merge-recursive: avoid clobbering untracked files with directory renames
[git/git-svn.git] / merge-recursive.c
blob87f2479e1a6d0b5f3f1d32c1e1f12b75ade1ffb7
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(int index_only,
341 struct tree *common,
342 struct tree *head,
343 struct tree *merge)
345 int rc;
346 struct tree_desc t[3];
347 struct unpack_trees_options opts;
349 memset(&opts, 0, sizeof(opts));
350 if (index_only)
351 opts.index_only = 1;
352 else
353 opts.update = 1;
354 opts.merge = 1;
355 opts.head_idx = 2;
356 opts.fn = threeway_merge;
357 opts.src_index = &the_index;
358 opts.dst_index = &the_index;
359 setup_unpack_trees_porcelain(&opts, "merge");
361 init_tree_desc_from_tree(t+0, common);
362 init_tree_desc_from_tree(t+1, head);
363 init_tree_desc_from_tree(t+2, merge);
365 rc = unpack_trees(3, t, &opts);
366 cache_tree_free(&active_cache_tree);
367 return rc;
370 struct tree *write_tree_from_memory(struct merge_options *o)
372 struct tree *result = NULL;
374 if (unmerged_cache()) {
375 int i;
376 fprintf(stderr, "BUG: There are unmerged index entries:\n");
377 for (i = 0; i < active_nr; i++) {
378 const struct cache_entry *ce = active_cache[i];
379 if (ce_stage(ce))
380 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
381 (int)ce_namelen(ce), ce->name);
383 die("BUG: unmerged index entries in merge-recursive.c");
386 if (!active_cache_tree)
387 active_cache_tree = cache_tree();
389 if (!cache_tree_fully_valid(active_cache_tree) &&
390 cache_tree_update(&the_index, 0) < 0) {
391 err(o, _("error building trees"));
392 return NULL;
395 result = lookup_tree(&active_cache_tree->oid);
397 return result;
400 static int save_files_dirs(const struct object_id *oid,
401 struct strbuf *base, const char *path,
402 unsigned int mode, int stage, void *context)
404 struct path_hashmap_entry *entry;
405 int baselen = base->len;
406 struct merge_options *o = context;
408 strbuf_addstr(base, path);
410 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
411 hashmap_entry_init(entry, path_hash(entry->path));
412 hashmap_add(&o->current_file_dir_set, entry);
414 strbuf_setlen(base, baselen);
415 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
418 static void get_files_dirs(struct merge_options *o, struct tree *tree)
420 struct pathspec match_all;
421 memset(&match_all, 0, sizeof(match_all));
422 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
425 static int get_tree_entry_if_blob(const struct object_id *tree,
426 const char *path,
427 struct object_id *hashy,
428 unsigned int *mode_o)
430 int ret;
432 ret = get_tree_entry(tree, path, hashy, mode_o);
433 if (S_ISDIR(*mode_o)) {
434 oidcpy(hashy, &null_oid);
435 *mode_o = 0;
437 return ret;
441 * Returns an index_entry instance which doesn't have to correspond to
442 * a real cache entry in Git's index.
444 static struct stage_data *insert_stage_data(const char *path,
445 struct tree *o, struct tree *a, struct tree *b,
446 struct string_list *entries)
448 struct string_list_item *item;
449 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
450 get_tree_entry_if_blob(&o->object.oid, path,
451 &e->stages[1].oid, &e->stages[1].mode);
452 get_tree_entry_if_blob(&a->object.oid, path,
453 &e->stages[2].oid, &e->stages[2].mode);
454 get_tree_entry_if_blob(&b->object.oid, path,
455 &e->stages[3].oid, &e->stages[3].mode);
456 item = string_list_insert(entries, path);
457 item->util = e;
458 return e;
462 * Create a dictionary mapping file names to stage_data objects. The
463 * dictionary contains one entry for every path with a non-zero stage entry.
465 static struct string_list *get_unmerged(void)
467 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
468 int i;
470 unmerged->strdup_strings = 1;
472 for (i = 0; i < active_nr; i++) {
473 struct string_list_item *item;
474 struct stage_data *e;
475 const struct cache_entry *ce = active_cache[i];
476 if (!ce_stage(ce))
477 continue;
479 item = string_list_lookup(unmerged, ce->name);
480 if (!item) {
481 item = string_list_insert(unmerged, ce->name);
482 item->util = xcalloc(1, sizeof(struct stage_data));
484 e = item->util;
485 e->stages[ce_stage(ce)].mode = ce->ce_mode;
486 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
489 return unmerged;
492 static int string_list_df_name_compare(const char *one, const char *two)
494 int onelen = strlen(one);
495 int twolen = strlen(two);
497 * Here we only care that entries for D/F conflicts are
498 * adjacent, in particular with the file of the D/F conflict
499 * appearing before files below the corresponding directory.
500 * The order of the rest of the list is irrelevant for us.
502 * To achieve this, we sort with df_name_compare and provide
503 * the mode S_IFDIR so that D/F conflicts will sort correctly.
504 * We use the mode S_IFDIR for everything else for simplicity,
505 * since in other cases any changes in their order due to
506 * sorting cause no problems for us.
508 int cmp = df_name_compare(one, onelen, S_IFDIR,
509 two, twolen, S_IFDIR);
511 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
512 * that 'foo' comes before 'foo/bar'.
514 if (cmp)
515 return cmp;
516 return onelen - twolen;
519 static void record_df_conflict_files(struct merge_options *o,
520 struct string_list *entries)
522 /* If there is a D/F conflict and the file for such a conflict
523 * currently exist in the working tree, we want to allow it to be
524 * removed to make room for the corresponding directory if needed.
525 * The files underneath the directories of such D/F conflicts will
526 * be processed before the corresponding file involved in the D/F
527 * conflict. If the D/F directory ends up being removed by the
528 * merge, then we won't have to touch the D/F file. If the D/F
529 * directory needs to be written to the working copy, then the D/F
530 * file will simply be removed (in make_room_for_path()) to make
531 * room for the necessary paths. Note that if both the directory
532 * and the file need to be present, then the D/F file will be
533 * reinstated with a new unique name at the time it is processed.
535 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
536 const char *last_file = NULL;
537 int last_len = 0;
538 int i;
541 * If we're merging merge-bases, we don't want to bother with
542 * any working directory changes.
544 if (o->call_depth)
545 return;
547 /* Ensure D/F conflicts are adjacent in the entries list. */
548 for (i = 0; i < entries->nr; i++) {
549 struct string_list_item *next = &entries->items[i];
550 string_list_append(&df_sorted_entries, next->string)->util =
551 next->util;
553 df_sorted_entries.cmp = string_list_df_name_compare;
554 string_list_sort(&df_sorted_entries);
556 string_list_clear(&o->df_conflict_file_set, 1);
557 for (i = 0; i < df_sorted_entries.nr; i++) {
558 const char *path = df_sorted_entries.items[i].string;
559 int len = strlen(path);
560 struct stage_data *e = df_sorted_entries.items[i].util;
563 * Check if last_file & path correspond to a D/F conflict;
564 * i.e. whether path is last_file+'/'+<something>.
565 * If so, record that it's okay to remove last_file to make
566 * room for path and friends if needed.
568 if (last_file &&
569 len > last_len &&
570 memcmp(path, last_file, last_len) == 0 &&
571 path[last_len] == '/') {
572 string_list_insert(&o->df_conflict_file_set, last_file);
576 * Determine whether path could exist as a file in the
577 * working directory as a possible D/F conflict. This
578 * will only occur when it exists in stage 2 as a
579 * file.
581 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
582 last_file = path;
583 last_len = len;
584 } else {
585 last_file = NULL;
588 string_list_clear(&df_sorted_entries, 0);
591 struct rename {
592 struct diff_filepair *pair;
594 * Purpose of src_entry and dst_entry:
596 * If 'before' is renamed to 'after' then src_entry will contain
597 * the versions of 'before' from the merge_base, HEAD, and MERGE in
598 * stages 1, 2, and 3; dst_entry will contain the respective
599 * versions of 'after' in corresponding locations. Thus, we have a
600 * total of six modes and oids, though some will be null. (Stage 0
601 * is ignored; we're interested in handling conflicts.)
603 * Since we don't turn on break-rewrites by default, neither
604 * src_entry nor dst_entry can have all three of their stages have
605 * non-null oids, meaning at most four of the six will be non-null.
606 * Also, since this is a rename, both src_entry and dst_entry will
607 * have at least one non-null oid, meaning at least two will be
608 * non-null. Of the six oids, a typical rename will have three be
609 * non-null. Only two implies a rename/delete, and four implies a
610 * rename/add.
612 struct stage_data *src_entry;
613 struct stage_data *dst_entry;
614 unsigned add_turned_into_rename:1;
615 unsigned processed:1;
618 static int update_stages(struct merge_options *opt, const char *path,
619 const struct diff_filespec *o,
620 const struct diff_filespec *a,
621 const struct diff_filespec *b)
625 * NOTE: It is usually a bad idea to call update_stages on a path
626 * before calling update_file on that same path, since it can
627 * sometimes lead to spurious "refusing to lose untracked file..."
628 * messages from update_file (via make_room_for path via
629 * would_lose_untracked). Instead, reverse the order of the calls
630 * (executing update_file first and then update_stages).
632 int clear = 1;
633 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
634 if (clear)
635 if (remove_file_from_cache(path))
636 return -1;
637 if (o)
638 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
639 return -1;
640 if (a)
641 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
642 return -1;
643 if (b)
644 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
645 return -1;
646 return 0;
649 static int update_stages_for_stage_data(struct merge_options *opt,
650 const char *path,
651 const struct stage_data *stage_data)
653 struct diff_filespec o, a, b;
655 o.mode = stage_data->stages[1].mode;
656 oidcpy(&o.oid, &stage_data->stages[1].oid);
658 a.mode = stage_data->stages[2].mode;
659 oidcpy(&a.oid, &stage_data->stages[2].oid);
661 b.mode = stage_data->stages[3].mode;
662 oidcpy(&b.oid, &stage_data->stages[3].oid);
664 return update_stages(opt, path,
665 is_null_oid(&o.oid) ? NULL : &o,
666 is_null_oid(&a.oid) ? NULL : &a,
667 is_null_oid(&b.oid) ? NULL : &b);
670 static void update_entry(struct stage_data *entry,
671 struct diff_filespec *o,
672 struct diff_filespec *a,
673 struct diff_filespec *b)
675 entry->processed = 0;
676 entry->stages[1].mode = o->mode;
677 entry->stages[2].mode = a->mode;
678 entry->stages[3].mode = b->mode;
679 oidcpy(&entry->stages[1].oid, &o->oid);
680 oidcpy(&entry->stages[2].oid, &a->oid);
681 oidcpy(&entry->stages[3].oid, &b->oid);
684 static int remove_file(struct merge_options *o, int clean,
685 const char *path, int no_wd)
687 int update_cache = o->call_depth || clean;
688 int update_working_directory = !o->call_depth && !no_wd;
690 if (update_cache) {
691 if (remove_file_from_cache(path))
692 return -1;
694 if (update_working_directory) {
695 if (ignore_case) {
696 struct cache_entry *ce;
697 ce = cache_file_exists(path, strlen(path), ignore_case);
698 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
699 return 0;
701 if (remove_path(path))
702 return -1;
704 return 0;
707 /* add a string to a strbuf, but converting "/" to "_" */
708 static void add_flattened_path(struct strbuf *out, const char *s)
710 size_t i = out->len;
711 strbuf_addstr(out, s);
712 for (; i < out->len; i++)
713 if (out->buf[i] == '/')
714 out->buf[i] = '_';
717 static char *unique_path(struct merge_options *o, const char *path, const char *branch)
719 struct path_hashmap_entry *entry;
720 struct strbuf newpath = STRBUF_INIT;
721 int suffix = 0;
722 size_t base_len;
724 strbuf_addf(&newpath, "%s~", path);
725 add_flattened_path(&newpath, branch);
727 base_len = newpath.len;
728 while (hashmap_get_from_hash(&o->current_file_dir_set,
729 path_hash(newpath.buf), newpath.buf) ||
730 (!o->call_depth && file_exists(newpath.buf))) {
731 strbuf_setlen(&newpath, base_len);
732 strbuf_addf(&newpath, "_%d", suffix++);
735 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
736 hashmap_entry_init(entry, path_hash(entry->path));
737 hashmap_add(&o->current_file_dir_set, entry);
738 return strbuf_detach(&newpath, NULL);
742 * Check whether a directory in the index is in the way of an incoming
743 * file. Return 1 if so. If check_working_copy is non-zero, also
744 * check the working directory. If empty_ok is non-zero, also return
745 * 0 in the case where the working-tree dir exists but is empty.
