Merge branch 'pe/cleanup'
[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 = xcalloc(src_size, sizeof(struct cache_entry *));
137
138 subposns = xcalloc(len, sizeof(struct tree_list_entry *));
139
140 if (cache_name && !strcmp(cache_name, first)) {
141 any_files = 1;
142 src[0] = active_cache[*indpos];
143 remove_cache_entry_at(*indpos);
144 }
145
146 for (i = 0; i < len; i++) {
147 struct cache_entry *ce;
148
149 if (!posns[i] ||
150 (posns[i] != &df_conflict_list &&
151 strcmp(first, posns[i]->name))) {
152 continue;
153 }
154
155 if (posns[i] == &df_conflict_list) {
156 src[i + merge] = &df_conflict_entry;
157 continue;
158 }
159
160 if (posns[i]->directory) {
161 any_dirs = 1;
162 parse_tree(posns[i]->item.tree);
163 subposns[i] = posns[i]->item.tree->entries;
164 posns[i] = posns[i]->next;
165 src[i + merge] = &df_conflict_entry;
166 continue;
167 }
168
169 if (!merge)
170 ce_stage = 0;
171 else if (i + 1 < head_idx)
172 ce_stage = 1;
173 else if (i + 1 > head_idx)
174 ce_stage = 3;
175 else
176 ce_stage = 2;
177
178 ce = xcalloc(1, ce_size);
179 ce->ce_mode = create_ce_mode(posns[i]->mode);
180 ce->ce_flags = create_ce_flags(baselen + pathlen,
181 ce_stage);
182 memcpy(ce->name, base, baselen);
183 memcpy(ce->name + baselen, first, pathlen + 1);
184
185 any_files = 1;
186
187 memcpy(ce->sha1, posns[i]->item.any->sha1, 20);
188 src[i + merge] = ce;
189 subposns[i] = &df_conflict_list;
190 posns[i] = posns[i]->next;
191 }
192 if (any_files) {
193 if (merge) {
194 int ret;
195
196 #if DBRT_DEBUG > 1
197 printf("%s:\n", first);
198 for (i = 0; i < src_size; i++) {
199 printf(" %d ", i);
200 if (src[i])
201 printf("%s\n", sha1_to_hex(src[i]->sha1));
202 else
203 printf("\n");
204 }
205 #endif
206 ret = fn(src);
207
208 #if DBRT_DEBUG > 1
209 printf("Added %d entries\n", ret);
210 #endif
211 *indpos += ret;
212 } else {
213 for (i = 0; i < src_size; i++) {
214 if (src[i]) {
215 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
216 }
217 }
218 }
219 }
220 if (any_dirs) {
221 char *newbase = xmalloc(baselen + 2 + pathlen);
222 memcpy(newbase, base, baselen);
223 memcpy(newbase + baselen, first, pathlen);
224 newbase[baselen + pathlen] = '/';
225 newbase[baselen + pathlen + 1] = '\0';
226 if (unpack_trees_rec(subposns, len, newbase, fn,
227 indpos))
228 return -1;
229 free(newbase);
230 }
231 free(subposns);
232 free(src);
233 } while (1);
234 }
235
236 static void reject_merge(struct cache_entry *ce)
237 {
238 die("Entry '%s' would be overwritten by merge. Cannot merge.",
239 ce->name);
240 }
241
242 /* Unlink the last component and attempt to remove leading
243 * directories, in case this unlink is the removal of the
244 * last entry in the directory -- empty directories are removed.
