More accurately detect header lines in read_one_header_line
[git/spearce.git] / read-tree.c
blob82e2a9a4d3663fad513f607e25fe84a8ccbf5460
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"
12 #include <sys/time.h>
13 #include <signal.h>
15 static int reset = 0;
16 static int merge = 0;
17 static int update = 0;
18 static int index_only = 0;
19 static int nontrivial_merge = 0;
20 static int trivial_merges_only = 0;
21 static int aggressive = 0;
22 static int verbose_update = 0;
23 static volatile int progress_update = 0;
25 static int head_idx = -1;
26 static int merge_size = 0;
28 static struct object_list *trees = NULL;
30 static struct cache_entry df_conflict_entry = {
33 static struct tree_entry_list df_conflict_list = {
34 .name = NULL,
35 .next = &df_conflict_list
38 typedef int (*merge_fn_t)(struct cache_entry **src);
40 static int entcmp(char *name1, int dir1, char *name2, int dir2)
42 int len1 = strlen(name1);
43 int len2 = strlen(name2);
44 int len = len1 < len2 ? len1 : len2;
45 int ret = memcmp(name1, name2, len);
46 unsigned char c1, c2;
47 if (ret)
48 return ret;
49 c1 = name1[len];
50 c2 = name2[len];
51 if (!c1 && dir1)
52 c1 = '/';
53 if (!c2 && dir2)
54 c2 = '/';
55 ret = (c1 < c2) ? -1 : (c1 > c2) ? 1 : 0;
56 if (c1 && c2 && !ret)
57 ret = len1 - len2;
58 return ret;
61 static int unpack_trees_rec(struct tree_entry_list **posns, int len,
62 const char *base, merge_fn_t fn, int *indpos)
64 int baselen = strlen(base);
65 int src_size = len + 1;
66 do {
67 int i;
68 char *first;
69 int firstdir = 0;
70 int pathlen;
71 unsigned ce_size;
72 struct tree_entry_list **subposns;
73 struct cache_entry **src;
74 int any_files = 0;
75 int any_dirs = 0;
76 char *cache_name;
77 int ce_stage;
79 /* Find the first name in the input. */
81 first = NULL;
82 cache_name = NULL;
84 /* Check the cache */
85 if (merge && *indpos < active_nr) {
86 /* This is a bit tricky: */
87 /* If the index has a subdirectory (with
88 * contents) as the first name, it'll get a
89 * filename like "foo/bar". But that's after
90 * "foo", so the entry in trees will get
91 * handled first, at which point we'll go into
92 * "foo", and deal with "bar" from the index,
93 * because the base will be "foo/". The only
94 * way we can actually have "foo/bar" first of
95 * all the things is if the trees don't
96 * contain "foo" at all, in which case we'll
97 * handle "foo/bar" without going into the
98 * directory, but that's fine (and will return
99 * an error anyway, with the added unknown
100 * file case.