747 static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
749 int pos;
750 struct strbuf dirpath = STRBUF_INIT;
751 struct stat st;
753 strbuf_addstr(&dirpath, path);
754 strbuf_addch(&dirpath, '/');
756 pos = cache_name_pos(dirpath.buf, dirpath.len);
758 if (pos < 0)
759 pos = -1 - pos;
760 if (pos < active_nr &&
761 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
762 strbuf_release(&dirpath);
763 return 1;
766 strbuf_release(&dirpath);
767 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
768 !(empty_ok && is_empty_dir(path));
771 static int was_tracked(const char *path)
773 int pos = cache_name_pos(path, strlen(path));
775 if (0 <= pos)
776 /* we have been tracking this path */
777 return 1;
780 * Look for an unmerged entry for the path,
781 * specifically stage #2, which would indicate
782 * that "our" side before the merge started
783 * had the path tracked (and resulted in a conflict).
785 for (pos = -1 - pos;
786 pos < active_nr && !strcmp(path, active_cache[pos]->name);
787 pos++)
788 if (ce_stage(active_cache[pos]) == 2)
789 return 1;
790 return 0;
793 static int would_lose_untracked(const char *path)
795 return !was_tracked(path) && file_exists(path);
798 static int make_room_for_path(struct merge_options *o, const char *path)
800 int status, i;
801 const char *msg = _("failed to create path '%s'%s");
803 /* Unlink any D/F conflict files that are in the way */
804 for (i = 0; i < o->df_conflict_file_set.nr; i++) {
805 const char *df_path = o->df_conflict_file_set.items[i].string;
806 size_t pathlen = strlen(path);
807 size_t df_pathlen = strlen(df_path);
808 if (df_pathlen < pathlen &&
809 path[df_pathlen] == '/' &&
810 strncmp(path, df_path, df_pathlen) == 0) {
811 output(o, 3,
812 _("Removing %s to make room for subdirectory\n"),
813 df_path);
814 unlink(df_path);
815 unsorted_string_list_delete_item(&o->df_conflict_file_set,
816 i, 0);
817 break;
821 /* Make sure leading directories are created */
822 status = safe_create_leading_directories_const(path);
823 if (status) {
824 if (status == SCLD_EXISTS)
825 /* something else exists */
826 return err(o, msg, path, _(": perhaps a D/F conflict?"));
827 return err(o, msg, path, "");
831 * Do not unlink a file in the work tree if we are not
832 * tracking it.
834 if (would_lose_untracked(path))
835 return err(o, _("refusing to lose untracked file at '%s'"),
836 path);
838 /* Successful unlink is good.. */
839 if (!unlink(path))
840 return 0;
841 /* .. and so is no existing file */
842 if (errno == ENOENT)
843 return 0;
844 /* .. but not some other error (who really cares what?) */
845 return err(o, msg, path, _(": perhaps a D/F conflict?"));
848 static int update_file_flags(struct merge_options *o,
849 const struct object_id *oid,
850 unsigned mode,
851 const char *path,
852 int update_cache,
853 int update_wd)
855 int ret = 0;
857 if (o->call_depth)
858 update_wd = 0;
860 if (update_wd) {
861 enum object_type type;
862 void *buf;
863 unsigned long size;
865 if (S_ISGITLINK(mode)) {
867 * We may later decide to recursively descend into
868 * the submodule directory and update its index
869 * and/or work tree, but we do not do that now.
871 update_wd = 0;
872 goto update_index;
875 buf = read_object_file(oid, &type, &size);
876 if (!buf)
877 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
878 if (type != OBJ_BLOB) {
879 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
880 goto free_buf;
882 if (S_ISREG(mode)) {
883 struct strbuf strbuf = STRBUF_INIT;
884 if (convert_to_working_tree(path, buf, size, &strbuf)) {
885 free(buf);
886 size = strbuf.len;
887 buf = strbuf_detach(&strbuf, NULL);
891 if (make_room_for_path(o, path) < 0) {
892 update_wd = 0;
893 goto free_buf;
895 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
896 int fd;
897 if (mode & 0100)
898 mode = 0777;
899 else
900 mode = 0666;
901 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
902 if (fd < 0) {
903 ret = err(o, _("failed to open '%s': %s"),
904 path, strerror(errno));
905 goto free_buf;
907 write_in_full(fd, buf, size);
908 close(fd);
909 } else if (S_ISLNK(mode)) {
910 char *lnk = xmemdupz(buf, size);
911 safe_create_leading_directories_const(path);
912 unlink(path);
913 if (symlink(lnk, path))
914 ret = err(o, _("failed to symlink '%s': %s"),
915 path, strerror(errno));
916 free(lnk);
917 } else
918 ret = err(o,
919 _("do not know what to do with %06o %s '%s'"),
920 mode, oid_to_hex(oid), path);
921 free_buf:
922 free(buf);
924 update_index:
925 if (!ret && update_cache)
926 add_cacheinfo(o, mode, oid, path, 0, update_wd, ADD_CACHE_OK_TO_ADD);
927 return ret;
930 static int update_file(struct merge_options *o,
931 int clean,
932 const struct object_id *oid,
933 unsigned mode,
934 const char *path)
936 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
939 /* Low level file merging, update and removal */
941 struct merge_file_info {
942 struct object_id oid;
943 unsigned mode;
944 unsigned clean:1,
945 merge:1;
948 static int merge_3way(struct merge_options *o,
949 mmbuffer_t *result_buf,
950 const struct diff_filespec *one,
951 const struct diff_filespec *a,
952 const struct diff_filespec *b,
953 const char *branch1,
954 const char *branch2)
956 mmfile_t orig, src1, src2;
957 struct ll_merge_options ll_opts = {0};
958 char *base_name, *name1, *name2;
959 int merge_status;
961 ll_opts.renormalize = o->renormalize;
962 ll_opts.xdl_opts = o->xdl_opts;
964 if (o->call_depth) {
965 ll_opts.virtual_ancestor = 1;
966 ll_opts.variant = 0;
967 } else {
968 switch (o->recursive_variant) {
969 case MERGE_RECURSIVE_OURS:
970 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
971 break;
972 case MERGE_RECURSIVE_THEIRS:
973 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
974 break;
975 default:
976 ll_opts.variant = 0;
977 break;
981 if (strcmp(a->path, b->path) ||
982 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
983 base_name = o->ancestor == NULL ? NULL :
984 mkpathdup("%s:%s", o->ancestor, one->path);
985 name1 = mkpathdup("%s:%s", branch1, a->path);
986 name2 = mkpathdup("%s:%s", branch2, b->path);
987 } else {
988 base_name = o->ancestor == NULL ? NULL :
989 mkpathdup("%s", o->ancestor);
990 name1 = mkpathdup("%s", branch1);
991 name2 = mkpathdup("%s", branch2);
994 read_mmblob(&orig, &one->oid);
995 read_mmblob(&src1, &a->oid);
996 read_mmblob(&src2, &b->oid);
998 merge_status = ll_merge(result_buf, a->path, &orig, base_name,
999 &src1, name1, &src2, name2, &ll_opts);
1001 free(base_name);
1002 free(name1);
1003 free(name2);
1004 free(orig.ptr);
1005 free(src1.ptr);
1006 free(src2.ptr);
1007 return merge_status;
1010 static int merge_file_1(struct merge_options *o,
1011 const struct diff_filespec *one,
1012 const struct diff_filespec *a,
1013 const struct diff_filespec *b,
1014 const char *branch1,
1015 const char *branch2,
1016 struct merge_file_info *result)
1018 result->merge = 0;
1019 result->clean = 1;
1021 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1022 result->clean = 0;
1023 if (S_ISREG(a->mode)) {
1024 result->mode = a->mode;
1025 oidcpy(&result->oid, &a->oid);
1026 } else {
1027 result->mode = b->mode;
1028 oidcpy(&result->oid, &b->oid);
1030 } else {
1031 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1032 result->merge = 1;
1035 * Merge modes
1037 if (a->mode == b->mode || a->mode == one->mode)
1038 result->mode = b->mode;
1039 else {
1040 result->mode = a->mode;
1041 if (b->mode != one->mode) {
1042 result->clean = 0;
1043 result->merge = 1;
1047 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1048 oidcpy(&result->oid, &b->oid);
1049 else if (oid_eq(&b->oid, &one->oid))
1050 oidcpy(&result->oid, &a->oid);
1051 else if (S_ISREG(a->mode)) {
1052 mmbuffer_t result_buf;
1053 int ret = 0, merge_status;
1055 merge_status = merge_3way(o, &result_buf, one, a, b,
1056 branch1, branch2);
1058 if ((merge_status < 0) || !result_buf.ptr)
1059 ret = err(o, _("Failed to execute internal merge"));
1061 if (!ret &&
1062 write_object_file(result_buf.ptr, result_buf.size,
1063 blob_type, &result->oid))
1064 ret = err(o, _("Unable to add %s to database"),
1065 a->path);
1067 free(result_buf.ptr);
1068 if (ret)
1069 return ret;
1070 result->clean = (merge_status == 0);
1071 } else if (S_ISGITLINK(a->mode)) {
1072 result->clean = merge_submodule(&result->oid,
1073 one->path,
1074 &one->oid,
1075 &a->oid,
1076 &b->oid,
1077 !o->call_depth);
1078 } else if (S_ISLNK(a->mode)) {
1079 switch (o->recursive_variant) {
1080 case MERGE_RECURSIVE_NORMAL:
1081 oidcpy(&result->oid, &a->oid);
1082 if (!oid_eq(&a->oid, &b->oid))
1083 result->clean = 0;
1084 break;
1085 case MERGE_RECURSIVE_OURS:
1086 oidcpy(&result->oid, &a->oid);
1087 break;
1088 case MERGE_RECURSIVE_THEIRS:
1089 oidcpy(&result->oid, &b->oid);
1090 break;
1092 } else
1093 die("BUG: unsupported object type in the tree");
1096 return 0;
1099 static int merge_file_special_markers(struct merge_options *o,
1100 const struct diff_filespec *one,
1101 const struct diff_filespec *a,
1102 const struct diff_filespec *b,
1103 const char *branch1,
1104 const char *filename1,
1105 const char *branch2,
1106 const char *filename2,
1107 struct merge_file_info *mfi)
1109 char *side1 = NULL;
1110 char *side2 = NULL;
1111 int ret;
1113 if (filename1)
1114 side1 = xstrfmt("%s:%s", branch1, filename1);
1115 if (filename2)
1116 side2 = xstrfmt("%s:%s", branch2, filename2);
1118 ret = merge_file_1(o, one, a, b,
1119 side1 ? side1 : branch1,
1120 side2 ? side2 : branch2, mfi);
1121 free(side1);
1122 free(side2);
1123 return ret;
1126 static int merge_file_one(struct merge_options *o,
1127 const char *path,
1128 const struct object_id *o_oid, int o_mode,
1129 const struct object_id *a_oid, int a_mode,
1130 const struct object_id *b_oid, int b_mode,
1131 const char *branch1,
1132 const char *branch2,
1133 struct merge_file_info *mfi)
1135 struct diff_filespec one, a, b;
1137 one.path = a.path = b.path = (char *)path;
1138 oidcpy(&one.oid, o_oid);
1139 one.mode = o_mode;
1140 oidcpy(&a.oid, a_oid);
1141 a.mode = a_mode;
1142 oidcpy(&b.oid, b_oid);
1143 b.mode = b_mode;
1144 return merge_file_1(o, &one, &a, &b, branch1, branch2, mfi);
1147 static int conflict_rename_dir(struct merge_options *o,
1148 struct diff_filepair *pair,
1149 const char *rename_branch,
1150 const char *other_branch)
1152 const struct diff_filespec *dest = pair->two;
1154 if (!o->call_depth && would_lose_untracked(dest->path)) {
1155 char *alt_path = unique_path(o, dest->path, rename_branch);
1157 output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1158 "writing to %s instead."),
1159 dest->path, alt_path);
1161 * Write the file in worktree at alt_path, but not in the
1162 * index. Instead, write to dest->path for the index but
1163 * only at the higher appropriate stage.
