Merge branch 'nd/worktree-prune'
[git.git] / merge-recursive.c
blob9c05eb7f700eed0dc3e20a4c22f3a60e0aa21488
1 /*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6 #include "cache.h"
7 #include "config.h"
8 #include "advice.h"
9 #include "lockfile.h"
10 #include "cache-tree.h"
11 #include "commit.h"
12 #include "blob.h"
13 #include "builtin.h"
14 #include "tree-walk.h"
15 #include "diff.h"
16 #include "diffcore.h"
17 #include "tag.h"
18 #include "unpack-trees.h"
19 #include "string-list.h"
20 #include "xdiff-interface.h"
21 #include "ll-merge.h"
22 #include "attr.h"
23 #include "merge-recursive.h"
24 #include "dir.h"
25 #include "submodule.h"
27 struct path_hashmap_entry {
28 struct hashmap_entry e;
29 char path[FLEX_ARRAY];
32 static int path_hashmap_cmp(const void *cmp_data,
33 const void *entry,
34 const void *entry_or_key,
35 const void *keydata)
37 const struct path_hashmap_entry *a = entry;
38 const struct path_hashmap_entry *b = entry_or_key;
39 const char *key = keydata;
41 if (ignore_case)
42 return strcasecmp(a->path, key ? key : b->path);
43 else
44 return strcmp(a->path, key ? key : b->path);
47 static unsigned int path_hash(const char *path)
49 return ignore_case ? strihash(path) : strhash(path);
52 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
53 char *dir)
55 struct dir_rename_entry key;
57 if (dir == NULL)
58 return NULL;
59 hashmap_entry_init(&key, strhash(dir));
60 key.dir = dir;
61 return hashmap_get(hashmap, &key, NULL);
64 static int dir_rename_cmp(const void *unused_cmp_data,
65 const void *entry,
66 const void *entry_or_key,
67 const void *unused_keydata)
69 const struct dir_rename_entry *e1 = entry;
70 const struct dir_rename_entry *e2 = entry_or_key;
72 return strcmp(e1->dir, e2->dir);
75 static void dir_rename_init(struct hashmap *map)
77 hashmap_init(map, dir_rename_cmp, NULL, 0);
80 static void dir_rename_entry_init(struct dir_rename_entry *entry,
81 char *directory)
83 hashmap_entry_init(entry, strhash(directory));
84 entry->dir = directory;
85 entry->non_unique_new_dir = 0;
86 strbuf_init(&entry->new_dir, 0);
87 string_list_init(&entry->possible_new_dirs, 0);
90 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
91 char *target_file)
93 struct collision_entry key;
95 hashmap_entry_init(&key, strhash(target_file));
96 key.target_file = target_file;
97 return hashmap_get(hashmap, &key, NULL);
100 static int collision_cmp(void *unused_cmp_data,
101 const struct collision_entry *e1,
102 const struct collision_entry *e2,
103 const void *unused_keydata)
105 return strcmp(e1->target_file, e2->target_file);
108 static void collision_init(struct hashmap *map)
110 hashmap_init(map, (hashmap_cmp_fn) collision_cmp, NULL, 0);
113 static void flush_output(struct merge_options *o)
115 if (o->buffer_output < 2 && o->obuf.len) {
116 fputs(o->obuf.buf, stdout);
117 strbuf_reset(&o->obuf);
121 static int err(struct merge_options *o, const char *err, ...)
123 va_list params;
125 if (o->buffer_output < 2)
126 flush_output(o);
127 else {
128 strbuf_complete(&o->obuf, '\n');
129 strbuf_addstr(&o->obuf, "error: ");
131 va_start(params, err);
132 strbuf_vaddf(&o->obuf, err, params);
133 va_end(params);
134 if (o->buffer_output > 1)
135 strbuf_addch(&o->obuf, '\n');
136 else {
137 error("%s", o->obuf.buf);
138 strbuf_reset(&o->obuf);
141 return -1;
144 static struct tree *shift_tree_object(struct tree *one, struct tree *two,
145 const char *subtree_shift)
147 struct object_id shifted;
149 if (!*subtree_shift) {
150 shift_tree(&one->object.oid, &two->object.oid, &shifted, 0);
151 } else {
152 shift_tree_by(&one->object.oid, &two->object.oid, &shifted,
153 subtree_shift);
155 if (!oidcmp(&two->object.oid, &shifted))
156 return two;
157 return lookup_tree(&shifted);
160 static struct commit *make_virtual_commit(struct tree *tree, const char *comment)
162 struct commit *commit = alloc_commit_node();
164 set_merge_remote_desc(commit, comment, (struct object *)commit);
165 commit->tree = tree;
166 commit->object.parsed = 1;
167 return commit;
171 * Since we use get_tree_entry(), which does not put the read object into
172 * the object pool, we cannot rely on a == b.
174 static int oid_eq(const struct object_id *a, const struct object_id *b)
176 if (!a && !b)
177 return 2;
178 return a && b && oidcmp(a, b) == 0;
181 enum rename_type {
182 RENAME_NORMAL = 0,
183 RENAME_DIR,
184 RENAME_DELETE,
185 RENAME_ONE_FILE_TO_ONE,
186 RENAME_ONE_FILE_TO_TWO,
187 RENAME_TWO_FILES_TO_ONE
190 struct rename_conflict_info {
191 enum rename_type rename_type;
192 struct diff_filepair *pair1;
193 struct diff_filepair *pair2;
194 const char *branch1;
195 const char *branch2;
196 struct stage_data *dst_entry1;
197 struct stage_data *dst_entry2;
198 struct diff_filespec ren1_other;
199 struct diff_filespec ren2_other;
203 * Since we want to write the index eventually, we cannot reuse the index
204 * for these (temporary) data.
206 struct stage_data {
207 struct {
208 unsigned mode;
209 struct object_id oid;
210 } stages[4];
211 struct rename_conflict_info *rename_conflict_info;
212 unsigned processed:1;
215 static inline void setup_rename_conflict_info(enum rename_type rename_type,
216 struct diff_filepair *pair1,
217 struct diff_filepair *pair2,
218 const char *branch1,
219 const char *branch2,
220 struct stage_data *dst_entry1,
221 struct stage_data *dst_entry2,
222 struct merge_options *o,
223 struct stage_data *src_entry1,
224 struct stage_data *src_entry2)
226 struct rename_conflict_info *ci = xcalloc(1, sizeof(struct rename_conflict_info));
227 ci->rename_type = rename_type;
228 ci->pair1 = pair1;
229 ci->branch1 = branch1;
230 ci->branch2 = branch2;
232 ci->dst_entry1 = dst_entry1;
233 dst_entry1->rename_conflict_info = ci;
234 dst_entry1->processed = 0;
236 assert(!pair2 == !dst_entry2);
237 if (dst_entry2) {
238 ci->dst_entry2 = dst_entry2;
239 ci->pair2 = pair2;
240 dst_entry2->rename_conflict_info = ci;
243 if (rename_type == RENAME_TWO_FILES_TO_ONE) {
245 * For each rename, there could have been
246 * modifications on the side of history where that
247 * file was not renamed.
249 int ostage1 = o->branch1 == branch1 ? 3 : 2;
250 int ostage2 = ostage1 ^ 1;
252 ci->ren1_other.path = pair1->one->path;
253 oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid);
254 ci->ren1_other.mode = src_entry1->stages[ostage1].mode;
256 ci->ren2_other.path = pair2->one->path;
257 oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid);
258 ci->ren2_other.mode = src_entry2->stages[ostage2].mode;
262 static int show(struct merge_options *o, int v)
264 return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5;
267 __attribute__((format (printf, 3, 4)))
268 static void output(struct merge_options *o, int v, const char *fmt, ...)
270 va_list ap;
272 if (!show(o, v))
273 return;
275 strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
277 va_start(ap, fmt);
278 strbuf_vaddf(&o->obuf, fmt, ap);
279 va_end(ap);
281 strbuf_addch(&o->obuf, '\n');
282 if (!o->buffer_output)
283 flush_output(o);
286 static void output_commit_title(struct merge_options *o, struct commit *commit)
288 strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
289 if (commit->util)
290 strbuf_addf(&o->obuf, "virtual %s\n",
291 merge_remote_util(commit)->name);
292 else {
293 strbuf_add_unique_abbrev(&o->obuf, &commit->object.oid,
294 DEFAULT_ABBREV);
295 strbuf_addch(&o->obuf, ' ');
296 if (parse_commit(commit) != 0)
297 strbuf_addstr(&o->obuf, _("(bad commit)\n"));
298 else {
299 const char *title;
300 const char *msg = get_commit_buffer(commit, NULL);
301 int len = find_commit_subject(msg, &title);
302 if (len)
303 strbuf_addf(&o->obuf, "%.*s\n", len, title);
304 unuse_commit_buffer(commit, msg);
307 flush_output(o);
310 static int add_cacheinfo(struct merge_options *o,
311 unsigned int mode, const struct object_id *oid,
312 const char *path, int stage, int refresh, int options)
314 struct cache_entry *ce;
315 int ret;
317 ce = make_cache_entry(mode, oid ? oid->hash : null_sha1, path, stage, 0);
318 if (!ce)
319 return err(o, _("addinfo_cache failed for path '%s'"), path);
321 ret = add_cache_entry(ce, options);
322 if (refresh) {
323 struct cache_entry *nce;
325 nce = refresh_cache_entry(ce, CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
326 if (!nce)
327 return err(o, _("addinfo_cache failed for path '%s'"), path);
328 if (nce != ce)
329 ret = add_cache_entry(nce, options);
331 return ret;
334 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
336 parse_tree(tree);
337 init_tree_desc(desc, tree->buffer, tree->size);
340 static int git_merge_trees(struct merge_options *o,
341 struct tree *common,
342 struct tree *head,
343 struct tree *merge)
345 int rc;
346 struct tree_desc t[3];
348 memset(&o->unpack_opts, 0, sizeof(o->unpack_opts));
349 if (o->call_depth)
350 o->unpack_opts.index_only = 1;
351 else
352 o->unpack_opts.update = 1;
353 o->unpack_opts.merge = 1;
354 o->unpack_opts.head_idx = 2;
355 o->unpack_opts.fn = threeway_merge;
356 o->unpack_opts.src_index = &the_index;
357 o->unpack_opts.dst_index = &the_index;
358 setup_unpack_trees_porcelain(&o->unpack_opts, "merge");
360 init_tree_desc_from_tree(t+0, common);
361 init_tree_desc_from_tree(t+1, head);
362 init_tree_desc_from_tree(t+2, merge);
364 rc = unpack_trees(3, t, &o->unpack_opts);
366 * unpack_trees NULLifies src_index, but it's used in verify_uptodate,
367 * so set to the new index which will usually have modification
368 * timestamp info copied over.
370 o->unpack_opts.src_index = &the_index;
371 cache_tree_free(&active_cache_tree);
372 return rc;
375 struct tree *write_tree_from_memory(struct merge_options *o)
377 struct tree *result = NULL;
379 if (unmerged_cache()) {
380 int i;
381 fprintf(stderr, "BUG: There are unmerged index entries:\n");
382 for (i = 0; i < active_nr; i++) {
383 const struct cache_entry *ce = active_cache[i];
384 if (ce_stage(ce))
385 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
386 (int)ce_namelen(ce), ce->name);
388 die("BUG: unmerged index entries in merge-recursive.c");
391 if (!active_cache_tree)
392 active_cache_tree = cache_tree();
394 if (!cache_tree_fully_valid(active_cache_tree) &&
395 cache_tree_update(&the_index, 0) < 0) {
396 err(o, _("error building trees"));
397 return NULL;
400 result = lookup_tree(&active_cache_tree->oid);
402 return result;
405 static int save_files_dirs(const struct object_id *oid,
406 struct strbuf *base, const char *path,
407 unsigned int mode, int stage, void *context)
409 struct path_hashmap_entry *entry;
410 int baselen = base->len;
411 struct merge_options *o = context;
413 strbuf_addstr(base, path);
415 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
416 hashmap_entry_init(entry, path_hash(entry->path));
417 hashmap_add(&o->current_file_dir_set, entry);
419 strbuf_setlen(base, baselen);
420 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
423 static void get_files_dirs(struct merge_options *o, struct tree *tree)
425 struct pathspec match_all;
426 memset(&match_all, 0, sizeof(match_all));
427 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
430 static int get_tree_entry_if_blob(struct tree *tree,
431 const char *path,
432 struct object_id *hashy,
433 unsigned int *mode_o)
435 int ret;
437 ret = get_tree_entry(&tree->object.oid, path, hashy, mode_o);
438 if (S_ISDIR(*mode_o)) {
439 oidcpy(hashy, &null_oid);
440 *mode_o = 0;
442 return ret;
446 * Returns an index_entry instance which doesn't have to correspond to
447 * a real cache entry in Git's index.