103 cache_name = active_cache[*indpos]->name;
104 if (strlen(cache_name) > baselen &&
105 !memcmp(cache_name, base, baselen)) {
106 cache_name += baselen;
107 first = cache_name;
108 } else {
109 cache_name = NULL;
113 #if DBRT_DEBUG > 1
114 if (first)
115 printf("index %s\n", first);
116 #endif
117 for (i = 0; i < len; i++) {
118 if (!posns[i] || posns[i] == &df_conflict_list)
119 continue;
120 #if DBRT_DEBUG > 1
121 printf("%d %s\n", i + 1, posns[i]->name);
122 #endif
123 if (!first || entcmp(first, firstdir,
124 posns[i]->name,
125 posns[i]->directory) > 0) {
126 first = posns[i]->name;
127 firstdir = posns[i]->directory;
130 /* No name means we're done */
131 if (!first)
132 return 0;
134 pathlen = strlen(first);
135 ce_size = cache_entry_size(baselen + pathlen);
137 src = xcalloc(src_size, sizeof(struct cache_entry *));
139 subposns = xcalloc(len, sizeof(struct tree_list_entry *));
141 if (cache_name && !strcmp(cache_name, first)) {
142 any_files = 1;
143 src[0] = active_cache[*indpos];
144 remove_cache_entry_at(*indpos);
147 for (i = 0; i < len; i++) {
148 struct cache_entry *ce;
150 if (!posns[i] ||
151 (posns[i] != &df_conflict_list &&
152 strcmp(first, posns[i]->name))) {
153 continue;
156 if (posns[i] == &df_conflict_list) {
157 src[i + merge] = &df_conflict_entry;
158 continue;
161 if (posns[i]->directory) {
162 any_dirs = 1;
163 parse_tree(posns[i]->item.tree);
164 subposns[i] = posns[i]->item.tree->entries;
165 posns[i] = posns[i]->next;
166 src[i + merge] = &df_conflict_entry;
167 continue;
170 if (!merge)
171 ce_stage = 0;
172 else if (i + 1 < head_idx)
173 ce_stage = 1;
174 else if (i + 1 > head_idx)
175 ce_stage = 3;
176 else
177 ce_stage = 2;
179 ce = xcalloc(1, ce_size);
180 ce->ce_mode = create_ce_mode(posns[i]->mode);
181 ce->ce_flags = create_ce_flags(baselen + pathlen,
182 ce_stage);
183 memcpy(ce->name, base, baselen);
184 memcpy(ce->name + baselen, first, pathlen + 1);
186 any_files = 1;
188 memcpy(ce->sha1, posns[i]->item.any->sha1, 20);
189 src[i + merge] = ce;
190 subposns[i] = &df_conflict_list;
191 posns[i] = posns[i]->next;
193 if (any_files) {
194 if (merge) {
195 int ret;
197 #if DBRT_DEBUG > 1
198 printf("%s:\n", first);
199 for (i = 0; i < src_size; i++) {
200 printf(" %d ", i);
201 if (src[i])
202 printf("%s\n", sha1_to_hex(src[i]->sha1));
203 else
204 printf("\n");
206 #endif
207 ret = fn(src);
209 #if DBRT_DEBUG > 1
210 printf("Added %d entries\n", ret);
211 #endif
212 *indpos += ret;
213 } else {
214 for (i = 0; i < src_size; i++) {
215 if (src[i]) {
216 add_cache_entry(src[i], ADD_CACHE_OK_TO_ADD|ADD_CACHE_SKIP_DFCHECK);
221 if (any_dirs) {
222 char *newbase = xmalloc(baselen + 2 + pathlen);
223 memcpy(newbase, base, baselen);
224 memcpy(newbase + baselen, first, pathlen);
225 newbase[baselen + pathlen] = '/';
226 newbase[baselen + pathlen + 1] = '\0';
227 if (unpack_trees_rec(subposns, len, newbase, fn,
228 indpos))
229 return -1;
230 free(newbase);
232 free(subposns);
233 free(src);
234 } while (1);
237 static void reject_merge(struct cache_entry *ce)
239 die("Entry '%s' would be overwritten by merge. Cannot merge.",
240 ce->name);
243 /* Unlink the last component and attempt to remove leading
244 * directories, in case this unlink is the removal of the
245 * last entry in the directory -- empty directories are removed.