1165 if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1166 return -1;
1167 free(alt_path);
1168 return update_stages(o, dest->path, NULL,
1169 rename_branch == o->branch1 ? dest : NULL,
1170 rename_branch == o->branch1 ? NULL : dest);
1173 /* Update dest->path both in index and in worktree */
1174 if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1175 return -1;
1176 return 0;
1179 static int handle_change_delete(struct merge_options *o,
1180 const char *path, const char *old_path,
1181 const struct object_id *o_oid, int o_mode,
1182 const struct object_id *changed_oid,
1183 int changed_mode,
1184 const char *change_branch,
1185 const char *delete_branch,
1186 const char *change, const char *change_past)
1188 char *alt_path = NULL;
1189 const char *update_path = path;
1190 int ret = 0;
1192 if (dir_in_way(path, !o->call_depth, 0) ||
1193 (!o->call_depth && would_lose_untracked(path))) {
1194 update_path = alt_path = unique_path(o, path, change_branch);
1197 if (o->call_depth) {
1199 * We cannot arbitrarily accept either a_sha or b_sha as
1200 * correct; since there is no true "middle point" between
1201 * them, simply reuse the base version for virtual merge base.
1203 ret = remove_file_from_cache(path);
1204 if (!ret)
1205 ret = update_file(o, 0, o_oid, o_mode, update_path);
1206 } else {
1207 if (!alt_path) {
1208 if (!old_path) {
1209 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1210 "and %s in %s. Version %s of %s left in tree."),
1211 change, path, delete_branch, change_past,
1212 change_branch, change_branch, path);
1213 } else {
1214 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1215 "and %s to %s in %s. Version %s of %s left in tree."),
1216 change, old_path, delete_branch, change_past, path,
1217 change_branch, change_branch, path);
1219 } else {
1220 if (!old_path) {
1221 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1222 "and %s in %s. Version %s of %s left in tree at %s."),
1223 change, path, delete_branch, change_past,
1224 change_branch, change_branch, path, alt_path);
1225 } else {
1226 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1227 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1228 change, old_path, delete_branch, change_past, path,
1229 change_branch, change_branch, path, alt_path);
1233 * No need to call update_file() on path when change_branch ==
1234 * o->branch1 && !alt_path, since that would needlessly touch
1235 * path. We could call update_file_flags() with update_cache=0
1236 * and update_wd=0, but that's a no-op.
1238 if (change_branch != o->branch1 || alt_path)
1239 ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1241 free(alt_path);
1243 return ret;
1246 static int conflict_rename_delete(struct merge_options *o,
1247 struct diff_filepair *pair,
1248 const char *rename_branch,
1249 const char *delete_branch)
1251 const struct diff_filespec *orig = pair->one;
1252 const struct diff_filespec *dest = pair->two;
1254 if (handle_change_delete(o,
1255 o->call_depth ? orig->path : dest->path,
1256 o->call_depth ? NULL : orig->path,
1257 &orig->oid, orig->mode,
1258 &dest->oid, dest->mode,
1259 rename_branch, delete_branch,
1260 _("rename"), _("renamed")))
1261 return -1;
1263 if (o->call_depth)
1264 return remove_file_from_cache(dest->path);
1265 else
1266 return update_stages(o, dest->path, NULL,
1267 rename_branch == o->branch1 ? dest : NULL,
1268 rename_branch == o->branch1 ? NULL : dest);
1271 static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1272 struct stage_data *entry,
1273 int stage)
1275 struct object_id *oid = &entry->stages[stage].oid;
1276 unsigned mode = entry->stages[stage].mode;
1277 if (mode == 0 || is_null_oid(oid))
1278 return NULL;
1279 oidcpy(&target->oid, oid);
1280 target->mode = mode;
1281 return target;
1284 static int handle_file(struct merge_options *o,
1285 struct diff_filespec *rename,
1286 int stage,
1287 struct rename_conflict_info *ci)
1289 char *dst_name = rename->path;
1290 struct stage_data *dst_entry;
1291 const char *cur_branch, *other_branch;
1292 struct diff_filespec other;
1293 struct diff_filespec *add;
1294 int ret;
1296 if (stage == 2) {
1297 dst_entry = ci->dst_entry1;
1298 cur_branch = ci->branch1;
1299 other_branch = ci->branch2;
1300 } else {
1301 dst_entry = ci->dst_entry2;
1302 cur_branch = ci->branch2;
1303 other_branch = ci->branch1;
1306 add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1307 if (add) {
1308 char *add_name = unique_path(o, rename->path, other_branch);
1309 if (update_file(o, 0, &add->oid, add->mode, add_name))
1310 return -1;
1312 remove_file(o, 0, rename->path, 0);
1313 dst_name = unique_path(o, rename->path, cur_branch);
1314 } else {
1315 if (dir_in_way(rename->path, !o->call_depth, 0)) {
1316 dst_name = unique_path(o, rename->path, cur_branch);
1317 output(o, 1, _("%s is a directory in %s adding as %s instead"),
1318 rename->path, other_branch, dst_name);
1319 } else if (!o->call_depth &&
1320 would_lose_untracked(rename->path)) {
1321 dst_name = unique_path(o, rename->path, cur_branch);
1322 output(o, 1, _("Refusing to lose untracked file at %s; "
1323 "adding as %s instead"),
1324 rename->path, dst_name);
1327 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1328 ; /* fall through, do allow dst_name to be released */
1329 else if (stage == 2)
1330 ret = update_stages(o, rename->path, NULL, rename, add);
1331 else
1332 ret = update_stages(o, rename->path, NULL, add, rename);
1334 if (dst_name != rename->path)
1335 free(dst_name);
1337 return ret;
1340 static int conflict_rename_rename_1to2(struct merge_options *o,
1341 struct rename_conflict_info *ci)
1343 /* One file was renamed in both branches, but to different names. */
1344 struct diff_filespec *one = ci->pair1->one;
1345 struct diff_filespec *a = ci->pair1->two;
1346 struct diff_filespec *b = ci->pair2->two;
1348 output(o, 1, _("CONFLICT (rename/rename): "
1349 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1350 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1351 one->path, a->path, ci->branch1,
1352 one->path, b->path, ci->branch2,
1353 o->call_depth ? _(" (left unresolved)") : "");
1354 if (o->call_depth) {
1355 struct merge_file_info mfi;
1356 struct diff_filespec other;
1357 struct diff_filespec *add;
1358 if (merge_file_one(o, one->path,
1359 &one->oid, one->mode,
1360 &a->oid, a->mode,
1361 &b->oid, b->mode,
1362 ci->branch1, ci->branch2, &mfi))
1363 return -1;
1366 * FIXME: For rename/add-source conflicts (if we could detect
1367 * such), this is wrong. We should instead find a unique
1368 * pathname and then either rename the add-source file to that
1369 * unique path, or use that unique path instead of src here.
1371 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1372 return -1;
1375 * Above, we put the merged content at the merge-base's
1376 * path. Now we usually need to delete both a->path and
1377 * b->path. However, the rename on each side of the merge
1378 * could also be involved in a rename/add conflict. In
1379 * such cases, we should keep the added file around,
1380 * resolving the conflict at that path in its favor.
1382 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1383 if (add) {
1384 if (update_file(o, 0, &add->oid, add->mode, a->path))
1385 return -1;
1387 else
1388 remove_file_from_cache(a->path);
1389 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1390 if (add) {
1391 if (update_file(o, 0, &add->oid, add->mode, b->path))
1392 return -1;
1394 else
1395 remove_file_from_cache(b->path);
1396 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1397 return -1;
1399 return 0;
1402 static int conflict_rename_rename_2to1(struct merge_options *o,
1403 struct rename_conflict_info *ci)
1405 /* Two files, a & b, were renamed to the same thing, c. */
1406 struct diff_filespec *a = ci->pair1->one;
1407 struct diff_filespec *b = ci->pair2->one;
1408 struct diff_filespec *c1 = ci->pair1->two;
1409 struct diff_filespec *c2 = ci->pair2->two;
1410 char *path = c1->path; /* == c2->path */
1411 struct merge_file_info mfi_c1;
1412 struct merge_file_info mfi_c2;
1413 int ret;
1415 output(o, 1, _("CONFLICT (rename/rename): "
1416 "Rename %s->%s in %s. "
1417 "Rename %s->%s in %s"),
1418 a->path, c1->path, ci->branch1,
1419 b->path, c2->path, ci->branch2);
1421 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1422 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1424 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1425 o->branch1, c1->path,
1426 o->branch2, ci->ren1_other.path, &mfi_c1) ||
1427 merge_file_special_markers(o, b, &ci->ren2_other, c2,
1428 o->branch1, ci->ren2_other.path,
1429 o->branch2, c2->path, &mfi_c2))
1430 return -1;
1432 if (o->call_depth) {
1434 * If mfi_c1.clean && mfi_c2.clean, then it might make
1435 * sense to do a two-way merge of those results. But, I
1436 * think in all cases, it makes sense to have the virtual
1437 * merge base just undo the renames; they can be detected
1438 * again later for the non-recursive merge.
1440 remove_file(o, 0, path, 0);
1441 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1442 if (!ret)
1443 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1444 b->path);
1445 } else {
1446 char *new_path1 = unique_path(o, path, ci->branch1);
1447 char *new_path2 = unique_path(o, path, ci->branch2);
1448 output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1449 a->path, new_path1, b->path, new_path2);
1450 if (would_lose_untracked(path))
1452 * Only way we get here is if both renames were from
1453 * a directory rename AND user had an untracked file
1454 * at the location where both files end up after the
1455 * two directory renames. See testcase 10d of t6043.
