3808d2398db1222f5f5e5da0610d8efaae676784
[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
13 static int merge = 0;
14 static int update = 0;
15
16 static int head_idx = -1;
17 static int merge_size = 0;
18
19 static struct object_list *trees = NULL;
20
21 static struct cache_entry df_conflict_entry = {
22 };
23
24 static struct tree_entry_list df_conflict_list = {
25 .name = NULL,
26 .next = &df_conflict_list
27 };
28
29 typedef int (*merge_fn_t)(struct cache_entry **src);
30
31 static int entcmp(char *name1, int dir1, char *name2, int dir2)
32 {
33 int len1 = strlen(name1);
34 int len2 = strlen(name2);
35 int len = len1 < len2 ? len1 : len2;
36 int ret = memcmp(name1, name2, len);
37 unsigned char c1, c2;
38 if (ret)
39 return ret;
40 c1 = name1[len];
41 c2 = name2[len];
42 if (!c1 && dir1)
43 c1 = '/';
44 if (!c2 && dir2)
45 c2 = '/';
46 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
47 if (c1 && c2 && !ret)
48 ret = len1 - len2;
49 return ret;
50 }
51
52 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
53 const char *base, merge_fn_t fn, int *indpos)
54 {
55 int baselen = strlen(base);
56 int src_size = len + 1;
57 do {
58 int i;
59 char *first;
60 int firstdir = 0;
61 int pathlen;
62 unsigned ce_size;
63 struct tree_entry_list **subposns;
64 struct cache_entry **src;
65 int any_files = 0;
66 int any_dirs = 0;
67 char *cache_name;
68 int ce_stage;
69
70 /* Find the first name in the input. */
71
72 first = NULL;
73 cache_name = NULL;
74
75 /* Check the cache */
76 if (merge && *indpos < active_nr) {
77 /* This is a bit tricky: */
78 /* If the index has a subdirectory (with
79 * contents) as the first name, it'll get a
80 * filename like "foo/bar". But that's after
81 * "foo", so the entry in trees will get
82 * handled first, at which point we'll go into
83 * "foo", and deal with "bar" from the index,
84 * because the base will be "foo/". The only
85 * way we can actually have "foo/bar" first of
86 * all the things is if the trees don't
87 * contain "foo" at all, in which case we'll
88 * handle "foo/bar" without going into the
89 * directory, but that's fine (and will return
90 * an error anyway, with the added unknown
91 * file case.
92 */
93
94 cache_name = active_cache[*indpos]->name;
95 if (strlen(cache_name) > baselen &&
96 !memcmp(cache_name, base, baselen)) {
97 cache_name += baselen;
98 first = cache_name;
99 } else {
100 cache_name = NULL;
101 }
102 }
103
104 #if DBRT_DEBUG > 1
105 if (first)
106 printf("index %s\n", first);
107 #endif
108 for (i = 0; i < len; i++) {
109 if (!posns[i] || posns[i] == &df_conflict_list)
110 continue;
111 #if DBRT_DEBUG > 1
112 printf("%d %s\n", i + 1, posns[i]->name);
113 #endif
114 if (!first || entcmp(first, firstdir,
115 posns[i]->name,
116 posns[i]->directory) > 0) {
117 first = posns[i]->name;
118 firstdir = posns[i]->directory;
119 }
120 }
121 /* No name means we're done */
122 if (!first)
123 return 0;
124
125 pathlen = strlen(first);
126 ce_size = cache_entry_size(baselen + pathlen);
127
128 src = xmalloc(sizeof(struct cache_entry *) * src_size);
129 memset(src, 0, sizeof(struct cache_entry *) * src_size);
130
131 subposns = xmalloc(sizeof(struct tree_list_entry *) * len);
132 memset(subposns, 0, sizeof(struct tree_list_entry *) * len);
133
134 if (cache_name && !strcmp(cache_name, first)) {
135 any_files = 1;
136 src[0] = active_cache[*indpos];
137 remove_cache_entry_at(*indpos);
138 }
139
140 for (i = 0; i < len; i++) {
141 struct cache_entry *ce;
142
143 if (!posns[i] ||
144 (posns[i] != &df_conflict_list &&
145 strcmp(first, posns[i]->name))) {
146 continue;
147 }
148
149 if (posns[i] == &df_conflict_list) {
150 src[i + merge] = &df_conflict_entry;
151 continue;
152 }
153
154 if (posns[i]->directory) {
155 any_dirs = 1;
156 parse_tree(posns[i]->item.tree);
157 subposns[i] = posns[i]->item.tree->entries;
158 posns[i] = posns[i]->next;