247 static void unlink_entry(char *name)
249 char *cp, *prev;
251 if (unlink(name))
252 return;
253 prev = NULL;
254 while (1) {
255 int status;
256 cp = strrchr(name, '/');
257 if (prev)
258 *prev = '/';
259 if (!cp)
260 break;
262 *cp = 0;
263 status = rmdir(name);
264 if (status) {
265 *cp = '/';
266 break;
268 prev = cp;
272 static void progress_interval(int signum)
274 progress_update = 1;
277 static void setup_progress_signal(void)
279 struct sigaction sa;
280 struct itimerval v;
282 memset(&sa, 0, sizeof(sa));
283 sa.sa_handler = progress_interval;
284 sigemptyset(&sa.sa_mask);
285 sa.sa_flags = SA_RESTART;
286 sigaction(SIGALRM, &sa, NULL);
288 v.it_interval.tv_sec = 1;
289 v.it_interval.tv_usec = 0;
290 v.it_value = v.it_interval;
291 setitimer(ITIMER_REAL, &v, NULL);
294 static void check_updates(struct cache_entry **src, int nr)
296 static struct checkout state = {
297 .base_dir = "",
298 .force = 1,
299 .quiet = 1,
300 .refresh_cache = 1,
302 unsigned short mask = htons(CE_UPDATE);
303 unsigned last_percent = 200, cnt = 0, total = 0;
305 if (update && verbose_update) {
306 for (total = cnt = 0; cnt < nr; cnt++) {
307 struct cache_entry *ce = src[cnt];
308 if (!ce->ce_mode || ce->ce_flags & mask)
309 total++;
312 /* Don't bother doing this for very small updates */
313 if (total < 250)
314 total = 0;
316 if (total) {
317 fprintf(stderr, "Checking files out...\n");
318 setup_progress_signal();
319 progress_update = 1;
321 cnt = 0;
324 while (nr--) {
325 struct cache_entry *ce = *src++;
327 if (total) {
328 if (!ce->ce_mode || ce->ce_flags & mask) {
329 unsigned percent;
330 cnt++;
331 percent = (cnt * 100) / total;
332 if (percent != last_percent ||
333 progress_update) {
334 fprintf(stderr, "%4u%% (%u/%u) done\r",
335 percent, cnt, total);
336 last_percent = percent;
340 if (!ce->ce_mode) {
341 if (update)
342 unlink_entry(ce->name);
343 continue;
345 if (ce->ce_flags & mask) {
346 ce->ce_flags &= ~mask;
347 if (update)
348 checkout_entry(ce, &state, NULL);
351 if (total) {
352 signal(SIGALRM, SIG_IGN);
353 fputc('\n', stderr);
357 static int unpack_trees(merge_fn_t fn)
359 int indpos = 0;
360 unsigned len = object_list_length(trees);
361 struct tree_entry_list **posns;
362 int i;
363 struct object_list *posn = trees;
364 merge_size = len;
366 if (len) {
367 posns = xmalloc(len * sizeof(struct tree_entry_list *));
368 for (i = 0; i < len; i++) {
369 posns[i] = ((struct tree *) posn->item)->entries;
370 posn = posn->next;
372 if (unpack_trees_rec(posns, len, "", fn, &indpos))
373 return -1;
376 if (trivial_merges_only && nontrivial_merge)
377 die("Merge requires file-level merging");
379 check_updates(active_cache, active_nr);
380 return 0;
383 static int list_tree(unsigned char *sha1)
385 struct tree *tree = parse_tree_indirect(sha1);
386 if (!tree)
387 return -1;
388 object_list_append(&tree->object, &trees);
389 return 0;
392 static int same(struct cache_entry *a, struct cache_entry *b)
394 if (!!a != !!b)
395 return 0;
396 if (!a && !b)
397 return 1;
398 return a->ce_mode == b->ce_mode &&
399 !memcmp(a->sha1, b->sha1, 20);
404 * When a CE gets turned into an unmerged entry, we
405 * want it to be up-to-date
407 static void verify_uptodate(struct cache_entry *ce)
409 struct stat st;
411 if (index_only || reset)
412 return;
414 if (!lstat(ce->name, &st)) {
415 unsigned changed = ce_match_stat(ce, &st, 1);
416 if (!changed)
417 return;
418 errno = 0;
420 if (reset) {
421 ce->ce_flags |= htons(CE_UPDATE);
422 return;
424 if (errno == ENOENT)
425 return;
426 die("Entry '%s' not uptodate. Cannot merge.", ce->name);
430 * We do not want to remove or overwrite a working tree file that
431 * is not tracked.