1457 output(o, 1, _("Refusing to lose untracked file at "
1458 "%s, even though it's in the way."),
1459 path);
1460 else
1461 remove_file(o, 0, path, 0);
1462 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1463 if (!ret)
1464 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1465 new_path2);
1467 * unpack_trees() actually populates the index for us for
1468 * "normal" rename/rename(2to1) situtations so that the
1469 * correct entries are at the higher stages, which would
1470 * make the call below to update_stages_for_stage_data
1471 * unnecessary. However, if either of the renames came
1472 * from a directory rename, then unpack_trees() will not
1473 * have gotten the right data loaded into the index, so we
1474 * need to do so now. (While it'd be tempting to move this
1475 * call to update_stages_for_stage_data() to
1476 * apply_directory_rename_modifications(), that would break
1477 * our intermediate calls to would_lose_untracked() since
1478 * those rely on the current in-memory index. See also the
1479 * big "NOTE" in update_stages()).
1481 if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1482 ret = -1;
1484 free(new_path2);
1485 free(new_path1);
1488 return ret;
1492 * Get the diff_filepairs changed between o_tree and tree.
1494 static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1495 struct tree *o_tree,
1496 struct tree *tree)
1498 struct diff_queue_struct *ret;
1499 struct diff_options opts;
1501 diff_setup(&opts);
1502 opts.flags.recursive = 1;
1503 opts.flags.rename_empty = 0;
1504 opts.detect_rename = DIFF_DETECT_RENAME;
1505 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1506 o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1507 1000;
1508 opts.rename_score = o->rename_score;
1509 opts.show_rename_progress = o->show_rename_progress;
1510 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1511 diff_setup_done(&opts);
1512 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1513 diffcore_std(&opts);
1514 if (opts.needed_rename_limit > o->needed_rename_limit)
1515 o->needed_rename_limit = opts.needed_rename_limit;
1517 ret = xmalloc(sizeof(*ret));
1518 *ret = diff_queued_diff;
1520 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1521 diff_queued_diff.nr = 0;
1522 diff_queued_diff.queue = NULL;
1523 diff_flush(&opts);
1524 return ret;
1527 static int tree_has_path(struct tree *tree, const char *path)
1529 struct object_id hashy;
1530 unsigned int mode_o;
1532 return !get_tree_entry(&tree->object.oid, path,
1533 &hashy, &mode_o);
1537 * Return a new string that replaces the beginning portion (which matches
1538 * entry->dir), with entry->new_dir. In perl-speak:
1539 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1540 * NOTE:
1541 * Caller must ensure that old_path starts with entry->dir + '/'.
1543 static char *apply_dir_rename(struct dir_rename_entry *entry,
1544 const char *old_path)
1546 struct strbuf new_path = STRBUF_INIT;
1547 int oldlen, newlen;
1549 if (entry->non_unique_new_dir)
1550 return NULL;
1552 oldlen = strlen(entry->dir);
1553 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1554 strbuf_grow(&new_path, newlen);
1555 strbuf_addbuf(&new_path, &entry->new_dir);
1556 strbuf_addstr(&new_path, &old_path[oldlen]);
1558 return strbuf_detach(&new_path, NULL);
1561 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1562 char **old_dir, char **new_dir)
1564 char *end_of_old, *end_of_new;
1565 int old_len, new_len;
1567 *old_dir = NULL;
1568 *new_dir = NULL;
1571 * For
1572 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1573 * the "e/foo.c" part is the same, we just want to know that
1574 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1575 * so, for this example, this function returns "a/b/c/d" in
1576 * *old_dir and "a/b/some/thing/else" in *new_dir.
1578 * Also, if the basename of the file changed, we don't care. We
1579 * want to know which portion of the directory, if any, changed.
1581 end_of_old = strrchr(old_path, '/');
1582 end_of_new = strrchr(new_path, '/');
1584 if (end_of_old == NULL || end_of_new == NULL)
1585 return;
1586 while (*--end_of_new == *--end_of_old &&
1587 end_of_old != old_path &&
1588 end_of_new != new_path)
1589 ; /* Do nothing; all in the while loop */
1591 * We've found the first non-matching character in the directory
1592 * paths. That means the current directory we were comparing
1593 * represents the rename. Move end_of_old and end_of_new back
1594 * to the full directory name.
1596 if (*end_of_old == '/')
1597 end_of_old++;
1598 if (*end_of_old != '/')
1599 end_of_new++;
1600 end_of_old = strchr(end_of_old, '/');
1601 end_of_new = strchr(end_of_new, '/');
1604 * It may have been the case that old_path and new_path were the same
1605 * directory all along. Don't claim a rename if they're the same.
1607 old_len = end_of_old - old_path;
1608 new_len = end_of_new - new_path;
1610 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1611 *old_dir = xstrndup(old_path, old_len);
1612 *new_dir = xstrndup(new_path, new_len);
1616 static void remove_hashmap_entries(struct hashmap *dir_renames,
1617 struct string_list *items_to_remove)
1619 int i;
1620 struct dir_rename_entry *entry;
1622 for (i = 0; i < items_to_remove->nr; i++) {
1623 entry = items_to_remove->items[i].util;
1624 hashmap_remove(dir_renames, entry, NULL);
1626 string_list_clear(items_to_remove, 0);
1630 * See if there is a directory rename for path, and if there are any file
1631 * level conflicts for the renamed location. If there is a rename and
1632 * there are no conflicts, return the new name. Otherwise, return NULL.
1634 static char *handle_path_level_conflicts(struct merge_options *o,
1635 const char *path,
1636 struct dir_rename_entry *entry,
1637 struct hashmap *collisions,
1638 struct tree *tree)
1640 char *new_path = NULL;
1641 struct collision_entry *collision_ent;
1642 int clean = 1;
1643 struct strbuf collision_paths = STRBUF_INIT;
1646 * entry has the mapping of old directory name to new directory name
1647 * that we want to apply to path.
1649 new_path = apply_dir_rename(entry, path);
1651 if (!new_path) {
1652 /* This should only happen when entry->non_unique_new_dir set */
1653 if (!entry->non_unique_new_dir)
1654 BUG("entry->non_unqiue_dir not set and !new_path");
1655 output(o, 1, _("CONFLICT (directory rename split): "
1656 "Unclear where to place %s because directory "
1657 "%s was renamed to multiple other directories, "
1658 "with no destination getting a majority of the "
1659 "files."),
1660 path, entry->dir);
1661 clean = 0;
1662 return NULL;
1666 * The caller needs to have ensured that it has pre-populated
1667 * collisions with all paths that map to new_path. Do a quick check
1668 * to ensure that's the case.
1670 collision_ent = collision_find_entry(collisions, new_path);
1671 if (collision_ent == NULL)
1672 BUG("collision_ent is NULL");
1675 * Check for one-sided add/add/.../add conflicts, i.e.
1676 * where implicit renames from the other side doing
1677 * directory rename(s) can affect this side of history
1678 * to put multiple paths into the same location. Warn
1679 * and bail on directory renames for such paths.
1681 if (collision_ent->reported_already) {
1682 clean = 0;
1683 } else if (tree_has_path(tree, new_path)) {
1684 collision_ent->reported_already = 1;
1685 strbuf_add_separated_string_list(&collision_paths, ", ",
1686 &collision_ent->source_files);
1687 output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1688 "file/dir at %s in the way of implicit "
1689 "directory rename(s) putting the following "
1690 "path(s) there: %s."),
1691 new_path, collision_paths.buf);
1692 clean = 0;
1693 } else if (collision_ent->source_files.nr > 1) {
1694 collision_ent->reported_already = 1;
1695 strbuf_add_separated_string_list(&collision_paths, ", ",
1696 &collision_ent->source_files);
1697 output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1698 "more than one path to %s; implicit directory "
1699 "renames tried to put these paths there: %s"),
1700 new_path, collision_paths.buf);
1701 clean = 0;
1704 /* Free memory we no longer need */
1705 strbuf_release(&collision_paths);
1706 if (!clean && new_path) {
1707 free(new_path);
1708 return NULL;
1711 return new_path;
1715 * There are a couple things we want to do at the directory level:
1716 * 1. Check for both sides renaming to the same thing, in order to avoid
1717 * implicit renaming of files that should be left in place. (See
1718 * testcase 6b in t6043 for details.)
1719 * 2. Prune directory renames if there are still files left in the
1720 * the original directory. These represent a partial directory rename,
1721 * i.e. a rename where only some of the files within the directory
1722 * were renamed elsewhere. (Technically, this could be done earlier
1723 * in get_directory_renames(), except that would prevent us from
1724 * doing the previous check and thus failing testcase 6b.)
1725 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
1726 * In the future, we could potentially record this info as well and
1727 * omit reporting rename/rename(1to2) conflicts for each path within
1728 * the affected directories, thus cleaning up the merge output.
1729 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1730 * directory level, because merging directories is fine. If it
1731 * causes conflicts for files within those merged directories, then
1732 * that should be detected at the individual path level.
1734 static void handle_directory_level_conflicts(struct merge_options *o,
1735 struct hashmap *dir_re_head,
1736 struct tree *head,
1737 struct hashmap *dir_re_merge,
1738 struct tree *merge)
1740 struct hashmap_iter iter;
1741 struct dir_rename_entry *head_ent;
1742 struct dir_rename_entry *merge_ent;
1744 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1745 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1747 hashmap_iter_init(dir_re_head, &iter);
1748 while ((head_ent = hashmap_iter_next(&iter))) {
1749 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1750 if (merge_ent &&
1751 !head_ent->non_unique_new_dir &&
1752 !merge_ent->non_unique_new_dir &&
1753 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1754 /* 1. Renamed identically; remove it from both sides */
1755 string_list_append(&remove_from_head,
1756 head_ent->dir)->util = head_ent;
1757 strbuf_release(&head_ent->new_dir);
1758 string_list_append(&remove_from_merge,
1759 merge_ent->dir)->util = merge_ent;
1760 strbuf_release(&merge_ent->new_dir);
1761 } else if (tree_has_path(head, head_ent->dir)) {
1762 /* 2. This wasn't a directory rename after all */
1763 string_list_append(&remove_from_head,
1764 head_ent->dir)->util = head_ent;
1765 strbuf_release(&head_ent->new_dir);
1769 remove_hashmap_entries(dir_re_head, &remove_from_head);
1770 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1772 hashmap_iter_init(dir_re_merge, &iter);
1773 while ((merge_ent = hashmap_iter_next(&iter))) {
1774 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1775 if (tree_has_path(merge, merge_ent->dir)) {
1776 /* 2. This wasn't a directory rename after all */
1777 string_list_append(&remove_from_merge,
1778 merge_ent->dir)->util = merge_ent;
1779 } else if (head_ent &&
1780 !head_ent->non_unique_new_dir &&
1781 !merge_ent->non_unique_new_dir) {
1782 /* 3. rename/rename(1to2) */
1784 * We can assume it's not rename/rename(1to1) because
1785 * that was case (1), already checked above. So we
1786 * know that head_ent->new_dir and merge_ent->new_dir
1787 * are different strings.
1789 output(o, 1, _("CONFLICT (rename/rename): "
1790 "Rename directory %s->%s in %s. "
1791 "Rename directory %s->%s in %s"),
1792 head_ent->dir, head_ent->new_dir.buf, o->branch1,
1793 head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1794 string_list_append(&remove_from_head,
1795 head_ent->dir)->util = head_ent;
1796 strbuf_release(&head_ent->new_dir);
1797 string_list_append(&remove_from_merge,
1798 merge_ent->dir)->util = merge_ent;
1799 strbuf_release(&merge_ent->new_dir);
1803 remove_hashmap_entries(dir_re_head, &remove_from_head);
1804 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1807 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1808 struct tree *tree)
1810 struct hashmap *dir_renames;
1811 struct hashmap_iter iter;
1812 struct dir_rename_entry *entry;
1813 int i;
1816 * Typically, we think of a directory rename as all files from a
1817 * certain directory being moved to a target directory. However,
1818 * what if someone first moved two files from the original
1819 * directory in one commit, and then renamed the directory
1820 * somewhere else in a later commit? At merge time, we just know
1821 * that files from the original directory went to two different
1822 * places, and that the bulk of them ended up in the same place.