159 src[i + merge] = &df_conflict_entry;
160 continue;
161 }
162
163 if (!merge)
164 ce_stage = 0;
165 else if (i + 1 < head_idx)
166 ce_stage = 1;
167 else if (i + 1 > head_idx)
168 ce_stage = 3;
169 else
170 ce_stage = 2;
171
172 ce = xmalloc(ce_size);
173 memset(ce, 0, ce_size);
174 ce->ce_mode = create_ce_mode(posns[i]->mode);
175 ce->ce_flags = create_ce_flags(baselen + pathlen,
176 ce_stage);
177 memcpy(ce->name, base, baselen);
178 memcpy(ce->name + baselen, first, pathlen + 1);
179
180 any_files = 1;
181
182 memcpy(ce->sha1, posns[i]->item.any->sha1, 20);
183 src[i + merge] = ce;
184 subposns[i] = &df_conflict_list;
185 posns[i] = posns[i]->next;
186 }
187 if (any_files) {
188 if (merge) {
189 int ret;
190
191 #if DBRT_DEBUG > 1
192 printf("%s:\n", first);
193 for (i = 0; i < src_size; i++) {
194 printf(" %d ", i);
195 if (src[i])
196 printf("%s\n", sha1_to_hex(src[i]->sha1));
197 else
198 printf("\n");
199 }
200 #endif
201 ret = fn(src);
202
203 #if DBRT_DEBUG > 1
204 printf("Added %d entries\n", ret);
205 #endif
206 *indpos += ret;
207 } else {
208 for (i = 0; i < src_size; i++) {
209 if (src[i]) {
210 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
211 }
212 }
213 }
214 }
215 if (any_dirs) {
216 char *newbase = xmalloc(baselen + 2 + pathlen);
217 memcpy(newbase, base, baselen);
218 memcpy(newbase + baselen, first, pathlen);
219 newbase[baselen + pathlen] = '/';
220 newbase[baselen + pathlen + 1] = '\0';
221 if (unpack_trees_rec(subposns, len, newbase, fn,
222 indpos))
223 return -1;
224 }
225 free(subposns);
226 free(src);
227 } while (1);
228 }
229
230 static void reject_merge(struct cache_entry *ce)
231 {
232 die("Entry '%s' would be overwritten by merge. Cannot merge.",
233 ce->name);
234 }
235
236 static void check_updates(struct cache_entry **src, int nr)
237 {
238 static struct checkout state = {
239 .base_dir = "",
240 .force = 1,
241 .quiet = 1,
242 .refresh_cache = 1,
243 };
244 unsigned short mask = htons(CE_UPDATE);
245 while (nr--) {
246 struct cache_entry *ce = *src++;
247 if (!ce->ce_mode) {
248 if (update)
249 unlink(ce->name);
250 continue;
251 }
252 if (ce->ce_flags & mask) {
253 ce->ce_flags &= ~mask;
254 if (update)
255 checkout_entry(ce, &state);
256 }
257 }
258 }
259
260 static int unpack_trees(merge_fn_t fn)
261 {
262 int indpos = 0;
263 unsigned len = object_list_length(trees);
264 struct tree_entry_list **posns =
265 xmalloc(len * sizeof(struct tree_entry_list *));
266 int i;
267 struct object_list *posn = trees;
268 merge_size = len;
269 for (i = 0; i < len; i++) {
270 posns[i] = ((struct tree *) posn->item)->entries;
271 posn = posn->next;
272 }
273 if (unpack_trees_rec(posns, len, "", fn, &indpos))
274 return -1;
275
276 check_updates(active_cache, active_nr);
277 return 0;
278 }
279
280 static int list_tree(unsigned char *sha1)
281 {
282 struct tree *tree = parse_tree_indirect(sha1);
283 if (!tree)
284 return -1;
285 object_list_append(&tree->object, &trees);
286 return 0;
287 }
288
289 static int same(struct cache_entry *a, struct cache_entry *b)
290 {
291 if (!!a != !!b)
292 return 0;
293 if (!a && !b)
294 return 1;
295 return a->ce_mode == b->ce_mode &&
296 !memcmp(a->sha1, b->sha1, 20);
297 }
298
299
300 /*
301 * When a CE gets turned into an unmerged entry, we
302 * want it to be up-to-date
303 */
304 static void verify_uptodate(struct cache_entry *ce)
305 {
306 struct stat st;
307
308 if (!lstat(ce->name, &st)) {
309 unsigned changed = ce_match_stat(ce, &st);
310 if (!changed)
311 return;
312 errno = 0;
313 }
314 if (errno == ENOENT)
315 return;
316 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
317 }
318
319 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
320 {
321 merge->ce_flags |= htons(CE_UPDATE);
322 if (old) {
323 /*
324 * See if we can re-use the old CE directly?
325 * That way we get the uptodate stat info.
326 *
327 * This also removes the UPDATE flag on
328 * a match.
