Make object creation in http fetch a bit safer.
[git/jrn.git] / read-tree.c
blob6a5c08c4c6903fec45cbe37a63f0bc42fbae53bc
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;
17 static int head_idx = -1;
18 static int merge_size = 0;
20 static struct object_list *trees = NULL;
22 static struct cache_entry df_conflict_entry = {
25 static struct tree_entry_list df_conflict_list = {
26 .name = NULL,
27 .next = &df_conflict_list
30 typedef int (*merge_fn_t)(struct cache_entry **src);
32 static int entcmp(char *name1, int dir1, char *name2, int dir2)
34 int len1 = strlen(name1);
35 int len2 = strlen(name2);
36 int len = len1 < len2 ? len1 : len2;
37 int ret = memcmp(name1, name2, len);
38 unsigned char c1, c2;
39 if (ret)
40 return ret;
41 c1 = name1[len];
42 c2 = name2[len];
43 if (!c1 && dir1)
44 c1 = '/';
45 if (!c2 && dir2)
46 c2 = '/';
47 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
48 if (c1 && c2 && !ret)
49 ret = len1 - len2;
50 return ret;
53 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
54 const char *base, merge_fn_t fn, int *indpos)
56 int baselen = strlen(base);
57 int src_size = len + 1;
58 do {
59 int i;
60 char *first;
61 int firstdir = 0;
62 int pathlen;
63 unsigned ce_size;
64 struct tree_entry_list **subposns;
65 struct cache_entry **src;
66 int any_files = 0;
67 int any_dirs = 0;
68 char *cache_name;
69 int ce_stage;
71 /* Find the first name in the input. */
73 first = NULL;
74 cache_name = NULL;
76 /* Check the cache */
77 if (merge && *indpos < active_nr) {
78 /* This is a bit tricky: */
79 /* If the index has a subdirectory (with
80 * contents) as the first name, it'll get a
81 * filename like "foo/bar". But that's after
82 * "foo", so the entry in trees will get
83 * handled first, at which point we'll go into
84 * "foo", and deal with "bar" from the index,
85 * because the base will be "foo/". The only
86 * way we can actually have "foo/bar" first of
87 * all the things is if the trees don't
88 * contain "foo" at all, in which case we'll
89 * handle "foo/bar" without going into the
90 * directory, but that's fine (and will return
91 * an error anyway, with the added unknown
92 * file case.
95 cache_name = active_cache[*indpos]->name;
96 if (strlen(cache_name) > baselen &&
97 !memcmp(cache_name, base, baselen)) {
98 cache_name += baselen;
99 first = cache_name;
100 } else {
101 cache_name = NULL;
105 #if DBRT_DEBUG > 1
106 if (first)
107 printf("index %s\n", first);
108 #endif
109 for (i = 0; i < len; i++) {
110 if (!posns[i] || posns[i] == &df_conflict_list)
111 continue;
112 #if DBRT_DEBUG > 1
113 printf("%d %s\n", i + 1, posns[i]->name);
114 #endif
115 if (!first || entcmp(first, firstdir,
116 posns[i]->name,
117 posns[i]->directory) > 0) {
118 first = posns[i]->name;
119 firstdir = posns[i]->directory;
122 /* No name means we're done */
123 if (!first)
124 return 0;
126 pathlen = strlen(first);
127 ce_size = cache_entry_size(baselen + pathlen);
129 src = xmalloc(sizeof(struct cache_entry *) * src_size);
130 memset(src, 0, sizeof(struct cache_entry *) * src_size);
132 subposns = xmalloc(sizeof(struct tree_list_entry *) * len);
133 memset(subposns, 0, sizeof(struct tree_list_entry *) * len);
135 if (cache_name && !strcmp(cache_name, first)) {
136 any_files = 1;
137 src[0] = active_cache[*indpos];
138 remove_cache_entry_at(*indpos);
141 for (i = 0; i < len; i++) {
142 struct cache_entry *ce;
144 if (!posns[i] ||
145 (posns[i] != &df_conflict_list &&
146 strcmp(first, posns[i]->name))) {
147 continue;
150 if (posns[i] == &df_conflict_list) {
151 src[i + merge] = &df_conflict_entry;
152 continue;
155 if (posns[i]->directory) {
156 any_dirs = 1;
157 parse_tree(posns[i]->item.tree);
158 subposns[i] = posns[i]->item.tree->entries;
