Merge branch 'lt/fix-sol-pack' into next
[git/git.git] / read-tree.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define DBRT_DEBUG 1
7
8 #include "cache.h"
9
10 #include "object.h"
11 #include "tree.h"
12 #include <sys/time.h>
13 #include <signal.h>
14
15 static int merge = 0;
16 static int update = 0;
17 static int index_only = 0;
18 static int nontrivial_merge = 0;
19 static int trivial_merges_only = 0;
20 static int aggressive = 0;
21 static int verbose_update = 0;
22 static volatile int progress_update = 0;
23
24 static int head_idx = -1;
25 static int merge_size = 0;
26
27 static struct object_list *trees = NULL;
28
29 static struct cache_entry df_conflict_entry = {
30 };
31
32 static struct tree_entry_list df_conflict_list = {
33 .name = NULL,
34 .next = &df_conflict_list
35 };
36
37 typedef int (*merge_fn_t)(struct cache_entry **src);
38
39 static int entcmp(char *name1, int dir1, char *name2, int dir2)
40 {
41 int len1 = strlen(name1);
42 int len2 = strlen(name2);
43 int len = len1 < len2 ? len1 : len2;
44 int ret = memcmp(name1, name2, len);
45 unsigned char c1, c2;
46 if (ret)
47 return ret;
48 c1 = name1[len];
49 c2 = name2[len];
50 if (!c1 && dir1)
51 c1 = '/';
52 if (!c2 && dir2)
53 c2 = '/';
54 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
55 if (c1 && c2 && !ret)
56 ret = len1 - len2;
57 return ret;
58 }
59
60 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
61 const char *base, merge_fn_t fn, int *indpos)
62 {
63 int baselen = strlen(base);
64 int src_size = len + 1;
65 do {
66 int i;
67 char *first;
68 int firstdir = 0;
69 int pathlen;
70 unsigned ce_size;
71 struct tree_entry_list **subposns;
72 struct cache_entry **src;
73 int any_files = 0;
74 int any_dirs = 0;
75 char *cache_name;
76 int ce_stage;
77
78 /* Find the first name in the input. */
79
80 first = NULL;
81 cache_name = NULL;
82
83 /* Check the cache */
84 if (merge && *indpos < active_nr) {
85 /* This is a bit tricky: */
86 /* If the index has a subdirectory (with
87 * contents) as the first name, it'll get a
88 * filename like "foo/bar". But that's after
89 * "foo", so the entry in trees will get
90 * handled first, at which point we'll go into
91 * "foo", and deal with "bar" from the index,
92 * because the base will be "foo/". The only
93 * way we can actually have "foo/bar" first of
94 * all the things is if the trees don't
95 * contain "foo" at all, in which case we'll
96 * handle "foo/bar" without going into the
97 * directory, but that's fine (and will return
98 * an error anyway, with the added unknown
99 * file case.