329 */
330 if (same(old, merge)) {
331 *merge = *old;
332 } else {
333 verify_uptodate(old);
334 }
335 }
336 merge->ce_flags &= ~htons(CE_STAGEMASK);
337 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
338 return 1;
339 }
340
341 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
342 {
343 if (old)
344 verify_uptodate(old);
345 ce->ce_mode = 0;
346 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
347 return 1;
348 }
349
350 static int keep_entry(struct cache_entry *ce)
351 {
352 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
353 return 1;
354 }
355
356 #if DBRT_DEBUG
357 static void show_stage_entry(FILE *o,
358 const char *label, const struct cache_entry *ce)
359 {
360 fprintf(stderr, "%s%06o %s %d\t%s\n",
361 label,
362 ntohl(ce->ce_mode),
363 sha1_to_hex(ce->sha1),
364 ce_stage(ce),
365 ce->name);
366 }
367 #endif
368
369 static int threeway_merge(struct cache_entry **stages)
370 {
371 struct cache_entry *index;
372 struct cache_entry *head;
373 struct cache_entry *remote = stages[head_idx + 1];
374 int count;
375 int head_match = 0;
376 int remote_match = 0;
377
378 int df_conflict_head = 0;
379 int df_conflict_remote = 0;
380
381 int any_anc_missing = 0;
382 int i;
383
384 for (i = 1; i < head_idx; i++) {
385 if (!stages[i])
386 any_anc_missing = 1;
387 }
388
389 index = stages[0];
390 head = stages[head_idx];
391
392 if (head == &df_conflict_entry) {
393 df_conflict_head = 1;
394 head = NULL;
395 }
396
397 if (remote == &df_conflict_entry) {
398 df_conflict_remote = 1;
399 remote = NULL;
400 }
401
402 /* First, if there's a #16 situation, note that to prevent #13
403 * and #14.
404 */
405 if (!same(remote, head)) {
406 for (i = 1; i < head_idx; i++) {
407 if (same(stages[i], head)) {
408 head_match = i;
409 }
410 if (same(stages[i], remote)) {
411 remote_match = i;
412 }
413 }
414 }
415
416 /* We start with cases where the index is allowed to match
417 * something other than the head: #14(ALT) and #2ALT, where it
418 * is permitted to match the result instead.
419 */
420 /* #14, #14ALT, #2ALT */
421 if (remote && !df_conflict_head && head_match && !remote_match) {
422 if (index && !same(index, remote) && !same(index, head))
423 reject_merge(index);
424 return merged_entry(remote, index);
425 }
426 /*
427 * If we have an entry in the index cache, then we want to
428 * make sure that it matches head.
429 */
430 if (index && !same(index, head)) {
431 reject_merge(index);
432 }
433
434 if (head) {
435 /* #5ALT, #15 */
436 if (same(head, remote))
437 return merged_entry(head, index);
438 /* #13, #3ALT */
439 if (!df_conflict_remote && remote_match && !head_match)
440 return merged_entry(head, index);
441 }
442
443 /* #1 */
444 if (!head && !remote && any_anc_missing)
445 return 0;
446
447 /* Below are "no merge" cases, which require that the index be
448 * up-to-date to avoid the files getting overwritten with
449 * conflict resolution files.
450 */
451 if (index) {
452 verify_uptodate(index);
453 }
454
455 /* #2, #3, #4, #6, #7, #9, #11. */
456 count = 0;
457 if (!head_match || !remote_match) {
458 for (i = 1; i < head_idx; i++) {
459 if (stages[i]) {
460 keep_entry(stages[i]);
461 count++;
462 break;
463 }
464 }
465 }
466 #if DBRT_DEBUG
467 else {
468 fprintf(stderr, "read-tree: warning #16 detected\n");
469 show_stage_entry(stderr, "head ", stages[head_match]);
470 show_stage_entry(stderr, "remote ", stages[remote_match]);
471 }
472 #endif
473 if (head) { count += keep_entry(head); }
474 if (remote) { count += keep_entry(remote); }
475 return count;
476 }
477
478 /*
479 * Two-way merge.