159 posns[i] = posns[i]->next;
160 src[i + merge] = &df_conflict_entry;
161 continue;
164 if (!merge)
165 ce_stage = 0;
166 else if (i + 1 < head_idx)
167 ce_stage = 1;
168 else if (i + 1 > head_idx)
169 ce_stage = 3;
170 else
171 ce_stage = 2;
173 ce = xmalloc(ce_size);
174 memset(ce, 0, ce_size);
175 ce->ce_mode = create_ce_mode(posns[i]->mode);
176 ce->ce_flags = create_ce_flags(baselen + pathlen,
177 ce_stage);
178 memcpy(ce->name, base, baselen);
179 memcpy(ce->name + baselen, first, pathlen + 1);
181 any_files = 1;
183 memcpy(ce->sha1, posns[i]->item.any->sha1, 20);
184 src[i + merge] = ce;
185 subposns[i] = &df_conflict_list;
186 posns[i] = posns[i]->next;
188 if (any_files) {
189 if (merge) {
190 int ret;
192 #if DBRT_DEBUG > 1
193 printf("%s:\n", first);
194 for (i = 0; i < src_size; i++) {
195 printf(" %d ", i);
196 if (src[i])
197 printf("%s\n", sha1_to_hex(src[i]->sha1));
198 else
199 printf("\n");
201 #endif
202 ret = fn(src);
204 #if DBRT_DEBUG > 1
205 printf("Added %d entries\n", ret);
206 #endif
207 *indpos += ret;
208 } else {
209 for (i = 0; i < src_size; i++) {
210 if (src[i]) {
211 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
216 if (any_dirs) {
217 char *newbase = xmalloc(baselen + 2 + pathlen);
218 memcpy(newbase, base, baselen);
219 memcpy(newbase + baselen, first, pathlen);
220 newbase[baselen + pathlen] = '/';
221 newbase[baselen + pathlen + 1] = '\0';
222 if (unpack_trees_rec(subposns, len, newbase, fn,
223 indpos))
224 return -1;
225 free(newbase);
227 free(subposns);
228 free(src);
229 } while (1);
232 static void reject_merge(struct cache_entry *ce)
234 die("Entry '%s' would be overwritten by merge. Cannot merge.",
235 ce->name);
238 static void check_updates(struct cache_entry **src, int nr)
240 static struct checkout state = {
241 .base_dir = "",
242 .force = 1,
243 .quiet = 1,
244 .refresh_cache = 1,
246 unsigned short mask = htons(CE_UPDATE);
247 while (nr--) {
248 struct cache_entry *ce = *src++;
249 if (!ce->ce_mode) {
250 if (update)
251 unlink(ce->name);
252 continue;
254 if (ce->ce_flags & mask) {
255 ce->ce_flags &= ~mask;
256 if (update)
257 checkout_entry(ce, &state);
262 static int unpack_trees(merge_fn_t fn)
264 int indpos = 0;
265 unsigned len = object_list_length(trees);
266 struct tree_entry_list **posns =
267 xmalloc(len * sizeof(struct tree_entry_list *));
268 int i;
269 struct object_list *posn = trees;
270 merge_size = len;
271 for (i = 0; i < len; i++) {
272 posns[i] = ((struct tree *) posn->item)->entries;
273 posn = posn->next;
275 if (unpack_trees_rec(posns, len, "", fn, &indpos))
276 return -1;
278 check_updates(active_cache, active_nr);
279 return 0;
282 static int list_tree(unsigned char *sha1)
284 struct tree *tree = parse_tree_indirect(sha1);
285 if (!tree)
286 return -1;
287 object_list_append(&tree->object, &trees);
288 return 0;
291 static int same(struct cache_entry *a, struct cache_entry *b)
293 if (!!a != !!b)
294 return 0;
295 if (!a && !b)
296 return 1;
297 return a->ce_mode == b->ce_mode &&
298 !memcmp(a->sha1, b->sha1, 20);
303 * When a CE gets turned into an unmerged entry, we
304 * want it to be up-to-date
306 static void verify_uptodate(struct cache_entry *ce)
308 struct stat st;
310 if (index_only)
311 return;
313 if (!lstat(ce->name, &st)) {
314 unsigned changed = ce_match_stat(ce, &st);
315 if (!changed)
316 return;
317 errno = 0;
319 if (errno == ENOENT)
320 return;
321 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
324 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
326 merge->ce_flags |= htons(CE_UPDATE);
327 if (old) {
329 * See if we can re-use the old CE directly?
330 * That way we get the uptodate stat info.
332 * This also removes the UPDATE flag on
333 * a match.