245 */
246 static void unlink_entry(char *name)
247 {
248 char *cp, *prev;
249
250 if (unlink(name))
251 return;
252 prev = NULL;
253 while (1) {
254 int status;
255 cp = strrchr(name, '/');
256 if (prev)
257 *prev = '/';
258 if (!cp)
259 break;
260
261 *cp = 0;
262 status = rmdir(name);
263 if (status) {
264 *cp = '/';
265 break;
266 }
267 prev = cp;
268 }
269 }
270
271 static void progress_interval(int signum)
272 {
273 progress_update = 1;
274 }
275
276 static void setup_progress_signal(void)
277 {
278 struct sigaction sa;
279 struct itimerval v;
280
281 memset(&sa, 0, sizeof(sa));
282 sa.sa_handler = progress_interval;
283 sigemptyset(&sa.sa_mask);
284 sa.sa_flags = SA_RESTART;
285 sigaction(SIGALRM, &sa, NULL);
286
287 v.it_interval.tv_sec = 1;
288 v.it_interval.tv_usec = 0;
289 v.it_value = v.it_interval;
290 setitimer(ITIMER_REAL, &v, NULL);
291 }
292
293 static void check_updates(struct cache_entry **src, int nr)
294 {
295 static struct checkout state = {
296 .base_dir = "",
297 .force = 1,
298 .quiet = 1,
299 .refresh_cache = 1,
300 };
301 unsigned short mask = htons(CE_UPDATE);
302 unsigned last_percent = 200, cnt = 0, total = 0;
303
304 if (update && verbose_update) {
305 for (total = cnt = 0; cnt < nr; cnt++) {
306 struct cache_entry *ce = src[cnt];
307 if (!ce->ce_mode || ce->ce_flags & mask)
308 total++;
309 }
310
311 /* Don't bother doing this for very small updates */
312 if (total < 250)
313 total = 0;
314
315 if (total) {
316 fprintf(stderr, "Checking files out...\n");
317 setup_progress_signal();
318 progress_update = 1;
319 }
320 cnt = 0;
321 }
322
323 while (nr--) {
324 struct cache_entry *ce = *src++;
325
326 if (total) {
327 if (!ce->ce_mode || ce->ce_flags & mask) {
328 unsigned percent;
329 cnt++;
330 percent = (cnt * 100) / total;
331 if (percent != last_percent ||
332 progress_update) {
333 fprintf(stderr, "%4u%% (%u/%u) done\r",
334 percent, cnt, total);
335 last_percent = percent;
336 }
337 }
338 }
339 if (!ce->ce_mode) {
340 if (update)
341 unlink_entry(ce->name);
342 continue;
343 }
344 if (ce->ce_flags & mask) {
345 ce->ce_flags &= ~mask;
346 if (update)
347 checkout_entry(ce, &state, NULL);
348 }
349 }
350 if (total) {
351 signal(SIGALRM, SIG_IGN);
352 fputc('\n', stderr);
353 }
354 }
355
356 static int unpack_trees(merge_fn_t fn)
357 {
358 int indpos = 0;
359 unsigned len = object_list_length(trees);
360 struct tree_entry_list **posns;
361 int i;
362 struct object_list *posn = trees;
363 merge_size = len;
364
365 if (len) {
366 posns = xmalloc(len * sizeof(struct tree_entry_list *));
367 for (i = 0; i < len; i++) {
368 posns[i] = ((struct tree *) posn->item)->entries;
369 posn = posn->next;
370 }
371 if (unpack_trees_rec(posns, len, "", fn, &indpos))
372 return -1;
373 }
374
375 if (trivial_merges_only && nontrivial_merge)
376 die("Merge requires file-level merging");
377
378 check_updates(active_cache, active_nr);
379 return 0;
380 }
381
382 static int list_tree(unsigned char *sha1)
383 {
384 struct tree *tree = parse_tree_indirect(sha1);
385 if (!tree)
386 return -1;
387 object_list_append(&tree->object, &trees);
388 return 0;
389 }
390
391 static int same(struct cache_entry *a, struct cache_entry *b)
392 {
393 if (!!a != !!b)
394 return 0;
395 if (!a && !b)
396 return 1;
397 return a->ce_mode == b->ce_mode &&
398 !memcmp(a->sha1, b->sha1, 20);
399 }
400
401
402 /*
403 * When a CE gets turned into an unmerged entry, we
404 * want it to be up-to-date
405 */
406 static void verify_uptodate(struct cache_entry *ce)
407 {
408 struct stat st;
409
410 if (index_only)
411 return;
412
413 if (!lstat(ce->name, &st)) {
414 unsigned changed = ce_match_stat(ce, &st, 1);
415 if (!changed)
416 return;
417 errno = 0;
418 }
419 if (errno == ENOENT)
420 return;
421 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
422 }
423
424 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
425 {
426 merge->ce_flags |= htons(CE_UPDATE);
427 if (old) {
428 /*
429 * See if we can re-use the old CE directly?
