Make "tree_entry" have a SHA1 instead of a union of object pointers
[git/jnareb-git.git] / builtin-read-tree.c
blobf0b8dad6eb5c469b78f08c12aa0f62befbb54e80
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 <sys/time.h>
13 #include <signal.h>
14 #include "builtin.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(const char *name1, int dir1, const 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 const 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 struct tree *tree = lookup_tree(posns[i]->sha1);
164 any_dirs = 1;
165 parse_tree(tree);
166 subposns[i] = 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]->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);
432 * We do not want to remove or overwrite a working tree file that
433 * is not tracked.
435 static void verify_absent(const char *path, const char *action)
437 struct stat st;
439 if (index_only || reset || !update)
440 return;
441 if (!lstat(path, &st))
442 die("Untracked working tree file '%s' "
443 "would be %s by merge.", path, action);
446 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
448 merge->ce_flags |= htons(CE_UPDATE);
449 if (old) {
451 * See if we can re-use the old CE directly?
452 * That way we get the uptodate stat info.
454 * This also removes the UPDATE flag on
455 * a match.
457 if (same(old, merge)) {
458 *merge = *old;
459 } else {
460 verify_uptodate(old);
463 else
464 verify_absent(merge->name, "overwritten");
466 merge->ce_flags &= ~htons(CE_STAGEMASK);
467 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
468 return 1;
471 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
473 if (old)
474 verify_uptodate(old);
475 else
476 verify_absent(ce->name, "removed");
477 ce->ce_mode = 0;
478 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
479 return 1;
482 static int keep_entry(struct cache_entry *ce)
484 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
485 return 1;
488 #if DBRT_DEBUG
489 static void show_stage_entry(FILE *o,
490 const char *label, const struct cache_entry *ce)
492 if (!ce)
493 fprintf(o, "%s (missing)\n", label);
494 else
495 fprintf(o, "%s%06o %s %d\t%s\n",
496 label,
497 ntohl(ce->ce_mode),
498 sha1_to_hex(ce->sha1),
499 ce_stage(ce),
500 ce->name);
502 #endif
504 static int threeway_merge(struct cache_entry **stages)
506 struct cache_entry *index;
507 struct cache_entry *head;
508 struct cache_entry *remote = stages[head_idx + 1];
509 int count;
510 int head_match = 0;
511 int remote_match = 0;
512 const char *path = NULL;
514 int df_conflict_head = 0;
515 int df_conflict_remote = 0;
517 int any_anc_missing = 0;
518 int no_anc_exists = 1;
519 int i;
521 for (i = 1; i < head_idx; i++) {
522 if (!stages[i])
523 any_anc_missing = 1;
524 else {
525 if (!path)
526 path = stages[i]->name;
527 no_anc_exists = 0;
531 index = stages[0];
532 head = stages[head_idx];
534 if (head == &df_conflict_entry) {
535 df_conflict_head = 1;
536 head = NULL;
539 if (remote == &df_conflict_entry) {
540 df_conflict_remote = 1;
541 remote = NULL;
544 if (!path && index)
545 path = index->name;
546 if (!path && head)
547 path = head->name;
548 if (!path && remote)
549 path = remote->name;
551 /* First, if there's a #16 situation, note that to prevent #13
552 * and #14.
554 if (!same(remote, head)) {
555 for (i = 1; i < head_idx; i++) {
556 if (same(stages[i], head)) {
557 head_match = i;
559 if (same(stages[i], remote)) {
560 remote_match = i;
565 /* We start with cases where the index is allowed to match
566 * something other than the head: #14(ALT) and #2ALT, where it
567 * is permitted to match the result instead.
569 /* #14, #14ALT, #2ALT */
570 if (remote && !df_conflict_head && head_match && !remote_match) {
571 if (index && !same(index, remote) && !same(index, head))
572 reject_merge(index);
573 return merged_entry(remote, index);
576 * If we have an entry in the index cache, then we want to
577 * make sure that it matches head.