433 static void verify_absent(const char *path, const char *action)
435 struct stat st;
437 if (index_only || reset || !update)
438 return;
439 if (!lstat(path, &st))
440 die("Untracked working tree file '%s' "
441 "would be %s by merge.", path, action);
444 static int merged_entry(struct cache_entry *merge, struct cache_entry *old)
446 merge->ce_flags |= htons(CE_UPDATE);
447 if (old) {
449 * See if we can re-use the old CE directly?
450 * That way we get the uptodate stat info.
452 * This also removes the UPDATE flag on
453 * a match.
455 if (same(old, merge)) {
456 *merge = *old;
457 } else {
458 verify_uptodate(old);
461 else
462 verify_absent(merge->name, "overwritten");
464 merge->ce_flags &= ~htons(CE_STAGEMASK);
465 add_cache_entry(merge, ADD_CACHE_OK_TO_ADD);
466 return 1;
469 static int deleted_entry(struct cache_entry *ce, struct cache_entry *old)
471 if (old)
472 verify_uptodate(old);
473 else
474 verify_absent(ce->name, "removed");
475 ce->ce_mode = 0;
476 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
477 return 1;
480 static int keep_entry(struct cache_entry *ce)
482 add_cache_entry(ce, ADD_CACHE_OK_TO_ADD);
483 return 1;
486 #if DBRT_DEBUG
487 static void show_stage_entry(FILE *o,
488 const char *label, const struct cache_entry *ce)
490 if (!ce)
491 fprintf(o, "%s (missing)\n", label);
492 else
493 fprintf(o, "%s%06o %s %d\t%s\n",
494 label,
495 ntohl(ce->ce_mode),
496 sha1_to_hex(ce->sha1),
497 ce_stage(ce),
498 ce->name);
500 #endif
502 static int threeway_merge(struct cache_entry **stages)
504 struct cache_entry *index;
505 struct cache_entry *head;
506 struct cache_entry *remote = stages[head_idx + 1];
507 int count;
508 int head_match = 0;
509 int remote_match = 0;
510 const char *path = NULL;
512 int df_conflict_head = 0;
513 int df_conflict_remote = 0;
515 int any_anc_missing = 0;
516 int no_anc_exists = 1;
517 int i;
519 for (i = 1; i < head_idx; i++) {
520 if (!stages[i])
521 any_anc_missing = 1;
522 else {
523 if (!path)
524 path = stages[i]->name;
525 no_anc_exists = 0;
529 index = stages[0];
530 head = stages[head_idx];
532 if (head == &df_conflict_entry) {
533 df_conflict_head = 1;
534 head = NULL;
537 if (remote == &df_conflict_entry) {
538 df_conflict_remote = 1;
539 remote = NULL;
542 if (!path && index)
543 path = index->name;
544 if (!path && head)
545 path = head->name;
546 if (!path && remote)
547 path = remote->name;
549 /* First, if there's a #16 situation, note that to prevent #13
550 * and #14.
552 if (!same(remote, head)) {
553 for (i = 1; i < head_idx; i++) {
554 if (same(stages[i], head)) {
555 head_match = i;
557 if (same(stages[i], remote)) {
558 remote_match = i;
563 /* We start with cases where the index is allowed to match
564 * something other than the head: #14(ALT) and #2ALT, where it
565 * is permitted to match the result instead.
567 /* #14, #14ALT, #2ALT */
568 if (remote && !df_conflict_head && head_match && !remote_match) {
569 if (index && !same(index, remote) && !same(index, head))
570 reject_merge(index);
571 return merged_entry(remote, index);
574 * If we have an entry in the index cache, then we want to
575 * make sure that it matches head.