1823 * We want each directory rename to represent where the bulk of the
1824 * files from that directory end up; this function exists to find
1825 * where the bulk of the files went.
1827 * The first loop below simply iterates through the list of file
1828 * renames, finding out how often each directory rename pair
1829 * possibility occurs.
1831 dir_renames = xmalloc(sizeof(*dir_renames));
1832 dir_rename_init(dir_renames);
1833 for (i = 0; i < pairs->nr; ++i) {
1834 struct string_list_item *item;
1835 int *count;
1836 struct diff_filepair *pair = pairs->queue[i];
1837 char *old_dir, *new_dir;
1839 /* File not part of directory rename if it wasn't renamed */
1840 if (pair->status != 'R')
1841 continue;
1843 get_renamed_dir_portion(pair->one->path, pair->two->path,
1844 &old_dir, &new_dir);
1845 if (!old_dir)
1846 /* Directory didn't change at all; ignore this one. */
1847 continue;
1849 entry = dir_rename_find_entry(dir_renames, old_dir);
1850 if (!entry) {
1851 entry = xmalloc(sizeof(*entry));
1852 dir_rename_entry_init(entry, old_dir);
1853 hashmap_put(dir_renames, entry);
1854 } else {
1855 free(old_dir);
1857 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1858 if (!item) {
1859 item = string_list_insert(&entry->possible_new_dirs,
1860 new_dir);
1861 item->util = xcalloc(1, sizeof(int));
1862 } else {
1863 free(new_dir);
1865 count = item->util;
1866 *count += 1;
1870 * For each directory with files moved out of it, we find out which
1871 * target directory received the most files so we can declare it to
1872 * be the "winning" target location for the directory rename. This
1873 * winner gets recorded in new_dir. If there is no winner
1874 * (multiple target directories received the same number of files),
1875 * we set non_unique_new_dir. Once we've determined the winner (or
1876 * that there is no winner), we no longer need possible_new_dirs.
1878 hashmap_iter_init(dir_renames, &iter);
1879 while ((entry = hashmap_iter_next(&iter))) {
1880 int max = 0;
1881 int bad_max = 0;
1882 char *best = NULL;
1884 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1885 int *count = entry->possible_new_dirs.items[i].util;
1887 if (*count == max)
1888 bad_max = max;
1889 else if (*count > max) {
1890 max = *count;
1891 best = entry->possible_new_dirs.items[i].string;
1894 if (bad_max == max)
1895 entry->non_unique_new_dir = 1;
1896 else {
1897 assert(entry->new_dir.len == 0);
1898 strbuf_addstr(&entry->new_dir, best);
1901 * The relevant directory sub-portion of the original full
1902 * filepaths were xstrndup'ed before inserting into
1903 * possible_new_dirs, and instead of manually iterating the
1904 * list and free'ing each, just lie and tell
1905 * possible_new_dirs that it did the strdup'ing so that it
1906 * will free them for us.
1908 entry->possible_new_dirs.strdup_strings = 1;
1909 string_list_clear(&entry->possible_new_dirs, 1);
1912 return dir_renames;
1915 static struct dir_rename_entry *check_dir_renamed(const char *path,
1916 struct hashmap *dir_renames)
1918 char temp[PATH_MAX];
1919 char *end;
1920 struct dir_rename_entry *entry;
1922 strcpy(temp, path);
1923 while ((end = strrchr(temp, '/'))) {
1924 *end = '\0';
1925 entry = dir_rename_find_entry(dir_renames, temp);
1926 if (entry)
1927 return entry;
1929 return NULL;
1932 static void compute_collisions(struct hashmap *collisions,
1933 struct hashmap *dir_renames,
1934 struct diff_queue_struct *pairs)
1936 int i;
1939 * Multiple files can be mapped to the same path due to directory
1940 * renames done by the other side of history. Since that other
1941 * side of history could have merged multiple directories into one,
1942 * if our side of history added the same file basename to each of
1943 * those directories, then all N of them would get implicitly
1944 * renamed by the directory rename detection into the same path,
1945 * and we'd get an add/add/.../add conflict, and all those adds
1946 * from *this* side of history. This is not representable in the
1947 * index, and users aren't going to easily be able to make sense of
1948 * it. So we need to provide a good warning about what's
1949 * happening, and fall back to no-directory-rename detection
1950 * behavior for those paths.
1952 * See testcases 9e and all of section 5 from t6043 for examples.
1954 collision_init(collisions);
1956 for (i = 0; i < pairs->nr; ++i) {
1957 struct dir_rename_entry *dir_rename_ent;
1958 struct collision_entry *collision_ent;
1959 char *new_path;
1960 struct diff_filepair *pair = pairs->queue[i];
1962 if (pair->status == 'D')
1963 continue;
1964 dir_rename_ent = check_dir_renamed(pair->two->path,
1965 dir_renames);
1966 if (!dir_rename_ent)
1967 continue;
1969 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
1970 if (!new_path)
1972 * dir_rename_ent->non_unique_new_path is true, which
1973 * means there is no directory rename for us to use,
1974 * which means it won't cause us any additional
1975 * collisions.
1977 continue;
1978 collision_ent = collision_find_entry(collisions, new_path);
1979 if (!collision_ent) {
1980 collision_ent = xcalloc(1,
1981 sizeof(struct collision_entry));
1982 hashmap_entry_init(collision_ent, strhash(new_path));
1983 hashmap_put(collisions, collision_ent);
1984 collision_ent->target_file = new_path;
1985 } else {
1986 free(new_path);
1988 string_list_insert(&collision_ent->source_files,
1989 pair->two->path);
1993 static char *check_for_directory_rename(struct merge_options *o,
1994 const char *path,
1995 struct tree *tree,
1996 struct hashmap *dir_renames,
1997 struct hashmap *dir_rename_exclusions,
1998 struct hashmap *collisions,
1999 int *clean_merge)
2001 char *new_path = NULL;
2002 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2003 struct dir_rename_entry *oentry = NULL;
2005 if (!entry)
2006 return new_path;
2009 * This next part is a little weird. We do not want to do an
2010 * implicit rename into a directory we renamed on our side, because
2011 * that will result in a spurious rename/rename(1to2) conflict. An
2012 * example:
2013 * Base commit: dumbdir/afile, otherdir/bfile
2014 * Side 1: smrtdir/afile, otherdir/bfile
2015 * Side 2: dumbdir/afile, dumbdir/bfile
2016 * Here, while working on Side 1, we could notice that otherdir was
2017 * renamed/merged to dumbdir, and change the diff_filepair for
2018 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2019 * 2 will notice the rename from dumbdir to smrtdir, and do the
2020 * transitive rename to move it from dumbdir/bfile to
2021 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2022 * smrtdir, a rename/rename(1to2) conflict. We really just want
2023 * the file to end up in smrtdir. And the way to achieve that is
2024 * to not let Side1 do the rename to dumbdir, since we know that is
2025 * the source of one of our directory renames.
2027 * That's why oentry and dir_rename_exclusions is here.
2029 * As it turns out, this also prevents N-way transient rename
2030 * confusion; See testcases 9c and 9d of t6043.
2032 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2033 if (oentry) {
2034 output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2035 "to %s, because %s itself was renamed."),
2036 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2037 } else {
2038 new_path = handle_path_level_conflicts(o, path, entry,
2039 collisions, tree);
2040 *clean_merge &= (new_path != NULL);
2043 return new_path;
2046 static void apply_directory_rename_modifications(struct merge_options *o,
2047 struct diff_filepair *pair,
2048 char *new_path,
2049 struct rename *re,
2050 struct tree *tree,
2051 struct tree *o_tree,
2052 struct tree *a_tree,
2053 struct tree *b_tree,
2054 struct string_list *entries,
2055 int *clean)
2057 struct string_list_item *item;
2058 int stage = (tree == a_tree ? 2 : 3);
2061 * In all cases where we can do directory rename detection,
2062 * unpack_trees() will have read pair->two->path into the
2063 * index and the working copy. We need to remove it so that
2064 * we can instead place it at new_path. It is guaranteed to
2065 * not be untracked (unpack_trees() would have errored out
2066 * saying the file would have been overwritten), but it might
2067 * be dirty, though.
2069 remove_file(o, 1, pair->two->path, 0 /* no_wd */);
2071 /* Find or create a new re->dst_entry */
2072 item = string_list_lookup(entries, new_path);
2073 if (item) {
2075 * Since we're renaming on this side of history, and it's
2076 * due to a directory rename on the other side of history
2077 * (which we only allow when the directory in question no
2078 * longer exists on the other side of history), the
2079 * original entry for re->dst_entry is no longer
2080 * necessary...
2082 re->dst_entry->processed = 1;
2085 * ...because we'll be using this new one.
2087 re->dst_entry = item->util;
2088 } else {
2090 * re->dst_entry is for the before-dir-rename path, and we
2091 * need it to hold information for the after-dir-rename
2092 * path. Before creating a new entry, we need to mark the
2093 * old one as unnecessary (...unless it is shared by
2094 * src_entry, i.e. this didn't use to be a rename, in which
2095 * case we can just allow the normal processing to happen
2096 * for it).
2098 if (pair->status == 'R')
2099 re->dst_entry->processed = 1;
2101 re->dst_entry = insert_stage_data(new_path,
2102 o_tree, a_tree, b_tree,
2103 entries);
2104 item = string_list_insert(entries, new_path);
2105 item->util = re->dst_entry;
2109 * Update the stage_data with the information about the path we are
2110 * moving into place. That slot will be empty and available for us
2111 * to write to because of the collision checks in
2112 * handle_path_level_conflicts(). In other words,
2113 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2114 * open for us to write to.
2116 * It may be tempting to actually update the index at this point as
2117 * well, using update_stages_for_stage_data(), but as per the big
2118 * "NOTE" in update_stages(), doing so will modify the current
2119 * in-memory index which will break calls to would_lose_untracked()
2120 * that we need to make. Instead, we need to just make sure that
2121 * the various conflict_rename_*() functions update the index
2122 * explicitly rather than relying on unpack_trees() to have done it.
2124 get_tree_entry(&tree->object.oid,
2125 pair->two->path,
2126 &re->dst_entry->stages[stage].oid,
2127 &re->dst_entry->stages[stage].mode);
2129 /* Update pair status */
2130 if (pair->status == 'A') {
2132 * Recording rename information for this add makes it look
2133 * like a rename/delete conflict. Make sure we can
2134 * correctly handle this as an add that was moved to a new
2135 * directory instead of reporting a rename/delete conflict.
2137 re->add_turned_into_rename = 1;
2140 * We don't actually look at pair->status again, but it seems
2141 * pedagogically correct to adjust it.
2143 pair->status = 'R';
2146 * Finally, record the new location.
2148 pair->two->path = new_path;
2152 * Get information of all renames which occurred in 'pairs', making use of
2153 * any implicit directory renames inferred from the other side of history.
2154 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2155 * to be able to associate the correct cache entries with the rename
2156 * information; tree is always equal to either a_tree or b_tree.