449 static struct stage_data *insert_stage_data(const char *path,
450 struct tree *o, struct tree *a, struct tree *b,
451 struct string_list *entries)
453 struct string_list_item *item;
454 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
455 get_tree_entry_if_blob(o, path,
456 &e->stages[1].oid, &e->stages[1].mode);
457 get_tree_entry_if_blob(a, path,
458 &e->stages[2].oid, &e->stages[2].mode);
459 get_tree_entry_if_blob(b, path,
460 &e->stages[3].oid, &e->stages[3].mode);
461 item = string_list_insert(entries, path);
462 item->util = e;
463 return e;
467 * Create a dictionary mapping file names to stage_data objects. The
468 * dictionary contains one entry for every path with a non-zero stage entry.
470 static struct string_list *get_unmerged(void)
472 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
473 int i;
475 unmerged->strdup_strings = 1;
477 for (i = 0; i < active_nr; i++) {
478 struct string_list_item *item;
479 struct stage_data *e;
480 const struct cache_entry *ce = active_cache[i];
481 if (!ce_stage(ce))
482 continue;
484 item = string_list_lookup(unmerged, ce->name);
485 if (!item) {
486 item = string_list_insert(unmerged, ce->name);
487 item->util = xcalloc(1, sizeof(struct stage_data));
489 e = item->util;
490 e->stages[ce_stage(ce)].mode = ce->ce_mode;
491 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
494 return unmerged;
497 static int string_list_df_name_compare(const char *one, const char *two)
499 int onelen = strlen(one);
500 int twolen = strlen(two);
502 * Here we only care that entries for D/F conflicts are
503 * adjacent, in particular with the file of the D/F conflict
504 * appearing before files below the corresponding directory.
505 * The order of the rest of the list is irrelevant for us.
507 * To achieve this, we sort with df_name_compare and provide
508 * the mode S_IFDIR so that D/F conflicts will sort correctly.
509 * We use the mode S_IFDIR for everything else for simplicity,
510 * since in other cases any changes in their order due to
511 * sorting cause no problems for us.
513 int cmp = df_name_compare(one, onelen, S_IFDIR,
514 two, twolen, S_IFDIR);
516 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
517 * that 'foo' comes before 'foo/bar'.
519 if (cmp)
520 return cmp;
521 return onelen - twolen;
524 static void record_df_conflict_files(struct merge_options *o,
525 struct string_list *entries)
527 /* If there is a D/F conflict and the file for such a conflict
528 * currently exist in the working tree, we want to allow it to be
529 * removed to make room for the corresponding directory if needed.
530 * The files underneath the directories of such D/F conflicts will
531 * be processed before the corresponding file involved in the D/F
532 * conflict. If the D/F directory ends up being removed by the
533 * merge, then we won't have to touch the D/F file. If the D/F
534 * directory needs to be written to the working copy, then the D/F
535 * file will simply be removed (in make_room_for_path()) to make
536 * room for the necessary paths. Note that if both the directory
537 * and the file need to be present, then the D/F file will be
538 * reinstated with a new unique name at the time it is processed.
540 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
541 const char *last_file = NULL;
542 int last_len = 0;
543 int i;
546 * If we're merging merge-bases, we don't want to bother with
547 * any working directory changes.
549 if (o->call_depth)
550 return;
552 /* Ensure D/F conflicts are adjacent in the entries list. */
553 for (i = 0; i < entries->nr; i++) {
554 struct string_list_item *next = &entries->items[i];
555 string_list_append(&df_sorted_entries, next->string)->util =
556 next->util;
558 df_sorted_entries.cmp = string_list_df_name_compare;
559 string_list_sort(&df_sorted_entries);
561 string_list_clear(&o->df_conflict_file_set, 1);
562 for (i = 0; i < df_sorted_entries.nr; i++) {
563 const char *path = df_sorted_entries.items[i].string;
564 int len = strlen(path);
565 struct stage_data *e = df_sorted_entries.items[i].util;
568 * Check if last_file & path correspond to a D/F conflict;
569 * i.e. whether path is last_file+'/'+<something>.
570 * If so, record that it's okay to remove last_file to make
571 * room for path and friends if needed.
573 if (last_file &&
574 len > last_len &&
575 memcmp(path, last_file, last_len) == 0 &&
576 path[last_len] == '/') {
577 string_list_insert(&o->df_conflict_file_set, last_file);
581 * Determine whether path could exist as a file in the
582 * working directory as a possible D/F conflict. This
583 * will only occur when it exists in stage 2 as a
584 * file.
586 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
587 last_file = path;
588 last_len = len;
589 } else {
590 last_file = NULL;
593 string_list_clear(&df_sorted_entries, 0);
596 struct rename {
597 struct diff_filepair *pair;
599 * Purpose of src_entry and dst_entry:
601 * If 'before' is renamed to 'after' then src_entry will contain
602 * the versions of 'before' from the merge_base, HEAD, and MERGE in
603 * stages 1, 2, and 3; dst_entry will contain the respective
604 * versions of 'after' in corresponding locations. Thus, we have a
605 * total of six modes and oids, though some will be null. (Stage 0
606 * is ignored; we're interested in handling conflicts.)
608 * Since we don't turn on break-rewrites by default, neither
609 * src_entry nor dst_entry can have all three of their stages have
610 * non-null oids, meaning at most four of the six will be non-null.
611 * Also, since this is a rename, both src_entry and dst_entry will
612 * have at least one non-null oid, meaning at least two will be
613 * non-null. Of the six oids, a typical rename will have three be
614 * non-null. Only two implies a rename/delete, and four implies a
615 * rename/add.
617 struct stage_data *src_entry;
618 struct stage_data *dst_entry;
619 unsigned add_turned_into_rename:1;
620 unsigned processed:1;
623 static int update_stages(struct merge_options *opt, const char *path,
624 const struct diff_filespec *o,
625 const struct diff_filespec *a,
626 const struct diff_filespec *b)
630 * NOTE: It is usually a bad idea to call update_stages on a path
631 * before calling update_file on that same path, since it can
632 * sometimes lead to spurious "refusing to lose untracked file..."
633 * messages from update_file (via make_room_for path via
634 * would_lose_untracked). Instead, reverse the order of the calls
635 * (executing update_file first and then update_stages).
637 int clear = 1;
638 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
639 if (clear)
640 if (remove_file_from_cache(path))
641 return -1;
642 if (o)
643 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
644 return -1;
645 if (a)
646 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
647 return -1;
648 if (b)
649 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
650 return -1;
651 return 0;
654 static int update_stages_for_stage_data(struct merge_options *opt,
655 const char *path,
656 const struct stage_data *stage_data)
658 struct diff_filespec o, a, b;
660 o.mode = stage_data->stages[1].mode;
661 oidcpy(&o.oid, &stage_data->stages[1].oid);
663 a.mode = stage_data->stages[2].mode;
664 oidcpy(&a.oid, &stage_data->stages[2].oid);
666 b.mode = stage_data->stages[3].mode;
667 oidcpy(&b.oid, &stage_data->stages[3].oid);
669 return update_stages(opt, path,
670 is_null_oid(&o.oid) ? NULL : &o,
671 is_null_oid(&a.oid) ? NULL : &a,
672 is_null_oid(&b.oid) ? NULL : &b);
675 static void update_entry(struct stage_data *entry,
676 struct diff_filespec *o,
677 struct diff_filespec *a,
678 struct diff_filespec *b)
680 entry->processed = 0;
681 entry->stages[1].mode = o->mode;
682 entry->stages[2].mode = a->mode;
683 entry->stages[3].mode = b->mode;
684 oidcpy(&entry->stages[1].oid, &o->oid);
685 oidcpy(&entry->stages[2].oid, &a->oid);
686 oidcpy(&entry->stages[3].oid, &b->oid);
689 static int remove_file(struct merge_options *o, int clean,
690 const char *path, int no_wd)
692 int update_cache = o->call_depth || clean;
693 int update_working_directory = !o->call_depth && !no_wd;
695 if (update_cache) {
696 if (remove_file_from_cache(path))
697 return -1;
699 if (update_working_directory) {
700 if (ignore_case) {
701 struct cache_entry *ce;
702 ce = cache_file_exists(path, strlen(path), ignore_case);
703 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
704 return 0;
706 if (remove_path(path))
707 return -1;
709 return 0;
712 /* add a string to a strbuf, but converting "/" to "_" */
713 static void add_flattened_path(struct strbuf *out, const char *s)
715 size_t i = out->len;
716 strbuf_addstr(out, s);
717 for (; i < out->len; i++)
718 if (out->buf[i] == '/')
719 out->buf[i] = '_';
722 static char *unique_path(struct merge_options *o, const char *path, const char *branch)
724 struct path_hashmap_entry *entry;
725 struct strbuf newpath = STRBUF_INIT;
726 int suffix = 0;
727 size_t base_len;
729 strbuf_addf(&newpath, "%s~", path);
730 add_flattened_path(&newpath, branch);
732 base_len = newpath.len;
733 while (hashmap_get_from_hash(&o->current_file_dir_set,
734 path_hash(newpath.buf), newpath.buf) ||
735 (!o->call_depth && file_exists(newpath.buf))) {
736 strbuf_setlen(&newpath, base_len);
737 strbuf_addf(&newpath, "_%d", suffix++);
740 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
741 hashmap_entry_init(entry, path_hash(entry->path));
742 hashmap_add(&o->current_file_dir_set, entry);
743 return strbuf_detach(&newpath, NULL);
747 * Check whether a directory in the index is in the way of an incoming
748 * file. Return 1 if so. If check_working_copy is non-zero, also
749 * check the working directory. If empty_ok is non-zero, also return
750 * 0 in the case where the working-tree dir exists but is empty.
752 static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
754 int pos;
755 struct strbuf dirpath = STRBUF_INIT;
756 struct stat st;
758 strbuf_addstr(&dirpath, path);
759 strbuf_addch(&dirpath, '/');
761 pos = cache_name_pos(dirpath.buf, dirpath.len);
763 if (pos < 0)
764 pos = -1 - pos;
765 if (pos < active_nr &&
766 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
767 strbuf_release(&dirpath);
768 return 1;
771 strbuf_release(&dirpath);
772 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
773 !(empty_ok && is_empty_dir(path));
776 static int was_tracked(const char *path)
778 int pos = cache_name_pos(path, strlen(path));
780 if (0 <= pos)
781 /* we have been tracking this path */
782 return 1;
785 * Look for an unmerged entry for the path,
786 * specifically stage #2, which would indicate
787 * that "our" side before the merge started
788 * had the path tracked (and resulted in a conflict).
790 for (pos = -1 - pos;
791 pos < active_nr && !strcmp(path, active_cache[pos]->name);
792 pos++)
793 if (ce_stage(active_cache[pos]) == 2)
794 return 1;
795 return 0;
798 static int would_lose_untracked(const char *path)
800 return !was_tracked(path) && file_exists(path);
803 static int was_dirty(struct merge_options *o, const char *path)
805 struct cache_entry *ce;
806 int dirty = 1;
808 if (o->call_depth || !was_tracked(path))
809 return !dirty;
811 ce = cache_file_exists(path, strlen(path), ignore_case);
812 dirty = (ce->ce_stat_data.sd_mtime.sec > 0 &&
813 verify_uptodate(ce, &o->unpack_opts) != 0);
814 return dirty;
817 static int make_room_for_path(struct merge_options *o, const char *path)
819 int status, i;
820 const char *msg = _("failed to create path '%s'%s");
822 /* Unlink any D/F conflict files that are in the way */
823 for (i = 0; i < o->df_conflict_file_set.nr; i++) {
824 const char *df_path = o->df_conflict_file_set.items[i].string;
825 size_t pathlen = strlen(path);
826 size_t df_pathlen = strlen(df_path);
827 if (df_pathlen < pathlen &&
828 path[df_pathlen] == '/' &&
829 strncmp(path, df_path, df_pathlen) == 0) {
830 output(o, 3,
831 _("Removing %s to make room for subdirectory\n"),
832 df_path);
833 unlink(df_path);
834 unsorted_string_list_delete_item(&o->df_conflict_file_set,
835 i, 0);
836 break;
840 /* Make sure leading directories are created */
841 status = safe_create_leading_directories_const(path);
842 if (status) {
843 if (status == SCLD_EXISTS)
844 /* something else exists */
845 return err(o, msg, path, _(": perhaps a D/F conflict?"));
846 return err(o, msg, path, "");
850 * Do not unlink a file in the work tree if we are not
851 * tracking it.
853 if (would_lose_untracked(path))
854 return err(o, _("refusing to lose untracked file at '%s'"),
855 path);
857 /* Successful unlink is good.. */
858 if (!unlink(path))
859 return 0;
860 /* .. and so is no existing file */
861 if (errno == ENOENT)
862 return 0;
863 /* .. but not some other error (who really cares what?) */
864 return err(o, msg, path, _(": perhaps a D/F conflict?"));
867 static int update_file_flags(struct merge_options *o,
868 const struct object_id *oid,
869 unsigned mode,
870 const char *path,
871 int update_cache,
872 int update_wd)
874 int ret = 0;
876 if (o->call_depth)
877 update_wd = 0;
879 if (update_wd) {
880 enum object_type type;
881 void *buf;
882 unsigned long size;
884 if (S_ISGITLINK(mode)) {
886 * We may later decide to recursively descend into
887 * the submodule directory and update its index
888 * and/or work tree, but we do not do that now.