430 * That way we get the uptodate stat info.
431 *
432 * This also removes the UPDATE flag on
433 * a match.
434 */
435 if (same(old, merge)) {
436 *merge = *old;
437 } else {
438 verify_uptodate(old);
439 }
440 }
441 merge->ce_flags &= ~htons(CE_STAGEMASK);
442 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
443 return 1;
444 }
445
446 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
447 {
448 if (old)
449 verify_uptodate(old);
450 ce->ce_mode = 0;
451 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
452 return 1;
453 }
454
455 static int keep_entry(struct cache_entry *ce)
456 {
457 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
458 return 1;
459 }
460
461 #if DBRT_DEBUG
462 static void show_stage_entry(FILE *o,
463 const char *label, const struct cache_entry *ce)
464 {
465 if (!ce)
466 fprintf(o, "%s (missing)\n", label);
467 else
468 fprintf(o, "%s%06o %s %d\t%s\n",
469 label,
470 ntohl(ce->ce_mode),
471 sha1_to_hex(ce->sha1),
472 ce_stage(ce),
473 ce->name);
474 }
475 #endif
476
477 static int threeway_merge(struct cache_entry **stages)
478 {
479 struct cache_entry *index;
480 struct cache_entry *head;
481 struct cache_entry *remote = stages[head_idx + 1];
482 int count;
483 int head_match = 0;
484 int remote_match = 0;
485
486 int df_conflict_head = 0;
487 int df_conflict_remote = 0;
488
489 int any_anc_missing = 0;
490 int no_anc_exists = 1;
491 int i;
492
493 for (i = 1; i < head_idx; i++) {
494 if (!stages[i])
495 any_anc_missing = 1;
496 else
497 no_anc_exists = 0;
498 }
499
500 index = stages[0];
501 head = stages[head_idx];
502
503 if (head == &df_conflict_entry) {
504 df_conflict_head = 1;
505 head = NULL;
506 }
507
508 if (remote == &df_conflict_entry) {
509 df_conflict_remote = 1;
510 remote = NULL;
511 }
512
513 /* First, if there's a #16 situation, note that to prevent #13
514 * and #14.
515 */
516 if (!same(remote, head)) {
517 for (i = 1; i < head_idx; i++) {
518 if (same(stages[i], head)) {
519 head_match = i;
520 }
521 if (same(stages[i], remote)) {
522 remote_match = i;
523 }
524 }
525 }
526
527 /* We start with cases where the index is allowed to match
528 * something other than the head: #14(ALT) and #2ALT, where it
529 * is permitted to match the result instead.
530 */
531 /* #14, #14ALT, #2ALT */
532 if (remote && !df_conflict_head && head_match && !remote_match) {
533 if (index && !same(index, remote) && !same(index, head))
534 reject_merge(index);
535 return merged_entry(remote, index);
536 }
537 /*
538 * If we have an entry in the index cache, then we want to
539 * make sure that it matches head.
540 */
541 if (index && !same(index, head)) {
542 reject_merge(index);
543 }
544
545 if (head) {
546 /* #5ALT, #15 */
547 if (same(head, remote))
548 return merged_entry(head, index);
549 /* #13, #3ALT */
550 if (!df_conflict_remote && remote_match && !head_match)
551 return merged_entry(head, index);
552 }
553
554 /* #1 */
555 if (!head && !remote && any_anc_missing)
556 return 0;
557
558 /* Under the new "aggressive" rule, we resolve mostly trivial
559 * cases that we historically had git-merge-one-file resolve.