2158 static struct string_list *get_renames(struct merge_options *o,
2159 struct diff_queue_struct *pairs,
2160 struct hashmap *dir_renames,
2161 struct hashmap *dir_rename_exclusions,
2162 struct tree *tree,
2163 struct tree *o_tree,
2164 struct tree *a_tree,
2165 struct tree *b_tree,
2166 struct string_list *entries,
2167 int *clean_merge)
2169 int i;
2170 struct hashmap collisions;
2171 struct hashmap_iter iter;
2172 struct collision_entry *e;
2173 struct string_list *renames;
2175 compute_collisions(&collisions, dir_renames, pairs);
2176 renames = xcalloc(1, sizeof(struct string_list));
2178 for (i = 0; i < pairs->nr; ++i) {
2179 struct string_list_item *item;
2180 struct rename *re;
2181 struct diff_filepair *pair = pairs->queue[i];
2182 char *new_path; /* non-NULL only with directory renames */
2184 if (pair->status == 'D') {
2185 diff_free_filepair(pair);
2186 continue;
2188 new_path = check_for_directory_rename(o, pair->two->path, tree,
2189 dir_renames,
2190 dir_rename_exclusions,
2191 &collisions,
2192 clean_merge);
2193 if (pair->status != 'R' && !new_path) {
2194 diff_free_filepair(pair);
2195 continue;
2198 re = xmalloc(sizeof(*re));
2199 re->processed = 0;
2200 re->add_turned_into_rename = 0;
2201 re->pair = pair;
2202 item = string_list_lookup(entries, re->pair->one->path);
2203 if (!item)
2204 re->src_entry = insert_stage_data(re->pair->one->path,
2205 o_tree, a_tree, b_tree, entries);
2206 else
2207 re->src_entry = item->util;
2209 item = string_list_lookup(entries, re->pair->two->path);
2210 if (!item)
2211 re->dst_entry = insert_stage_data(re->pair->two->path,
2212 o_tree, a_tree, b_tree, entries);
2213 else
2214 re->dst_entry = item->util;
2215 item = string_list_insert(renames, pair->one->path);
2216 item->util = re;
2217 if (new_path)
2218 apply_directory_rename_modifications(o, pair, new_path,
2219 re, tree, o_tree,
2220 a_tree, b_tree,
2221 entries,
2222 clean_merge);
2225 hashmap_iter_init(&collisions, &iter);
2226 while ((e = hashmap_iter_next(&iter))) {
2227 free(e->target_file);
2228 string_list_clear(&e->source_files, 0);
2230 hashmap_free(&collisions, 1);
2231 return renames;
2234 static int process_renames(struct merge_options *o,
2235 struct string_list *a_renames,
2236 struct string_list *b_renames)
2238 int clean_merge = 1, i, j;
2239 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2240 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2241 const struct rename *sre;
2243 for (i = 0; i < a_renames->nr; i++) {
2244 sre = a_renames->items[i].util;
2245 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2246 = (void *)sre;
2248 for (i = 0; i < b_renames->nr; i++) {
2249 sre = b_renames->items[i].util;
2250 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2251 = (void *)sre;
2254 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2255 struct string_list *renames1, *renames2Dst;
2256 struct rename *ren1 = NULL, *ren2 = NULL;
2257 const char *branch1, *branch2;
2258 const char *ren1_src, *ren1_dst;
2259 struct string_list_item *lookup;
2261 if (i >= a_renames->nr) {
2262 ren2 = b_renames->items[j++].util;
2263 } else if (j >= b_renames->nr) {
2264 ren1 = a_renames->items[i++].util;
2265 } else {
2266 int compare = strcmp(a_renames->items[i].string,
2267 b_renames->items[j].string);
2268 if (compare <= 0)
2269 ren1 = a_renames->items[i++].util;
2270 if (compare >= 0)
2271 ren2 = b_renames->items[j++].util;
2274 /* TODO: refactor, so that 1/2 are not needed */
2275 if (ren1) {
2276 renames1 = a_renames;
2277 renames2Dst = &b_by_dst;
2278 branch1 = o->branch1;
2279 branch2 = o->branch2;
2280 } else {
2281 renames1 = b_renames;
2282 renames2Dst = &a_by_dst;
2283 branch1 = o->branch2;
2284 branch2 = o->branch1;
2285 SWAP(ren2, ren1);
2288 if (ren1->processed)
2289 continue;
2290 ren1->processed = 1;
2291 ren1->dst_entry->processed = 1;
2292 /* BUG: We should only mark src_entry as processed if we
2293 * are not dealing with a rename + add-source case.
2295 ren1->src_entry->processed = 1;
2297 ren1_src = ren1->pair->one->path;
2298 ren1_dst = ren1->pair->two->path;
2300 if (ren2) {
2301 /* One file renamed on both sides */
2302 const char *ren2_src = ren2->pair->one->path;
2303 const char *ren2_dst = ren2->pair->two->path;
2304 enum rename_type rename_type;
2305 if (strcmp(ren1_src, ren2_src) != 0)
2306 die("BUG: ren1_src != ren2_src");
2307 ren2->dst_entry->processed = 1;
2308 ren2->processed = 1;
2309 if (strcmp(ren1_dst, ren2_dst) != 0) {
2310 rename_type = RENAME_ONE_FILE_TO_TWO;
2311 clean_merge = 0;
2312 } else {
2313 rename_type = RENAME_ONE_FILE_TO_ONE;
2314 /* BUG: We should only remove ren1_src in
2315 * the base stage (think of rename +
2316 * add-source cases).
2318 remove_file(o, 1, ren1_src, 1);
2319 update_entry(ren1->dst_entry,
2320 ren1->pair->one,
2321 ren1->pair->two,
2322 ren2->pair->two);
2324 setup_rename_conflict_info(rename_type,
2325 ren1->pair,
2326 ren2->pair,
2327 branch1,
2328 branch2,
2329 ren1->dst_entry,
2330 ren2->dst_entry,
2332 NULL,
2333 NULL);
2334 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2335 /* Two different files renamed to the same thing */
2336 char *ren2_dst;
2337 ren2 = lookup->util;
2338 ren2_dst = ren2->pair->two->path;
2339 if (strcmp(ren1_dst, ren2_dst) != 0)
2340 die("BUG: ren1_dst != ren2_dst");
2342 clean_merge = 0;
2343 ren2->processed = 1;
2345 * BUG: We should only mark src_entry as processed
2346 * if we are not dealing with a rename + add-source
2347 * case.
2349 ren2->src_entry->processed = 1;
2351 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2352 ren1->pair,
2353 ren2->pair,
2354 branch1,
2355 branch2,
2356 ren1->dst_entry,
2357 ren2->dst_entry,
2359 ren1->src_entry,
2360 ren2->src_entry);
2362 } else {
2363 /* Renamed in 1, maybe changed in 2 */
2364 /* we only use sha1 and mode of these */
2365 struct diff_filespec src_other, dst_other;
2366 int try_merge;
2369 * unpack_trees loads entries from common-commit
2370 * into stage 1, from head-commit into stage 2, and
2371 * from merge-commit into stage 3. We keep track
2372 * of which side corresponds to the rename.
2374 int renamed_stage = a_renames == renames1 ? 2 : 3;
2375 int other_stage = a_renames == renames1 ? 3 : 2;
2377 /* BUG: We should only remove ren1_src in the base
2378 * stage and in other_stage (think of rename +
2379 * add-source case).
2381 remove_file(o, 1, ren1_src,
2382 renamed_stage == 2 || !was_tracked(ren1_src));
2384 oidcpy(&src_other.oid,
2385 &ren1->src_entry->stages[other_stage].oid);
2386 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2387 oidcpy(&dst_other.oid,
2388 &ren1->dst_entry->stages[other_stage].oid);
2389 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2390 try_merge = 0;
2392 if (oid_eq(&src_other.oid, &null_oid) &&
2393 ren1->add_turned_into_rename) {
2394 setup_rename_conflict_info(RENAME_DIR,
2395 ren1->pair,
2396 NULL,
2397 branch1,
2398 branch2,
2399 ren1->dst_entry,
2400 NULL,
2402 NULL,
2403 NULL);
2404 } else if (oid_eq(&src_other.oid, &null_oid)) {
2405 setup_rename_conflict_info(RENAME_DELETE,
2406 ren1->pair,
2407 NULL,
2408 branch1,
2409 branch2,
2410 ren1->dst_entry,
2411 NULL,
2413 NULL,
2414 NULL);
2415 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2416 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2418 * Added file on the other side identical to
2419 * the file being renamed: clean merge.
2420 * Also, there is no need to overwrite the
2421 * file already in the working copy, so call
2422 * update_file_flags() instead of
2423 * update_file().
2425 if (update_file_flags(o,
2426 &ren1->pair->two->oid,
2427 ren1->pair->two->mode,
2428 ren1_dst,
2429 1, /* update_cache */
2430 0 /* update_wd */))
2431 clean_merge = -1;
2432 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2433 clean_merge = 0;
2434 try_merge = 1;
2435 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2436 "%s added in %s"),
2437 ren1_src, ren1_dst, branch1,
2438 ren1_dst, branch2);
2439 if (o->call_depth) {
2440 struct merge_file_info mfi;
2441 if (merge_file_one(o, ren1_dst, &null_oid, 0,
2442 &ren1->pair->two->oid,
2443 ren1->pair->two->mode,
2444 &dst_other.oid,
2445 dst_other.mode,
2446 branch1, branch2, &mfi)) {
2447 clean_merge = -1;
2448 goto cleanup_and_return;
2450 output(o, 1, _("Adding merged %s"), ren1_dst);
2451 if (update_file(o, 0, &mfi.oid,
2452 mfi.mode, ren1_dst))
2453 clean_merge = -1;
2454 try_merge = 0;
2455 } else {
2456 char *new_path = unique_path(o, ren1_dst, branch2);
2457 output(o, 1, _("Adding as %s instead"), new_path);
2458 if (update_file(o, 0, &dst_other.oid,
2459 dst_other.mode, new_path))
2460 clean_merge = -1;
2461 free(new_path);
2463 } else
2464 try_merge = 1;
2466 if (clean_merge < 0)
2467 goto cleanup_and_return;
2468 if (try_merge) {
2469 struct diff_filespec *one, *a, *b;
2470 src_other.path = (char *)ren1_src;
2472 one = ren1->pair->one;
2473 if (a_renames == renames1) {
2474 a = ren1->pair->two;
2475 b = &src_other;
2476 } else {
2477 b = ren1->pair->two;
2478 a = &src_other;
2480 update_entry(ren1->dst_entry, one, a, b);
2481 setup_rename_conflict_info(RENAME_NORMAL,
2482 ren1->pair,
2483 NULL,
2484 branch1,
2485 NULL,
2486 ren1->dst_entry,
2487 NULL,
2489 NULL,
2490 NULL);
2494 cleanup_and_return:
2495 string_list_clear(&a_by_dst, 0);
2496 string_list_clear(&b_by_dst, 0);
2498 return clean_merge;
2501 struct rename_info {
2502 struct string_list *head_renames;
2503 struct string_list *merge_renames;
2506 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2507 struct hashmap *dir_renames)
2509 struct hashmap_iter iter;
2510 struct dir_rename_entry *e;
2512 hashmap_iter_init(dir_renames, &iter);
2513 while ((e = hashmap_iter_next(&iter))) {
2514 free(e->dir);
2515 strbuf_release(&e->new_dir);
2516 /* possible_new_dirs already cleared in get_directory_renames */
2518 hashmap_free(dir_renames, 1);
2519 free(dir_renames);
2521 free(pairs->queue);
2522 free(pairs);
2525 static int handle_renames(struct merge_options *o,
2526 struct tree *common,
2527 struct tree *head,
2528 struct tree *merge,
2529 struct string_list *entries,
2530 struct rename_info *ri)
2532 struct diff_queue_struct *head_pairs, *merge_pairs;
2533 struct hashmap *dir_re_head, *dir_re_merge;
2534 int clean = 1;
2536 ri->head_renames = NULL;
2537 ri->merge_renames = NULL;
2539 if (!o->detect_rename)
2540 return 1;
2542 head_pairs = get_diffpairs(o, common, head);
2543 merge_pairs = get_diffpairs(o, common, merge);
2545 dir_re_head = get_directory_renames(head_pairs, head);
2546 dir_re_merge = get_directory_renames(merge_pairs, merge);
2548 handle_directory_level_conflicts(o,
2549 dir_re_head, head,
2550 dir_re_merge, merge);
2552 ri->head_renames = get_renames(o, head_pairs,
2553 dir_re_merge, dir_re_head, head,
2554 common, head, merge, entries,
2555 &clean);
2556 if (clean < 0)
2557 goto cleanup;
2558 ri->merge_renames = get_renames(o, merge_pairs,
2559 dir_re_head, dir_re_merge, merge,
2560 common, head, merge, entries,
2561 &clean);
2562 if (clean < 0)
2563 goto cleanup;
2564 clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2566 cleanup:
2568 * Some cleanup is deferred until cleanup_renames() because the
2569 * data structures are still needed and referenced in
2570 * process_entry(). But there are a few things we can free now.