890 update_wd = 0;
891 goto update_index;
894 buf = read_object_file(oid, &type, &size);
895 if (!buf)
896 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
897 if (type != OBJ_BLOB) {
898 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
899 goto free_buf;
901 if (S_ISREG(mode)) {
902 struct strbuf strbuf = STRBUF_INIT;
903 if (convert_to_working_tree(path, buf, size, &strbuf)) {
904 free(buf);
905 size = strbuf.len;
906 buf = strbuf_detach(&strbuf, NULL);
910 if (make_room_for_path(o, path) < 0) {
911 update_wd = 0;
912 goto free_buf;
914 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
915 int fd;
916 if (mode & 0100)
917 mode = 0777;
918 else
919 mode = 0666;
920 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
921 if (fd < 0) {
922 ret = err(o, _("failed to open '%s': %s"),
923 path, strerror(errno));
924 goto free_buf;
926 write_in_full(fd, buf, size);
927 close(fd);
928 } else if (S_ISLNK(mode)) {
929 char *lnk = xmemdupz(buf, size);
930 safe_create_leading_directories_const(path);
931 unlink(path);
932 if (symlink(lnk, path))
933 ret = err(o, _("failed to symlink '%s': %s"),
934 path, strerror(errno));
935 free(lnk);
936 } else
937 ret = err(o,
938 _("do not know what to do with %06o %s '%s'"),
939 mode, oid_to_hex(oid), path);
940 free_buf:
941 free(buf);
943 update_index:
944 if (!ret && update_cache)
945 add_cacheinfo(o, mode, oid, path, 0, update_wd, ADD_CACHE_OK_TO_ADD);
946 return ret;
949 static int update_file(struct merge_options *o,
950 int clean,
951 const struct object_id *oid,
952 unsigned mode,
953 const char *path)
955 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
958 /* Low level file merging, update and removal */
960 struct merge_file_info {
961 struct object_id oid;
962 unsigned mode;
963 unsigned clean:1,
964 merge:1;
967 static int merge_3way(struct merge_options *o,
968 mmbuffer_t *result_buf,
969 const struct diff_filespec *one,
970 const struct diff_filespec *a,
971 const struct diff_filespec *b,
972 const char *branch1,
973 const char *branch2)
975 mmfile_t orig, src1, src2;
976 struct ll_merge_options ll_opts = {0};
977 char *base_name, *name1, *name2;
978 int merge_status;
980 ll_opts.renormalize = o->renormalize;
981 ll_opts.xdl_opts = o->xdl_opts;
983 if (o->call_depth) {
984 ll_opts.virtual_ancestor = 1;
985 ll_opts.variant = 0;
986 } else {
987 switch (o->recursive_variant) {
988 case MERGE_RECURSIVE_OURS:
989 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
990 break;
991 case MERGE_RECURSIVE_THEIRS:
992 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
993 break;
994 default:
995 ll_opts.variant = 0;
996 break;
1000 if (strcmp(a->path, b->path) ||
1001 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1002 base_name = o->ancestor == NULL ? NULL :
1003 mkpathdup("%s:%s", o->ancestor, one->path);
1004 name1 = mkpathdup("%s:%s", branch1, a->path);
1005 name2 = mkpathdup("%s:%s", branch2, b->path);
1006 } else {
1007 base_name = o->ancestor == NULL ? NULL :
1008 mkpathdup("%s", o->ancestor);
1009 name1 = mkpathdup("%s", branch1);
1010 name2 = mkpathdup("%s", branch2);
1013 read_mmblob(&orig, &one->oid);
1014 read_mmblob(&src1, &a->oid);
1015 read_mmblob(&src2, &b->oid);
1017 merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1018 &src1, name1, &src2, name2, &ll_opts);
1020 free(base_name);
1021 free(name1);
1022 free(name2);
1023 free(orig.ptr);
1024 free(src1.ptr);
1025 free(src2.ptr);
1026 return merge_status;
1029 static int merge_file_1(struct merge_options *o,
1030 const struct diff_filespec *one,
1031 const struct diff_filespec *a,
1032 const struct diff_filespec *b,
1033 const char *branch1,
1034 const char *branch2,
1035 struct merge_file_info *result)
1037 result->merge = 0;
1038 result->clean = 1;
1040 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1041 result->clean = 0;
1042 if (S_ISREG(a->mode)) {
1043 result->mode = a->mode;
1044 oidcpy(&result->oid, &a->oid);
1045 } else {
1046 result->mode = b->mode;
1047 oidcpy(&result->oid, &b->oid);
1049 } else {
1050 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1051 result->merge = 1;
1054 * Merge modes
1056 if (a->mode == b->mode || a->mode == one->mode)
1057 result->mode = b->mode;
1058 else {
1059 result->mode = a->mode;
1060 if (b->mode != one->mode) {
1061 result->clean = 0;
1062 result->merge = 1;
1066 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1067 oidcpy(&result->oid, &b->oid);
1068 else if (oid_eq(&b->oid, &one->oid))
1069 oidcpy(&result->oid, &a->oid);
1070 else if (S_ISREG(a->mode)) {
1071 mmbuffer_t result_buf;
1072 int ret = 0, merge_status;
1074 merge_status = merge_3way(o, &result_buf, one, a, b,
1075 branch1, branch2);
1077 if ((merge_status < 0) || !result_buf.ptr)
1078 ret = err(o, _("Failed to execute internal merge"));
1080 if (!ret &&
1081 write_object_file(result_buf.ptr, result_buf.size,
1082 blob_type, &result->oid))
1083 ret = err(o, _("Unable to add %s to database"),
1084 a->path);
1086 free(result_buf.ptr);
1087 if (ret)
1088 return ret;
1089 result->clean = (merge_status == 0);
1090 } else if (S_ISGITLINK(a->mode)) {
1091 result->clean = merge_submodule(&result->oid,
1092 one->path,
1093 &one->oid,
1094 &a->oid,
1095 &b->oid,
1096 !o->call_depth);
1097 } else if (S_ISLNK(a->mode)) {
1098 switch (o->recursive_variant) {
1099 case MERGE_RECURSIVE_NORMAL:
1100 oidcpy(&result->oid, &a->oid);
1101 if (!oid_eq(&a->oid, &b->oid))
1102 result->clean = 0;
1103 break;
1104 case MERGE_RECURSIVE_OURS:
1105 oidcpy(&result->oid, &a->oid);
1106 break;
1107 case MERGE_RECURSIVE_THEIRS:
1108 oidcpy(&result->oid, &b->oid);
1109 break;
1111 } else
1112 die("BUG: unsupported object type in the tree");
1115 return 0;
1118 static int merge_file_special_markers(struct merge_options *o,
1119 const struct diff_filespec *one,
1120 const struct diff_filespec *a,
1121 const struct diff_filespec *b,
1122 const char *branch1,
1123 const char *filename1,
1124 const char *branch2,
1125 const char *filename2,
1126 struct merge_file_info *mfi)
1128 char *side1 = NULL;
1129 char *side2 = NULL;
1130 int ret;
1132 if (filename1)
1133 side1 = xstrfmt("%s:%s", branch1, filename1);
1134 if (filename2)
1135 side2 = xstrfmt("%s:%s", branch2, filename2);
1137 ret = merge_file_1(o, one, a, b,
1138 side1 ? side1 : branch1,
1139 side2 ? side2 : branch2, mfi);
1140 free(side1);
1141 free(side2);
1142 return ret;
1145 static int merge_file_one(struct merge_options *o,
1146 const char *path,
1147 const struct object_id *o_oid, int o_mode,
1148 const struct object_id *a_oid, int a_mode,
1149 const struct object_id *b_oid, int b_mode,
1150 const char *branch1,
1151 const char *branch2,
1152 struct merge_file_info *mfi)
1154 struct diff_filespec one, a, b;
1156 one.path = a.path = b.path = (char *)path;
1157 oidcpy(&one.oid, o_oid);
1158 one.mode = o_mode;
1159 oidcpy(&a.oid, a_oid);
1160 a.mode = a_mode;
1161 oidcpy(&b.oid, b_oid);
1162 b.mode = b_mode;
1163 return merge_file_1(o, &one, &a, &b, branch1, branch2, mfi);
1166 static int conflict_rename_dir(struct merge_options *o,
1167 struct diff_filepair *pair,
1168 const char *rename_branch,
1169 const char *other_branch)
1171 const struct diff_filespec *dest = pair->two;
1173 if (!o->call_depth && would_lose_untracked(dest->path)) {
1174 char *alt_path = unique_path(o, dest->path, rename_branch);
1176 output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1177 "writing to %s instead."),
1178 dest->path, alt_path);
1180 * Write the file in worktree at alt_path, but not in the
1181 * index. Instead, write to dest->path for the index but
1182 * only at the higher appropriate stage.
1184 if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1185 return -1;
1186 free(alt_path);
1187 return update_stages(o, dest->path, NULL,
1188 rename_branch == o->branch1 ? dest : NULL,
1189 rename_branch == o->branch1 ? NULL : dest);
1192 /* Update dest->path both in index and in worktree */
1193 if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1194 return -1;
1195 return 0;
1198 static int handle_change_delete(struct merge_options *o,
1199 const char *path, const char *old_path,
1200 const struct object_id *o_oid, int o_mode,
1201 const struct object_id *changed_oid,
1202 int changed_mode,
1203 const char *change_branch,
1204 const char *delete_branch,
1205 const char *change, const char *change_past)
1207 char *alt_path = NULL;
1208 const char *update_path = path;
1209 int ret = 0;
1211 if (dir_in_way(path, !o->call_depth, 0) ||
1212 (!o->call_depth && would_lose_untracked(path))) {
1213 update_path = alt_path = unique_path(o, path, change_branch);
1216 if (o->call_depth) {
1218 * We cannot arbitrarily accept either a_sha or b_sha as
1219 * correct; since there is no true "middle point" between
1220 * them, simply reuse the base version for virtual merge base.
1222 ret = remove_file_from_cache(path);
1223 if (!ret)
1224 ret = update_file(o, 0, o_oid, o_mode, update_path);
1225 } else {
1226 if (!alt_path) {
1227 if (!old_path) {
1228 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1229 "and %s in %s. Version %s of %s left in tree."),
1230 change, path, delete_branch, change_past,
1231 change_branch, change_branch, path);
1232 } else {
1233 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1234 "and %s to %s in %s. Version %s of %s left in tree."),
1235 change, old_path, delete_branch, change_past, path,
1236 change_branch, change_branch, path);
1238 } else {
1239 if (!old_path) {
1240 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1241 "and %s in %s. Version %s of %s left in tree at %s."),
1242 change, path, delete_branch, change_past,
1243 change_branch, change_branch, path, alt_path);
1244 } else {
1245 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1246 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1247 change, old_path, delete_branch, change_past, path,
1248 change_branch, change_branch, path, alt_path);
1252 * No need to call update_file() on path when change_branch ==
1253 * o->branch1 && !alt_path, since that would needlessly touch
1254 * path. We could call update_file_flags() with update_cache=0
1255 * and update_wd=0, but that's a no-op.
1257 if (change_branch != o->branch1 || alt_path)
1258 ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1260 free(alt_path);
1262 return ret;
1265 static int conflict_rename_delete(struct merge_options *o,
1266 struct diff_filepair *pair,
1267 const char *rename_branch,
1268 const char *delete_branch)
1270 const struct diff_filespec *orig = pair->one;
1271 const struct diff_filespec *dest = pair->two;
1273 if (handle_change_delete(o,
1274 o->call_depth ? orig->path : dest->path,
1275 o->call_depth ? NULL : orig->path,
1276 &orig->oid, orig->mode,
1277 &dest->oid, dest->mode,
1278 rename_branch, delete_branch,
1279 _("rename"), _("renamed")))
1280 return -1;
1282 if (o->call_depth)
1283 return remove_file_from_cache(dest->path);
1284 else
1285 return update_stages(o, dest->path, NULL,
1286 rename_branch == o->branch1 ? dest : NULL,
1287 rename_branch == o->branch1 ? NULL : dest);
1290 static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1291 struct stage_data *entry,
1292 int stage)
1294 struct object_id *oid = &entry->stages[stage].oid;
1295 unsigned mode = entry->stages[stage].mode;
1296 if (mode == 0 || is_null_oid(oid))
1297 return NULL;
1298 oidcpy(&target->oid, oid);
1299 target->mode = mode;
1300 return target;
1303 static int handle_file(struct merge_options *o,
1304 struct diff_filespec *rename,
1305 int stage,
1306 struct rename_conflict_info *ci)
1308 char *dst_name = rename->path;
1309 struct stage_data *dst_entry;
1310 const char *cur_branch, *other_branch;
1311 struct diff_filespec other;
1312 struct diff_filespec *add;
1313 int ret;
1315 if (stage == 2) {
1316 dst_entry = ci->dst_entry1;
1317 cur_branch = ci->branch1;
1318 other_branch = ci->branch2;
1319 } else {
1320 dst_entry = ci->dst_entry2;
1321 cur_branch = ci->branch2;
1322 other_branch = ci->branch1;
1325 add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1326 if (add) {
1327 int ren_src_was_dirty = was_dirty(o, rename->path);
1328 char *add_name = unique_path(o, rename->path, other_branch);
1329 if (update_file(o, 0, &add->oid, add->mode, add_name))
1330 return -1;
1332 if (ren_src_was_dirty) {
1333 output(o, 1, _("Refusing to lose dirty file at %s"),
1334 rename->path);
1337 * Because the double negatives somehow keep confusing me...
