Merge branch 'js/sequencer-and-root-commits'
[git.git] / merge-recursive.c
blob35df695fa420073d04997a3792f1c3a45d278b5d
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->maybe_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, _("add_cacheinfo failed for path '%s'; merge aborting."), 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, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
328 if (nce != ce)
329 ret = add_cache_entry(nce, options);
331 return ret;
334 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
336 parse_tree(tree);
337 init_tree_desc(desc, tree->buffer, tree->size);
340 static int git_merge_trees(struct merge_options *o,
341 struct tree *common,
342 struct tree *head,
343 struct tree *merge)
345 int rc;
346 struct tree_desc t[3];
347 struct index_state tmp_index = { NULL };
349 memset(&o->unpack_opts, 0, sizeof(o->unpack_opts));
350 if (o->call_depth)
351 o->unpack_opts.index_only = 1;
352 else
353 o->unpack_opts.update = 1;
354 o->unpack_opts.merge = 1;
355 o->unpack_opts.head_idx = 2;
356 o->unpack_opts.fn = threeway_merge;
357 o->unpack_opts.src_index = &the_index;
358 o->unpack_opts.dst_index = &tmp_index;
359 setup_unpack_trees_porcelain(&o->unpack_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, &o->unpack_opts);
366 cache_tree_free(&active_cache_tree);
369 * Update the_index to match the new results, AFTER saving a copy
370 * in o->orig_index. Update src_index to point to the saved copy.
371 * (verify_uptodate() checks src_index, and the original index is
372 * the one that had the necessary modification timestamps.)
374 o->orig_index = the_index;
375 the_index = tmp_index;
376 o->unpack_opts.src_index = &o->orig_index;
378 return rc;
381 struct tree *write_tree_from_memory(struct merge_options *o)
383 struct tree *result = NULL;
385 if (unmerged_cache()) {
386 int i;
387 fprintf(stderr, "BUG: There are unmerged index entries:\n");
388 for (i = 0; i < active_nr; i++) {
389 const struct cache_entry *ce = active_cache[i];
390 if (ce_stage(ce))
391 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
392 (int)ce_namelen(ce), ce->name);
394 die("BUG: unmerged index entries in merge-recursive.c");
397 if (!active_cache_tree)
398 active_cache_tree = cache_tree();
400 if (!cache_tree_fully_valid(active_cache_tree) &&
401 cache_tree_update(&the_index, 0) < 0) {
402 err(o, _("error building trees"));
403 return NULL;
406 result = lookup_tree(&active_cache_tree->oid);
408 return result;
411 static int save_files_dirs(const struct object_id *oid,
412 struct strbuf *base, const char *path,
413 unsigned int mode, int stage, void *context)
415 struct path_hashmap_entry *entry;
416 int baselen = base->len;
417 struct merge_options *o = context;
419 strbuf_addstr(base, path);
421 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
422 hashmap_entry_init(entry, path_hash(entry->path));
423 hashmap_add(&o->current_file_dir_set, entry);
425 strbuf_setlen(base, baselen);
426 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
429 static void get_files_dirs(struct merge_options *o, struct tree *tree)
431 struct pathspec match_all;
432 memset(&match_all, 0, sizeof(match_all));
433 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
436 static int get_tree_entry_if_blob(const struct object_id *tree,
437 const char *path,
438 struct object_id *hashy,
439 unsigned int *mode_o)
441 int ret;
443 ret = get_tree_entry(tree, path, hashy, mode_o);
444 if (S_ISDIR(*mode_o)) {
445 oidcpy(hashy, &null_oid);
446 *mode_o = 0;
448 return ret;
452 * Returns an index_entry instance which doesn't have to correspond to
453 * a real cache entry in Git's index.
455 static struct stage_data *insert_stage_data(const char *path,
456 struct tree *o, struct tree *a, struct tree *b,
457 struct string_list *entries)
459 struct string_list_item *item;
460 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
461 get_tree_entry_if_blob(&o->object.oid, path,
462 &e->stages[1].oid, &e->stages[1].mode);
463 get_tree_entry_if_blob(&a->object.oid, path,
464 &e->stages[2].oid, &e->stages[2].mode);
465 get_tree_entry_if_blob(&b->object.oid, path,
466 &e->stages[3].oid, &e->stages[3].mode);
467 item = string_list_insert(entries, path);
468 item->util = e;
469 return e;
473 * Create a dictionary mapping file names to stage_data objects. The
474 * dictionary contains one entry for every path with a non-zero stage entry.
476 static struct string_list *get_unmerged(void)
478 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
479 int i;
481 unmerged->strdup_strings = 1;
483 for (i = 0; i < active_nr; i++) {
484 struct string_list_item *item;
485 struct stage_data *e;
486 const struct cache_entry *ce = active_cache[i];
487 if (!ce_stage(ce))
488 continue;
490 item = string_list_lookup(unmerged, ce->name);
491 if (!item) {
492 item = string_list_insert(unmerged, ce->name);
493 item->util = xcalloc(1, sizeof(struct stage_data));
495 e = item->util;
496 e->stages[ce_stage(ce)].mode = ce->ce_mode;
497 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
500 return unmerged;
503 static int string_list_df_name_compare(const char *one, const char *two)
505 int onelen = strlen(one);
506 int twolen = strlen(two);
508 * Here we only care that entries for D/F conflicts are
509 * adjacent, in particular with the file of the D/F conflict
510 * appearing before files below the corresponding directory.
