Teach git-status about spaces in file names also on MacOSX
[git/jrn.git] / read-tree.c
blob5fdf58d240e68f2c4b332dbdd02be357641127f8
1 /*
2 * GIT - The information manager from hell
4 * Copyright (C) Linus Torvalds, 2005
5 */
6 #define DBRT_DEBUG 1
8 #include "cache.h"
10 #include "object.h"
11 #include "tree.h"
13 static int merge = 0;
14 static int update = 0;
15 static int index_only = 0;
16 static int nontrivial_merge = 0;
17 static int trivial_merges_only = 0;
19 static int head_idx = -1;
20 static int merge_size = 0;
22 static struct object_list *trees = NULL;
24 static struct cache_entry df_conflict_entry = {
27 static struct tree_entry_list df_conflict_list = {
28 .name = NULL,
29 .next = &df_conflict_list
32 typedef int (*merge_fn_t)(struct cache_entry **src);
34 static int entcmp(char *name1, int dir1, char *name2, int dir2)
36 int len1 = strlen(name1);
37 int len2 = strlen(name2);
38 int len = len1 < len2 ? len1 : len2;
39 int ret = memcmp(name1, name2, len);
40 unsigned char c1, c2;
41 if (ret)
42 return ret;
43 c1 = name1[len];
44 c2 = name2[len];
45 if (!c1 && dir1)
46 c1 = '/';
47 if (!c2 && dir2)
48 c2 = '/';
49 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
50 if (c1 && c2 && !ret)
51 ret = len1 - len2;
52 return ret;
55 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
56 const char *base, merge_fn_t fn, int *indpos)
58 int baselen = strlen(base);
59 int src_size = len + 1;
60 do {
61 int i;
62 char *first;
63 int firstdir = 0;
64 int pathlen;
65 unsigned ce_size;
66 struct tree_entry_list **subposns;
67 struct cache_entry **src;
68 int any_files = 0;
69 int any_dirs = 0;
70 char *cache_name;
71 int ce_stage;
73 /* Find the first name in the input. */
75 first = NULL;
76 cache_name = NULL;
78 /* Check the cache */
79 if (merge && *indpos < active_nr) {
80 /* This is a bit tricky: */
81 /* If the index has a subdirectory (with
82 * contents) as the first name, it'll get a
83 * filename like "foo/bar". But that's after
84 * "foo", so the entry in trees will get
85 * handled first, at which point we'll go into
86 * "foo", and deal with "bar" from the index,
87 * because the base will be "foo/". The only
88 * way we can actually have "foo/bar" first of
89 * all the things is if the trees don't
90 * contain "foo" at all, in which case we'll
91 * handle "foo/bar" without going into the
92 * directory, but that's fine (and will return
93 * an error anyway, with the added unknown
94 * file case.
97 cache_name = active_cache[*indpos]->name;
98 if (strlen(cache_name) > baselen &&
99 !memcmp(cache_name, base, baselen)) {
100 cache_name += baselen;
101 first = cache_name;
102 } else {
103 cache_name = NULL;
107 #if DBRT_DEBUG > 1
108 if (first)
109 printf("index %s\n", first);
110 #endif
111 for (i = 0; i < len; i++) {
112 if (!posns[i] || posns[i] == &df_conflict_list)
113 continue;
114 #if DBRT_DEBUG > 1
115 printf("%d %s\n", i + 1, posns[i]->name);
116 #endif
117 if (!first || entcmp(first, firstdir,
118 posns[i]->name,
119 posns[i]->directory) > 0) {
120 first = posns[i]->name;
121 firstdir = posns[i]->directory;
124 /* No name means we're done */
125 if (!first)
126 return 0;
128 pathlen = strlen(first);
129 ce_size = cache_entry_size(baselen + pathlen);
131 src = xmalloc(sizeof(struct cache_entry *) * src_size);
132 memset(src, 0, sizeof(struct cache_entry *) * src_size);
134 subposns = xmalloc(sizeof(struct tree_list_entry *) * len);
135 memset(subposns, 0, sizeof(struct tree_list_entry *) * len);
137 if (cache_name && !strcmp(cache_name, first)) {
138 any_files = 1;
139 src[0] = active_cache[*indpos];
140 remove_cache_entry_at(*indpos);
143 for (i = 0; i < len; i++) {
144 struct cache_entry *ce;
146 if (!posns[i] ||
147 (posns[i] != &df_conflict_list &&
148 strcmp(first, posns[i]->name))) {
149 continue;
152 if (posns[i] == &df_conflict_list) {
153 src[i + merge] = &df_conflict_entry;