1338 * 1) update_wd iff !ren_src_was_dirty.
1339 * 2) no_wd iff !update_wd
1340 * 3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty
1342 remove_file(o, 0, rename->path, ren_src_was_dirty);
1343 dst_name = unique_path(o, rename->path, cur_branch);
1344 } else {
1345 if (dir_in_way(rename->path, !o->call_depth, 0)) {
1346 dst_name = unique_path(o, rename->path, cur_branch);
1347 output(o, 1, _("%s is a directory in %s adding as %s instead"),
1348 rename->path, other_branch, dst_name);
1349 } else if (!o->call_depth &&
1350 would_lose_untracked(rename->path)) {
1351 dst_name = unique_path(o, rename->path, cur_branch);
1352 output(o, 1, _("Refusing to lose untracked file at %s; "
1353 "adding as %s instead"),
1354 rename->path, dst_name);
1357 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1358 ; /* fall through, do allow dst_name to be released */
1359 else if (stage == 2)
1360 ret = update_stages(o, rename->path, NULL, rename, add);
1361 else
1362 ret = update_stages(o, rename->path, NULL, add, rename);
1364 if (dst_name != rename->path)
1365 free(dst_name);
1367 return ret;
1370 static int conflict_rename_rename_1to2(struct merge_options *o,
1371 struct rename_conflict_info *ci)
1373 /* One file was renamed in both branches, but to different names. */
1374 struct diff_filespec *one = ci->pair1->one;
1375 struct diff_filespec *a = ci->pair1->two;
1376 struct diff_filespec *b = ci->pair2->two;
1378 output(o, 1, _("CONFLICT (rename/rename): "
1379 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1380 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1381 one->path, a->path, ci->branch1,
1382 one->path, b->path, ci->branch2,
1383 o->call_depth ? _(" (left unresolved)") : "");
1384 if (o->call_depth) {
1385 struct merge_file_info mfi;
1386 struct diff_filespec other;
1387 struct diff_filespec *add;
1388 if (merge_file_one(o, one->path,
1389 &one->oid, one->mode,
1390 &a->oid, a->mode,
1391 &b->oid, b->mode,
1392 ci->branch1, ci->branch2, &mfi))
1393 return -1;
1396 * FIXME: For rename/add-source conflicts (if we could detect
1397 * such), this is wrong. We should instead find a unique
1398 * pathname and then either rename the add-source file to that
1399 * unique path, or use that unique path instead of src here.
1401 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1402 return -1;
1405 * Above, we put the merged content at the merge-base's
1406 * path. Now we usually need to delete both a->path and
1407 * b->path. However, the rename on each side of the merge
1408 * could also be involved in a rename/add conflict. In
1409 * such cases, we should keep the added file around,
1410 * resolving the conflict at that path in its favor.
1412 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1413 if (add) {
1414 if (update_file(o, 0, &add->oid, add->mode, a->path))
1415 return -1;
1417 else
1418 remove_file_from_cache(a->path);
1419 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1420 if (add) {
1421 if (update_file(o, 0, &add->oid, add->mode, b->path))
1422 return -1;
1424 else
1425 remove_file_from_cache(b->path);
1426 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1427 return -1;
1429 return 0;
1432 static int conflict_rename_rename_2to1(struct merge_options *o,
1433 struct rename_conflict_info *ci)
1435 /* Two files, a & b, were renamed to the same thing, c. */
1436 struct diff_filespec *a = ci->pair1->one;
1437 struct diff_filespec *b = ci->pair2->one;
1438 struct diff_filespec *c1 = ci->pair1->two;
1439 struct diff_filespec *c2 = ci->pair2->two;
1440 char *path = c1->path; /* == c2->path */
1441 struct merge_file_info mfi_c1;
1442 struct merge_file_info mfi_c2;
1443 int ret;
1445 output(o, 1, _("CONFLICT (rename/rename): "
1446 "Rename %s->%s in %s. "
1447 "Rename %s->%s in %s"),
1448 a->path, c1->path, ci->branch1,
1449 b->path, c2->path, ci->branch2);
1451 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1452 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1454 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1455 o->branch1, c1->path,
1456 o->branch2, ci->ren1_other.path, &mfi_c1) ||
1457 merge_file_special_markers(o, b, &ci->ren2_other, c2,
1458 o->branch1, ci->ren2_other.path,
1459 o->branch2, c2->path, &mfi_c2))
1460 return -1;
1462 if (o->call_depth) {
1464 * If mfi_c1.clean && mfi_c2.clean, then it might make
1465 * sense to do a two-way merge of those results. But, I
1466 * think in all cases, it makes sense to have the virtual
1467 * merge base just undo the renames; they can be detected
1468 * again later for the non-recursive merge.
1470 remove_file(o, 0, path, 0);
1471 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1472 if (!ret)
1473 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1474 b->path);
1475 } else {
1476 char *new_path1 = unique_path(o, path, ci->branch1);
1477 char *new_path2 = unique_path(o, path, ci->branch2);
1478 output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1479 a->path, new_path1, b->path, new_path2);
1480 if (was_dirty(o, path))
1481 output(o, 1, _("Refusing to lose dirty file at %s"),
1482 path);
1483 else if (would_lose_untracked(path))
1485 * Only way we get here is if both renames were from
1486 * a directory rename AND user had an untracked file
1487 * at the location where both files end up after the
1488 * two directory renames. See testcase 10d of t6043.
1490 output(o, 1, _("Refusing to lose untracked file at "
1491 "%s, even though it's in the way."),
1492 path);
1493 else
1494 remove_file(o, 0, path, 0);
1495 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1496 if (!ret)
1497 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1498 new_path2);
1500 * unpack_trees() actually populates the index for us for
1501 * "normal" rename/rename(2to1) situtations so that the
1502 * correct entries are at the higher stages, which would
1503 * make the call below to update_stages_for_stage_data
1504 * unnecessary. However, if either of the renames came
1505 * from a directory rename, then unpack_trees() will not
1506 * have gotten the right data loaded into the index, so we
1507 * need to do so now. (While it'd be tempting to move this
1508 * call to update_stages_for_stage_data() to
1509 * apply_directory_rename_modifications(), that would break
1510 * our intermediate calls to would_lose_untracked() since
1511 * those rely on the current in-memory index. See also the
1512 * big "NOTE" in update_stages()).
1514 if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1515 ret = -1;
1517 free(new_path2);
1518 free(new_path1);
1521 return ret;
1525 * Get the diff_filepairs changed between o_tree and tree.
1527 static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1528 struct tree *o_tree,
1529 struct tree *tree)
1531 struct diff_queue_struct *ret;
1532 struct diff_options opts;
1534 diff_setup(&opts);
1535 opts.flags.recursive = 1;
1536 opts.flags.rename_empty = 0;
1537 opts.detect_rename = DIFF_DETECT_RENAME;
1538 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1539 o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1540 1000;
1541 opts.rename_score = o->rename_score;
1542 opts.show_rename_progress = o->show_rename_progress;
1543 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1544 diff_setup_done(&opts);
1545 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1546 diffcore_std(&opts);
1547 if (opts.needed_rename_limit > o->needed_rename_limit)
1548 o->needed_rename_limit = opts.needed_rename_limit;
1550 ret = xmalloc(sizeof(*ret));
1551 *ret = diff_queued_diff;
1553 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1554 diff_queued_diff.nr = 0;
1555 diff_queued_diff.queue = NULL;
1556 diff_flush(&opts);
1557 return ret;
1560 static int tree_has_path(struct tree *tree, const char *path)
1562 struct object_id hashy;
1563 unsigned int mode_o;
1565 return !get_tree_entry(&tree->object.oid, path,
1566 &hashy, &mode_o);
1570 * Return a new string that replaces the beginning portion (which matches
1571 * entry->dir), with entry->new_dir. In perl-speak:
1572 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1573 * NOTE:
1574 * Caller must ensure that old_path starts with entry->dir + '/'.
1576 static char *apply_dir_rename(struct dir_rename_entry *entry,
1577 const char *old_path)
1579 struct strbuf new_path = STRBUF_INIT;
1580 int oldlen, newlen;
1582 if (entry->non_unique_new_dir)
1583 return NULL;
1585 oldlen = strlen(entry->dir);
1586 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1587 strbuf_grow(&new_path, newlen);
1588 strbuf_addbuf(&new_path, &entry->new_dir);
1589 strbuf_addstr(&new_path, &old_path[oldlen]);
1591 return strbuf_detach(&new_path, NULL);
1594 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1595 char **old_dir, char **new_dir)
1597 char *end_of_old, *end_of_new;
1598 int old_len, new_len;
1600 *old_dir = NULL;
1601 *new_dir = NULL;
1604 * For
1605 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1606 * the "e/foo.c" part is the same, we just want to know that
1607 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1608 * so, for this example, this function returns "a/b/c/d" in
1609 * *old_dir and "a/b/some/thing/else" in *new_dir.
1611 * Also, if the basename of the file changed, we don't care. We
1612 * want to know which portion of the directory, if any, changed.
1614 end_of_old = strrchr(old_path, '/');
1615 end_of_new = strrchr(new_path, '/');
1617 if (end_of_old == NULL || end_of_new == NULL)
1618 return;
1619 while (*--end_of_new == *--end_of_old &&
1620 end_of_old != old_path &&
1621 end_of_new != new_path)
1622 ; /* Do nothing; all in the while loop */
1624 * We've found the first non-matching character in the directory
1625 * paths. That means the current directory we were comparing
1626 * represents the rename. Move end_of_old and end_of_new back
1627 * to the full directory name.
1629 if (*end_of_old == '/')
1630 end_of_old++;
1631 if (*end_of_old != '/')
1632 end_of_new++;
1633 end_of_old = strchr(end_of_old, '/');
1634 end_of_new = strchr(end_of_new, '/');
1637 * It may have been the case that old_path and new_path were the same
1638 * directory all along. Don't claim a rename if they're the same.
1640 old_len = end_of_old - old_path;
1641 new_len = end_of_new - new_path;
1643 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1644 *old_dir = xstrndup(old_path, old_len);
1645 *new_dir = xstrndup(new_path, new_len);
1649 static void remove_hashmap_entries(struct hashmap *dir_renames,
1650 struct string_list *items_to_remove)
1652 int i;
1653 struct dir_rename_entry *entry;
1655 for (i = 0; i < items_to_remove->nr; i++) {
1656 entry = items_to_remove->items[i].util;
1657 hashmap_remove(dir_renames, entry, NULL);
1659 string_list_clear(items_to_remove, 0);
1663 * See if there is a directory rename for path, and if there are any file
1664 * level conflicts for the renamed location. If there is a rename and
1665 * there are no conflicts, return the new name. Otherwise, return NULL.
1667 static char *handle_path_level_conflicts(struct merge_options *o,
1668 const char *path,
1669 struct dir_rename_entry *entry,
1670 struct hashmap *collisions,
1671 struct tree *tree)
1673 char *new_path = NULL;
1674 struct collision_entry *collision_ent;
1675 int clean = 1;
1676 struct strbuf collision_paths = STRBUF_INIT;
1679 * entry has the mapping of old directory name to new directory name
1680 * that we want to apply to path.
1682 new_path = apply_dir_rename(entry, path);
1684 if (!new_path) {
1685 /* This should only happen when entry->non_unique_new_dir set */
1686 if (!entry->non_unique_new_dir)
1687 BUG("entry->non_unqiue_dir not set and !new_path");
1688 output(o, 1, _("CONFLICT (directory rename split): "
1689 "Unclear where to place %s because directory "
1690 "%s was renamed to multiple other directories, "
1691 "with no destination getting a majority of the "
1692 "files."),
1693 path, entry->dir);
1694 clean = 0;
1695 return NULL;
1699 * The caller needs to have ensured that it has pre-populated
1700 * collisions with all paths that map to new_path. Do a quick check
1701 * to ensure that's the case.
1703 collision_ent = collision_find_entry(collisions, new_path);
1704 if (collision_ent == NULL)
1705 BUG("collision_ent is NULL");
1708 * Check for one-sided add/add/.../add conflicts, i.e.
1709 * where implicit renames from the other side doing
1710 * directory rename(s) can affect this side of history
1711 * to put multiple paths into the same location. Warn
1712 * and bail on directory renames for such paths.