100 */
101
102 cache_name = active_cache[*indpos]->name;
103 if (strlen(cache_name) > baselen &&
104 !memcmp(cache_name, base, baselen)) {
105 cache_name += baselen;
106 first = cache_name;
107 } else {
108 cache_name = NULL;
109 }
110 }
111
112 #if DBRT_DEBUG > 1
113 if (first)
114 printf("index %s\n", first);
115 #endif
116 for (i = 0; i < len; i++) {
117 if (!posns[i] || posns[i] == &df_conflict_list)
118 continue;
119 #if DBRT_DEBUG > 1
120 printf("%d %s\n", i + 1, posns[i]->name);
121 #endif
122 if (!first || entcmp(first, firstdir,
123 posns[i]->name,
124 posns[i]->directory) > 0) {
125 first = posns[i]->name;
126 firstdir = posns[i]->directory;
127 }
128 }
129 /* No name means we're done */
130 if (!first)
131 return 0;
132
133 pathlen = strlen(first);
134 ce_size = cache_entry_size(baselen + pathlen);
135
136 src = xmalloc(sizeof(struct cache_entry *) * src_size);
137 memset(src, 0, sizeof(struct cache_entry *) * src_size);
138
139 subposns = xmalloc(sizeof(struct tree_list_entry *) * len);
140 memset(subposns, 0, sizeof(struct tree_list_entry *) * len);
141
142 if (cache_name && !strcmp(cache_name, first)) {
143 any_files = 1;
144 src[0] = active_cache[*indpos];
145 remove_cache_entry_at(*indpos);
146 }
147
148 for (i = 0; i < len; i++) {
149 struct cache_entry *ce;
150
151 if (!posns[i] ||
152 (posns[i] != &df_conflict_list &&
153 strcmp(first, posns[i]->name))) {
154 continue;
155 }
156
157 if (posns[i] == &df_conflict_list) {
158 src[i + merge] = &df_conflict_entry;
159 continue;
160 }
161
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;
169 }
170
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;
179
180 ce = xmalloc(ce_size);
181 memset(ce, 0, 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);
187
188 any_files = 1;
189
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;
194 }
195 if (any_files) {
196 if (merge) {
197 int ret;
198
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");
207 }
208 #endif
209 ret = fn(src);
210
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);
219 }
220 }
221 }
222 }
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);
233 }
234 free(subposns);
235 free(src);
236 } while (1);
237 }
238
239 static void reject_merge(struct cache_entry *ce)
240 {
241 die("Entry '%s' would be overwritten by merge. Cannot merge.",
242 ce->name);
243 }
244
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.
248 */
249 static void unlink_entry(char *name)
250 {
251 char *cp, *prev;
252
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;
263
264 *cp = 0;
265 status = rmdir(name);
266 if (status) {
267 *cp = '/';
268 break;
269 }
270 prev = cp;
271 }
272 }
273
274 static void progress_interval(int signum)
275 {
276 progress_update = 1;
277 }
278
279 static void setup_progress_signal(void)
280 {
281 struct sigaction sa;
282 struct itimerval v;
283
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);
289
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);
294 }
295
296 static void check_updates(struct cache_entry **src, int nr)
297 {
298 static struct checkout state = {
299 .base_dir = "",
300 .force = 1,
301 .quiet = 1,
302 .refresh_cache = 1,
303 };
304 unsigned short mask = htons(CE_UPDATE);
305 unsigned last_percent = 200, cnt = 0, total = 0;
306
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++;
312 }
313
314 /* Don't bother doing this for very small updates */
315 if (total < 250)
316 total = 0;
317
318 if (total) {
319 fprintf(stderr, "Checking files out...\n");
320 setup_progress_signal();
321 progress_update = 1;
322 }
323 cnt = 0;
324 }
325
326 while (nr--) {
327 struct cache_entry *ce = *src++;
328
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;
339 }
340 }
341 }
342 if (!ce->ce_mode) {
343 if (update)
344 unlink_entry(ce->name);
345 continue;
346 }
347 if (ce->ce_flags & mask) {
348 ce->ce_flags &= ~mask;
349 if (update)
350 checkout_entry(ce, &state, NULL);
351 }
352 }
353 if (total) {
354 signal(SIGALRM, SIG_IGN);
355 fputc('\n', stderr);
356 }
357 }
358
359 static int unpack_trees(merge_fn_t fn)
360 {
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;
367
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;
373 }
374 if (unpack_trees_rec(posns, len, "", fn, &indpos))
375 return -1;
376 }
377
378 if (trivial_merges_only && nontrivial_merge)
379 die("Merge requires file-level merging");
380
381 check_updates(active_cache, active_nr);
382 return 0;
383 }
384
385 static int list_tree(unsigned char *sha1)
386 {
387 struct tree *tree = parse_tree_indirect(sha1);
388 if (!tree)
389 return -1;
390 object_list_append(&tree->object, &trees);
391 return 0;
392 }
393
394 static int same(struct cache_entry *a, struct cache_entry *b)
395 {
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);
402 }
403
404
405 /*
406 * When a CE gets turned into an unmerged entry, we
407 * want it to be up-to-date
408 */
409 static void verify_uptodate(struct cache_entry *ce)
410 {
411 struct stat st;
412
413 if (index_only)
414 return;
415
416 if (!lstat(ce->name, &st)) {
417 unsigned changed = ce_match_stat(ce, &st, 1);
418 if (!changed)
419 return;
420 errno = 0;
421 }
422 if (errno == ENOENT)
423 return;
424 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
425 }
426
427 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
428 {
429 merge->ce_flags |= htons(CE_UPDATE);
430 if (old) {
431 /*
432 * See if we can re-use the old CE directly?