511 * The order of the rest of the list is irrelevant for us.
513 * To achieve this, we sort with df_name_compare and provide
514 * the mode S_IFDIR so that D/F conflicts will sort correctly.
515 * We use the mode S_IFDIR for everything else for simplicity,
516 * since in other cases any changes in their order due to
517 * sorting cause no problems for us.
519 int cmp = df_name_compare(one, onelen, S_IFDIR,
520 two, twolen, S_IFDIR);
522 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
523 * that 'foo' comes before 'foo/bar'.
525 if (cmp)
526 return cmp;
527 return onelen - twolen;
530 static void record_df_conflict_files(struct merge_options *o,
531 struct string_list *entries)
533 /* If there is a D/F conflict and the file for such a conflict
534 * currently exist in the working tree, we want to allow it to be
535 * removed to make room for the corresponding directory if needed.
536 * The files underneath the directories of such D/F conflicts will
537 * be processed before the corresponding file involved in the D/F
538 * conflict. If the D/F directory ends up being removed by the
539 * merge, then we won't have to touch the D/F file. If the D/F
540 * directory needs to be written to the working copy, then the D/F
541 * file will simply be removed (in make_room_for_path()) to make
542 * room for the necessary paths. Note that if both the directory
543 * and the file need to be present, then the D/F file will be
544 * reinstated with a new unique name at the time it is processed.
546 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
547 const char *last_file = NULL;
548 int last_len = 0;
549 int i;
552 * If we're merging merge-bases, we don't want to bother with
553 * any working directory changes.
555 if (o->call_depth)
556 return;
558 /* Ensure D/F conflicts are adjacent in the entries list. */
559 for (i = 0; i < entries->nr; i++) {
560 struct string_list_item *next = &entries->items[i];
561 string_list_append(&df_sorted_entries, next->string)->util =
562 next->util;
564 df_sorted_entries.cmp = string_list_df_name_compare;
565 string_list_sort(&df_sorted_entries);
567 string_list_clear(&o->df_conflict_file_set, 1);
568 for (i = 0; i < df_sorted_entries.nr; i++) {
569 const char *path = df_sorted_entries.items[i].string;
570 int len = strlen(path);
571 struct stage_data *e = df_sorted_entries.items[i].util;
574 * Check if last_file & path correspond to a D/F conflict;
575 * i.e. whether path is last_file+'/'+<something>.
576 * If so, record that it's okay to remove last_file to make
577 * room for path and friends if needed.
579 if (last_file &&
580 len > last_len &&
581 memcmp(path, last_file, last_len) == 0 &&
582 path[last_len] == '/') {
583 string_list_insert(&o->df_conflict_file_set, last_file);
587 * Determine whether path could exist as a file in the
588 * working directory as a possible D/F conflict. This
589 * will only occur when it exists in stage 2 as a
590 * file.
592 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
593 last_file = path;
594 last_len = len;
595 } else {
596 last_file = NULL;
599 string_list_clear(&df_sorted_entries, 0);
602 struct rename {
603 struct diff_filepair *pair;
605 * Purpose of src_entry and dst_entry:
607 * If 'before' is renamed to 'after' then src_entry will contain
608 * the versions of 'before' from the merge_base, HEAD, and MERGE in
609 * stages 1, 2, and 3; dst_entry will contain the respective
610 * versions of 'after' in corresponding locations. Thus, we have a
611 * total of six modes and oids, though some will be null. (Stage 0
612 * is ignored; we're interested in handling conflicts.)
614 * Since we don't turn on break-rewrites by default, neither
615 * src_entry nor dst_entry can have all three of their stages have
616 * non-null oids, meaning at most four of the six will be non-null.
617 * Also, since this is a rename, both src_entry and dst_entry will
618 * have at least one non-null oid, meaning at least two will be
619 * non-null. Of the six oids, a typical rename will have three be
620 * non-null. Only two implies a rename/delete, and four implies a
621 * rename/add.
623 struct stage_data *src_entry;
624 struct stage_data *dst_entry;
625 unsigned add_turned_into_rename:1;
626 unsigned processed:1;
629 static int update_stages(struct merge_options *opt, const char *path,
630 const struct diff_filespec *o,
631 const struct diff_filespec *a,
632 const struct diff_filespec *b)
636 * NOTE: It is usually a bad idea to call update_stages on a path
637 * before calling update_file on that same path, since it can
638 * sometimes lead to spurious "refusing to lose untracked file..."
639 * messages from update_file (via make_room_for path via
640 * would_lose_untracked). Instead, reverse the order of the calls
641 * (executing update_file first and then update_stages).