2572 initial_cleanup_rename(head_pairs, dir_re_head);
2573 initial_cleanup_rename(merge_pairs, dir_re_merge);
2575 return clean;
2578 static void final_cleanup_rename(struct string_list *rename)
2580 const struct rename *re;
2581 int i;
2583 if (rename == NULL)
2584 return;
2586 for (i = 0; i < rename->nr; i++) {
2587 re = rename->items[i].util;
2588 diff_free_filepair(re->pair);
2590 string_list_clear(rename, 1);
2591 free(rename);
2594 static void final_cleanup_renames(struct rename_info *re_info)
2596 final_cleanup_rename(re_info->head_renames);
2597 final_cleanup_rename(re_info->merge_renames);
2600 static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2602 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2605 static int read_oid_strbuf(struct merge_options *o,
2606 const struct object_id *oid, struct strbuf *dst)
2608 void *buf;
2609 enum object_type type;
2610 unsigned long size;
2611 buf = read_object_file(oid, &type, &size);
2612 if (!buf)
2613 return err(o, _("cannot read object %s"), oid_to_hex(oid));
2614 if (type != OBJ_BLOB) {
2615 free(buf);
2616 return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2618 strbuf_attach(dst, buf, size, size + 1);
2619 return 0;
2622 static int blob_unchanged(struct merge_options *opt,
2623 const struct object_id *o_oid,
2624 unsigned o_mode,
2625 const struct object_id *a_oid,
2626 unsigned a_mode,
2627 int renormalize, const char *path)
2629 struct strbuf o = STRBUF_INIT;
2630 struct strbuf a = STRBUF_INIT;
2631 int ret = 0; /* assume changed for safety */
2633 if (a_mode != o_mode)
2634 return 0;
2635 if (oid_eq(o_oid, a_oid))
2636 return 1;
2637 if (!renormalize)
2638 return 0;
2640 assert(o_oid && a_oid);
2641 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2642 goto error_return;
2644 * Note: binary | is used so that both renormalizations are
2645 * performed. Comparison can be skipped if both files are
2646 * unchanged since their sha1s have already been compared.
2648 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2649 renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2650 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2652 error_return:
2653 strbuf_release(&o);
2654 strbuf_release(&a);
2655 return ret;
2658 static int handle_modify_delete(struct merge_options *o,
2659 const char *path,
2660 struct object_id *o_oid, int o_mode,
2661 struct object_id *a_oid, int a_mode,
2662 struct object_id *b_oid, int b_mode)
2664 const char *modify_branch, *delete_branch;
2665 struct object_id *changed_oid;
2666 int changed_mode;
2668 if (a_oid) {
2669 modify_branch = o->branch1;
2670 delete_branch = o->branch2;
2671 changed_oid = a_oid;
2672 changed_mode = a_mode;
2673 } else {
2674 modify_branch = o->branch2;
2675 delete_branch = o->branch1;
2676 changed_oid = b_oid;
2677 changed_mode = b_mode;
2680 return handle_change_delete(o,
2681 path, NULL,
2682 o_oid, o_mode,
2683 changed_oid, changed_mode,
2684 modify_branch, delete_branch,
2685 _("modify"), _("modified"));
2688 static int merge_content(struct merge_options *o,
2689 const char *path,
2690 struct object_id *o_oid, int o_mode,
2691 struct object_id *a_oid, int a_mode,
2692 struct object_id *b_oid, int b_mode,
2693 struct rename_conflict_info *rename_conflict_info)
2695 const char *reason = _("content");
2696 const char *path1 = NULL, *path2 = NULL;
2697 struct merge_file_info mfi;
2698 struct diff_filespec one, a, b;
2699 unsigned df_conflict_remains = 0;
2701 if (!o_oid) {
2702 reason = _("add/add");
2703 o_oid = (struct object_id *)&null_oid;
2705 one.path = a.path = b.path = (char *)path;
2706 oidcpy(&one.oid, o_oid);
2707 one.mode = o_mode;
2708 oidcpy(&a.oid, a_oid);
2709 a.mode = a_mode;
2710 oidcpy(&b.oid, b_oid);
2711 b.mode = b_mode;
2713 if (rename_conflict_info) {
2714 struct diff_filepair *pair1 = rename_conflict_info->pair1;
2716 path1 = (o->branch1 == rename_conflict_info->branch1) ?
2717 pair1->two->path : pair1->one->path;
2718 /* If rename_conflict_info->pair2 != NULL, we are in
2719 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
2720 * normal rename.
2722 path2 = (rename_conflict_info->pair2 ||
2723 o->branch2 == rename_conflict_info->branch1) ?
2724 pair1->two->path : pair1->one->path;
2726 if (dir_in_way(path, !o->call_depth,
2727 S_ISGITLINK(pair1->two->mode)))
2728 df_conflict_remains = 1;
2730 if (merge_file_special_markers(o, &one, &a, &b,
2731 o->branch1, path1,
2732 o->branch2, path2, &mfi))
2733 return -1;
2735 if (mfi.clean && !df_conflict_remains &&
2736 oid_eq(&mfi.oid, a_oid) && mfi.mode == a_mode) {
2737 int path_renamed_outside_HEAD;
2738 output(o, 3, _("Skipped %s (merged same as existing)"), path);
2740 * The content merge resulted in the same file contents we
2741 * already had. We can return early if those file contents
2742 * are recorded at the correct path (which may not be true
2743 * if the merge involves a rename).
2745 path_renamed_outside_HEAD = !path2 || !strcmp(path, path2);
2746 if (!path_renamed_outside_HEAD) {
2747 add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2748 0, (!o->call_depth), 0);
2749 return mfi.clean;
2751 } else
2752 output(o, 2, _("Auto-merging %s"), path);
2754 if (!mfi.clean) {
2755 if (S_ISGITLINK(mfi.mode))
2756 reason = _("submodule");
2757 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2758 reason, path);
2759 if (rename_conflict_info && !df_conflict_remains)
2760 if (update_stages(o, path, &one, &a, &b))
2761 return -1;
2764 if (df_conflict_remains) {
2765 char *new_path;
2766 if (o->call_depth) {
2767 remove_file_from_cache(path);
2768 } else {
2769 if (!mfi.clean) {
2770 if (update_stages(o, path, &one, &a, &b))
2771 return -1;
2772 } else {
2773 int file_from_stage2 = was_tracked(path);
2774 struct diff_filespec merged;
2775 oidcpy(&merged.oid, &mfi.oid);
2776 merged.mode = mfi.mode;
2778 if (update_stages(o, path, NULL,
2779 file_from_stage2 ? &merged : NULL,
2780 file_from_stage2 ? NULL : &merged))
2781 return -1;
2785 new_path = unique_path(o, path, rename_conflict_info->branch1);
2786 output(o, 1, _("Adding as %s instead"), new_path);
2787 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2788 free(new_path);
2789 return -1;
2791 free(new_path);
2792 mfi.clean = 0;
2793 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2794 return -1;
2795 return mfi.clean;
2798 /* Per entry merge function */
2799 static int process_entry(struct merge_options *o,
2800 const char *path, struct stage_data *entry)
2802 int clean_merge = 1;
2803 int normalize = o->renormalize;
2804 unsigned o_mode = entry->stages[1].mode;
2805 unsigned a_mode = entry->stages[2].mode;
2806 unsigned b_mode = entry->stages[3].mode;
2807 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2808 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2809 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2811 entry->processed = 1;
2812 if (entry->rename_conflict_info) {
2813 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2814 switch (conflict_info->rename_type) {
2815 case RENAME_NORMAL:
2816 case RENAME_ONE_FILE_TO_ONE:
2817 clean_merge = merge_content(o, path,
2818 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2819 conflict_info);
2820 break;
2821 case RENAME_DIR:
2822 clean_merge = 1;
2823 if (conflict_rename_dir(o,
2824 conflict_info->pair1,
2825 conflict_info->branch1,
2826 conflict_info->branch2))
2827 clean_merge = -1;
2828 break;
2829 case RENAME_DELETE:
2830 clean_merge = 0;
2831 if (conflict_rename_delete(o,
2832 conflict_info->pair1,
2833 conflict_info->branch1,
2834 conflict_info->branch2))
2835 clean_merge = -1;
2836 break;
2837 case RENAME_ONE_FILE_TO_TWO:
2838 clean_merge = 0;
2839 if (conflict_rename_rename_1to2(o, conflict_info))
2840 clean_merge = -1;
2841 break;
2842 case RENAME_TWO_FILES_TO_ONE:
2843 clean_merge = 0;
2844 if (conflict_rename_rename_2to1(o, conflict_info))
2845 clean_merge = -1;
2846 break;
2847 default:
2848 entry->processed = 0;
2849 break;
2851 } else if (o_oid && (!a_oid || !b_oid)) {
2852 /* Case A: Deleted in one */
2853 if ((!a_oid && !b_oid) ||
2854 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2855 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2856 /* Deleted in both or deleted in one and
2857 * unchanged in the other */
2858 if (a_oid)
2859 output(o, 2, _("Removing %s"), path);
2860 /* do not touch working file if it did not exist */
2861 remove_file(o, 1, path, !a_oid);
2862 } else {
2863 /* Modify/delete; deleted side may have put a directory in the way */
2864 clean_merge = 0;
2865 if (handle_modify_delete(o, path, o_oid, o_mode,
2866 a_oid, a_mode, b_oid, b_mode))
2867 clean_merge = -1;
2869 } else if ((!o_oid && a_oid && !b_oid) ||
2870 (!o_oid && !a_oid && b_oid)) {
2871 /* Case B: Added in one. */
2872 /* [nothing|directory] -> ([nothing|directory], file) */
2874 const char *add_branch;
2875 const char *other_branch;
2876 unsigned mode;
2877 const struct object_id *oid;
2878 const char *conf;
2880 if (a_oid) {
2881 add_branch = o->branch1;
2882 other_branch = o->branch2;
2883 mode = a_mode;
2884 oid = a_oid;
2885 conf = _("file/directory");
2886 } else {
2887 add_branch = o->branch2;
2888 other_branch = o->branch1;
2889 mode = b_mode;
2890 oid = b_oid;
2891 conf = _("directory/file");
2893 if (dir_in_way(path,
2894 !o->call_depth && !S_ISGITLINK(a_mode),
2895 0)) {
2896 char *new_path = unique_path(o, path, add_branch);
2897 clean_merge = 0;
2898 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
2899 "Adding %s as %s"),
2900 conf, path, other_branch, path, new_path);
2901 if (update_file(o, 0, oid, mode, new_path))
2902 clean_merge = -1;
2903 else if (o->call_depth)
2904 remove_file_from_cache(path);
2905 free(new_path);
2906 } else {
2907 output(o, 2, _("Adding %s"), path);
2908 /* do not overwrite file if already present */
2909 if (update_file_flags(o, oid, mode, path, 1, !a_oid))
2910 clean_merge = -1;
2912 } else if (a_oid && b_oid) {
2913 /* Case C: Added in both (check for same permissions) and */
2914 /* case D: Modified in both, but differently. */
2915 clean_merge = merge_content(o, path,
2916 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2917 NULL);
2918 } else if (!o_oid && !a_oid && !b_oid) {
2920 * this entry was deleted altogether. a_mode == 0 means
2921 * we had that path and want to actively remove it.