154 continue;
157 if (posns[i]->directory) {
158 any_dirs = 1;
159 parse_tree(posns[i]->item.tree);
160 subposns[i] = posns[i]->item.tree->entries;
161 posns[i] = posns[i]->next;
162 src[i + merge] = &df_conflict_entry;
163 continue;
166 if (!merge)
167 ce_stage = 0;
168 else if (i + 1 < head_idx)
169 ce_stage = 1;
170 else if (i + 1 > head_idx)
171 ce_stage = 3;
172 else
173 ce_stage = 2;
175 ce = xmalloc(ce_size);
176 memset(ce, 0, ce_size);
177 ce->ce_mode = create_ce_mode(posns[i]->mode);
178 ce->ce_flags = create_ce_flags(baselen + pathlen,
179 ce_stage);
180 memcpy(ce->name, base, baselen);
181 memcpy(ce->name + baselen, first, pathlen + 1);
183 any_files = 1;
185 memcpy(ce->sha1, posns[i]->item.any->sha1, 20);
186 src[i + merge] = ce;
187 subposns[i] = &df_conflict_list;
188 posns[i] = posns[i]->next;
190 if (any_files) {
191 if (merge) {
192 int ret;
194 #if DBRT_DEBUG > 1
195 printf("%s:\n", first);
196 for (i = 0; i < src_size; i++) {
197 printf(" %d ", i);
198 if (src[i])
199 printf("%s\n", sha1_to_hex(src[i]->sha1));
200 else
201 printf("\n");
203 #endif
204 ret = fn(src);
206 #if DBRT_DEBUG > 1
207 printf("Added %d entries\n", ret);
208 #endif
209 *indpos += ret;
210 } else {
211 for (i = 0; i < src_size; i++) {
212 if (src[i]) {
213 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
218 if (any_dirs) {
219 char *newbase = xmalloc(baselen + 2 + pathlen);
220 memcpy(newbase, base, baselen);
221 memcpy(newbase + baselen, first, pathlen);
222 newbase[baselen + pathlen] = '/';
223 newbase[baselen + pathlen + 1] = '\0';
224 if (unpack_trees_rec(subposns, len, newbase, fn,
225 indpos))
226 return -1;
227 free(newbase);
229 free(subposns);
230 free(src);
231 } while (1);
234 static void reject_merge(struct cache_entry *ce)
236 die("Entry '%s' would be overwritten by merge. Cannot merge.",
237 ce->name);
240 static void check_updates(struct cache_entry **src, int nr)
242 static struct checkout state = {
243 .base_dir = "",
244 .force = 1,
245 .quiet = 1,
246 .refresh_cache = 1,
248 unsigned short mask = htons(CE_UPDATE);
249 while (nr--) {
250 struct cache_entry *ce = *src++;
251 if (!ce->ce_mode) {
252 if (update)
253 unlink(ce->name);
254 continue;
256 if (ce->ce_flags & mask) {
257 ce->ce_flags &= ~mask;
258 if (update)
259 checkout_entry(ce, &state);
264 static int unpack_trees(merge_fn_t fn)
266 int indpos = 0;
267 unsigned len = object_list_length(trees);
268 struct tree_entry_list **posns =
269 xmalloc(len * sizeof(struct tree_entry_list *));
270 int i;
271 struct object_list *posn = trees;
272 merge_size = len;
273 for (i = 0; i < len; i++) {
274 posns[i] = ((struct tree *) posn->item)->entries;
275 posn = posn->next;
277 if (unpack_trees_rec(posns, len, "", fn, &indpos))
278 return -1;
280 if (trivial_merges_only && nontrivial_merge)
281 die("Merge requires file-level merging");
283 check_updates(active_cache, active_nr);
284 return 0;
287 static int list_tree(unsigned char *sha1)
289 struct tree *tree = parse_tree_indirect(sha1);
290 if (!tree)
291 return -1;
292 object_list_append(&tree->object, &trees);
293 return 0;
296 static int same(struct cache_entry *a, struct cache_entry *b)
298 if (!!a != !!b)
299 return 0;
300 if (!a && !b)
301 return 1;
302 return a->ce_mode == b->ce_mode &&
303 !memcmp(a->sha1, b->sha1, 20);
308 * When a CE gets turned into an unmerged entry, we
309 * want it to be up-to-date
311 static void verify_uptodate(struct cache_entry *ce)
313 struct stat st;
315 if (index_only)
316 return;
318 if (!lstat(ce->name, &st)) {
319 unsigned changed = ce_match_stat(ce, &st);
320 if (!changed)
321 return;
322 errno = 0;
324 if (errno == ENOENT)
325 return;
326 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
329 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
331 merge->ce_flags |= htons(CE_UPDATE);
332 if (old) {
334 * See if we can re-use the old CE directly?
