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#include "builtin.h"
15
16static int reset = 0;
17static int merge = 0;
18static int update = 0;
19static int index_only = 0;
20static int nontrivial_merge = 0;
21static int trivial_merges_only = 0;
22static int aggressive = 0;
23static int verbose_update = 0;
24static volatile int progress_update = 0;
25
26static int head_idx = -1;
27static int merge_size = 0;
28
29static struct object_list *trees = NULL;
30
31static struct cache_entry df_conflict_entry = {
32};
33
34static struct tree_entry_list df_conflict_list = {
35 .name = NULL,
36 .next = &df_conflict_list
37};
38
39typedef int (*merge_fn_t)(struct cache_entry **src);
40
41static int entcmp(char *name1, int dir1, char *name2, int dir2)
42{
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;
60}
61
62static int unpack_trees_rec(struct tree_entry_list **posns, int len,
63 const char *base, merge_fn_t fn, int *indpos)
64{
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;
79
80 /* Find the first name in the input. */
81
82 first = NULL;
83 cache_name = NULL;
84
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.
102 */
103
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;
111 }
112 }
113
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;
129 }
130 }
131 /* No name means we're done */
132 if (!first)
133 return 0;
134
135 pathlen = strlen(first);
136 ce_size = cache_entry_size(baselen + pathlen);
137
138 src = xcalloc(src_size, sizeof(struct cache_entry *));
139
140 subposns = xcalloc(len, sizeof(struct tree_list_entry *));
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 = 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);
186
187 any_files = 1;
188
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;
193 }
194 if (any_files) {
195 if (merge) {
196 int ret;
197
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");
206 }
207#endif
208 ret = fn(src);
209
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);
218 }
219 }
220 }
221 }
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);
232 }
233 free(subposns);
234 free(src);
235 } while (1);
236}
237
238static void reject_merge(struct cache_entry *ce)
239{
240 die("Entry '%s' would be overwritten by merge. Cannot merge.",
241 ce->name);
242}
243
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.
247 */
248static void unlink_entry(char *name)
249{
250 char *cp, *prev;
251
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;
262
263 *cp = 0;
264 status = rmdir(name);
265 if (status) {
266 *cp = '/';
267 break;
268 }
269 prev = cp;
270 }
271}
272
273static void progress_interval(int signum)
274{
275 progress_update = 1;
276}
277
278static void setup_progress_signal(void)
279{
280 struct sigaction sa;
281 struct itimerval v;
282
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);
288
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);
293}
294
295static void check_updates(struct cache_entry **src, int nr)
296{
297 static struct checkout state = {
298 .base_dir = "",
299 .force = 1,
300 .quiet = 1,
301 .refresh_cache = 1,
302 };
303 unsigned short mask = htons(CE_UPDATE);
304 unsigned last_percent = 200, cnt = 0, total = 0;
305
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++;
311 }
312
313 /* Don't bother doing this for very small updates */
314 if (total < 250)
315 total = 0;
316
317 if (total) {
318 fprintf(stderr, "Checking files out...\n");
319 setup_progress_signal();
320 progress_update = 1;
321 }
322 cnt = 0;
323 }
324
325 while (nr--) {
326 struct cache_entry *ce = *src++;
327
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;
338 }
339 }
340 }
341 if (!ce->ce_mode) {
342 if (update)
343 unlink_entry(ce->name);
344 continue;
345 }
346 if (ce->ce_flags & mask) {
347 ce->ce_flags &= ~mask;
348 if (update)
349 checkout_entry(ce, &state, NULL);
350 }
351 }
352 if (total) {
353 signal(SIGALRM, SIG_IGN);
354 fputc('\n', stderr);
355 }
356}
357
358static int unpack_trees(merge_fn_t fn)
359{
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;
366
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;
372 }
373 if (unpack_trees_rec(posns, len, "", fn, &indpos))
374 return -1;
375 }
376
377 if (trivial_merges_only && nontrivial_merge)
378 die("Merge requires file-level merging");
379
380 check_updates(active_cache, active_nr);
381 return 0;
382}
383
384static int list_tree(unsigned char *sha1)
385{
386 struct tree *tree = parse_tree_indirect(sha1);
387 if (!tree)
388 return -1;
389 object_list_append(&tree->object, &trees);
390 return 0;
391}
392
393static int same(struct cache_entry *a, struct cache_entry *b)
394{
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);
401}
402
403
404/*
405 * When a CE gets turned into an unmerged entry, we
406 * want it to be up-to-date
407 */
408static void verify_uptodate(struct cache_entry *ce)
409{
410 struct stat st;
411
412 if (index_only || reset)
413 return;
414
415 if (!lstat(ce->name, &st)) {
416 unsigned changed = ce_match_stat(ce, &st, 1);
417 if (!changed)
418 return;
419 errno = 0;
420 }
421 if (reset) {
422 ce->ce_flags |= htons(CE_UPDATE);
423 return;
424 }
425 if (errno == ENOENT)
426 return;
427 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
428}
429
430/*
431 * We do not want to remove or overwrite a working tree file that
432 * is not tracked.