1714 if (collision_ent->reported_already) {
1715 clean = 0;
1716 } else if (tree_has_path(tree, new_path)) {
1717 collision_ent->reported_already = 1;
1718 strbuf_add_separated_string_list(&collision_paths, ", ",
1719 &collision_ent->source_files);
1720 output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1721 "file/dir at %s in the way of implicit "
1722 "directory rename(s) putting the following "
1723 "path(s) there: %s."),
1724 new_path, collision_paths.buf);
1725 clean = 0;
1726 } else if (collision_ent->source_files.nr > 1) {
1727 collision_ent->reported_already = 1;
1728 strbuf_add_separated_string_list(&collision_paths, ", ",
1729 &collision_ent->source_files);
1730 output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1731 "more than one path to %s; implicit directory "
1732 "renames tried to put these paths there: %s"),
1733 new_path, collision_paths.buf);
1734 clean = 0;
1737 /* Free memory we no longer need */
1738 strbuf_release(&collision_paths);
1739 if (!clean && new_path) {
1740 free(new_path);
1741 return NULL;
1744 return new_path;
1748 * There are a couple things we want to do at the directory level:
1749 * 1. Check for both sides renaming to the same thing, in order to avoid
1750 * implicit renaming of files that should be left in place. (See
1751 * testcase 6b in t6043 for details.)
1752 * 2. Prune directory renames if there are still files left in the
1753 * the original directory. These represent a partial directory rename,
1754 * i.e. a rename where only some of the files within the directory
1755 * were renamed elsewhere. (Technically, this could be done earlier
1756 * in get_directory_renames(), except that would prevent us from
1757 * doing the previous check and thus failing testcase 6b.)
1758 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
1759 * In the future, we could potentially record this info as well and
1760 * omit reporting rename/rename(1to2) conflicts for each path within
1761 * the affected directories, thus cleaning up the merge output.
1762 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1763 * directory level, because merging directories is fine. If it
1764 * causes conflicts for files within those merged directories, then
1765 * that should be detected at the individual path level.
1767 static void handle_directory_level_conflicts(struct merge_options *o,
1768 struct hashmap *dir_re_head,
1769 struct tree *head,
1770 struct hashmap *dir_re_merge,
1771 struct tree *merge)
1773 struct hashmap_iter iter;
1774 struct dir_rename_entry *head_ent;
1775 struct dir_rename_entry *merge_ent;
1777 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1778 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1780 hashmap_iter_init(dir_re_head, &iter);
1781 while ((head_ent = hashmap_iter_next(&iter))) {
1782 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1783 if (merge_ent &&
1784 !head_ent->non_unique_new_dir &&
1785 !merge_ent->non_unique_new_dir &&
1786 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1787 /* 1. Renamed identically; remove it from both sides */
1788 string_list_append(&remove_from_head,
1789 head_ent->dir)->util = head_ent;
1790 strbuf_release(&head_ent->new_dir);
1791 string_list_append(&remove_from_merge,
1792 merge_ent->dir)->util = merge_ent;
1793 strbuf_release(&merge_ent->new_dir);
1794 } else if (tree_has_path(head, head_ent->dir)) {
1795 /* 2. This wasn't a directory rename after all */
1796 string_list_append(&remove_from_head,
1797 head_ent->dir)->util = head_ent;
1798 strbuf_release(&head_ent->new_dir);
1802 remove_hashmap_entries(dir_re_head, &remove_from_head);
1803 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1805 hashmap_iter_init(dir_re_merge, &iter);
1806 while ((merge_ent = hashmap_iter_next(&iter))) {
1807 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1808 if (tree_has_path(merge, merge_ent->dir)) {
1809 /* 2. This wasn't a directory rename after all */
1810 string_list_append(&remove_from_merge,
1811 merge_ent->dir)->util = merge_ent;
1812 } else if (head_ent &&
1813 !head_ent->non_unique_new_dir &&
1814 !merge_ent->non_unique_new_dir) {
1815 /* 3. rename/rename(1to2) */
1817 * We can assume it's not rename/rename(1to1) because
1818 * that was case (1), already checked above. So we
1819 * know that head_ent->new_dir and merge_ent->new_dir
1820 * are different strings.
1822 output(o, 1, _("CONFLICT (rename/rename): "
1823 "Rename directory %s->%s in %s. "
1824 "Rename directory %s->%s in %s"),
1825 head_ent->dir, head_ent->new_dir.buf, o->branch1,
1826 head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1827 string_list_append(&remove_from_head,
1828 head_ent->dir)->util = head_ent;
1829 strbuf_release(&head_ent->new_dir);
1830 string_list_append(&remove_from_merge,
1831 merge_ent->dir)->util = merge_ent;
1832 strbuf_release(&merge_ent->new_dir);
1836 remove_hashmap_entries(dir_re_head, &remove_from_head);
1837 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1840 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1841 struct tree *tree)
1843 struct hashmap *dir_renames;
1844 struct hashmap_iter iter;
1845 struct dir_rename_entry *entry;
1846 int i;
1849 * Typically, we think of a directory rename as all files from a
1850 * certain directory being moved to a target directory. However,
1851 * what if someone first moved two files from the original
1852 * directory in one commit, and then renamed the directory
1853 * somewhere else in a later commit? At merge time, we just know
1854 * that files from the original directory went to two different
1855 * places, and that the bulk of them ended up in the same place.
1856 * We want each directory rename to represent where the bulk of the
1857 * files from that directory end up; this function exists to find
1858 * where the bulk of the files went.
1860 * The first loop below simply iterates through the list of file
1861 * renames, finding out how often each directory rename pair
1862 * possibility occurs.
1864 dir_renames = xmalloc(sizeof(struct hashmap));
1865 dir_rename_init(dir_renames);
1866 for (i = 0; i < pairs->nr; ++i) {
1867 struct string_list_item *item;
1868 int *count;
1869 struct diff_filepair *pair = pairs->queue[i];
1870 char *old_dir, *new_dir;
1872 /* File not part of directory rename if it wasn't renamed */
1873 if (pair->status != 'R')
1874 continue;
1876 get_renamed_dir_portion(pair->one->path, pair->two->path,
1877 &old_dir, &new_dir);
1878 if (!old_dir)
1879 /* Directory didn't change at all; ignore this one. */
1880 continue;
1882 entry = dir_rename_find_entry(dir_renames, old_dir);
1883 if (!entry) {
1884 entry = xmalloc(sizeof(struct dir_rename_entry));
1885 dir_rename_entry_init(entry, old_dir);
1886 hashmap_put(dir_renames, entry);
1887 } else {
1888 free(old_dir);
1890 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1891 if (!item) {
1892 item = string_list_insert(&entry->possible_new_dirs,
1893 new_dir);
1894 item->util = xcalloc(1, sizeof(int));
1895 } else {
1896 free(new_dir);
1898 count = item->util;
1899 *count += 1;
1903 * For each directory with files moved out of it, we find out which
1904 * target directory received the most files so we can declare it to
1905 * be the "winning" target location for the directory rename. This
1906 * winner gets recorded in new_dir. If there is no winner
1907 * (multiple target directories received the same number of files),
1908 * we set non_unique_new_dir. Once we've determined the winner (or
1909 * that there is no winner), we no longer need possible_new_dirs.
1911 hashmap_iter_init(dir_renames, &iter);
1912 while ((entry = hashmap_iter_next(&iter))) {
1913 int max = 0;
1914 int bad_max = 0;
1915 char *best = NULL;
1917 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1918 int *count = entry->possible_new_dirs.items[i].util;
1920 if (*count == max)
1921 bad_max = max;
1922 else if (*count > max) {
1923 max = *count;
1924 best = entry->possible_new_dirs.items[i].string;
1927 if (bad_max == max)
1928 entry->non_unique_new_dir = 1;
1929 else {
1930 assert(entry->new_dir.len == 0);
1931 strbuf_addstr(&entry->new_dir, best);
1934 * The relevant directory sub-portion of the original full
1935 * filepaths were xstrndup'ed before inserting into
1936 * possible_new_dirs, and instead of manually iterating the
1937 * list and free'ing each, just lie and tell
1938 * possible_new_dirs that it did the strdup'ing so that it
1939 * will free them for us.
1941 entry->possible_new_dirs.strdup_strings = 1;
1942 string_list_clear(&entry->possible_new_dirs, 1);
1945 return dir_renames;
1948 static struct dir_rename_entry *check_dir_renamed(const char *path,
1949 struct hashmap *dir_renames)
1951 char temp[PATH_MAX];
1952 char *end;
1953 struct dir_rename_entry *entry;
1955 strcpy(temp, path);
1956 while ((end = strrchr(temp, '/'))) {
1957 *end = '\0';
1958 entry = dir_rename_find_entry(dir_renames, temp);
1959 if (entry)
1960 return entry;
1962 return NULL;
1965 static void compute_collisions(struct hashmap *collisions,
1966 struct hashmap *dir_renames,
1967 struct diff_queue_struct *pairs)
1969 int i;
1972 * Multiple files can be mapped to the same path due to directory
1973 * renames done by the other side of history. Since that other
1974 * side of history could have merged multiple directories into one,
1975 * if our side of history added the same file basename to each of
1976 * those directories, then all N of them would get implicitly
1977 * renamed by the directory rename detection into the same path,
1978 * and we'd get an add/add/.../add conflict, and all those adds
1979 * from *this* side of history. This is not representable in the
1980 * index, and users aren't going to easily be able to make sense of
1981 * it. So we need to provide a good warning about what's
1982 * happening, and fall back to no-directory-rename detection
1983 * behavior for those paths.
1985 * See testcases 9e and all of section 5 from t6043 for examples.
1987 collision_init(collisions);
1989 for (i = 0; i < pairs->nr; ++i) {
1990 struct dir_rename_entry *dir_rename_ent;
1991 struct collision_entry *collision_ent;
1992 char *new_path;
1993 struct diff_filepair *pair = pairs->queue[i];
1995 if (pair->status != 'A' && pair->status != 'R')
1996 continue;
1997 dir_rename_ent = check_dir_renamed(pair->two->path,
1998 dir_renames);
1999 if (!dir_rename_ent)
2000 continue;
2002 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2003 if (!new_path)
2005 * dir_rename_ent->non_unique_new_path is true, which
2006 * means there is no directory rename for us to use,
2007 * which means it won't cause us any additional
2008 * collisions.
2010 continue;
2011 collision_ent = collision_find_entry(collisions, new_path);
2012 if (!collision_ent) {
2013 collision_ent = xcalloc(1,
2014 sizeof(struct collision_entry));
2015 hashmap_entry_init(collision_ent, strhash(new_path));
2016 hashmap_put(collisions, collision_ent);
2017 collision_ent->target_file = new_path;
2018 } else {
2019 free(new_path);
2021 string_list_insert(&collision_ent->source_files,
2022 pair->two->path);
2026 static char *check_for_directory_rename(struct merge_options *o,
2027 const char *path,
2028 struct tree *tree,
2029 struct hashmap *dir_renames,
2030 struct hashmap *dir_rename_exclusions,
2031 struct hashmap *collisions,
2032 int *clean_merge)
2034 char *new_path = NULL;
2035 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2036 struct dir_rename_entry *oentry = NULL;
2038 if (!entry)
2039 return new_path;
2042 * This next part is a little weird. We do not want to do an
2043 * implicit rename into a directory we renamed on our side, because
2044 * that will result in a spurious rename/rename(1to2) conflict. An
2045 * example:
2046 * Base commit: dumbdir/afile, otherdir/bfile
2047 * Side 1: smrtdir/afile, otherdir/bfile
2048 * Side 2: dumbdir/afile, dumbdir/bfile
2049 * Here, while working on Side 1, we could notice that otherdir was
2050 * renamed/merged to dumbdir, and change the diff_filepair for
2051 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2052 * 2 will notice the rename from dumbdir to smrtdir, and do the
2053 * transitive rename to move it from dumbdir/bfile to
2054 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2055 * smrtdir, a rename/rename(1to2) conflict. We really just want
2056 * the file to end up in smrtdir. And the way to achieve that is
2057 * to not let Side1 do the rename to dumbdir, since we know that is
2058 * the source of one of our directory renames.
2060 * That's why oentry and dir_rename_exclusions is here.
2062 * As it turns out, this also prevents N-way transient rename
2063 * confusion; See testcases 9c and 9d of t6043.
2065 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2066 if (oentry) {
2067 output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2068 "to %s, because %s itself was renamed."),
2069 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2070 } else {
2071 new_path = handle_path_level_conflicts(o, path, entry,
2072 collisions, tree);
2073 *clean_merge &= (new_path != NULL);
2076 return new_path;
2079 static void apply_directory_rename_modifications(struct merge_options *o,
2080 struct diff_filepair *pair,
2081 char *new_path,
2082 struct rename *re,
2083 struct tree *tree,
2084 struct tree *o_tree,
2085 struct tree *a_tree,
2086 struct tree *b_tree,
2087 struct string_list *entries,
2088 int *clean)
2090 struct string_list_item *item;
2091 int stage = (tree == a_tree ? 2 : 3);
2092 int update_wd;
2095 * In all cases where we can do directory rename detection,
2096 * unpack_trees() will have read pair->two->path into the
2097 * index and the working copy. We need to remove it so that
2098 * we can instead place it at new_path. It is guaranteed to
2099 * not be untracked (unpack_trees() would have errored out
2100 * saying the file would have been overwritten), but it might
2101 * be dirty, though.