643 int clear = 1;
644 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
645 if (clear)
646 if (remove_file_from_cache(path))
647 return -1;
648 if (o)
649 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
650 return -1;
651 if (a)
652 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
653 return -1;
654 if (b)
655 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
656 return -1;
657 return 0;
660 static int update_stages_for_stage_data(struct merge_options *opt,
661 const char *path,
662 const struct stage_data *stage_data)
664 struct diff_filespec o, a, b;
666 o.mode = stage_data->stages[1].mode;
667 oidcpy(&o.oid, &stage_data->stages[1].oid);
669 a.mode = stage_data->stages[2].mode;
670 oidcpy(&a.oid, &stage_data->stages[2].oid);
672 b.mode = stage_data->stages[3].mode;
673 oidcpy(&b.oid, &stage_data->stages[3].oid);
675 return update_stages(opt, path,
676 is_null_oid(&o.oid) ? NULL : &o,
677 is_null_oid(&a.oid) ? NULL : &a,
678 is_null_oid(&b.oid) ? NULL : &b);
681 static void update_entry(struct stage_data *entry,
682 struct diff_filespec *o,
683 struct diff_filespec *a,
684 struct diff_filespec *b)
686 entry->processed = 0;
687 entry->stages[1].mode = o->mode;
688 entry->stages[2].mode = a->mode;
689 entry->stages[3].mode = b->mode;
690 oidcpy(&entry->stages[1].oid, &o->oid);
691 oidcpy(&entry->stages[2].oid, &a->oid);
692 oidcpy(&entry->stages[3].oid, &b->oid);
695 static int remove_file(struct merge_options *o, int clean,
696 const char *path, int no_wd)
698 int update_cache = o->call_depth || clean;
699 int update_working_directory = !o->call_depth && !no_wd;
701 if (update_cache) {
702 if (remove_file_from_cache(path))
703 return -1;
705 if (update_working_directory) {
706 if (ignore_case) {
707 struct cache_entry *ce;
708 ce = cache_file_exists(path, strlen(path), ignore_case);
709 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
710 return 0;
712 if (remove_path(path))
713 return -1;
715 return 0;
718 /* add a string to a strbuf, but converting "/" to "_" */
719 static void add_flattened_path(struct strbuf *out, const char *s)
721 size_t i = out->len;
722 strbuf_addstr(out, s);
723 for (; i < out->len; i++)
724 if (out->buf[i] == '/')
725 out->buf[i] = '_';
728 static char *unique_path(struct merge_options *o, const char *path, const char *branch)
730 struct path_hashmap_entry *entry;
731 struct strbuf newpath = STRBUF_INIT;
732 int suffix = 0;
733 size_t base_len;
735 strbuf_addf(&newpath, "%s~", path);
736 add_flattened_path(&newpath, branch);
738 base_len = newpath.len;
739 while (hashmap_get_from_hash(&o->current_file_dir_set,
740 path_hash(newpath.buf), newpath.buf) ||
741 (!o->call_depth && file_exists(newpath.buf))) {
742 strbuf_setlen(&newpath, base_len);
743 strbuf_addf(&newpath, "_%d", suffix++);
746 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
747 hashmap_entry_init(entry, path_hash(entry->path));
748 hashmap_add(&o->current_file_dir_set, entry);
749 return strbuf_detach(&newpath, NULL);
753 * Check whether a directory in the index is in the way of an incoming
754 * file. Return 1 if so. If check_working_copy is non-zero, also
755 * check the working directory. If empty_ok is non-zero, also return
756 * 0 in the case where the working-tree dir exists but is empty.
758 static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
760 int pos;
761 struct strbuf dirpath = STRBUF_INIT;
762 struct stat st;
764 strbuf_addstr(&dirpath, path);
765 strbuf_addch(&dirpath, '/');
767 pos = cache_name_pos(dirpath.buf, dirpath.len);
769 if (pos < 0)
770 pos = -1 - pos;
771 if (pos < active_nr &&
772 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
773 strbuf_release(&dirpath);
774 return 1;
777 strbuf_release(&dirpath);
778 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
779 !(empty_ok && is_empty_dir(path));
783 * Returns whether path was tracked in the index before the merge started,
784 * and its oid and mode match the specified values
786 static int was_tracked_and_matches(struct merge_options *o, const char *path,
787 const struct object_id *oid, unsigned mode)
789 int pos = index_name_pos(&o->orig_index, path, strlen(path));
790 struct cache_entry *ce;
792 if (0 > pos)
793 /* we were not tracking this path before the merge */
794 return 0;
796 /* See if the file we were tracking before matches */
797 ce = o->orig_index.cache[pos];
798 return (oid_eq(&ce->oid, oid) && ce->ce_mode == mode);
802 * Returns whether path was tracked in the index before the merge started
804 static int was_tracked(struct merge_options *o, const char *path)
806 int pos = index_name_pos(&o->orig_index, path, strlen(path));
808 if (0 <= pos)
809 /* we were tracking this path before the merge */
810 return 1;
812 return 0;
815 static int would_lose_untracked(const char *path)
818 * This may look like it can be simplified to:
819 * return !was_tracked(o, path) && file_exists(path)
820 * but it can't. This function needs to know whether path was in
821 * the working tree due to EITHER having been tracked in the index
822 * before the merge OR having been put into the working copy and
823 * index by unpack_trees(). Due to that either-or requirement, we
824 * check the current index instead of the original one.