433 */
434static void verify_absent(const char *path, const char *action)
435{
436 struct stat st;
437
438 if (index_only || reset || !update)
439 return;
440 if (!lstat(path, &st))
441 die("Untracked working tree file '%s' "
442 "would be %s by merge.", path, action);
443}
444
445static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
446{
447 merge->ce_flags |= htons(CE_UPDATE);
448 if (old) {
449 /*
450 * See if we can re-use the old CE directly?
451 * That way we get the uptodate stat info.
452 *
453 * This also removes the UPDATE flag on
454 * a match.
455 */
456 if (same(old, merge)) {
457 *merge = *old;
458 } else {
459 verify_uptodate(old);
460 }
461 }
462 else
463 verify_absent(merge->name, "overwritten");
464
465 merge->ce_flags &= ~htons(CE_STAGEMASK);
466 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
467 return 1;
468}
469
470static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
471{
472 if (old)
473 verify_uptodate(old);
474 else
475 verify_absent(ce->name, "removed");
476 ce->ce_mode = 0;
477 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
478 return 1;
479}
480
481static int keep_entry(struct cache_entry *ce)
482{
483 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
484 return 1;
485}
486
487#if DBRT_DEBUG
488static void show_stage_entry(FILE *o,
489 const char *label, const struct cache_entry *ce)
490{
491 if (!ce)
492 fprintf(o, "%s (missing)\n", label);
493 else
494 fprintf(o, "%s%06o %s %d\t%s\n",
495 label,
496 ntohl(ce->ce_mode),
497 sha1_to_hex(ce->sha1),
498 ce_stage(ce),
499 ce->name);
500}
501#endif
502
503static int threeway_merge(struct cache_entry **stages)
504{
505 struct cache_entry *index;
506 struct cache_entry *head;
507 struct cache_entry *remote = stages[head_idx + 1];
508 int count;
509 int head_match = 0;
510 int remote_match = 0;
511 const char *path = NULL;
512
513 int df_conflict_head = 0;
514 int df_conflict_remote = 0;
515
516 int any_anc_missing = 0;
517 int no_anc_exists = 1;
518 int i;
519
520 for (i = 1; i < head_idx; i++) {
521 if (!stages[i])
522 any_anc_missing = 1;
523 else {
524 if (!path)
525 path = stages[i]->name;
526 no_anc_exists = 0;
527 }
528 }
529
530 index = stages[0];
531 head = stages[head_idx];
532
533 if (head == &df_conflict_entry) {
534 df_conflict_head = 1;
535 head = NULL;
536 }
537
538 if (remote == &df_conflict_entry) {
539 df_conflict_remote = 1;
540 remote = NULL;
541 }
542
543 if (!path && index)
544 path = index->name;
545 if (!path && head)
546 path = head->name;
547 if (!path && remote)
548 path = remote->name;
549
550 /* First, if there's a #16 situation, note that to prevent #13
551 * and #14.
552 */
553 if (!same(remote, head)) {
554 for (i = 1; i < head_idx; i++) {
555 if (same(stages[i], head)) {
556 head_match = i;
557 }
558 if (same(stages[i], remote)) {
559 remote_match = i;
560 }
561 }
562 }
563
564 /* We start with cases where the index is allowed to match
565 * something other than the head: #14(ALT) and #2ALT, where it
566 * is permitted to match the result instead.
567 */
568 /* #14, #14ALT, #2ALT */
569 if (remote && !df_conflict_head && head_match && !remote_match) {
570 if (index && !same(index, remote) && !same(index, head))
571 reject_merge(index);
572 return merged_entry(remote, index);
573 }
574 /*
575 * If we have an entry in the index cache, then we want to
576 * make sure that it matches head.