433 * That way we get the uptodate stat info.
434 *
435 * This also removes the UPDATE flag on
436 * a match.
437 */
438 if (same(old, merge)) {
439 *merge = *old;
440 } else {
441 verify_uptodate(old);
442 }
443 }
444 merge->ce_flags &= ~htons(CE_STAGEMASK);
445 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
446 return 1;
447 }
448
449 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
450 {
451 if (old)
452 verify_uptodate(old);
453 ce->ce_mode = 0;
454 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
455 return 1;
456 }
457
458 static int keep_entry(struct cache_entry *ce)
459 {
460 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
461 return 1;
462 }
463
464 #if DBRT_DEBUG
465 static void show_stage_entry(FILE *o,
466 const char *label, const struct cache_entry *ce)
467 {
468 if (!ce)
469 fprintf(o, "%s (missing)\n", label);
470 else
471 fprintf(o, "%s%06o %s %d\t%s\n",
472 label,
473 ntohl(ce->ce_mode),
474 sha1_to_hex(ce->sha1),
475 ce_stage(ce),
476 ce->name);
477 }
478 #endif
479
480 static int threeway_merge(struct cache_entry **stages)
481 {
482 struct cache_entry *index;
483 struct cache_entry *head;
484 struct cache_entry *remote = stages[head_idx + 1];
485 int count;
486 int head_match = 0;
487 int remote_match = 0;
488
489 int df_conflict_head = 0;
490 int df_conflict_remote = 0;
491
492 int any_anc_missing = 0;
493 int no_anc_exists = 1;
494 int i;
495
496 for (i = 1; i < head_idx; i++) {
497 if (!stages[i])
498 any_anc_missing = 1;
499 else
500 no_anc_exists = 0;
501 }
502
503 index = stages[0];
504 head = stages[head_idx];
505
506 if (head == &df_conflict_entry) {
507 df_conflict_head = 1;
508 head = NULL;
509 }
510
511 if (remote == &df_conflict_entry) {
512 df_conflict_remote = 1;
513 remote = NULL;
514 }
515
516 /* First, if there's a #16 situation, note that to prevent #13
517 * and #14.
518 */
519 if (!same(remote, head)) {
520 for (i = 1; i < head_idx; i++) {
521 if (same(stages[i], head)) {
522 head_match = i;
523 }
524 if (same(stages[i], remote)) {
525 remote_match = i;
526 }
527 }
528 }
529
530 /* We start with cases where the index is allowed to match
531 * something other than the head: #14(ALT) and #2ALT, where it
532 * is permitted to match the result instead.
533 */
534 /* #14, #14ALT, #2ALT */
535 if (remote && !df_conflict_head && head_match && !remote_match) {
536 if (index && !same(index, remote) && !same(index, head))
537 reject_merge(index);
538 return merged_entry(remote, index);
539 }
540 /*
541 * If we have an entry in the index cache, then we want to
542 * make sure that it matches head.
543 */
544 if (index && !same(index, head)) {
545 reject_merge(index);
546 }
547
548 if (head) {
549 /* #5ALT, #15 */
550 if (same(head, remote))
551 return merged_entry(head, index);
552 /* #13, #3ALT */
553 if (!df_conflict_remote && remote_match && !head_match)
554 return merged_entry(head, index);
555 }
556
557 /* #1 */
558 if (!head && !remote && any_anc_missing)
559 return 0;
560
561 /* Under the new "aggressive" rule, we resolve mostly trivial
562 * cases that we historically had git-merge-one-file resolve.