579 if (index && !same(index, head)) {
580 reject_merge(index);
583 if (head) {
584 /* #5ALT, #15 */
585 if (same(head, remote))
586 return merged_entry(head, index);
587 /* #13, #3ALT */
588 if (!df_conflict_remote && remote_match && !head_match)
589 return merged_entry(head, index);
592 /* #1 */
593 if (!head && !remote && any_anc_missing)
594 return 0;
596 /* Under the new "aggressive" rule, we resolve mostly trivial
597 * cases that we historically had git-merge-one-file resolve.
599 if (aggressive) {
600 int head_deleted = !head && !df_conflict_head;
601 int remote_deleted = !remote && !df_conflict_remote;
603 * Deleted in both.
604 * Deleted in one and unchanged in the other.
606 if ((head_deleted && remote_deleted) ||
607 (head_deleted && remote && remote_match) ||
608 (remote_deleted && head && head_match)) {
609 if (index)
610 return deleted_entry(index, index);
611 else if (path)
612 verify_absent(path, "removed");
613 return 0;
616 * Added in both, identically.
618 if (no_anc_exists && head && remote && same(head, remote))
619 return merged_entry(head, index);
623 /* Below are "no merge" cases, which require that the index be
624 * up-to-date to avoid the files getting overwritten with
625 * conflict resolution files.
627 if (index) {
628 verify_uptodate(index);
630 else if (path)
631 verify_absent(path, "overwritten");
633 nontrivial_merge = 1;
635 /* #2, #3, #4, #6, #7, #9, #11. */
636 count = 0;
637 if (!head_match || !remote_match) {
638 for (i = 1; i < head_idx; i++) {
639 if (stages[i]) {
640 keep_entry(stages[i]);
641 count++;
642 break;
646 #if DBRT_DEBUG
647 else {
648 fprintf(stderr, "read-tree: warning #16 detected\n");
649 show_stage_entry(stderr, "head ", stages[head_match]);
650 show_stage_entry(stderr, "remote ", stages[remote_match]);
652 #endif
653 if (head) { count += keep_entry(head); }
654 if (remote) { count += keep_entry(remote); }
655 return count;
659 * Two-way merge.
661 * The rule is to "carry forward" what is in the index without losing
662 * information across a "fast forward", favoring a successful merge
663 * over a merge failure when it makes sense. For details of the
664 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
667 static int twoway_merge(struct cache_entry **src)
669 struct cache_entry *current = src[0];
670 struct cache_entry *oldtree = src[1], *newtree = src[2];
672 if (merge_size != 2)
673 return error("Cannot do a twoway merge of %d trees",
674 merge_size);
676 if (current) {
677 if ((!oldtree && !newtree) || /* 4 and 5 */
678 (!oldtree && newtree &&
679 same(current, newtree)) || /* 6 and 7 */
680 (oldtree && newtree &&
681 same(oldtree, newtree)) || /* 14 and 15 */
682 (oldtree && newtree &&
683 !same(oldtree, newtree) && /* 18 and 19*/
684 same(current, newtree))) {
685 return keep_entry(current);
687 else if (oldtree && !newtree && same(current, oldtree)) {
688 /* 10 or 11 */
689 return deleted_entry(oldtree, current);
691 else if (oldtree && newtree &&
692 same(current, oldtree) && !same(current, newtree)) {
693 /* 20 or 21 */
694 return merged_entry(newtree, current);
696 else {
697 /* all other failures */
698 if (oldtree)
699 reject_merge(oldtree);
700 if (current)
701 reject_merge(current);
702 if (newtree)
703 reject_merge(newtree);
704 return -1;
707 else if (newtree)
708 return merged_entry(newtree, current);
709 else
710 return deleted_entry(oldtree, current);
714 * One-way merge.