2103 update_wd = !was_dirty(o, pair->two->path);
2104 if (!update_wd)
2105 output(o, 1, _("Refusing to lose dirty file at %s"),
2106 pair->two->path);
2107 remove_file(o, 1, pair->two->path, !update_wd);
2109 /* Find or create a new re->dst_entry */
2110 item = string_list_lookup(entries, new_path);
2111 if (item) {
2113 * Since we're renaming on this side of history, and it's
2114 * due to a directory rename on the other side of history
2115 * (which we only allow when the directory in question no
2116 * longer exists on the other side of history), the
2117 * original entry for re->dst_entry is no longer
2118 * necessary...
2120 re->dst_entry->processed = 1;
2123 * ...because we'll be using this new one.
2125 re->dst_entry = item->util;
2126 } else {
2128 * re->dst_entry is for the before-dir-rename path, and we
2129 * need it to hold information for the after-dir-rename
2130 * path. Before creating a new entry, we need to mark the
2131 * old one as unnecessary (...unless it is shared by
2132 * src_entry, i.e. this didn't use to be a rename, in which
2133 * case we can just allow the normal processing to happen
2134 * for it).
2136 if (pair->status == 'R')
2137 re->dst_entry->processed = 1;
2139 re->dst_entry = insert_stage_data(new_path,
2140 o_tree, a_tree, b_tree,
2141 entries);
2142 item = string_list_insert(entries, new_path);
2143 item->util = re->dst_entry;
2147 * Update the stage_data with the information about the path we are
2148 * moving into place. That slot will be empty and available for us
2149 * to write to because of the collision checks in
2150 * handle_path_level_conflicts(). In other words,
2151 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2152 * open for us to write to.
2154 * It may be tempting to actually update the index at this point as
2155 * well, using update_stages_for_stage_data(), but as per the big
2156 * "NOTE" in update_stages(), doing so will modify the current
2157 * in-memory index which will break calls to would_lose_untracked()
2158 * that we need to make. Instead, we need to just make sure that
2159 * the various conflict_rename_*() functions update the index
2160 * explicitly rather than relying on unpack_trees() to have done it.
2162 get_tree_entry(&tree->object.oid,
2163 pair->two->path,
2164 &re->dst_entry->stages[stage].oid,
2165 &re->dst_entry->stages[stage].mode);
2167 /* Update pair status */
2168 if (pair->status == 'A') {
2170 * Recording rename information for this add makes it look
2171 * like a rename/delete conflict. Make sure we can
2172 * correctly handle this as an add that was moved to a new
2173 * directory instead of reporting a rename/delete conflict.
2175 re->add_turned_into_rename = 1;
2178 * We don't actually look at pair->status again, but it seems
2179 * pedagogically correct to adjust it.
2181 pair->status = 'R';
2184 * Finally, record the new location.
2186 pair->two->path = new_path;
2190 * Get information of all renames which occurred in 'pairs', making use of
2191 * any implicit directory renames inferred from the other side of history.
2192 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2193 * to be able to associate the correct cache entries with the rename
2194 * information; tree is always equal to either a_tree or b_tree.
2196 static struct string_list *get_renames(struct merge_options *o,
2197 struct diff_queue_struct *pairs,
2198 struct hashmap *dir_renames,
2199 struct hashmap *dir_rename_exclusions,
2200 struct tree *tree,
2201 struct tree *o_tree,
2202 struct tree *a_tree,
2203 struct tree *b_tree,
2204 struct string_list *entries,
2205 int *clean_merge)
2207 int i;
2208 struct hashmap collisions;
2209 struct hashmap_iter iter;
2210 struct collision_entry *e;
2211 struct string_list *renames;
2213 compute_collisions(&collisions, dir_renames, pairs);
2214 renames = xcalloc(1, sizeof(struct string_list));
2216 for (i = 0; i < pairs->nr; ++i) {
2217 struct string_list_item *item;
2218 struct rename *re;
2219 struct diff_filepair *pair = pairs->queue[i];
2220 char *new_path; /* non-NULL only with directory renames */
2222 if (pair->status != 'A' && pair->status != 'R') {
2223 diff_free_filepair(pair);
2224 continue;
2226 new_path = check_for_directory_rename(o, pair->two->path, tree,
2227 dir_renames,
2228 dir_rename_exclusions,
2229 &collisions,
2230 clean_merge);
2231 if (pair->status != 'R' && !new_path) {
2232 diff_free_filepair(pair);
2233 continue;
2236 re = xmalloc(sizeof(*re));
2237 re->processed = 0;
2238 re->add_turned_into_rename = 0;
2239 re->pair = pair;
2240 item = string_list_lookup(entries, re->pair->one->path);
2241 if (!item)
2242 re->src_entry = insert_stage_data(re->pair->one->path,
2243 o_tree, a_tree, b_tree, entries);
2244 else
2245 re->src_entry = item->util;
2247 item = string_list_lookup(entries, re->pair->two->path);
2248 if (!item)
2249 re->dst_entry = insert_stage_data(re->pair->two->path,
2250 o_tree, a_tree, b_tree, entries);
2251 else
2252 re->dst_entry = item->util;
2253 item = string_list_insert(renames, pair->one->path);
2254 item->util = re;
2255 if (new_path)
2256 apply_directory_rename_modifications(o, pair, new_path,
2257 re, tree, o_tree,
2258 a_tree, b_tree,
2259 entries,
2260 clean_merge);
2263 hashmap_iter_init(&collisions, &iter);
2264 while ((e = hashmap_iter_next(&iter))) {
2265 free(e->target_file);
2266 string_list_clear(&e->source_files, 0);
2268 hashmap_free(&collisions, 1);
2269 return renames;
2272 static int process_renames(struct merge_options *o,
2273 struct string_list *a_renames,
2274 struct string_list *b_renames)
2276 int clean_merge = 1, i, j;
2277 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2278 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2279 const struct rename *sre;
2281 for (i = 0; i < a_renames->nr; i++) {
2282 sre = a_renames->items[i].util;
2283 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2284 = (void *)sre;
2286 for (i = 0; i < b_renames->nr; i++) {
2287 sre = b_renames->items[i].util;
2288 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2289 = (void *)sre;
2292 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2293 struct string_list *renames1, *renames2Dst;
2294 struct rename *ren1 = NULL, *ren2 = NULL;
2295 const char *branch1, *branch2;
2296 const char *ren1_src, *ren1_dst;
2297 struct string_list_item *lookup;
2299 if (i >= a_renames->nr) {
2300 ren2 = b_renames->items[j++].util;
2301 } else if (j >= b_renames->nr) {
2302 ren1 = a_renames->items[i++].util;
2303 } else {
2304 int compare = strcmp(a_renames->items[i].string,
2305 b_renames->items[j].string);
2306 if (compare <= 0)
2307 ren1 = a_renames->items[i++].util;
2308 if (compare >= 0)
2309 ren2 = b_renames->items[j++].util;
2312 /* TODO: refactor, so that 1/2 are not needed */
2313 if (ren1) {
2314 renames1 = a_renames;
2315 renames2Dst = &b_by_dst;
2316 branch1 = o->branch1;
2317 branch2 = o->branch2;
2318 } else {
2319 renames1 = b_renames;
2320 renames2Dst = &a_by_dst;
2321 branch1 = o->branch2;
2322 branch2 = o->branch1;
2323 SWAP(ren2, ren1);
2326 if (ren1->processed)
2327 continue;
2328 ren1->processed = 1;
2329 ren1->dst_entry->processed = 1;
2330 /* BUG: We should only mark src_entry as processed if we
2331 * are not dealing with a rename + add-source case.
2333 ren1->src_entry->processed = 1;
2335 ren1_src = ren1->pair->one->path;
2336 ren1_dst = ren1->pair->two->path;
2338 if (ren2) {
2339 /* One file renamed on both sides */
2340 const char *ren2_src = ren2->pair->one->path;
2341 const char *ren2_dst = ren2->pair->two->path;
2342 enum rename_type rename_type;
2343 if (strcmp(ren1_src, ren2_src) != 0)
2344 die("BUG: ren1_src != ren2_src");
2345 ren2->dst_entry->processed = 1;
2346 ren2->processed = 1;
2347 if (strcmp(ren1_dst, ren2_dst) != 0) {
2348 rename_type = RENAME_ONE_FILE_TO_TWO;
2349 clean_merge = 0;
2350 } else {
2351 rename_type = RENAME_ONE_FILE_TO_ONE;
2352 /* BUG: We should only remove ren1_src in
2353 * the base stage (think of rename +
2354 * add-source cases).
2356 remove_file(o, 1, ren1_src, 1);
2357 update_entry(ren1->dst_entry,
2358 ren1->pair->one,
2359 ren1->pair->two,
2360 ren2->pair->two);
2362 setup_rename_conflict_info(rename_type,
2363 ren1->pair,
2364 ren2->pair,
2365 branch1,
2366 branch2,
2367 ren1->dst_entry,
2368 ren2->dst_entry,
2370 NULL,
2371 NULL);
2372 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2373 /* Two different files renamed to the same thing */
2374 char *ren2_dst;
2375 ren2 = lookup->util;
2376 ren2_dst = ren2->pair->two->path;
2377 if (strcmp(ren1_dst, ren2_dst) != 0)
2378 die("BUG: ren1_dst != ren2_dst");
2380 clean_merge = 0;
2381 ren2->processed = 1;
2383 * BUG: We should only mark src_entry as processed
2384 * if we are not dealing with a rename + add-source
2385 * case.
2387 ren2->src_entry->processed = 1;
2389 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2390 ren1->pair,
2391 ren2->pair,
2392 branch1,
2393 branch2,
2394 ren1->dst_entry,
2395 ren2->dst_entry,
2397 ren1->src_entry,
2398 ren2->src_entry);
2400 } else {
2401 /* Renamed in 1, maybe changed in 2 */
2402 /* we only use sha1 and mode of these */
2403 struct diff_filespec src_other, dst_other;
2404 int try_merge;
2407 * unpack_trees loads entries from common-commit
2408 * into stage 1, from head-commit into stage 2, and
2409 * from merge-commit into stage 3. We keep track
2410 * of which side corresponds to the rename.
2412 int renamed_stage = a_renames == renames1 ? 2 : 3;
2413 int other_stage = a_renames == renames1 ? 3 : 2;
2415 /* BUG: We should only remove ren1_src in the base
2416 * stage and in other_stage (think of rename +
2417 * add-source case).
2419 remove_file(o, 1, ren1_src,
2420 renamed_stage == 2 || !was_tracked(ren1_src));
2422 oidcpy(&src_other.oid,
2423 &ren1->src_entry->stages[other_stage].oid);
2424 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2425 oidcpy(&dst_other.oid,
2426 &ren1->dst_entry->stages[other_stage].oid);
2427 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2428 try_merge = 0;
2430 if (oid_eq(&src_other.oid, &null_oid) &&
2431 ren1->add_turned_into_rename) {
2432 setup_rename_conflict_info(RENAME_DIR,
2433 ren1->pair,
2434 NULL,
2435 branch1,
2436 branch2,
2437 ren1->dst_entry,
2438 NULL,
2440 NULL,
2441 NULL);
2442 } else if (oid_eq(&src_other.oid, &null_oid)) {
2443 setup_rename_conflict_info(RENAME_DELETE,
2444 ren1->pair,
2445 NULL,
2446 branch1,
2447 branch2,
2448 ren1->dst_entry,
2449 NULL,
2451 NULL,
2452 NULL);
2453 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2454 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2456 * Added file on the other side identical to
2457 * the file being renamed: clean merge.
2458 * Also, there is no need to overwrite the
2459 * file already in the working copy, so call
2460 * update_file_flags() instead of
2461 * update_file().