335 * That way we get the uptodate stat info.
337 * This also removes the UPDATE flag on
338 * a match.
340 if (same(old, merge)) {
341 *merge = *old;
342 } else {
343 verify_uptodate(old);
346 merge->ce_flags &= ~htons(CE_STAGEMASK);
347 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
348 return 1;
351 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
353 if (old)
354 verify_uptodate(old);
355 ce->ce_mode = 0;
356 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
357 return 1;
360 static int keep_entry(struct cache_entry *ce)
362 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
363 return 1;
366 #if DBRT_DEBUG
367 static void show_stage_entry(FILE *o,
368 const char *label, const struct cache_entry *ce)
370 if (!ce)
371 fprintf(o, "%s (missing)\n", label);
372 else
373 fprintf(o, "%s%06o %s %d\t%s\n",
374 label,
375 ntohl(ce->ce_mode),
376 sha1_to_hex(ce->sha1),
377 ce_stage(ce),
378 ce->name);
380 #endif
382 static int threeway_merge(struct cache_entry **stages)
384 struct cache_entry *index;
385 struct cache_entry *head;
386 struct cache_entry *remote = stages[head_idx + 1];
387 int count;
388 int head_match = 0;
389 int remote_match = 0;
391 int df_conflict_head = 0;
392 int df_conflict_remote = 0;
394 int any_anc_missing = 0;
395 int i;
397 for (i = 1; i < head_idx; i++) {
398 if (!stages[i])
399 any_anc_missing = 1;
402 index = stages[0];
403 head = stages[head_idx];
405 if (head == &df_conflict_entry) {
406 df_conflict_head = 1;
407 head = NULL;
410 if (remote == &df_conflict_entry) {
411 df_conflict_remote = 1;
412 remote = NULL;
415 /* First, if there's a #16 situation, note that to prevent #13
416 * and #14.
418 if (!same(remote, head)) {
419 for (i = 1; i < head_idx; i++) {
420 if (same(stages[i], head)) {
421 head_match = i;
423 if (same(stages[i], remote)) {
424 remote_match = i;
429 /* We start with cases where the index is allowed to match
430 * something other than the head: #14(ALT) and #2ALT, where it
431 * is permitted to match the result instead.
433 /* #14, #14ALT, #2ALT */
434 if (remote && !df_conflict_head && head_match && !remote_match) {
435 if (index && !same(index, remote) && !same(index, head))
436 reject_merge(index);
437 return merged_entry(remote, index);
440 * If we have an entry in the index cache, then we want to
441 * make sure that it matches head.
443 if (index && !same(index, head)) {
444 reject_merge(index);
447 if (head) {
448 /* #5ALT, #15 */
449 if (same(head, remote))
450 return merged_entry(head, index);
451 /* #13, #3ALT */
452 if (!df_conflict_remote && remote_match && !head_match)
453 return merged_entry(head, index);
456 /* #1 */
457 if (!head && !remote && any_anc_missing)
458 return 0;
460 /* Below are "no merge" cases, which require that the index be
461 * up-to-date to avoid the files getting overwritten with
462 * conflict resolution files.
464 if (index) {
465 verify_uptodate(index);
468 nontrivial_merge = 1;
470 /* #2, #3, #4, #6, #7, #9, #11. */
471 count = 0;
472 if (!head_match || !remote_match) {
473 for (i = 1; i < head_idx; i++) {
474 if (stages[i]) {
475 keep_entry(stages[i]);
476 count++;
477 break;
481 #if DBRT_DEBUG
482 else {
483 fprintf(stderr, "read-tree: warning #16 detected\n");
484 show_stage_entry(stderr, "head ", stages[head_match]);
485 show_stage_entry(stderr, "remote ", stages[remote_match]);
487 #endif
488 if (head) { count += keep_entry(head); }
489 if (remote) { count += keep_entry(remote); }
490 return count;
494 * Two-way merge.
