fed2a3c544c0090d04175042fe5999d9c86bc1a9
[git/git.git] / read-tree.c
1 /*
2 * GIT - The information manager from hell
3 *
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #include "cache.h"
7
8 #include "object.h"
9 #include "tree.h"
10
11 static int merge = 0;
12 static int update = 0;
13
14 static int head_idx = -1;
15 static int merge_size = 0;
16
17 static struct object_list *trees = NULL;
18
19 static struct cache_entry df_conflict_entry = {
20 };
21
22 static struct tree_entry_list df_conflict_list = {
23 .name = NULL,
24 .next = &df_conflict_list
25 };
26
27 typedef int (*merge_fn_t)(struct cache_entry **src);
28
29 static int entcmp(char *name1, int dir1, char *name2, int dir2)
30 {
31 int len1 = strlen(name1);
32 int len2 = strlen(name2);
33 int len = len1 < len2 ? len1 : len2;
34 int ret = memcmp(name1, name2, len);
35 unsigned char c1, c2;
36 if (ret)
37 return ret;
38 c1 = name1[len];
39 c2 = name2[len];
40 if (!c1 && dir1)
41 c1 = '/';
42 if (!c2 && dir2)
43 c2 = '/';
44 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
45 if (c1 && c2 && !ret)
46 ret = len1 - len2;
47 return ret;
48 }
49
50 #define DBRT_DEBUG 0
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
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
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
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
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 static int threeway_merge(struct cache_entry **stages)
357 {
358 struct cache_entry *index;
359 struct cache_entry *head;
360 struct cache_entry *remote = stages[head_idx + 1];
361 int count;
362 int head_match = 0;
363 int remote_match = 0;
364
365 int df_conflict_head = 0;
366 int df_conflict_remote = 0;
367
368 int any_anc_missing = 0;
369 int i;
370
371 for (i = 1; i < head_idx; i++) {
372 if (!stages[i])
373 any_anc_missing = 1;
374 }
375
376 index = stages[0];
377 head = stages[head_idx];
378
379 if (head == &df_conflict_entry) {
380 df_conflict_head = 1;
381 head = NULL;
382 }
383
384 if (remote == &df_conflict_entry) {
385 df_conflict_remote = 1;
386 remote = NULL;
387 }
388
389 /* First, if there's a #16 situation, note that to prevent #13
390 * and #14.
391 */
392 if (!same(remote, head)) {
393 for (i = 1; i < head_idx; i++) {
394 if (same(stages[i], head)) {
395 head_match = 1;
396 }
397 if (same(stages[i], remote)) {
398 remote_match = 1;
399 }
400 }
401 }
402
403 /* We start with cases where the index is allowed to match
404 * something other than the head: #14(ALT) and #2ALT, where it
405 * is permitted to match the result instead.
406 */
407 /* #14, #14ALT, #2ALT */
408 if (remote && !df_conflict_head && head_match && !remote_match) {
409 if (index && !same(index, remote) && !same(index, head))
410 reject_merge(index);
411 return merged_entry(remote, index);
412 }
413 /*
414 * If we have an entry in the index cache, then we want to
415 * make sure that it matches head.
416 */
417 if (index && !same(index, head)) {
418 reject_merge(index);
419 }
420
421 if (head) {
422 /* #5ALT, #15 */
423 if (same(head, remote))
424 return merged_entry(head, index);
425 /* #13, #3ALT */
426 if (!df_conflict_remote && remote_match && !head_match)
427 return merged_entry(head, index);
428 }
429
430 /* #1 */
431 if (!head && !remote && any_anc_missing)
432 return 0;
433
434 /* Below are "no merge" cases, which require that the index be
435 * up-to-date to avoid the files getting overwritten with
436 * conflict resolution files.
437 */
438 if (index) {
439 verify_uptodate(index);
440 }
441
442 /* #2, #3, #4, #6, #7, #9, #11. */
443 count = 0;
444 if (!head_match || !remote_match) {
445 for (i = 1; i < head_idx; i++) {
446 if (stages[i]) {
447 keep_entry(stages[i]);
448 count++;
449 break;
450 }
451 }
452 }
453 if (head) { count += keep_entry(head); }
454 if (remote) { count += keep_entry(remote); }
455 return count;
456 }
457
458 /*
459 * Two-way merge.