577 */
578 if (index && !same(index, head)) {
579 reject_merge(index);
580 }
581
582 if (head) {
583 /* #5ALT, #15 */
584 if (same(head, remote))
585 return merged_entry(head, index);
586 /* #13, #3ALT */
587 if (!df_conflict_remote && remote_match && !head_match)
588 return merged_entry(head, index);
589 }
590
591 /* #1 */
592 if (!head && !remote && any_anc_missing)
593 return 0;
594
595 /* Under the new "aggressive" rule, we resolve mostly trivial
596 * cases that we historically had git-merge-one-file resolve.
597 */
598 if (aggressive) {
599 int head_deleted = !head && !df_conflict_head;
600 int remote_deleted = !remote && !df_conflict_remote;
601 /*
602 * Deleted in both.
603 * Deleted in one and unchanged in the other.
604 */
605 if ((head_deleted && remote_deleted) ||
606 (head_deleted && remote && remote_match) ||
607 (remote_deleted && head && head_match)) {
608 if (index)
609 return deleted_entry(index, index);
610 else if (path)
611 verify_absent(path, "removed");
612 return 0;
613 }
614 /*
615 * Added in both, identically.
616 */
617 if (no_anc_exists && head && remote && same(head, remote))
618 return merged_entry(head, index);
619
620 }
621
622 /* Below are "no merge" cases, which require that the index be
623 * up-to-date to avoid the files getting overwritten with
624 * conflict resolution files.
625 */
626 if (index) {
627 verify_uptodate(index);
628 }
629 else if (path)
630 verify_absent(path, "overwritten");
631
632 nontrivial_merge = 1;
633
634 /* #2, #3, #4, #6, #7, #9, #11. */
635 count = 0;
636 if (!head_match || !remote_match) {
637 for (i = 1; i < head_idx; i++) {
638 if (stages[i]) {
639 keep_entry(stages[i]);
640 count++;
641 break;
642 }
643 }
644 }
645#if DBRT_DEBUG
646 else {
647 fprintf(stderr, "read-tree: warning #16 detected\n");
648 show_stage_entry(stderr, "head ", stages[head_match]);
649 show_stage_entry(stderr, "remote ", stages[remote_match]);
650 }
651#endif
652 if (head) { count += keep_entry(head); }
653 if (remote) { count += keep_entry(remote); }
654 return count;
655}
656
657/*
658 * Two-way merge.
659 *
660 * The rule is to "carry forward" what is in the index without losing
661 * information across a "fast forward", favoring a successful merge
662 * over a merge failure when it makes sense. For details of the
663 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
664 *
665 */
666static int twoway_merge(struct cache_entry **src)
667{
668 struct cache_entry *current = src[0];
669 struct cache_entry *oldtree = src[1], *newtree = src[2];
670
671 if (merge_size != 2)
672 return error("Cannot do a twoway merge of %d trees",
673 merge_size);
674
675 if (current) {
676 if ((!oldtree && !newtree) || /* 4 and 5 */
677 (!oldtree && newtree &&
678 same(current, newtree)) || /* 6 and 7 */
679 (oldtree && newtree &&
680 same(oldtree, newtree)) || /* 14 and 15 */
681 (oldtree && newtree &&
682 !same(oldtree, newtree) && /* 18 and 19*/
683 same(current, newtree))) {
684 return keep_entry(current);
685 }
686 else if (oldtree && !newtree && same(current, oldtree)) {
687 /* 10 or 11 */
688 return deleted_entry(oldtree, current);
689 }
690 else if (oldtree && newtree &&
691 same(current, oldtree) && !same(current, newtree)) {
692 /* 20 or 21 */
693 return merged_entry(newtree, current);
694 }
695 else {
696 /* all other failures */
697 if (oldtree)
698 reject_merge(oldtree);
699 if (current)
700 reject_merge(current);
701 if (newtree)
702 reject_merge(newtree);
703 return -1;
704 }
705 }
706 else if (newtree)
707 return merged_entry(newtree, current);
708 else
709 return deleted_entry(oldtree, current);
710}
711
712/*
713 * One-way merge.
