_tree_usage);
877 if (read_cache_unmerged())
878 die("you need to resolve your current index first");
879 stage = 1;
880 merge = 1;
881 continue;
884 /* using -u and -i at the same time makes no sense */
885 if (1 < index_only + update)
886 usage(read_tree_usage);
888 if (get_sha1(arg, sha1))
889 die("Not a valid object name %s", arg);
890 if (list_tree(sha1) < 0)
891 die("failed to unpack tree object %s", arg);
892 stage++;
894 if ((update||index_only) && !merge)
895 usage(read_tree_usage);
897 if (merge) {
898 if (stage < 2)
899 die("just how do you expect me to merge %d trees?", stage-1);
900 switch (stage - 1) {
901 case 1:
902 fn = oneway_merge;
903 break;
904 case 2:
905 fn = twoway_merge;
906 break;
907 case 3:
908 default:
909 fn = threeway_merge;
910 cache_tree_free(&active_cache_tree);
911 break;
914 if (stage - 1 >= 3)
915 head_idx = stage - 2;
916 else
917 head_idx = 1;
920 unpack_trees(fn);
923 * When reading only one tree (either the most basic form,
924 * "-m ent" or "--reset ent" form), we can obtain a fully
925 * valid cache-tree because the index must match exactly
926 * what came from the tree.
928 if (trees && trees->item && (!merge || (stage == 2))) {
929 cache_tree_free(&active_cache_tree);
930 prime_cache_tree();
933 if (write_cache(newfd, active_cache, active_nr) ||
934 commit_index_file(&cache_file))
935 die("unable to write new index file");
936 return 0;