563 */
564 if (aggressive) {
565 int head_deleted = !head && !df_conflict_head;
566 int remote_deleted = !remote && !df_conflict_remote;
567 /*
568 * Deleted in both.
569 * Deleted in one and unchanged in the other.
570 */
571 if ((head_deleted && remote_deleted) ||
572 (head_deleted && remote && remote_match) ||
573 (remote_deleted && head && head_match)) {
574 if (index)
575 return deleted_entry(index, index);
576 return 0;
577 }
578 /*
579 * Added in both, identically.
580 */
581 if (no_anc_exists && head && remote && same(head, remote))
582 return merged_entry(head, index);
583
584 }
585
586 /* Below are "no merge" cases, which require that the index be
587 * up-to-date to avoid the files getting overwritten with
588 * conflict resolution files.
589 */
590 if (index) {
591 verify_uptodate(index);
592 }
593
594 nontrivial_merge = 1;
595
596 /* #2, #3, #4, #6, #7, #9, #11. */
597 count = 0;
598 if (!head_match || !remote_match) {
599 for (i = 1; i < head_idx; i++) {
600 if (stages[i]) {
601 keep_entry(stages[i]);
602 count++;
603 break;
604 }
605 }
606 }
607 #if DBRT_DEBUG
608 else {
609 fprintf(stderr, "read-tree: warning #16 detected\n");
610 show_stage_entry(stderr, "head ", stages[head_match]);
611 show_stage_entry(stderr, "remote ", stages[remote_match]);
612 }
613 #endif
614 if (head) { count += keep_entry(head); }
615 if (remote) { count += keep_entry(remote); }
616 return count;
617 }
618
619 /*
620 * Two-way merge.
621 *
622 * The rule is to "carry forward" what is in the index without losing
623 * information across a "fast forward", favoring a successful merge
624 * over a merge failure when it makes sense. For details of the
625 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
626 *
627 */
628 static int twoway_merge(struct cache_entry **src)
629 {
630 struct cache_entry *current = src[0];
631 struct cache_entry *oldtree = src[1], *newtree = src[2];
632
633 if (merge_size != 2)
634 return error("Cannot do a twoway merge of %d trees",
635 merge_size);
636
637 if (current) {
638 if ((!oldtree && !newtree) || /* 4 and 5 */
639 (!oldtree && newtree &&
640 same(current, newtree)) || /* 6 and 7 */
641 (oldtree && newtree &&
642 same(oldtree, newtree)) || /* 14 and 15 */
643 (oldtree && newtree &&
644 !same(oldtree, newtree) && /* 18 and 19*/
645 same(current, newtree))) {
646 return keep_entry(current);
647 }
648 else if (oldtree && !newtree && same(current, oldtree)) {
649 /* 10 or 11 */
650 return deleted_entry(oldtree, current);
651 }
652 else if (oldtree && newtree &&
653 same(current, oldtree) && !same(current, newtree)) {
654 /* 20 or 21 */
655 return merged_entry(newtree, current);
656 }
657 else {
658 /* all other failures */
659 if (oldtree)
660 reject_merge(oldtree);
661 if (current)
662 reject_merge(current);
663 if (newtree)
664 reject_merge(newtree);
665 return -1;
666 }
667 }
668 else if (newtree)
669 return merged_entry(newtree, current);
670 else
671 return deleted_entry(oldtree, current);
672 }
673
674 /*
675 * One-way merge.