335 if (same(old, merge)) {
336 *merge = *old;
337 } else {
338 verify_uptodate(old);
341 merge->ce_flags &= ~htons(CE_STAGEMASK);
342 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
343 return 1;
346 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
348 if (old)
349 verify_uptodate(old);
350 ce->ce_mode = 0;
351 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
352 return 1;
355 static int keep_entry(struct cache_entry *ce)
357 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
358 return 1;
361 #if DBRT_DEBUG
362 static void show_stage_entry(FILE *o,
363 const char *label, const struct cache_entry *ce)
365 if (!ce)
366 fprintf(o, "%s (missing)\n", label);
367 else
368 fprintf(o, "%s%06o %s %d\t%s\n",
369 label,
370 ntohl(ce->ce_mode),
371 sha1_to_hex(ce->sha1),
372 ce_stage(ce),
373 ce->name);
375 #endif
377 static int threeway_merge(struct cache_entry **stages)
379 struct cache_entry *index;
380 struct cache_entry *head;
381 struct cache_entry *remote = stages[head_idx + 1];
382 int count;
383 int head_match = 0;
384 int remote_match = 0;
386 int df_conflict_head = 0;
387 int df_conflict_remote = 0;
389 int any_anc_missing = 0;
390 int i;
392 for (i = 1; i < head_idx; i++) {
393 if (!stages[i])
394 any_anc_missing = 1;
397 index = stages[0];
398 head = stages[head_idx];
400 if (head == &df_conflict_entry) {
401 df_conflict_head = 1;
402 head = NULL;
405 if (remote == &df_conflict_entry) {
406 df_conflict_remote = 1;
407 remote = NULL;
410 /* First, if there's a #16 situation, note that to prevent #13
411 * and #14.
413 if (!same(remote, head)) {
414 for (i = 1; i < head_idx; i++) {
415 if (same(stages[i], head)) {
416 head_match = i;
418 if (same(stages[i], remote)) {
419 remote_match = i;
424 /* We start with cases where the index is allowed to match
425 * something other than the head: #14(ALT) and #2ALT, where it
426 * is permitted to match the result instead.
428 /* #14, #14ALT, #2ALT */
429 if (remote && !df_conflict_head && head_match && !remote_match) {
430 if (index && !same(index, remote) && !same(index, head))
431 reject_merge(index);
432 return merged_entry(remote, index);
435 * If we have an entry in the index cache, then we want to
436 * make sure that it matches head.
438 if (index && !same(index, head)) {
439 reject_merge(index);
442 if (head) {
443 /* #5ALT, #15 */
444 if (same(head, remote))
445 return merged_entry(head, index);
446 /* #13, #3ALT */
447 if (!df_conflict_remote && remote_match && !head_match)
448 return merged_entry(head, index);
451 /* #1 */
452 if (!head && !remote && any_anc_missing)
453 return 0;
455 /* Below are "no merge" cases, which require that the index be
456 * up-to-date to avoid the files getting overwritten with
457 * conflict resolution files.
459 if (index) {
460 verify_uptodate(index);
463 /* #2, #3, #4, #6, #7, #9, #11. */
464 count = 0;
465 if (!head_match || !remote_match) {
466 for (i = 1; i < head_idx; i++) {
467 if (stages[i]) {
468 keep_entry(stages[i]);
469 count++;
470 break;
474 #if DBRT_DEBUG
475 else {
476 fprintf(stderr, "read-tree: warning #16 detected\n");
477 show_stage_entry(stderr, "head ", stages[head_match]);
478 show_stage_entry(stderr, "remote ", stages[remote_match]);
480 #endif
481 if (head) { count += keep_entry(head); }
482 if (remote) { count += keep_entry(remote); }
483 return count;
487 * Two-way merge.
489 * The rule is to "carry forward" what is in the index without losing
490 * information across a "fast forward", favoring a successful merge
491 * over a merge failure when it makes sense. For details of the
492 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
495 static int twoway_merge(struct cache_entry **src)
497 struct cache_entry *current = src[0];
498 struct cache_entry *oldtree = src[1], *newtree = src[2];
500 if (merge_size != 2)
501 return error("Cannot do a twoway merge of %d trees\n",
502 merge_size);
504 if (current) {
505 if ((!oldtree && !newtree) || /* 4 and 5 */
506 (!oldtree && newtree &&
507 same(current, newtree)) || /* 6 and 7 */
508 (oldtree && newtree &&
509 same(oldtree, newtree)) || /* 14 and 15 */
510 (oldtree && newtree &&
511 !same(oldtree, newtree) && /* 18 and 19*/
512 same(current, newtree))) {
513 return keep_entry(current);
515 else if (oldtree && !newtree && same(current, oldtree)) {
516 /* 10 or 11 */
517 return deleted_entry(oldtree, current);
519 else if (oldtree && newtree &&
520 same(current, oldtree) && !same(current, newtree)) {
521 /* 20 or 21 */
522 return merged_entry(newtree, current);
524 else {
525 /* all other failures */
526 if (oldtree)
527 reject_merge(oldtree);
528 if (current)
529 reject_merge(current);
530 if (newtree)
531 reject_merge(newtree);
532 return -1;
535 else if (newtree)
536 return merged_entry(newtree, current);
537 else
538 return deleted_entry(oldtree, current);
542 * One-way merge.