826 * Note that we do not need to worry about merge-recursive itself
827 * updating the index after unpack_trees() and before calling this
828 * function, because we strictly require all code paths in
829 * merge-recursive to update the working tree first and the index
830 * second. Doing otherwise would break
831 * update_file()/would_lose_untracked(); see every comment in this
832 * file which mentions "update_stages".
834 int pos = cache_name_pos(path, strlen(path));
836 if (pos < 0)
837 pos = -1 - pos;
838 while (pos < active_nr &&
839 !strcmp(path, active_cache[pos]->name)) {
841 * If stage #0, it is definitely tracked.
842 * If it has stage #2 then it was tracked
843 * before this merge started. All other
844 * cases the path was not tracked.
846 switch (ce_stage(active_cache[pos])) {
847 case 0:
848 case 2:
849 return 0;
851 pos++;
853 return file_exists(path);
856 static int was_dirty(struct merge_options *o, const char *path)
858 struct cache_entry *ce;
859 int dirty = 1;
861 if (o->call_depth || !was_tracked(o, path))
862 return !dirty;
864 ce = index_file_exists(o->unpack_opts.src_index,
865 path, strlen(path), ignore_case);
866 dirty = verify_uptodate(ce, &o->unpack_opts) != 0;
867 return dirty;
870 static int make_room_for_path(struct merge_options *o, const char *path)
872 int status, i;
873 const char *msg = _("failed to create path '%s'%s");
875 /* Unlink any D/F conflict files that are in the way */
876 for (i = 0; i < o->df_conflict_file_set.nr; i++) {
877 const char *df_path = o->df_conflict_file_set.items[i].string;
878 size_t pathlen = strlen(path);
879 size_t df_pathlen = strlen(df_path);
880 if (df_pathlen < pathlen &&
881 path[df_pathlen] == '/' &&
882 strncmp(path, df_path, df_pathlen) == 0) {
883 output(o, 3,
884 _("Removing %s to make room for subdirectory\n"),
885 df_path);
886 unlink(df_path);
887 unsorted_string_list_delete_item(&o->df_conflict_file_set,
888 i, 0);
889 break;
893 /* Make sure leading directories are created */
894 status = safe_create_leading_directories_const(path);
895 if (status) {
896 if (status == SCLD_EXISTS)
897 /* something else exists */
898 return err(o, msg, path, _(": perhaps a D/F conflict?"));
899 return err(o, msg, path, "");
903 * Do not unlink a file in the work tree if we are not
904 * tracking it.
906 if (would_lose_untracked(path))
907 return err(o, _("refusing to lose untracked file at '%s'"),
908 path);
910 /* Successful unlink is good.. */
911 if (!unlink(path))
912 return 0;
913 /* .. and so is no existing file */
914 if (errno == ENOENT)
915 return 0;
916 /* .. but not some other error (who really cares what?) */
917 return err(o, msg, path, _(": perhaps a D/F conflict?"));
920 static int update_file_flags(struct merge_options *o,
921 const struct object_id *oid,
922 unsigned mode,
923 const char *path,
924 int update_cache,
925 int update_wd)
927 int ret = 0;
929 if (o->call_depth)
930 update_wd = 0;
932 if (update_wd) {
933 enum object_type type;
934 void *buf;
935 unsigned long size;
937 if (S_ISGITLINK(mode)) {
939 * We may later decide to recursively descend into
940 * the submodule directory and update its index
941 * and/or work tree, but we do not do that now.
943 update_wd = 0;
944 goto update_index;
947 buf = read_object_file(oid, &type, &size);
948 if (!buf)
949 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
950 if (type != OBJ_BLOB) {
951 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
952 goto free_buf;
954 if (S_ISREG(mode)) {
955 struct strbuf strbuf = STRBUF_INIT;
956 if (convert_to_working_tree(path, buf, size, &strbuf)) {
957 free(buf);
958 size = strbuf.len;
959 buf = strbuf_detach(&strbuf, NULL);
963 if (make_room_for_path(o, path) < 0) {
964 update_wd = 0;
965 goto free_buf;
967 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
968 int fd;
969 if (mode & 0100)
970 mode = 0777;
971 else
972 mode = 0666;
973 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
974 if (fd < 0) {
975 ret = err(o, _("failed to open '%s': %s"),
976 path, strerror(errno));
977 goto free_buf;
979 write_in_full(fd, buf, size);
980 close(fd);
981 } else if (S_ISLNK(mode)) {
982 char *lnk = xmemdupz(buf, size);
983 safe_create_leading_directories_const(path);
984 unlink(path);
985 if (symlink(lnk, path))
986 ret = err(o, _("failed to symlink '%s': %s"),
987 path, strerror(errno));
988 free(lnk);
989 } else
990 ret = err(o,
991 _("do not know what to do with %06o %s '%s'"),
992 mode, oid_to_hex(oid), path);
993 free_buf:
994 free(buf);
996 update_index:
997 if (!ret && update_cache)
998 if (add_cacheinfo(o, mode, oid, path, 0, update_wd,
999 ADD_CACHE_OK_TO_ADD))
1000 return -1;
1001 return ret;
1004 static int update_file(struct merge_options *o,
1005 int clean,
1006 const struct object_id *oid,
1007 unsigned mode,
1008 const char *path)
1010 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
1013 /* Low level file merging, update and removal */
1015 struct merge_file_info {
1016 struct object_id oid;
1017 unsigned mode;
1018 unsigned clean:1,
1019 merge:1;
1022 static int merge_3way(struct merge_options *o,
1023 mmbuffer_t *result_buf,
1024 const struct diff_filespec *one,
1025 const struct diff_filespec *a,
1026 const struct diff_filespec *b,
1027 const char *branch1,
1028 const char *branch2)
1030 mmfile_t orig, src1, src2;
1031 struct ll_merge_options ll_opts = {0};
1032 char *base_name, *name1, *name2;
1033 int merge_status;
1035 ll_opts.renormalize = o->renormalize;
1036 ll_opts.xdl_opts = o->xdl_opts;
1038 if (o->call_depth) {
1039 ll_opts.virtual_ancestor = 1;
1040 ll_opts.variant = 0;
1041 } else {
1042 switch (o->recursive_variant) {
1043 case MERGE_RECURSIVE_OURS:
1044 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1045 break;
1046 case MERGE_RECURSIVE_THEIRS:
1047 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1048 break;
1049 default:
1050 ll_opts.variant = 0;
1051 break;
1055 if (strcmp(a->path, b->path) ||
1056 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1057 base_name = o->ancestor == NULL ? NULL :
1058 mkpathdup("%s:%s", o->ancestor, one->path);
1059 name1 = mkpathdup("%s:%s", branch1, a->path);
1060 name2 = mkpathdup("%s:%s", branch2, b->path);
1061 } else {
1062 base_name = o->ancestor == NULL ? NULL :
1063 mkpathdup("%s", o->ancestor);
1064 name1 = mkpathdup("%s", branch1);
1065 name2 = mkpathdup("%s", branch2);
1068 read_mmblob(&orig, &one->oid);
1069 read_mmblob(&src1, &a->oid);
1070 read_mmblob(&src2, &b->oid);
1072 merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1073 &src1, name1, &src2, name2, &ll_opts);
1075 free(base_name);
1076 free(name1);
1077 free(name2);
1078 free(orig.ptr);
1079 free(src1.ptr);
1080 free(src2.ptr);
1081 return merge_status;
1084 static int merge_file_1(struct merge_options *o,
1085 const struct diff_filespec *one,
1086 const struct diff_filespec *a,
1087 const struct diff_filespec *b,
1088 const char *filename,
1089 const char *branch1,
1090 const char *branch2,
1091 struct merge_file_info *result)
1093 result->merge = 0;
1094 result->clean = 1;
1096 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1097 result->clean = 0;
1098 if (S_ISREG(a->mode)) {
1099 result->mode = a->mode;
1100 oidcpy(&result->oid, &a->oid);
1101 } else {
1102 result->mode = b->mode;
1103 oidcpy(&result->oid, &b->oid);
1105 } else {
1106 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1107 result->merge = 1;
1110 * Merge modes
1112 if (a->mode == b->mode || a->mode == one->mode)
1113 result->mode = b->mode;
1114 else {
1115 result->mode = a->mode;
1116 if (b->mode != one->mode) {
1117 result->clean = 0;
1118 result->merge = 1;
1122 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1123 oidcpy(&result->oid, &b->oid);
1124 else if (oid_eq(&b->oid, &one->oid))
1125 oidcpy(&result->oid, &a->oid);
1126 else if (S_ISREG(a->mode)) {
1127 mmbuffer_t result_buf;
1128 int ret = 0, merge_status;
1130 merge_status = merge_3way(o, &result_buf, one, a, b,
1131 branch1, branch2);
1133 if ((merge_status < 0) || !result_buf.ptr)
1134 ret = err(o, _("Failed to execute internal merge"));
1136 if (!ret &&
1137 write_object_file(result_buf.ptr, result_buf.size,
1138 blob_type, &result->oid))
1139 ret = err(o, _("Unable to add %s to database"),
1140 a->path);
1142 free(result_buf.ptr);
1143 if (ret)
1144 return ret;
1145 result->clean = (merge_status == 0);
1146 } else if (S_ISGITLINK(a->mode)) {
1147 result->clean = merge_submodule(&result->oid,
1148 one->path,
1149 &one->oid,
1150 &a->oid,
1151 &b->oid,
1152 !o->call_depth);
1153 } else if (S_ISLNK(a->mode)) {
1154 switch (o->recursive_variant) {
1155 case MERGE_RECURSIVE_NORMAL:
1156 oidcpy(&result->oid, &a->oid);
1157 if (!oid_eq(&a->oid, &b->oid))
1158 result->clean = 0;
1159 break;
1160 case MERGE_RECURSIVE_OURS:
1161 oidcpy(&result->oid, &a->oid);
1162 break;
1163 case MERGE_RECURSIVE_THEIRS:
1164 oidcpy(&result->oid, &b->oid);
1165 break;
1167 } else
1168 die("BUG: unsupported object type in the tree");
1171 if (result->merge)
1172 output(o, 2, _("Auto-merging %s"), filename);
1174 return 0;