676 *
677 * The rule is:
678 * - take the stat information from stage0, take the data from stage1
679 */
680 static int oneway_merge(struct cache_entry **src)
681 {
682 struct cache_entry *old = src[0];
683 struct cache_entry *a = src[1];
684
685 if (merge_size != 1)
686 return error("Cannot do a oneway merge of %d trees",
687 merge_size);
688
689 if (!a)
690 return 0;
691 if (old && same(old, a)) {
692 return keep_entry(old);
693 }
694 return merged_entry(a, NULL);
695 }
696
697 static int read_cache_unmerged(void)
698 {
699 int i, deleted;
700 struct cache_entry **dst;
701
702 read_cache();
703 dst = active_cache;
704 deleted = 0;
705 for (i = 0; i < active_nr; i++) {
706 struct cache_entry *ce = active_cache[i];
707 if (ce_stage(ce)) {
708 deleted++;
709 continue;
710 }
711 if (deleted)
712 *dst = ce;
713 dst++;
714 }
715 active_nr -= deleted;
716 return deleted;
717 }
718
719 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
720
721 static struct cache_file cache_file;
722
723 int main(int argc, char **argv)
724 {
725 int i, newfd, reset, stage = 0;
726 unsigned char sha1[20];
727 merge_fn_t fn = NULL;
728
729 setup_git_directory();
730 git_config(git_default_config);
731
732 newfd = hold_index_file_for_update(&cache_file, get_index_file());
733 if (newfd < 0)
734 die("unable to create new cachefile");
735
736 git_config(git_default_config);
737
738 merge = 0;
739 reset = 0;
740 for (i = 1; i < argc; i++) {
741 const char *arg = argv[i];
742
743 /* "-u" means "update", meaning that a merge will update
744 * the working tree.
745 */
746 if (!strcmp(arg, "-u")) {
747 update = 1;
748 continue;
749 }
750
751 if (!strcmp(arg, "-v")) {
752 verbose_update = 1;
753 continue;
754 }
755
756 /* "-i" means "index only", meaning that a merge will
757 * not even look at the working tree.
758 */
759 if (!strcmp(arg, "-i")) {
760 index_only = 1;
761 continue;
762 }
763
764 /* This differs from "-m" in that we'll silently ignore unmerged entries */
765 if (!strcmp(arg, "--reset")) {
766 if (stage || merge)
767 usage(read_tree_usage);
768 reset = 1;
769 merge = 1;
770 stage = 1;
771 read_cache_unmerged();
772 continue;
773 }
774
775 if (!strcmp(arg, "--trivial")) {
776 trivial_merges_only = 1;
777 continue;
778 }
779
780 if (!strcmp(arg, "--aggressive")) {
781 aggressive = 1;
782 continue;
783 }
784
785 /* "-m" stands for "merge", meaning we start in stage 1 */
786 if (!strcmp(arg, "-m")) {
787 if (stage || merge)
788 usage(read_tree_usage);
789 if (read_cache_unmerged())
790 die("you need to resolve your current index first");
791 stage = 1;
792 merge = 1;
793 continue;
794 }
795
796 /* using -u and -i at the same time makes no sense */
797 if (1 < index_only + update)
798 usage(read_tree_usage);
799
800 if (get_sha1(arg, sha1) < 0)
801 usage(read_tree_usage);
802 if (list_tree(sha1) < 0)
803 die("failed to unpack tree object %s", arg);
804 stage++;
805 }
806 if ((update||index_only) && !merge)
807 usage(read_tree_usage);
808
809 if (merge) {
810 if (stage < 2)
811 die("just how do you expect me to merge %d trees?", stage-1);
812 switch (stage - 1) {
813 case 1:
814 fn = oneway_merge;
815 break;
816 case 2:
817 fn = twoway_merge;
818 break;
819 case 3:
820 fn = threeway_merge;
821 break;
822 default:
823 fn = threeway_merge;
824 break;
825 }
826
827 if (stage - 1 >= 3)
828 head_idx = stage - 2;
829 else
830 head_idx = 1;
831 }
832
833 unpack_trees(fn);
834 if (write_cache(newfd, active_cache, active_nr) ||
835 commit_index_file(&cache_file))
836 die("unable to write new index file");
837 return 0;
838 }