544 * The rule is:
545 * - take the stat information from stage0, take the data from stage1
547 static int oneway_merge(struct cache_entry **src)
549 struct cache_entry *old = src[0];
550 struct cache_entry *a = src[1];
552 if (merge_size != 1)
553 return error("Cannot do a oneway merge of %d trees\n",
554 merge_size);
556 if (!a)
557 return 0;
558 if (old && same(old, a)) {
559 return keep_entry(old);
561 return merged_entry(a, NULL);
564 static int read_cache_unmerged(void)
566 int i, deleted;
567 struct cache_entry **dst;
569 read_cache();
570 dst = active_cache;
571 deleted = 0;
572 for (i = 0; i < active_nr; i++) {
573 struct cache_entry *ce = active_cache[i];
574 if (ce_stage(ce)) {
575 deleted++;
576 continue;
578 if (deleted)
579 *dst = ce;
580 dst++;
582 active_nr -= deleted;
583 return deleted;
586 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [-u | -i] <sha1> [<sha2> [<sha3>]])";
588 static struct cache_file cache_file;
590 int main(int argc, char **argv)
592 int i, newfd, reset, stage = 0;
593 unsigned char sha1[20];
594 merge_fn_t fn = NULL;
596 newfd = hold_index_file_for_update(&cache_file, get_index_file());
597 if (newfd < 0)
598 die("unable to create new cachefile");
600 merge = 0;
601 reset = 0;
602 for (i = 1; i < argc; i++) {
603 const char *arg = argv[i];
605 /* "-u" means "update", meaning that a merge will update
606 * the working tree.
608 if (!strcmp(arg, "-u")) {
609 update = 1;
610 continue;
613 /* "-i" means "index only", meaning that a merge will
614 * not even look at the working tree.
616 if (!strcmp(arg, "-i")) {
617 index_only = 1;
618 continue;
621 /* This differs from "-m" in that we'll silently ignore unmerged entries */
622 if (!strcmp(arg, "--reset")) {
623 if (stage || merge)
624 usage(read_tree_usage);
625 reset = 1;
626 merge = 1;
627 stage = 1;
628 read_cache_unmerged();
629 continue;
632 if (!strcmp(arg, "--head")) {
633 head_idx = stage - 1;
634 fn = threeway_merge;
637 /* "-m" stands for "merge", meaning we start in stage 1 */
638 if (!strcmp(arg, "-m")) {
639 if (stage || merge)
640 usage(read_tree_usage);
641 if (read_cache_unmerged())
642 die("you need to resolve your current index first");
643 stage = 1;
644 merge = 1;
645 continue;
648 /* using -u and -i at the same time makes no sense */
649 if (1 < index_only + update)
650 usage(read_tree_usage);
652 if (get_sha1(arg, sha1) < 0)
653 usage(read_tree_usage);
654 if (list_tree(sha1) < 0)
655 die("failed to unpack tree object %s", arg);
656 stage++;
658 if (update && !merge)
659 usage(read_tree_usage);
660 if (merge && !fn) {
661 if (stage < 2)
662 die("just how do you expect me to merge %d trees?", stage-1);
663 switch (stage - 1) {
664 case 1:
665 fn = oneway_merge;
666 break;
667 case 2:
668 fn = twoway_merge;
669 break;
670 case 3:
671 fn = threeway_merge;
672 break;
673 default:
674 fn = threeway_merge;
675 break;
679 if (head_idx < 0) {
680 if (stage - 1 >= 3)
681 head_idx = stage - 2;
682 else
683 head_idx = 1;
686 unpack_trees(fn);
687 if (write_cache(newfd, active_cache, active_nr) ||
688 commit_index_file(&cache_file))
689 die("unable to write new index file");
690 return 0;