496 * The rule is to "carry forward" what is in the index without losing
497 * information across a "fast forward", favoring a successful merge
498 * over a merge failure when it makes sense. For details of the
499 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
502 static int twoway_merge(struct cache_entry **src)
504 struct cache_entry *current = src[0];
505 struct cache_entry *oldtree = src[1], *newtree = src[2];
507 if (merge_size != 2)
508 return error("Cannot do a twoway merge of %d trees\n",
509 merge_size);
511 if (current) {
512 if ((!oldtree && !newtree) || /* 4 and 5 */
513 (!oldtree && newtree &&
514 same(current, newtree)) || /* 6 and 7 */
515 (oldtree && newtree &&
516 same(oldtree, newtree)) || /* 14 and 15 */
517 (oldtree && newtree &&
518 !same(oldtree, newtree) && /* 18 and 19*/
519 same(current, newtree))) {
520 return keep_entry(current);
522 else if (oldtree && !newtree && same(current, oldtree)) {
523 /* 10 or 11 */
524 return deleted_entry(oldtree, current);
526 else if (oldtree && newtree &&
527 same(current, oldtree) && !same(current, newtree)) {
528 /* 20 or 21 */
529 return merged_entry(newtree, current);
531 else {
532 /* all other failures */
533 if (oldtree)
534 reject_merge(oldtree);
535 if (current)
536 reject_merge(current);
537 if (newtree)
538 reject_merge(newtree);
539 return -1;
542 else if (newtree)
543 return merged_entry(newtree, current);
544 else
545 return deleted_entry(oldtree, current);
549 * One-way merge.
551 * The rule is:
552 * - take the stat information from stage0, take the data from stage1
554 static int oneway_merge(struct cache_entry **src)
556 struct cache_entry *old = src[0];
557 struct cache_entry *a = src[1];
559 if (merge_size != 1)
560 return error("Cannot do a oneway merge of %d trees\n",
561 merge_size);
563 if (!a)
564 return 0;
565 if (old && same(old, a)) {
566 return keep_entry(old);
568 return merged_entry(a, NULL);
571 static int read_cache_unmerged(void)
573 int i, deleted;
574 struct cache_entry **dst;
576 read_cache();
577 dst = active_cache;
578 deleted = 0;
579 for (i = 0; i < active_nr; i++) {
580 struct cache_entry *ce = active_cache[i];
581 if (ce_stage(ce)) {
582 deleted++;
583 continue;
585 if (deleted)
586 *dst = ce;
587 dst++;
589 active_nr -= deleted;
590 return deleted;
593 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [-u | -i] <sha1> [<sha2> [<sha3>]])";
595 static struct cache_file cache_file;
597 int main(int argc, char **argv)
599 int i, newfd, reset, stage = 0;
600 unsigned char sha1[20];
601 merge_fn_t fn = NULL;
603 newfd = hold_index_file_for_update(&cache_file, get_index_file());
604 if (newfd < 0)
605 die("unable to create new cachefile");
607 merge = 0;
608 reset = 0;
609 for (i = 1; i < argc; i++) {
610 const char *arg = argv[i];
612 /* "-u" means "update", meaning that a merge will update
613 * the working tree.
615 if (!strcmp(arg, "-u")) {
616 update = 1;
617 continue;
620 /* "-i" means "index only", meaning that a merge will
621 * not even look at the working tree.
623 if (!strcmp(arg, "-i")) {
624 index_only = 1;
625 continue;
628 /* This differs from "-m" in that we'll silently ignore unmerged entries */
629 if (!strcmp(arg, "--reset")) {
630 if (stage || merge)
631 usage(read_tree_usage);
632 reset = 1;
633 merge = 1;
634 stage = 1;
635 read_cache_unmerged();
636 continue;
639 if (!strcmp(arg, "--trivial")) {
640 trivial_merges_only = 1;
641 continue;
644 /* "-m" stands for "merge", meaning we start in stage 1 */
645 if (!strcmp(arg, "-m")) {
646 if (stage || merge)
647 usage(read_tree_usage);
648 if (read_cache_unmerged())
649 die("you need to resolve your current index first");
650 stage = 1;
651 merge = 1;
652 continue;
655 /* using -u and -i at the same time makes no sense */
656 if (1 < index_only + update)
657 usage(read_tree_usage);
659 if (get_sha1(arg, sha1) < 0)
660 usage(read_tree_usage);
661 if (list_tree(sha1) < 0)
662 die("failed to unpack tree object %s", arg);
663 stage++;
665 if ((update||index_only) && !merge)
666 usage(read_tree_usage);
668 if (merge) {
669 if (stage < 2)
670 die("just how do you expect me to merge %d trees?", stage-1);
671 switch (stage - 1) {
672 case 1:
673 fn = oneway_merge;
674 break;
675 case 2:
676 fn = twoway_merge;
677 break;
678 case 3:
679 fn = threeway_merge;
680 break;
681 default:
682 fn = threeway_merge;
683 break;
686 if (stage - 1 >= 3)
687 head_idx = stage - 2;
688 else
689 head_idx = 1;
692 unpack_trees(fn);
693 if (write_cache(newfd, active_cache, active_nr) ||
694 commit_index_file(&cache_file))
695 die("unable to write new index file");
696 return 0;