1177 static int merge_file_special_markers(struct merge_options *o,
1178 const struct diff_filespec *one,
1179 const struct diff_filespec *a,
1180 const struct diff_filespec *b,
1181 const char *target_filename,
1182 const char *branch1,
1183 const char *filename1,
1184 const char *branch2,
1185 const char *filename2,
1186 struct merge_file_info *mfi)
1188 char *side1 = NULL;
1189 char *side2 = NULL;
1190 int ret;
1192 if (filename1)
1193 side1 = xstrfmt("%s:%s", branch1, filename1);
1194 if (filename2)
1195 side2 = xstrfmt("%s:%s", branch2, filename2);
1197 ret = merge_file_1(o, one, a, b, target_filename,
1198 side1 ? side1 : branch1,
1199 side2 ? side2 : branch2, mfi);
1201 free(side1);
1202 free(side2);
1203 return ret;
1206 static int merge_file_one(struct merge_options *o,
1207 const char *path,
1208 const struct object_id *o_oid, int o_mode,
1209 const struct object_id *a_oid, int a_mode,
1210 const struct object_id *b_oid, int b_mode,
1211 const char *branch1,
1212 const char *branch2,
1213 struct merge_file_info *mfi)
1215 struct diff_filespec one, a, b;
1217 one.path = a.path = b.path = (char *)path;
1218 oidcpy(&one.oid, o_oid);
1219 one.mode = o_mode;
1220 oidcpy(&a.oid, a_oid);
1221 a.mode = a_mode;
1222 oidcpy(&b.oid, b_oid);
1223 b.mode = b_mode;
1224 return merge_file_1(o, &one, &a, &b, path, branch1, branch2, mfi);
1227 static int conflict_rename_dir(struct merge_options *o,
1228 struct diff_filepair *pair,
1229 const char *rename_branch,
1230 const char *other_branch)
1232 const struct diff_filespec *dest = pair->two;
1234 if (!o->call_depth && would_lose_untracked(dest->path)) {
1235 char *alt_path = unique_path(o, dest->path, rename_branch);
1237 output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1238 "writing to %s instead."),
1239 dest->path, alt_path);
1241 * Write the file in worktree at alt_path, but not in the
1242 * index. Instead, write to dest->path for the index but
1243 * only at the higher appropriate stage.
1245 if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1246 return -1;
1247 free(alt_path);
1248 return update_stages(o, dest->path, NULL,
1249 rename_branch == o->branch1 ? dest : NULL,
1250 rename_branch == o->branch1 ? NULL : dest);
1253 /* Update dest->path both in index and in worktree */
1254 if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1255 return -1;
1256 return 0;
1259 static int handle_change_delete(struct merge_options *o,
1260 const char *path, const char *old_path,
1261 const struct object_id *o_oid, int o_mode,
1262 const struct object_id *changed_oid,
1263 int changed_mode,
1264 const char *change_branch,
1265 const char *delete_branch,
1266 const char *change, const char *change_past)
1268 char *alt_path = NULL;
1269 const char *update_path = path;
1270 int ret = 0;
1272 if (dir_in_way(path, !o->call_depth, 0) ||
1273 (!o->call_depth && would_lose_untracked(path))) {
1274 update_path = alt_path = unique_path(o, path, change_branch);
1277 if (o->call_depth) {
1279 * We cannot arbitrarily accept either a_sha or b_sha as
1280 * correct; since there is no true "middle point" between
1281 * them, simply reuse the base version for virtual merge base.
1283 ret = remove_file_from_cache(path);
1284 if (!ret)
1285 ret = update_file(o, 0, o_oid, o_mode, update_path);
1286 } else {
1287 if (!alt_path) {
1288 if (!old_path) {
1289 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1290 "and %s in %s. Version %s of %s left in tree."),
1291 change, path, delete_branch, change_past,
1292 change_branch, change_branch, path);
1293 } else {
1294 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1295 "and %s to %s in %s. Version %s of %s left in tree."),
1296 change, old_path, delete_branch, change_past, path,
1297 change_branch, change_branch, path);
1299 } else {
1300 if (!old_path) {
1301 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1302 "and %s in %s. Version %s of %s left in tree at %s."),
1303 change, path, delete_branch, change_past,
1304 change_branch, change_branch, path, alt_path);
1305 } else {
1306 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1307 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1308 change, old_path, delete_branch, change_past, path,
1309 change_branch, change_branch, path, alt_path);
1313 * No need to call update_file() on path when change_branch ==
1314 * o->branch1 && !alt_path, since that would needlessly touch
1315 * path. We could call update_file_flags() with update_cache=0
1316 * and update_wd=0, but that's a no-op.