714 *
715 * The rule is:
716 * - take the stat information from stage0, take the data from stage1
717 */
718static int oneway_merge(struct cache_entry **src)
719{
720 struct cache_entry *old = src[0];
721 struct cache_entry *a = src[1];
722
723 if (merge_size != 1)
724 return error("Cannot do a oneway merge of %d trees",
725 merge_size);
726
727 if (!a)
728 return deleted_entry(old, old);
729 if (old && same(old, a)) {
730 if (reset) {
731 struct stat st;
732 if (lstat(old->name, &st) ||
733 ce_match_stat(old, &st, 1))
734 old->ce_flags |= htons(CE_UPDATE);
735 }
736 return keep_entry(old);
737 }
738 return merged_entry(a, old);
739}
740
741static int read_cache_unmerged(void)
742{
743 int i, deleted;
744 struct cache_entry **dst;
745
746 read_cache();
747 dst = active_cache;
748 deleted = 0;
749 for (i = 0; i < active_nr; i++) {
750 struct cache_entry *ce = active_cache[i];
751 if (ce_stage(ce)) {
752 deleted++;
753 continue;
754 }
755 if (deleted)
756 *dst = ce;
757 dst++;
758 }
759 active_nr -= deleted;
760 return deleted;
761}
762
763static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
764
765static struct cache_file cache_file;
766
767int cmd_read_tree(int argc, const char **argv, char **envp)
768{
769 int i, newfd, stage = 0;
770 unsigned char sha1[20];
771 merge_fn_t fn = NULL;
772
773 setup_git_directory();
774 git_config(git_default_config);
775
776 newfd = hold_index_file_for_update(&cache_file, get_index_file());
777 if (newfd < 0)
778 die("unable to create new cachefile");
779
780 git_config(git_default_config);
781
782 merge = 0;
783 reset = 0;
784 for (i = 1; i < argc; i++) {
785 const char *arg = argv[i];
786
787 /* "-u" means "update", meaning that a merge will update
788 * the working tree.
789 */
790 if (!strcmp(arg, "-u")) {
791 update = 1;
792 continue;
793 }
794
795 if (!strcmp(arg, "-v")) {
796 verbose_update = 1;
797 continue;
798 }
799
800 /* "-i" means "index only", meaning that a merge will
801 * not even look at the working tree.
802 */
803 if (!strcmp(arg, "-i")) {
804 index_only = 1;
805 continue;
806 }
807
808 /* This differs from "-m" in that we'll silently ignore unmerged entries */
809 if (!strcmp(arg, "--reset")) {
810 if (stage || merge)
811 usage(read_tree_usage);
812 reset = 1;
813 merge = 1;
814 stage = 1;
815 read_cache_unmerged();
816 continue;
817 }
818
819 if (!strcmp(arg, "--trivial")) {
820 trivial_merges_only = 1;
821 continue;
822 }
823
824 if (!strcmp(arg, "--aggressive")) {
825 aggressive = 1;
826 continue;
827 }
828
829 /* "-m" stands for "merge", meaning we start in stage 1 */
830 if (!strcmp(arg, "-m")) {
831 if (stage || merge)
832 usage(read_tree_usage);
833 if (read_cache_unmerged())
834 die("you need to resolve your current index first");
835 stage = 1;
836 merge = 1;
837 continue;
838 }
839
840 /* using -u and -i at the same time makes no sense */
841 if (1 < index_only + update)
842 usage(read_tree_usage);
843
844 if (get_sha1(arg, sha1))
845 die("Not a valid object name %s", arg);
846 if (list_tree(sha1) < 0)
847 die("failed to unpack tree object %s", arg);
848 stage++;
849 }
850 if ((update||index_only) && !merge)
851 usage(read_tree_usage);
852
853 if (merge) {
854 if (stage < 2)
855 die("just how do you expect me to merge %d trees?", stage-1);
856 switch (stage - 1) {
857 case 1:
858 fn = oneway_merge;
859 break;
860 case 2:
861 fn = twoway_merge;
862 break;
863 case 3:
864 fn = threeway_merge;
865 break;
866 default:
867 fn = threeway_merge;
868 break;
869 }
870
871 if (stage - 1 >= 3)
872 head_idx = stage - 2;
873 else
874 head_idx = 1;
875 }
876
877 unpack_trees(fn);
878 if (write_cache(newfd, active_cache, active_nr) ||
879 commit_index_file(&cache_file))
880 die("unable to write new index file");
881 return 0;
882}