2463 if (update_file_flags(o,
2464 &ren1->pair->two->oid,
2465 ren1->pair->two->mode,
2466 ren1_dst,
2467 1, /* update_cache */
2468 0 /* update_wd */))
2469 clean_merge = -1;
2470 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2471 clean_merge = 0;
2472 try_merge = 1;
2473 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2474 "%s added in %s"),
2475 ren1_src, ren1_dst, branch1,
2476 ren1_dst, branch2);
2477 if (o->call_depth) {
2478 struct merge_file_info mfi;
2479 if (merge_file_one(o, ren1_dst, &null_oid, 0,
2480 &ren1->pair->two->oid,
2481 ren1->pair->two->mode,
2482 &dst_other.oid,
2483 dst_other.mode,
2484 branch1, branch2, &mfi)) {
2485 clean_merge = -1;
2486 goto cleanup_and_return;
2488 output(o, 1, _("Adding merged %s"), ren1_dst);
2489 if (update_file(o, 0, &mfi.oid,
2490 mfi.mode, ren1_dst))
2491 clean_merge = -1;
2492 try_merge = 0;
2493 } else {
2494 char *new_path = unique_path(o, ren1_dst, branch2);
2495 output(o, 1, _("Adding as %s instead"), new_path);
2496 if (update_file(o, 0, &dst_other.oid,
2497 dst_other.mode, new_path))
2498 clean_merge = -1;
2499 free(new_path);
2501 } else
2502 try_merge = 1;
2504 if (clean_merge < 0)
2505 goto cleanup_and_return;
2506 if (try_merge) {
2507 struct diff_filespec *one, *a, *b;
2508 src_other.path = (char *)ren1_src;
2510 one = ren1->pair->one;
2511 if (a_renames == renames1) {
2512 a = ren1->pair->two;
2513 b = &src_other;
2514 } else {
2515 b = ren1->pair->two;
2516 a = &src_other;
2518 update_entry(ren1->dst_entry, one, a, b);
2519 setup_rename_conflict_info(RENAME_NORMAL,
2520 ren1->pair,
2521 NULL,
2522 branch1,
2523 NULL,
2524 ren1->dst_entry,
2525 NULL,
2527 NULL,
2528 NULL);
2532 cleanup_and_return:
2533 string_list_clear(&a_by_dst, 0);
2534 string_list_clear(&b_by_dst, 0);
2536 return clean_merge;
2539 struct rename_info {
2540 struct string_list *head_renames;
2541 struct string_list *merge_renames;
2544 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2545 struct hashmap *dir_renames)
2547 struct hashmap_iter iter;
2548 struct dir_rename_entry *e;
2550 hashmap_iter_init(dir_renames, &iter);
2551 while ((e = hashmap_iter_next(&iter))) {
2552 free(e->dir);
2553 strbuf_release(&e->new_dir);
2554 /* possible_new_dirs already cleared in get_directory_renames */
2556 hashmap_free(dir_renames, 1);
2557 free(dir_renames);
2559 free(pairs->queue);
2560 free(pairs);
2563 static int handle_renames(struct merge_options *o,
2564 struct tree *common,
2565 struct tree *head,
2566 struct tree *merge,
2567 struct string_list *entries,
2568 struct rename_info *ri)
2570 struct diff_queue_struct *head_pairs, *merge_pairs;
2571 struct hashmap *dir_re_head, *dir_re_merge;
2572 int clean = 1;
2574 ri->head_renames = NULL;
2575 ri->merge_renames = NULL;
2577 if (!o->detect_rename)
2578 return 1;
2580 head_pairs = get_diffpairs(o, common, head);
2581 merge_pairs = get_diffpairs(o, common, merge);
2583 dir_re_head = get_directory_renames(head_pairs, head);
2584 dir_re_merge = get_directory_renames(merge_pairs, merge);
2586 handle_directory_level_conflicts(o,
2587 dir_re_head, head,
2588 dir_re_merge, merge);
2590 ri->head_renames = get_renames(o, head_pairs,
2591 dir_re_merge, dir_re_head, head,
2592 common, head, merge, entries,
2593 &clean);
2594 if (clean < 0)
2595 goto cleanup;
2596 ri->merge_renames = get_renames(o, merge_pairs,
2597 dir_re_head, dir_re_merge, merge,
2598 common, head, merge, entries,
2599 &clean);
2600 if (clean < 0)
2601 goto cleanup;
2602 clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2604 cleanup:
2606 * Some cleanup is deferred until cleanup_renames() because the
2607 * data structures are still needed and referenced in
2608 * process_entry(). But there are a few things we can free now.
2610 initial_cleanup_rename(head_pairs, dir_re_head);
2611 initial_cleanup_rename(merge_pairs, dir_re_merge);
2613 return clean;
2616 static void final_cleanup_rename(struct string_list *rename)
2618 const struct rename *re;
2619 int i;
2621 if (rename == NULL)
2622 return;
2624 for (i = 0; i < rename->nr; i++) {
2625 re = rename->items[i].util;
2626 diff_free_filepair(re->pair);
2628 string_list_clear(rename, 1);
2629 free(rename);
2632 static void final_cleanup_renames(struct rename_info *re_info)
2634 final_cleanup_rename(re_info->head_renames);
2635 final_cleanup_rename(re_info->merge_renames);
2638 static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2640 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2643 static int read_oid_strbuf(struct merge_options *o,
2644 const struct object_id *oid, struct strbuf *dst)
2646 void *buf;
2647 enum object_type type;
2648 unsigned long size;
2649 buf = read_object_file(oid, &type, &size);
2650 if (!buf)
2651 return err(o, _("cannot read object %s"), oid_to_hex(oid));
2652 if (type != OBJ_BLOB) {
2653 free(buf);
2654 return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2656 strbuf_attach(dst, buf, size, size + 1);
2657 return 0;
2660 static int blob_unchanged(struct merge_options *opt,
2661 const struct object_id *o_oid,
2662 unsigned o_mode,
2663 const struct object_id *a_oid,
2664 unsigned a_mode,
2665 int renormalize, const char *path)
2667 struct strbuf o = STRBUF_INIT;
2668 struct strbuf a = STRBUF_INIT;
2669 int ret = 0; /* assume changed for safety */
2671 if (a_mode != o_mode)
2672 return 0;
2673 if (oid_eq(o_oid, a_oid))
2674 return 1;
2675 if (!renormalize)
2676 return 0;
2678 assert(o_oid && a_oid);
2679 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2680 goto error_return;
2682 * Note: binary | is used so that both renormalizations are
2683 * performed. Comparison can be skipped if both files are
2684 * unchanged since their sha1s have already been compared.
2686 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2687 renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2688 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2690 error_return:
2691 strbuf_release(&o);
2692 strbuf_release(&a);
2693 return ret;
2696 static int handle_modify_delete(struct merge_options *o,
2697 const char *path,
2698 struct object_id *o_oid, int o_mode,
2699 struct object_id *a_oid, int a_mode,
2700 struct object_id *b_oid, int b_mode)
2702 const char *modify_branch, *delete_branch;
2703 struct object_id *changed_oid;
2704 int changed_mode;
2706 if (a_oid) {
2707 modify_branch = o->branch1;
2708 delete_branch = o->branch2;
2709 changed_oid = a_oid;
2710 changed_mode = a_mode;
2711 } else {
2712 modify_branch = o->branch2;
2713 delete_branch = o->branch1;
2714 changed_oid = b_oid;
2715 changed_mode = b_mode;
2718 return handle_change_delete(o,
2719 path, NULL,
2720 o_oid, o_mode,
2721 changed_oid, changed_mode,
2722 modify_branch, delete_branch,
2723 _("modify"), _("modified"));
2726 static int merge_content(struct merge_options *o,
2727 const char *path,
2728 int file_in_way,
2729 struct object_id *o_oid, int o_mode,
2730 struct object_id *a_oid, int a_mode,
2731 struct object_id *b_oid, int b_mode,
2732 struct rename_conflict_info *rename_conflict_info)
2734 const char *reason = _("content");
2735 const char *path1 = NULL, *path2 = NULL;
2736 struct merge_file_info mfi;
2737 struct diff_filespec one, a, b;
2738 unsigned df_conflict_remains = 0;
2740 if (!o_oid) {
2741 reason = _("add/add");
2742 o_oid = (struct object_id *)&null_oid;
2744 one.path = a.path = b.path = (char *)path;
2745 oidcpy(&one.oid, o_oid);
2746 one.mode = o_mode;
2747 oidcpy(&a.oid, a_oid);
2748 a.mode = a_mode;
2749 oidcpy(&b.oid, b_oid);
2750 b.mode = b_mode;
2752 if (rename_conflict_info) {
2753 struct diff_filepair *pair1 = rename_conflict_info->pair1;
2755 path1 = (o->branch1 == rename_conflict_info->branch1) ?
2756 pair1->two->path : pair1->one->path;
2757 /* If rename_conflict_info->pair2 != NULL, we are in
2758 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
2759 * normal rename.
2761 path2 = (rename_conflict_info->pair2 ||
2762 o->branch2 == rename_conflict_info->branch1) ?
2763 pair1->two->path : pair1->one->path;
2765 if (dir_in_way(path, !o->call_depth,
2766 S_ISGITLINK(pair1->two->mode)))
2767 df_conflict_remains = 1;
2769 if (merge_file_special_markers(o, &one, &a, &b,
2770 o->branch1, path1,
2771 o->branch2, path2, &mfi))
2772 return -1;
2774 if (mfi.clean && !df_conflict_remains &&
2775 oid_eq(&mfi.oid, a_oid) && mfi.mode == a_mode) {
2776 output(o, 3, _("Skipped %s (merged same as existing)"), path);
2778 * The content merge resulted in the same file contents we
2779 * already had. We can return early if those file contents
2780 * are recorded at the correct path (which may not be true
2781 * if the merge involves a rename).
2783 if (was_tracked(path)) {
2784 add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2785 0, (!o->call_depth), 0);
2786 return mfi.clean;
2788 } else
2789 output(o, 2, _("Auto-merging %s"), path);
2791 if (!mfi.clean) {
2792 if (S_ISGITLINK(mfi.mode))
2793 reason = _("submodule");
2794 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2795 reason, path);
2796 if (rename_conflict_info && !df_conflict_remains)
2797 if (update_stages(o, path, &one, &a, &b))
2798 return -1;
2801 if (df_conflict_remains || file_in_way) {
2802 char *new_path;
2803 if (o->call_depth) {
2804 remove_file_from_cache(path);
2805 } else {
2806 if (!mfi.clean) {
2807 if (update_stages(o, path, &one, &a, &b))
2808 return -1;
2809 } else {
2810 int file_from_stage2 = was_tracked(path);
2811 struct diff_filespec merged;
2812 oidcpy(&merged.oid, &mfi.oid);
2813 merged.mode = mfi.mode;
2815 if (update_stages(o, path, NULL,
2816 file_from_stage2 ? &merged : NULL,
2817 file_from_stage2 ? NULL : &merged))
2818 return -1;
2822 new_path = unique_path(o, path, rename_conflict_info->branch1);
2823 output(o, 1, _("Adding as %s instead"), new_path);
2824 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2825 free(new_path);
2826 return -1;
2828 free(new_path);
2829 mfi.clean = 0;
2830 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2831 return -1;
2832 return mfi.clean;
2835 static int conflict_rename_normal(struct merge_options *o,
2836 const char *path,
2837 struct object_id *o_oid, unsigned int o_mode,
2838 struct object_id *a_oid, unsigned int a_mode,
2839 struct object_id *b_oid, unsigned int b_mode,
2840 struct rename_conflict_info *ci)
2842 int clean_merge;
2843 int file_in_the_way = 0;
2845 if (was_dirty(o, path)) {
2846 file_in_the_way = 1;
2847 output(o, 1, _("Refusing to lose dirty file at %s"), path);
2850 /* Merge the content and write it out */
2851 clean_merge = merge_content(o, path, file_in_the_way,
2852 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2853 ci);
2854 if (clean_merge > 0 && file_in_the_way)
2855 clean_merge = 0;
2856 return clean_merge;
2859 /* Per entry merge function */
2860 static int process_entry(struct merge_options *o,
2861 const char *path, struct stage_data *entry)
2863 int clean_merge = 1;
2864 int normalize = o->renormalize;
2865 unsigned o_mode = entry->stages[1].mode;
2866 unsigned a_mode = entry->stages[2].mode;
2867 unsigned b_mode = entry->stages[3].mode;
2868 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2869 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2870 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2872 entry->processed = 1;
2873 if (entry->rename_conflict_info) {
2874 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2875 switch (conflict_info->rename_type) {
2876 case RENAME_NORMAL:
2877 case RENAME_ONE_FILE_TO_ONE:
2878 clean_merge = conflict_rename_normal(o,
2879 path,
2880 o_oid, o_mode,
2881 a_oid, a_mode,
2882 b_oid, b_mode,
2883 conflict_info);
2884 break;
2885 case RENAME_DIR:
2886 clean_merge = 1;
2887 if (conflict_rename_dir(o,
2888 conflict_info->pair1,
2889 conflict_info->branch1,
2890 conflict_info->branch2))
2891 clean_merge = -1;
2892 break;
2893 case RENAME_DELETE:
2894 clean_merge = 0;
2895 if (conflict_rename_delete(o,
2896 conflict_info->pair1,
2897 conflict_info->branch1,
2898 conflict_info->branch2))
2899 clean_merge = -1;
2900 break;
2901 case RENAME_ONE_FILE_TO_TWO:
2902 clean_merge = 0;
2903 if (conflict_rename_rename_1to2(o, conflict_info))
2904 clean_merge = -1;
2905 break;
2906 case RENAME_TWO_FILES_TO_ONE:
2907 clean_merge = 0;
2908 if (conflict_rename_rename_2to1(o, conflict_info))
2909 clean_merge = -1;
2910 break;
2911 default:
2912 entry->processed = 0;
2913 break;
2915 } else if (o_oid && (!a_oid || !b_oid)) {
2916 /* Case A: Deleted in one */
2917 if ((!a_oid && !b_oid) ||
2918 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2919 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2920 /* Deleted in both or deleted in one and
2921 * unchanged in the other */
2922 if (a_oid)
2923 output(o, 2, _("Removing %s"), path);
2924 /* do not touch working file if it did not exist */
2925 remove_file(o, 1, path, !a_oid);
2926 } else {
2927 /* Modify/delete; deleted side may have put a directory in the way */
2928 clean_merge = 0;
2929 if (handle_modify_delete(o, path, o_oid, o_mode,
2930 a_oid, a_mode, b_oid, b_mode))
2931 clean_merge = -1;
2933 } else if ((!o_oid && a_oid && !b_oid) ||
2934 (!o_oid && !a_oid && b_oid)) {
2935 /* Case B: Added in one. */
2936 /* [nothing|directory] -> ([nothing|directory], file) */
2938 const char *add_branch;
2939 const char *other_branch;
2940 unsigned mode;
2941 const struct object_id *oid;
2942 const char *conf;
2944 if (a_oid) {
2945 add_branch = o->branch1;
2946 other_branch = o->branch2;
2947 mode = a_mode;
2948 oid = a_oid;
2949 conf = _("file/directory");
2950 } else {
2951 add_branch = o->branch2;
2952 other_branch = o->branch1;
2953 mode = b_mode;
2954 oid = b_oid;
2955 conf = _("directory/file");
2957 if (dir_in_way(path,
2958 !o->call_depth && !S_ISGITLINK(a_mode),
2959 0)) {
2960 char *new_path = unique_path(o, path, add_branch);
2961 clean_merge = 0;
2962 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
2963 "Adding %s as %s"),
2964 conf, path, other_branch, path, new_path);
2965 if (update_file(o, 0, oid, mode, new_path))
2966 clean_merge = -1;
2967 else if (o->call_depth)
2968 remove_file_from_cache(path);
2969 free(new_path);
2970 } else {
2971 output(o, 2, _("Adding %s"), path);
2972 /* do not overwrite file if already present */
2973 if (update_file_flags(o, oid, mode, path, 1, !a_oid))
2974 clean_merge = -1;
2976 } else if (a_oid && b_oid) {
2977 /* Case C: Added in both (check for same permissions) and */
2978 /* case D: Modified in both, but differently. */
2979 clean_merge = merge_content(o, path, 0 /* file_in_way */,
2980 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2981 NULL);
2982 } else if (!o_oid && !a_oid && !b_oid) {
2984 * this entry was deleted altogether. a_mode == 0 means
2985 * we had that path and want to actively remove it.