480 *
481 * The rule is to "carry forward" what is in the index without losing
482 * information across a "fast forward", favoring a successful merge
483 * over a merge failure when it makes sense. For details of the
484 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
485 *
486 */
487 static int twoway_merge(struct cache_entry **src)
488 {
489 struct cache_entry *current = src[0];
490 struct cache_entry *oldtree = src[1], *newtree = src[2];
491
492 if (merge_size != 2)
493 return error("Cannot do a twoway merge of %d trees\n",
494 merge_size);
495
496 if (current) {
497 if ((!oldtree && !newtree) || /* 4 and 5 */
498 (!oldtree && newtree &&
499 same(current, newtree)) || /* 6 and 7 */
500 (oldtree && newtree &&
501 same(oldtree, newtree)) || /* 14 and 15 */
502 (oldtree && newtree &&
503 !same(oldtree, newtree) && /* 18 and 19*/
504 same(current, newtree))) {
505 return keep_entry(current);
506 }
507 else if (oldtree && !newtree && same(current, oldtree)) {
508 /* 10 or 11 */
509 return deleted_entry(oldtree, current);
510 }
511 else if (oldtree && newtree &&
512 same(current, oldtree) && !same(current, newtree)) {
513 /* 20 or 21 */
514 return merged_entry(newtree, current);
515 }
516 else {
517 /* all other failures */
518 if (oldtree)
519 reject_merge(oldtree);
520 if (current)
521 reject_merge(current);
522 if (newtree)
523 reject_merge(newtree);
524 return -1;
525 }
526 }
527 else if (newtree)
528 return merged_entry(newtree, current);
529 else
530 return deleted_entry(oldtree, current);
531 }
532
533 /*
534 * One-way merge.
535 *
536 * The rule is:
537 * - take the stat information from stage0, take the data from stage1
538 */
539 static int oneway_merge(struct cache_entry **src)
540 {
541 struct cache_entry *old = src[0];
542 struct cache_entry *a = src[1];
543
544 if (merge_size != 1)
545 return error("Cannot do a oneway merge of %d trees\n",
546 merge_size);
547
548 if (!a)
549 return 0;
550 if (old && same(old, a)) {
551 return keep_entry(old);
552 }
553 return merged_entry(a, NULL);
554 }
555
556 static int read_cache_unmerged(void)
557 {
558 int i, deleted;
559 struct cache_entry **dst;
560
561 read_cache();
562 dst = active_cache;
563 deleted = 0;
564 for (i = 0; i < active_nr; i++) {
565 struct cache_entry *ce = active_cache[i];
566 if (ce_stage(ce)) {
567 deleted++;
568 continue;
569 }
570 if (deleted)
571 *dst = ce;
572 dst++;
573 }
574 active_nr -= deleted;
575 return deleted;
576 }
577
578 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [-u] <sha1> [<sha2> [<sha3>]])";
579
580 static struct cache_file cache_file;
581
582 int main(int argc, char **argv)
583 {
584 int i, newfd, reset, stage = 0;
585 unsigned char sha1[20];
586 merge_fn_t fn = NULL;
587
588 newfd = hold_index_file_for_update(&cache_file, get_index_file());
589 if (newfd < 0)
590 die("unable to create new cachefile");
591
592 merge = 0;
593 reset = 0;
594 for (i = 1; i < argc; i++) {
595 const char *arg = argv[i];
596
597 /* "-u" means "update", meaning that a merge will update the working directory */
598 if (!strcmp(arg, "-u")) {
599 update = 1;
600 continue;
601 }
602
603 /* This differs from "-m" in that we'll silently ignore unmerged entries */
604 if (!strcmp(arg, "--reset")) {
605 if (stage || merge)
606 usage(read_tree_usage);
607 reset = 1;
608 merge = 1;
609 stage = 1;
610 read_cache_unmerged();
611 continue;
612 }
613
614 if (!strcmp(arg, "--head")) {
615 head_idx = stage - 1;
616 fn = threeway_merge;
617 }
618
619 /* "-m" stands for "merge", meaning we start in stage 1 */
620 if (!strcmp(arg, "-m")) {
621 if (stage || merge)
622 usage(read_tree_usage);
623 if (read_cache_unmerged())
624 die("you need to resolve your current index first");
625 stage = 1;
626 merge = 1;
627 continue;
628 }
629
630 if (get_sha1(arg, sha1) < 0)
631 usage(read_tree_usage);
632 if (list_tree(sha1) < 0)
633 die("failed to unpack tree object %s", arg);
634 stage++;
635 }
636 if (update && !merge)
637 usage(read_tree_usage);
638 if (merge && !fn) {
639 if (stage < 2)
640 die("just how do you expect me to merge %d trees?", stage-1);
641 switch (stage - 1) {
642 case 1:
643 fn = oneway_merge;
644 break;
645 case 2:
646 fn = twoway_merge;
647 break;
648 case 3:
649 fn = threeway_merge;
650 break;
651 default:
652 fn = threeway_merge;
653 break;
654 }
655 }
656
657 if (head_idx < 0) {
658 if (stage - 1 >= 3)
659 head_idx = stage - 2;
660 else
661 head_idx = 1;
662 }
663
664 unpack_trees(fn);
665 if (write_cache(newfd, active_cache, active_nr) ||
666 commit_index_file(&cache_file))
667 die("unable to write new index file");
668 return 0;
669 }