2923 remove_file(o, 1, path, !a_mode);
2924 } else
2925 die("BUG: fatal merge failure, shouldn't happen.");
2927 return clean_merge;
2930 int merge_trees(struct merge_options *o,
2931 struct tree *head,
2932 struct tree *merge,
2933 struct tree *common,
2934 struct tree **result)
2936 int code, clean;
2938 if (o->subtree_shift) {
2939 merge = shift_tree_object(head, merge, o->subtree_shift);
2940 common = shift_tree_object(head, common, o->subtree_shift);
2943 if (oid_eq(&common->object.oid, &merge->object.oid)) {
2944 struct strbuf sb = STRBUF_INIT;
2946 if (!o->call_depth && index_has_changes(&sb)) {
2947 err(o, _("Dirty index: cannot merge (dirty: %s)"),
2948 sb.buf);
2949 return 0;
2951 output(o, 0, _("Already up to date!"));
2952 *result = head;
2953 return 1;
2956 code = git_merge_trees(o->call_depth, common, head, merge);
2958 if (code != 0) {
2959 if (show(o, 4) || o->call_depth)
2960 err(o, _("merging of trees %s and %s failed"),
2961 oid_to_hex(&head->object.oid),
2962 oid_to_hex(&merge->object.oid));
2963 return -1;
2966 if (unmerged_cache()) {
2967 struct string_list *entries;
2968 struct rename_info re_info;
2969 int i;
2971 * Only need the hashmap while processing entries, so
2972 * initialize it here and free it when we are done running
2973 * through the entries. Keeping it in the merge_options as
2974 * opposed to decaring a local hashmap is for convenience
2975 * so that we don't have to pass it to around.
2977 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
2978 get_files_dirs(o, head);
2979 get_files_dirs(o, merge);
2981 entries = get_unmerged();
2982 clean = handle_renames(o, common, head, merge, entries,
2983 &re_info);
2984 record_df_conflict_files(o, entries);
2985 if (clean < 0)
2986 goto cleanup;
2987 for (i = entries->nr-1; 0 <= i; i--) {
2988 const char *path = entries->items[i].string;
2989 struct stage_data *e = entries->items[i].util;
2990 if (!e->processed) {
2991 int ret = process_entry(o, path, e);
2992 if (!ret)
2993 clean = 0;
2994 else if (ret < 0) {
2995 clean = ret;
2996 goto cleanup;
3000 for (i = 0; i < entries->nr; i++) {
3001 struct stage_data *e = entries->items[i].util;
3002 if (!e->processed)
3003 die("BUG: unprocessed path??? %s",
3004 entries->items[i].string);
3007 cleanup:
3008 final_cleanup_renames(&re_info);
3010 string_list_clear(entries, 1);
3011 free(entries);
3013 hashmap_free(&o->current_file_dir_set, 1);
3015 if (clean < 0)
3016 return clean;
3018 else
3019 clean = 1;
3021 if (o->call_depth && !(*result = write_tree_from_memory(o)))
3022 return -1;
3024 return clean;
3027 static struct commit_list *reverse_commit_list(struct commit_list *list)
3029 struct commit_list *next = NULL, *current, *backup;
3030 for (current = list; current; current = backup) {
3031 backup = current->next;
3032 current->next = next;
3033 next = current;
3035 return next;
3039 * Merge the commits h1 and h2, return the resulting virtual
3040 * commit object and a flag indicating the cleanness of the merge.
3042 int merge_recursive(struct merge_options *o,
3043 struct commit *h1,
3044 struct commit *h2,
3045 struct commit_list *ca,
3046 struct commit **result)
3048 struct commit_list *iter;
3049 struct commit *merged_common_ancestors;
3050 struct tree *mrtree;
3051 int clean;
3053 if (show(o, 4)) {
3054 output(o, 4, _("Merging:"));
3055 output_commit_title(o, h1);
3056 output_commit_title(o, h2);
3059 if (!ca) {
3060 ca = get_merge_bases(h1, h2);
3061 ca = reverse_commit_list(ca);
3064 if (show(o, 5)) {
3065 unsigned cnt = commit_list_count(ca);
3067 output(o, 5, Q_("found %u common ancestor:",
3068 "found %u common ancestors:", cnt), cnt);
3069 for (iter = ca; iter; iter = iter->next)
3070 output_commit_title(o, iter->item);
3073 merged_common_ancestors = pop_commit(&ca);
3074 if (merged_common_ancestors == NULL) {
3075 /* if there is no common ancestor, use an empty tree */
3076 struct tree *tree;
3078 tree = lookup_tree(the_hash_algo->empty_tree);
3079 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3082 for (iter = ca; iter; iter = iter->next) {
3083 const char *saved_b1, *saved_b2;
3084 o->call_depth++;
3086 * When the merge fails, the result contains files
3087 * with conflict markers. The cleanness flag is
3088 * ignored (unless indicating an error), it was never
3089 * actually used, as result of merge_trees has always
3090 * overwritten it: the committed "conflicts" were
3091 * already resolved.
3093 discard_cache();
3094 saved_b1 = o->branch1;
3095 saved_b2 = o->branch2;
3096 o->branch1 = "Temporary merge branch 1";
3097 o->branch2 = "Temporary merge branch 2";
3098 if (merge_recursive(o, merged_common_ancestors, iter->item,
3099 NULL, &merged_common_ancestors) < 0)
3100 return -1;
3101 o->branch1 = saved_b1;
3102 o->branch2 = saved_b2;
3103 o->call_depth--;
3105 if (!merged_common_ancestors)
3106 return err(o, _("merge returned no commit"));
3109 discard_cache();
3110 if (!o->call_depth)
3111 read_cache();
3113 o->ancestor = "merged common ancestors";
3114 clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,
3115 &mrtree);
3116 if (clean < 0) {
3117 flush_output(o);
3118 return clean;
3121 if (o->call_depth) {
3122 *result = make_virtual_commit(mrtree, "merged tree");
3123 commit_list_insert(h1, &(*result)->parents);
3124 commit_list_insert(h2, &(*result)->parents->next);
3126 flush_output(o);
3127 if (!o->call_depth && o->buffer_output < 2)
3128 strbuf_release(&o->obuf);
3129 if (show(o, 2))
3130 diff_warn_rename_limit("merge.renamelimit",
3131 o->needed_rename_limit, 0);
3132 return clean;
3135 static struct commit *get_ref(const struct object_id *oid, const char *name)
3137 struct object *object;
3139 object = deref_tag(parse_object(oid), name, strlen(name));
3140 if (!object)
3141 return NULL;
3142 if (object->type == OBJ_TREE)
3143 return make_virtual_commit((struct tree*)object, name);
3144 if (object->type != OBJ_COMMIT)
3145 return NULL;
3146 if (parse_commit((struct commit *)object))
3147 return NULL;
3148 return (struct commit *)object;
3151 int merge_recursive_generic(struct merge_options *o,
3152 const struct object_id *head,
3153 const struct object_id *merge,
3154 int num_base_list,
3155 const struct object_id **base_list,
3156 struct commit **result)
3158 int clean;
3159 struct lock_file lock = LOCK_INIT;
3160 struct commit *head_commit = get_ref(head, o->branch1);
3161 struct commit *next_commit = get_ref(merge, o->branch2);
3162 struct commit_list *ca = NULL;
3164 if (base_list) {
3165 int i;
3166 for (i = 0; i < num_base_list; ++i) {
3167 struct commit *base;
3168 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3169 return err(o, _("Could not parse object '%s'"),
3170 oid_to_hex(base_list[i]));
3171 commit_list_insert(base, &ca);
3175 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3176 clean = merge_recursive(o, head_commit, next_commit, ca,
3177 result);
3178 if (clean < 0) {
3179 rollback_lock_file(&lock);
3180 return clean;
3183 if (write_locked_index(&the_index, &lock,
3184 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3185 return err(o, _("Unable to write index."));
3187 return clean ? 0 : 1;
3190 static void merge_recursive_config(struct merge_options *o)
3192 git_config_get_int("merge.verbosity", &o->verbosity);
3193 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3194 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3195 git_config(git_xmerge_config, NULL);
3198 void init_merge_options(struct merge_options *o)
3200 const char *merge_verbosity;
3201 memset(o, 0, sizeof(struct merge_options));
3202 o->verbosity = 2;
3203 o->buffer_output = 1;
3204 o->diff_rename_limit = -1;
3205 o->merge_rename_limit = -1;
3206 o->renormalize = 0;
3207 o->detect_rename = 1;
3208 merge_recursive_config(o);
3209 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3210 if (merge_verbosity)
3211 o->verbosity = strtol(merge_verbosity, NULL, 10);
3212 if (o->verbosity >= 5)
3213 o->buffer_output = 0;
3214 strbuf_init(&o->obuf, 0);
3215 string_list_init(&o->df_conflict_file_set, 1);
3218 int parse_merge_opt(struct merge_options *o, const char *s)
3220 const char *arg;
3222 if (!s || !*s)
3223 return -1;
3224 if (!strcmp(s, "ours"))
3225 o->recursive_variant = MERGE_RECURSIVE_OURS;
3226 else if (!strcmp(s, "theirs"))
3227 o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3228 else if (!strcmp(s, "subtree"))
3229 o->subtree_shift = "";
3230 else if (skip_prefix(s, "subtree=", &arg))
3231 o->subtree_shift = arg;
3232 else if (!strcmp(s, "patience"))
3233 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3234 else if (!strcmp(s, "histogram"))
3235 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3236 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3237 long value = parse_algorithm_value(arg);
3238 if (value < 0)
3239 return -1;
3240 /* clear out previous settings */
3241 DIFF_XDL_CLR(o, NEED_MINIMAL);
3242 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3243 o->xdl_opts |= value;
3245 else if (!strcmp(s, "ignore-space-change"))
3246 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3247 else if (!strcmp(s, "ignore-all-space"))
3248 DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3249 else if (!strcmp(s, "ignore-space-at-eol"))
3250 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3251 else if (!strcmp(s, "ignore-cr-at-eol"))
3252 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3253 else if (!strcmp(s, "renormalize"))
3254 o->renormalize = 1;
3255 else if (!strcmp(s, "no-renormalize"))
3256 o->renormalize = 0;
3257 else if (!strcmp(s, "no-renames"))
3258 o->detect_rename = 0;
3259 else if (!strcmp(s, "find-renames")) {
3260 o->detect_rename = 1;
3261 o->rename_score = 0;
3263 else if (skip_prefix(s, "find-renames=", &arg) ||
3264 skip_prefix(s, "rename-threshold=", &arg)) {
3265 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3266 return -1;
3267 o->detect_rename = 1;
3269 else
3270 return -1;
3271 return 0;