716 * The rule is:
717 * - take the stat information from stage0, take the data from stage1
719 static int oneway_merge(struct cache_entry **src)
721 struct cache_entry *old = src[0];
722 struct cache_entry *a = src[1];
724 if (merge_size != 1)
725 return error("Cannot do a oneway merge of %d trees",
726 merge_size);
728 if (!a)
729 return deleted_entry(old, old);
730 if (old && same(old, a)) {
731 if (reset) {
732 struct stat st;
733 if (lstat(old->name, &st) ||
734 ce_match_stat(old, &st, 1))
735 old->ce_flags |= htons(CE_UPDATE);
737 return keep_entry(old);
739 return merged_entry(a, old);
742 static int read_cache_unmerged(void)
744 int i, deleted;
745 struct cache_entry **dst;
747 read_cache();
748 dst = active_cache;
749 deleted = 0;
750 for (i = 0; i < active_nr; i++) {
751 struct cache_entry *ce = active_cache[i];
752 if (ce_stage(ce)) {
753 deleted++;
754 continue;
756 if (deleted)
757 *dst = ce;
758 dst++;
760 active_nr -= deleted;
761 return deleted;
764 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
766 static struct cache_file cache_file;
768 int cmd_read_tree(int argc, const char **argv, char **envp)
770 int i, newfd, stage = 0;
771 unsigned char sha1[20];
772 merge_fn_t fn = NULL;
774 setup_git_directory();
775 git_config(git_default_config);
777 newfd = hold_index_file_for_update(&cache_file, get_index_file());
778 if (newfd < 0)
779 die("unable to create new cachefile");
781 git_config(git_default_config);
783 merge = 0;
784 reset = 0;
785 for (i = 1; i < argc; i++) {
786 const char *arg = argv[i];
788 /* "-u" means "update", meaning that a merge will update
789 * the working tree.
791 if (!strcmp(arg, "-u")) {
792 update = 1;
793 continue;
796 if (!strcmp(arg, "-v")) {
797 verbose_update = 1;
798 continue;
801 /* "-i" means "index only", meaning that a merge will
802 * not even look at the working tree.
804 if (!strcmp(arg, "-i")) {
805 index_only = 1;
806 continue;
809 /* This differs from "-m" in that we'll silently ignore unmerged entries */
810 if (!strcmp(arg, "--reset")) {
811 if (stage || merge)
812 usage(read_tree_usage);
813 reset = 1;
814 merge = 1;
815 stage = 1;
816 read_cache_unmerged();
817 continue;
820 if (!strcmp(arg, "--trivial")) {
821 trivial_merges_only = 1;
822 continue;
825 if (!strcmp(arg, "--aggressive")) {
826 aggressive = 1;
827 continue;
830 /* "-m" stands for "merge", meaning we start in stage 1 */
831 if (!strcmp(arg, "-m")) {
832 if (stage || merge)
833 usage(read_tree_usage);
834 if (read_cache_unmerged())
835 die("you need to resolve your current index first");
836 stage = 1;
837 merge = 1;
838 continue;
841 /* using -u and -i at the same time makes no sense */
842 if (1 < index_only + update)
843 usage(read_tree_usage);
845 if (get_sha1(arg, sha1))
846 die("Not a valid object name %s", arg);
847 if (list_tree(sha1) < 0)
848 die("failed to unpack tree object %s", arg);
849 stage++;
851 if ((update||index_only) && !merge)
852 usage(read_tree_usage);
854 if (merge) {
855 if (stage < 2)
856 die("just how do you expect me to merge %d trees?", stage-1);
857 switch (stage - 1) {
858 case 1:
859 fn = oneway_merge;
860 break;
861 case 2:
862 fn = twoway_merge;
863 break;
864 case 3:
865 fn = threeway_merge;
866 break;
867 default:
868 fn = threeway_merge;
869 break;
872 if (stage - 1 >= 3)
873 head_idx = stage - 2;
874 else
875 head_idx = 1;
878 unpack_trees(fn);
879 if (write_cache(newfd, active_cache, active_nr) ||
880 commit_index_file(&cache_file))
881 die("unable to write new index file");
882 return 0;