577 if (index && !same(index, head)) {
578 reject_merge(index);
581 if (head) {
582 /* #5ALT, #15 */
583 if (same(head, remote))
584 return merged_entry(head, index);
585 /* #13, #3ALT */
586 if (!df_conflict_remote && remote_match && !head_match)
587 return merged_entry(head, index);
590 /* #1 */
591 if (!head && !remote && any_anc_missing)
592 return 0;
594 /* Under the new "aggressive" rule, we resolve mostly trivial
595 * cases that we historically had git-merge-one-file resolve.
597 if (aggressive) {
598 int head_deleted = !head && !df_conflict_head;
599 int remote_deleted = !remote && !df_conflict_remote;
601 * Deleted in both.
602 * Deleted in one and unchanged in the other.
604 if ((head_deleted && remote_deleted) ||
605 (head_deleted && remote && remote_match) ||
606 (remote_deleted && head && head_match)) {
607 if (index)
608 return deleted_entry(index, index);
609 else if (path)
610 verify_absent(path, "removed");
611 return 0;
614 * Added in both, identically.
616 if (no_anc_exists && head && remote && same(head, remote))
617 return merged_entry(head, index);
621 /* Below are "no merge" cases, which require that the index be
622 * up-to-date to avoid the files getting overwritten with
623 * conflict resolution files.
625 if (index) {
626 verify_uptodate(index);
628 else if (path)
629 verify_absent(path, "overwritten");
631 nontrivial_merge = 1;
633 /* #2, #3, #4, #6, #7, #9, #11. */
634 count = 0;
635 if (!head_match || !remote_match) {
636 for (i = 1; i < head_idx; i++) {
637 if (stages[i]) {
638 keep_entry(stages[i]);
639 count++;
640 break;
644 #if DBRT_DEBUG
645 else {
646 fprintf(stderr, "read-tree: warning #16 detected\n");
647 show_stage_entry(stderr, "head ", stages[head_match]);
648 show_stage_entry(stderr, "remote ", stages[remote_match]);
650 #endif
651 if (head) { count += keep_entry(head); }
652 if (remote) { count += keep_entry(remote); }
653 return count;
657 * Two-way merge.
659 * The rule is to "carry forward" what is in the index without losing
660 * information across a "fast forward", favoring a successful merge
661 * over a merge failure when it makes sense. For details of the
662 * "carry forward" rule, please see <Documentation/git-read-tree.txt>.
665 static int twoway_merge(struct cache_entry **src)
667 struct cache_entry *current = src[0];
668 struct cache_entry *oldtree = src[1], *newtree = src[2];
670 if (merge_size != 2)
671 return error("Cannot do a twoway merge of %d trees",
672 merge_size);
674 if (current) {
675 if ((!oldtree && !newtree) || /* 4 and 5 */
676 (!oldtree && newtree &&
677 same(current, newtree)) || /* 6 and 7 */
678 (oldtree && newtree &&
679 same(oldtree, newtree)) || /* 14 and 15 */
680 (oldtree && newtree &&
681 !same(oldtree, newtree) && /* 18 and 19*/
682 same(current, newtree))) {
683 return keep_entry(current);
685 else if (oldtree && !newtree && same(current, oldtree)) {
686 /* 10 or 11 */
687 return deleted_entry(oldtree, current);
689 else if (oldtree && newtree &&
690 same(current, oldtree) && !same(current, newtree)) {
691 /* 20 or 21 */
692 return merged_entry(newtree, current);
694 else {
695 /* all other failures */
696 if (oldtree)
697 reject_merge(oldtree);
698 if (current)
699 reject_merge(current);
700 if (newtree)
701 reject_merge(newtree);
702 return -1;
705 else if (newtree)
706 return merged_entry(newtree, current);
707 else
708 return deleted_entry(oldtree, current);
712 * One-way merge.