1318 if (change_branch != o->branch1 || alt_path)
1319 ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1321 free(alt_path);
1323 return ret;
1326 static int conflict_rename_delete(struct merge_options *o,
1327 struct diff_filepair *pair,
1328 const char *rename_branch,
1329 const char *delete_branch)
1331 const struct diff_filespec *orig = pair->one;
1332 const struct diff_filespec *dest = pair->two;
1334 if (handle_change_delete(o,
1335 o->call_depth ? orig->path : dest->path,
1336 o->call_depth ? NULL : orig->path,
1337 &orig->oid, orig->mode,
1338 &dest->oid, dest->mode,
1339 rename_branch, delete_branch,
1340 _("rename"), _("renamed")))
1341 return -1;
1343 if (o->call_depth)
1344 return remove_file_from_cache(dest->path);
1345 else
1346 return update_stages(o, dest->path, NULL,
1347 rename_branch == o->branch1 ? dest : NULL,
1348 rename_branch == o->branch1 ? NULL : dest);
1351 static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1352 struct stage_data *entry,
1353 int stage)
1355 struct object_id *oid = &entry->stages[stage].oid;
1356 unsigned mode = entry->stages[stage].mode;
1357 if (mode == 0 || is_null_oid(oid))
1358 return NULL;
1359 oidcpy(&target->oid, oid);
1360 target->mode = mode;
1361 return target;
1364 static int handle_file(struct merge_options *o,
1365 struct diff_filespec *rename,
1366 int stage,
1367 struct rename_conflict_info *ci)
1369 char *dst_name = rename->path;
1370 struct stage_data *dst_entry;
1371 const char *cur_branch, *other_branch;
1372 struct diff_filespec other;
1373 struct diff_filespec *add;
1374 int ret;
1376 if (stage == 2) {
1377 dst_entry = ci->dst_entry1;
1378 cur_branch = ci->branch1;
1379 other_branch = ci->branch2;
1380 } else {
1381 dst_entry = ci->dst_entry2;
1382 cur_branch = ci->branch2;
1383 other_branch = ci->branch1;
1386 add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1387 if (add) {
1388 int ren_src_was_dirty = was_dirty(o, rename->path);
1389 char *add_name = unique_path(o, rename->path, other_branch);
1390 if (update_file(o, 0, &add->oid, add->mode, add_name))
1391 return -1;
1393 if (ren_src_was_dirty) {
1394 output(o, 1, _("Refusing to lose dirty file at %s"),
1395 rename->path);
1398 * Because the double negatives somehow keep confusing me...
1399 * 1) update_wd iff !ren_src_was_dirty.
1400 * 2) no_wd iff !update_wd
1401 * 3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty
1403 remove_file(o, 0, rename->path, ren_src_was_dirty);
1404 dst_name = unique_path(o, rename->path, cur_branch);
1405 } else {
1406 if (dir_in_way(rename->path, !o->call_depth, 0)) {
1407 dst_name = unique_path(o, rename->path, cur_branch);
1408 output(o, 1, _("%s is a directory in %s adding as %s instead"),
1409 rename->path, other_branch, dst_name);
1410 } else if (!o->call_depth &&
1411 would_lose_untracked(rename->path)) {
1412 dst_name = unique_path(o, rename->path, cur_branch);
1413 output(o, 1, _("Refusing to lose untracked file at %s; "
1414 "adding as %s instead"),
1415 rename->path, dst_name);
1418 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1419 ; /* fall through, do allow dst_name to be released */
1420 else if (stage == 2)
1421 ret = update_stages(o, rename->path, NULL, rename, add);
1422 else
1423 ret = update_stages(o, rename->path, NULL, add, rename);
1425 if (dst_name != rename->path)
1426 free(dst_name);
1428 return ret;
1431 static int conflict_rename_rename_1to2(struct merge_options *o,
1432 struct rename_conflict_info *ci)
1434 /* One file was renamed in both branches, but to different names. */
1435 struct diff_filespec *one = ci->pair1->one;
1436 struct diff_filespec *a = ci->pair1->two;
1437 struct diff_filespec *b = ci->pair2->two;
1439 output(o, 1, _("CONFLICT (rename/rename): "
1440 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1441 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1442 one->path, a->path, ci->branch1,
1443 one->path, b->path, ci->branch2,
1444 o->call_depth ? _(" (left unresolved)") : "");
1445 if (o->call_depth) {
1446 struct merge_file_info mfi;
1447 struct diff_filespec other;
1448 struct diff_filespec *add;
1449 if (merge_file_one(o, one->path,
1450 &one->oid, one->mode,
1451 &a->oid, a->mode,
1452 &b->oid, b->mode,
1453 ci->branch1, ci->branch2, &mfi))
1454 return -1;
1457 * FIXME: For rename/add-source conflicts (if we could detect
1458 * such), this is wrong. We should instead find a unique
1459 * pathname and then either rename the add-source file to that
1460 * unique path, or use that unique path instead of src here.
1462 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1463 return -1;
1466 * Above, we put the merged content at the merge-base's
1467 * path. Now we usually need to delete both a->path and
1468 * b->path. However, the rename on each side of the merge
1469 * could also be involved in a rename/add conflict. In
1470 * such cases, we should keep the added file around,
1471 * resolving the conflict at that path in its favor.