560 */
561 if (aggressive) {
562 int head_deleted = !head && !df_conflict_head;
563 int remote_deleted = !remote && !df_conflict_remote;
564 /*
565 * Deleted in both.
566 * Deleted in one and unchanged in the other.
567 */
568 if ((head_deleted && remote_deleted) ||
569 (head_deleted && remote && remote_match) ||
570 (remote_deleted && head && head_match)) {
571 if (index)
572 return deleted_entry(index, index);
573 return 0;
574 }
575 /*
576 * Added in both, identically.
577 */
578 if (no_anc_exists && head && remote && same(head, remote))
579 return merged_entry(head, index);
580
581 }
582
583 /* Below are "no merge" cases, which require that the index be
584 * up-to-date to avoid the files getting overwritten with
585 * conflict resolution files.
586 */
587 if (index) {
588 verify_uptodate(index);
589 }
590
591 nontrivial_merge = 1;
592
593 /* #2, #3, #4, #6, #7, #9, #11. */
594 count = 0;
595 if (!head_match || !remote_match) {
596 for (i = 1; i < head_idx; i++) {
597 if (stages[i]) {
598 keep_entry(stages[i]);
599 count++;
600 break;
601 }
602 }
603 }
604 #if DBRT_DEBUG
605 else {
606 fprintf(stderr, "read-tree: warning #16 detected\n");
607 show_stage_entry(stderr, "head ", stages[head_match]);
608 show_stage_entry(stderr, "remote ", stages[remote_match]);
609 }
610 #endif
611 if (head) { count += keep_entry(head); }
612 if (remote) { count += keep_entry(remote); }
613 return count;
614 }
615
616 /*
617 * Two-way merge.
618 *
619 * The rule is to "carry forward" what is in the index without losing
620 * information across a "fast forward", favoring a successful merge
621 * over a merge failure when it makes sense. For details of the
622 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
623 *
624 */
625 static int twoway_merge(struct cache_entry **src)
626 {
627 struct cache_entry *current = src[0];
628 struct cache_entry *oldtree = src[1], *newtree = src[2];
629
630 if (merge_size != 2)
631 return error("Cannot do a twoway merge of %d trees",
632 merge_size);
633
634 if (current) {
635 if ((!oldtree && !newtree) || /* 4 and 5 */
636 (!oldtree && newtree &&
637 same(current, newtree)) || /* 6 and 7 */
638 (oldtree && newtree &&
639 same(oldtree, newtree)) || /* 14 and 15 */
640 (oldtree && newtree &&
641 !same(oldtree, newtree) && /* 18 and 19*/
642 same(current, newtree))) {
643 return keep_entry(current);
644 }
645 else if (oldtree && !newtree && same(current, oldtree)) {
646 /* 10 or 11 */
647 return deleted_entry(oldtree, current);
648 }
649 else if (oldtree && newtree &&
650 same(current, oldtree) && !same(current, newtree)) {
651 /* 20 or 21 */
652 return merged_entry(newtree, current);
653 }
654 else {
655 /* all other failures */
656 if (oldtree)
657 reject_merge(oldtree);
658 if (current)
659 reject_merge(current);
660 if (newtree)
661 reject_merge(newtree);
662 return -1;
663 }
664 }
665 else if (newtree)
666 return merged_entry(newtree, current);
667 else
668 return deleted_entry(oldtree, current);
669 }
670
671 /*
672 * One-way merge.