714 * The rule is:
715 * - take the stat information from stage0, take the data from stage1
717 static int oneway_merge(struct cache_entry **src)
719 struct cache_entry *old = src[0];
720 struct cache_entry *a = src[1];
722 if (merge_size != 1)
723 return error("Cannot do a oneway merge of %d trees",
724 merge_size);
726 if (!a)
727 return deleted_entry(old, old);
728 if (old && same(old, a)) {
729 if (reset) {
730 struct stat st;
731 if (lstat(old->name, &st) ||
732 ce_match_stat(old, &st, 1))
733 old->ce_flags |= htons(CE_UPDATE);
735 return keep_entry(old);
737 return merged_entry(a, old);
740 static int read_cache_unmerged(void)
742 int i, deleted;
743 struct cache_entry **dst;
745 read_cache();
746 dst = active_cache;
747 deleted = 0;
748 for (i = 0; i < active_nr; i++) {
749 struct cache_entry *ce = active_cache[i];
750 if (ce_stage(ce)) {
751 deleted++;
752 continue;
754 if (deleted)
755 *dst = ce;
756 dst++;
758 active_nr -= deleted;
759 return deleted;
762 static const char read_tree_usage[] = "git-read-tree (<sha> | -m [--aggressive] [-u | -i] <sha1> [<sha2> [<sha3>]])";
764 static struct cache_file cache_file;
766 int main(int argc, char **argv)
768 int i, newfd, stage = 0;
769 unsigned char sha1[20];
770 merge_fn_t fn = NULL;
772 setup_git_directory();
773 git_config(git_default_config);
775 newfd = hold_index_file_for_update(&cache_file, get_index_file());
776 if (newfd < 0)
777 die("unable to create new cachefile");
779 git_config(git_default_config);
781 merge = 0;
782 reset = 0;
783 for (i = 1; i < argc; i++) {
784 const char *arg = argv[i];
786 /* "-u" means "update", meaning that a merge will update
787 * the working tree.
789 if (!strcmp(arg, "-u")) {
790 update = 1;
791 continue;
794 if (!strcmp(arg, "-v")) {
795 verbose_update = 1;
796 continue;
799 /* "-i" means "index only", meaning that a merge will
800 * not even look at the working tree.
802 if (!strcmp(arg, "-i")) {
803 index_only = 1;
804 continue;
807 /* This differs from "-m" in that we'll silently ignore unmerged entries */
808 if (!strcmp(arg, "--reset")) {
809 if (stage || merge)
810 usage(read_tree_usage);
811 reset = 1;
812 merge = 1;
813 stage = 1;
814 read_cache_unmerged();
815 continue;
818 if (!strcmp(arg, "--trivial")) {
819 trivial_merges_only = 1;
820 continue;
823 if (!strcmp(arg, "--aggressive")) {
824 aggressive = 1;
825 continue;
828 /* "-m" stands for "merge", meaning we start in stage 1 */
829 if (!strcmp(arg, "-m")) {
830 if (stage || merge)
831 usage(read_tree_usage);
832 if (read_cache_unmerged())
833 die("you need to resolve your current index first");
834 stage = 1;
835 merge = 1;
836 continue;
839 /* using -u and -i at the same time makes no sense */
840 if (1 < index_only + update)
841 usage(read_tree_usage);
843 if (get_sha1(arg, sha1))
844 die("Not a valid object name %s", arg);
845 if (list_tree(sha1) < 0)
846 die("failed to unpack tree object %s", arg);
847 stage++;
849 if ((update||index_only) && !merge)
850 usage(read_tree_usage);
852 if (merge) {
853 if (stage < 2)
854 die("just how do you expect me to merge %d trees?", stage-1);
855 switch (stage - 1) {
856 case 1:
857 fn = oneway_merge;
858 break;
859 case 2:
860 fn = twoway_merge;
861 break;
862 case 3:
863 fn = threeway_merge;
864 break;
865 default:
866 fn = threeway_merge;
867 break;
870 if (stage - 1 >= 3)
871 head_idx = stage - 2;
872 else
873 head_idx = 1;
876 unpack_trees(fn);
877 if (write_cache(newfd, active_cache, active_nr) ||
878 commit_index_file(&cache_file))
879 die("unable to write new index file");
880 return 0;