460 *
461 * The rule is to "carry forward" what is in the index without losing
462 * information across a "fast forward", favoring a successful merge
463 * over a merge failure when it makes sense. For details of the
464 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
465 *
466 */
467 static int twoway_merge(struct cache_entry **src)
468 {
469 struct cache_entry *current = src[0];
470 struct cache_entry *oldtree = src[1], *newtree = src[2];
471
472 if (merge_size != 2)
473 return error("Cannot do a twoway merge of %d trees\n",
474 merge_size);
475
476 if (current) {
477 if ((!oldtree && !newtree) || /* 4 and 5 */
478 (!oldtree && newtree &&
479 same(current, newtree)) || /* 6 and 7 */
480 (oldtree && newtree &&
481 same(oldtree, newtree)) || /* 14 and 15 */
482 (oldtree && newtree &&
483 !same(oldtree, newtree) && /* 18 and 19*/
484 same(current, newtree))) {
485 return keep_entry(current);
486 }
487 else if (oldtree && !newtree && same(current, oldtree)) {
488 /* 10 or 11 */
489 return deleted_entry(oldtree, current);
490 }
491 else if (oldtree && newtree &&
492 same(current, oldtree) && !same(current, newtree)) {
493 /* 20 or 21 */
494 return merged_entry(newtree, current);
495 }
496 else {
497 /* all other failures */
498 if (oldtree)
499 reject_merge(oldtree);
500 if (current)
501 reject_merge(current);
502 if (newtree)
503 reject_merge(newtree);
504 return -1;
505 }
506 }
507 else if (newtree)
508 return merged_entry(newtree, current);
509 else
510 return deleted_entry(oldtree, current);
511 }
512
513 /*
514 * One-way merge.
515 *
516 * The rule is:
517 * - take the stat information from stage0, take the data from stage1
518 */
519 static int oneway_merge(struct cache_entry **src)
520 {
521 struct cache_entry *old = src[0];
522 struct cache_entry *a = src[1];
523
524 if (merge_size != 1)
525 return error("Cannot do a oneway merge of %d trees\n",
526 merge_size);
527
528 if (!a)
529 return 0;
530 if (old && same(old, a)) {
531 return keep_entry(old);
532 }
533 return merged_entry(a, NULL);
534 }
535
536 static int read_cache_unmerged(void)
537 {
538 int i, deleted;
539 struct cache_entry **dst;
540
541 read_cache();
542 dst = active_cache;
543 deleted = 0;
544 for (i = 0; i < active_nr; i++) {
545 struct cache_entry *ce = active_cache[i];
546 if (ce_stage(ce)) {
547 deleted++;
548 continue;
549 }
550 if (deleted)
551 *dst = ce;
552 dst++;
553 }
554 active_nr -= deleted;
555 return deleted;
556 }
557
558 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [-u] <sha1> [<sha2> [<sha3>]])";
559
560 static struct cache_file cache_file;
561
562 int main(int argc, char **argv)
563 {
564 int i, newfd, reset, stage = 0;
565 unsigned char sha1[20];
566 merge_fn_t fn = NULL;
567
568 newfd = hold_index_file_for_update(&cache_file, get_index_file());
569 if (newfd < 0)
570 die("unable to create new cachefile");
571
572 merge = 0;
573 reset = 0;
574 for (i = 1; i < argc; i++) {
575 const char *arg = argv[i];
576
577 /* "-u" means "update", meaning that a merge will update the working directory */
578 if (!strcmp(arg, "-u")) {
579 update = 1;
580 continue;
581 }
582
583 /* This differs from "-m" in that we'll silently ignore unmerged entries */
584 if (!strcmp(arg, "--reset")) {
585 if (stage || merge)
586 usage(read_tree_usage);
587 reset = 1;
588 merge = 1;
589 stage = 1;
590 read_cache_unmerged();
591 continue;
592 }
593
594 if (!strcmp(arg, "--head")) {
595 head_idx = stage - 1;
596 fn = threeway_merge;
597 }
598
599 /* "-m" stands for "merge", meaning we start in stage 1 */
600 if (!strcmp(arg, "-m")) {
601 if (stage || merge)
602 usage(read_tree_usage);
603 if (read_cache_unmerged())
604 die("you need to resolve your current index first");
605 stage = 1;
606 merge = 1;
607 continue;
608 }
609
610 if (get_sha1(arg, sha1) < 0)
611 usage(read_tree_usage);
612 if (list_tree(sha1) < 0)
613 die("failed to unpack tree object %s", arg);
614 stage++;
615 }
616 if (update && !merge)
617 usage(read_tree_usage);
618 if (merge && !fn) {
619 if (stage < 2)
620 die("just how do you expect me to merge %d trees?", stage-1);
621 switch (stage - 1) {
622 case 1:
623 fn = oneway_merge;
624 break;
625 case 2:
626 fn = twoway_merge;
627 break;
628 case 3:
629 fn = threeway_merge;
630 break;
631 default:
632 fn = threeway_merge;
633 break;
634 }
635 }
636
637 if (head_idx < 0) {
638 if (stage - 1 >= 3)
639 head_idx = stage - 2;
640 else
641 head_idx = 1;
642 }
643
644 unpack_trees(fn);
645 if (write_cache(newfd, active_cache, active_nr) ||
646 commit_index_file(&cache_file))
647 die("unable to write new index file");
648 return 0;
649 }