673 *
674 * The rule is:
675 * - take the stat information from stage0, take the data from stage1
676 */
677 static int oneway_merge(struct cache_entry **src)
678 {
679 struct cache_entry *old = src[0];
680 struct cache_entry *a = src[1];
681
682 if (merge_size != 1)
683 return error("Cannot do a oneway merge of %d trees",
684 merge_size);
685
686 if (!a)
687 return 0;
688 if (old && same(old, a)) {
689 return keep_entry(old);
690 }
691 return merged_entry(a, NULL);
692 }
693
694 static int read_cache_unmerged(void)
695 {
696 int i, deleted;
697 struct cache_entry **dst;
698
699 read_cache();
700 dst = active_cache;
701 deleted = 0;
702 for (i = 0; i < active_nr; i++) {
703 struct cache_entry *ce = active_cache[i];
704 if (ce_stage(ce)) {
705 deleted++;
706 continue;
707 }
708 if (deleted)
709 *dst = ce;
710 dst++;
711 }
712 active_nr -= deleted;
713 return deleted;
714 }
715
716 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
717
718 static struct cache_file cache_file;
719
720 int main(int argc, char **argv)
721 {
722 int i, newfd, reset, stage = 0;
723 unsigned char sha1[20];
724 merge_fn_t fn = NULL;
725
726 setup_git_directory();
727 git_config(git_default_config);
728
729 newfd = hold_index_file_for_update(&cache_file, get_index_file());
730 if (newfd < 0)
731 die("unable to create new cachefile");
732
733 git_config(git_default_config);
734
735 merge = 0;
736 reset = 0;
737 for (i = 1; i < argc; i++) {
738 const char *arg = argv[i];
739
740 /* "-u" means "update", meaning that a merge will update
741 * the working tree.
742 */
743 if (!strcmp(arg, "-u")) {
744 update = 1;
745 continue;
746 }
747
748 if (!strcmp(arg, "-v")) {
749 verbose_update = 1;
750 continue;
751 }
752
753 /* "-i" means "index only", meaning that a merge will
754 * not even look at the working tree.
755 */
756 if (!strcmp(arg, "-i")) {
757 index_only = 1;
758 continue;
759 }
760
761 /* This differs from "-m" in that we'll silently ignore unmerged entries */
762 if (!strcmp(arg, "--reset")) {
763 if (stage || merge)
764 usage(read_tree_usage);
765 reset = 1;
766 merge = 1;
767 stage = 1;
768 read_cache_unmerged();
769 continue;
770 }
771
772 if (!strcmp(arg, "--trivial")) {
773 trivial_merges_only = 1;
774 continue;
775 }
776
777 if (!strcmp(arg, "--aggressive")) {
778 aggressive = 1;
779 continue;
780 }
781
782 /* "-m" stands for "merge", meaning we start in stage 1 */
783 if (!strcmp(arg, "-m")) {
784 if (stage || merge)
785 usage(read_tree_usage);
786 if (read_cache_unmerged())
787 die("you need to resolve your current index first");
788 stage = 1;
789 merge = 1;
790 continue;
791 }
792
793 /* using -u and -i at the same time makes no sense */
794 if (1 < index_only + update)
795 usage(read_tree_usage);
796
797 if (get_sha1(arg, sha1) < 0)
798 usage(read_tree_usage);
799 if (list_tree(sha1) < 0)
800 die("failed to unpack tree object %s", arg);
801 stage++;
802 }
803 if ((update||index_only) && !merge)
804 usage(read_tree_usage);
805
806 if (merge) {
807 if (stage < 2)
808 die("just how do you expect me to merge %d trees?", stage-1);
809 switch (stage - 1) {
810 case 1:
811 fn = oneway_merge;
812 break;
813 case 2:
814 fn = twoway_merge;
815 break;
816 case 3:
817 fn = threeway_merge;
818 break;
819 default:
820 fn = threeway_merge;
821 break;
822 }
823
824 if (stage - 1 >= 3)
825 head_idx = stage - 2;
826 else
827 head_idx = 1;
828 }
829
830 unpack_trees(fn);
831 if (write_cache(newfd, active_cache, active_nr) ||
832 commit_index_file(&cache_file))
833 die("unable to write new index file");
834 return 0;
835 }