2987 remove_file(o, 1, path, !a_mode);
2988 } else
2989 die("BUG: fatal merge failure, shouldn't happen.");
2991 return clean_merge;
2994 int merge_trees(struct merge_options *o,
2995 struct tree *head,
2996 struct tree *merge,
2997 struct tree *common,
2998 struct tree **result)
3000 int code, clean;
3002 if (o->subtree_shift) {
3003 merge = shift_tree_object(head, merge, o->subtree_shift);
3004 common = shift_tree_object(head, common, o->subtree_shift);
3007 if (oid_eq(&common->object.oid, &merge->object.oid)) {
3008 struct strbuf sb = STRBUF_INIT;
3010 if (!o->call_depth && index_has_changes(&sb)) {
3011 err(o, _("Dirty index: cannot merge (dirty: %s)"),
3012 sb.buf);
3013 return 0;
3015 output(o, 0, _("Already up to date!"));
3016 *result = head;
3017 return 1;
3020 code = git_merge_trees(o, common, head, merge);
3022 if (code != 0) {
3023 if (show(o, 4) || o->call_depth)
3024 err(o, _("merging of trees %s and %s failed"),
3025 oid_to_hex(&head->object.oid),
3026 oid_to_hex(&merge->object.oid));
3027 return -1;
3030 if (unmerged_cache()) {
3031 struct string_list *entries;
3032 struct rename_info re_info;
3033 int i;
3035 * Only need the hashmap while processing entries, so
3036 * initialize it here and free it when we are done running
3037 * through the entries. Keeping it in the merge_options as
3038 * opposed to decaring a local hashmap is for convenience
3039 * so that we don't have to pass it to around.
3041 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3042 get_files_dirs(o, head);
3043 get_files_dirs(o, merge);
3045 entries = get_unmerged();
3046 clean = handle_renames(o, common, head, merge, entries,
3047 &re_info);
3048 record_df_conflict_files(o, entries);
3049 if (clean < 0)
3050 goto cleanup;
3051 for (i = entries->nr-1; 0 <= i; i--) {
3052 const char *path = entries->items[i].string;
3053 struct stage_data *e = entries->items[i].util;
3054 if (!e->processed) {
3055 int ret = process_entry(o, path, e);
3056 if (!ret)
3057 clean = 0;
3058 else if (ret < 0) {
3059 clean = ret;
3060 goto cleanup;
3064 for (i = 0; i < entries->nr; i++) {
3065 struct stage_data *e = entries->items[i].util;
3066 if (!e->processed)
3067 die("BUG: unprocessed path??? %s",
3068 entries->items[i].string);
3071 cleanup:
3072 final_cleanup_renames(&re_info);
3074 string_list_clear(entries, 1);
3075 free(entries);
3077 hashmap_free(&o->current_file_dir_set, 1);
3079 if (clean < 0)
3080 return clean;
3082 else
3083 clean = 1;
3085 if (o->call_depth && !(*result = write_tree_from_memory(o)))
3086 return -1;
3088 return clean;
3091 static struct commit_list *reverse_commit_list(struct commit_list *list)
3093 struct commit_list *next = NULL, *current, *backup;
3094 for (current = list; current; current = backup) {
3095 backup = current->next;
3096 current->next = next;
3097 next = current;
3099 return next;
3103 * Merge the commits h1 and h2, return the resulting virtual
3104 * commit object and a flag indicating the cleanness of the merge.
3106 int merge_recursive(struct merge_options *o,
3107 struct commit *h1,
3108 struct commit *h2,
3109 struct commit_list *ca,
3110 struct commit **result)
3112 struct commit_list *iter;
3113 struct commit *merged_common_ancestors;
3114 struct tree *mrtree;
3115 int clean;
3117 if (show(o, 4)) {
3118 output(o, 4, _("Merging:"));
3119 output_commit_title(o, h1);
3120 output_commit_title(o, h2);
3123 if (!ca) {
3124 ca = get_merge_bases(h1, h2);
3125 ca = reverse_commit_list(ca);
3128 if (show(o, 5)) {
3129 unsigned cnt = commit_list_count(ca);
3131 output(o, 5, Q_("found %u common ancestor:",
3132 "found %u common ancestors:", cnt), cnt);
3133 for (iter = ca; iter; iter = iter->next)
3134 output_commit_title(o, iter->item);
3137 merged_common_ancestors = pop_commit(&ca);
3138 if (merged_common_ancestors == NULL) {
3139 /* if there is no common ancestor, use an empty tree */
3140 struct tree *tree;
3142 tree = lookup_tree(the_hash_algo->empty_tree);
3143 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3146 for (iter = ca; iter; iter = iter->next) {
3147 const char *saved_b1, *saved_b2;
3148 o->call_depth++;
3150 * When the merge fails, the result contains files
3151 * with conflict markers. The cleanness flag is
3152 * ignored (unless indicating an error), it was never
3153 * actually used, as result of merge_trees has always
3154 * overwritten it: the committed "conflicts" were
3155 * already resolved.
3157 discard_cache();
3158 saved_b1 = o->branch1;
3159 saved_b2 = o->branch2;
3160 o->branch1 = "Temporary merge branch 1";
3161 o->branch2 = "Temporary merge branch 2";
3162 if (merge_recursive(o, merged_common_ancestors, iter->item,
3163 NULL, &merged_common_ancestors) < 0)
3164 return -1;
3165 o->branch1 = saved_b1;
3166 o->branch2 = saved_b2;
3167 o->call_depth--;
3169 if (!merged_common_ancestors)
3170 return err(o, _("merge returned no commit"));
3173 discard_cache();
3174 if (!o->call_depth)
3175 read_cache();
3177 o->ancestor = "merged common ancestors";
3178 clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,
3179 &mrtree);
3180 if (clean < 0) {
3181 flush_output(o);
3182 return clean;
3185 if (o->call_depth) {
3186 *result = make_virtual_commit(mrtree, "merged tree");
3187 commit_list_insert(h1, &(*result)->parents);
3188 commit_list_insert(h2, &(*result)->parents->next);
3190 flush_output(o);
3191 if (!o->call_depth && o->buffer_output < 2)
3192 strbuf_release(&o->obuf);
3193 if (show(o, 2))
3194 diff_warn_rename_limit("merge.renamelimit",
3195 o->needed_rename_limit, 0);
3196 return clean;
3199 static struct commit *get_ref(const struct object_id *oid, const char *name)
3201 struct object *object;
3203 object = deref_tag(parse_object(oid), name, strlen(name));
3204 if (!object)
3205 return NULL;
3206 if (object->type == OBJ_TREE)
3207 return make_virtual_commit((struct tree*)object, name);
3208 if (object->type != OBJ_COMMIT)
3209 return NULL;
3210 if (parse_commit((struct commit *)object))
3211 return NULL;
3212 return (struct commit *)object;
3215 int merge_recursive_generic(struct merge_options *o,
3216 const struct object_id *head,
3217 const struct object_id *merge,
3218 int num_base_list,
3219 const struct object_id **base_list,
3220 struct commit **result)
3222 int clean;
3223 struct lock_file lock = LOCK_INIT;
3224 struct commit *head_commit = get_ref(head, o->branch1);
3225 struct commit *next_commit = get_ref(merge, o->branch2);
3226 struct commit_list *ca = NULL;
3228 if (base_list) {
3229 int i;
3230 for (i = 0; i < num_base_list; ++i) {
3231 struct commit *base;
3232 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3233 return err(o, _("Could not parse object '%s'"),
3234 oid_to_hex(base_list[i]));
3235 commit_list_insert(base, &ca);
3239 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3240 clean = merge_recursive(o, head_commit, next_commit, ca,
3241 result);
3242 if (clean < 0) {
3243 rollback_lock_file(&lock);
3244 return clean;
3247 if (write_locked_index(&the_index, &lock,
3248 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3249 return err(o, _("Unable to write index."));
3251 return clean ? 0 : 1;
3254 static void merge_recursive_config(struct merge_options *o)
3256 git_config_get_int("merge.verbosity", &o->verbosity);
3257 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3258 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3259 git_config(git_xmerge_config, NULL);
3262 void init_merge_options(struct merge_options *o)
3264 const char *merge_verbosity;
3265 memset(o, 0, sizeof(struct merge_options));
3266 o->verbosity = 2;
3267 o->buffer_output = 1;
3268 o->diff_rename_limit = -1;
3269 o->merge_rename_limit = -1;
3270 o->renormalize = 0;
3271 o->detect_rename = 1;
3272 merge_recursive_config(o);
3273 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3274 if (merge_verbosity)
3275 o->verbosity = strtol(merge_verbosity, NULL, 10);
3276 if (o->verbosity >= 5)
3277 o->buffer_output = 0;
3278 strbuf_init(&o->obuf, 0);
3279 string_list_init(&o->df_conflict_file_set, 1);
3282 int parse_merge_opt(struct merge_options *o, const char *s)
3284 const char *arg;
3286 if (!s || !*s)
3287 return -1;
3288 if (!strcmp(s, "ours"))
3289 o->recursive_variant = MERGE_RECURSIVE_OURS;
3290 else if (!strcmp(s, "theirs"))
3291 o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3292 else if (!strcmp(s, "subtree"))
3293 o->subtree_shift = "";
3294 else if (skip_prefix(s, "subtree=", &arg))
3295 o->subtree_shift = arg;
3296 else if (!strcmp(s, "patience"))
3297 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3298 else if (!strcmp(s, "histogram"))
3299 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3300 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3301 long value = parse_algorithm_value(arg);
3302 if (value < 0)
3303 return -1;
3304 /* clear out previous settings */
3305 DIFF_XDL_CLR(o, NEED_MINIMAL);
3306 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3307 o->xdl_opts |= value;
3309 else if (!strcmp(s, "ignore-space-change"))
3310 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3311 else if (!strcmp(s, "ignore-all-space"))
3312 DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3313 else if (!strcmp(s, "ignore-space-at-eol"))
3314 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3315 else if (!strcmp(s, "ignore-cr-at-eol"))
3316 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3317 else if (!strcmp(s, "renormalize"))
3318 o->renormalize = 1;
3319 else if (!strcmp(s, "no-renormalize"))
3320 o->renormalize = 0;
3321 else if (!strcmp(s, "no-renames"))
3322 o->detect_rename = 0;
3323 else if (!strcmp(s, "find-renames")) {
3324 o->detect_rename = 1;
3325 o->rename_score = 0;
3327 else if (skip_prefix(s, "find-renames=", &arg) ||
3328 skip_prefix(s, "rename-threshold=", &arg)) {
3329 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3330 return -1;
3331 o->detect_rename = 1;
3333 else
3334 return -1;
3335 return 0;