1473 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1474 if (add) {
1475 if (update_file(o, 0, &add->oid, add->mode, a->path))
1476 return -1;
1478 else
1479 remove_file_from_cache(a->path);
1480 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1481 if (add) {
1482 if (update_file(o, 0, &add->oid, add->mode, b->path))
1483 return -1;
1485 else
1486 remove_file_from_cache(b->path);
1487 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1488 return -1;
1490 return 0;
1493 static int conflict_rename_rename_2to1(struct merge_options *o,
1494 struct rename_conflict_info *ci)
1496 /* Two files, a & b, were renamed to the same thing, c. */
1497 struct diff_filespec *a = ci->pair1->one;
1498 struct diff_filespec *b = ci->pair2->one;
1499 struct diff_filespec *c1 = ci->pair1->two;
1500 struct diff_filespec *c2 = ci->pair2->two;
1501 char *path = c1->path; /* == c2->path */
1502 char *path_side_1_desc;
1503 char *path_side_2_desc;
1504 struct merge_file_info mfi_c1;
1505 struct merge_file_info mfi_c2;
1506 int ret;
1508 output(o, 1, _("CONFLICT (rename/rename): "
1509 "Rename %s->%s in %s. "
1510 "Rename %s->%s in %s"),
1511 a->path, c1->path, ci->branch1,
1512 b->path, c2->path, ci->branch2);
1514 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1515 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1517 path_side_1_desc = xstrfmt("%s (was %s)", path, a->path);
1518 path_side_2_desc = xstrfmt("%s (was %s)", path, b->path);
1519 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1520 path_side_1_desc,
1521 o->branch1, c1->path,
1522 o->branch2, ci->ren1_other.path, &mfi_c1) ||
1523 merge_file_special_markers(o, b, &ci->ren2_other, c2,
1524 path_side_2_desc,
1525 o->branch1, ci->ren2_other.path,
1526 o->branch2, c2->path, &mfi_c2))
1527 return -1;
1528 free(path_side_1_desc);
1529 free(path_side_2_desc);
1531 if (o->call_depth) {
1533 * If mfi_c1.clean && mfi_c2.clean, then it might make
1534 * sense to do a two-way merge of those results. But, I
1535 * think in all cases, it makes sense to have the virtual
1536 * merge base just undo the renames; they can be detected
1537 * again later for the non-recursive merge.
1539 remove_file(o, 0, path, 0);
1540 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1541 if (!ret)
1542 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1543 b->path);
1544 } else {
1545 char *new_path1 = unique_path(o, path, ci->branch1);
1546 char *new_path2 = unique_path(o, path, ci->branch2);
1547 output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1548 a->path, new_path1, b->path, new_path2);
1549 if (was_dirty(o, path))
1550 output(o, 1, _("Refusing to lose dirty file at %s"),
1551 path);
1552 else if (would_lose_untracked(path))
1554 * Only way we get here is if both renames were from
1555 * a directory rename AND user had an untracked file
1556 * at the location where both files end up after the
1557 * two directory renames. See testcase 10d of t6043.
1559 output(o, 1, _("Refusing to lose untracked file at "
1560 "%s, even though it's in the way."),
1561 path);
1562 else
1563 remove_file(o, 0, path, 0);
1564 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1565 if (!ret)
1566 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1567 new_path2);
1569 * unpack_trees() actually populates the index for us for
1570 * "normal" rename/rename(2to1) situtations so that the
1571 * correct entries are at the higher stages, which would
1572 * make the call below to update_stages_for_stage_data
1573 * unnecessary. However, if either of the renames came
1574 * from a directory rename, then unpack_trees() will not
1575 * have gotten the right data loaded into the index, so we
1576 * need to do so now. (While it'd be tempting to move this
1577 * call to update_stages_for_stage_data() to
1578 * apply_directory_rename_modifications(), that would break
1579 * our intermediate calls to would_lose_untracked() since
1580 * those rely on the current in-memory index. See also the
1581 * big "NOTE" in update_stages()).
1583 if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1584 ret = -1;
1586 free(new_path2);
1587 free(new_path1);
1590 return ret;
1594 * Get the diff_filepairs changed between o_tree and tree.
1596 static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1597 struct tree *o_tree,
1598 struct tree *tree)
1600 struct diff_queue_struct *ret;
1601 struct diff_options opts;
1603 diff_setup(&opts);
1604 opts.flags.recursive = 1;
1605 opts.flags.rename_empty = 0;
1606 opts.detect_rename = DIFF_DETECT_RENAME;
1607 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1608 o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1609 1000;
1610 opts.rename_score = o->rename_score;
1611 opts.show_rename_progress = o->show_rename_progress;
1612 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1613 diff_setup_done(&opts);
1614 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1615 diffcore_std(&opts);
1616 if (opts.needed_rename_limit > o->needed_rename_limit)
1617 o->needed_rename_limit = opts.needed_rename_limit;
1619 ret = xmalloc(sizeof(*ret));
1620 *ret = diff_queued_diff;
1622 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1623 diff_queued_diff.nr = 0;
1624 diff_queued_diff.queue = NULL;
1625 diff_flush(&opts);
1626 return ret;
1629 static int tree_has_path(struct tree *tree, const char *path)
1631 struct object_id hashy;
1632 unsigned int mode_o;
1634 return !get_tree_entry(&tree->object.oid, path,
1635 &hashy, &mode_o);
1639 * Return a new string that replaces the beginning portion (which matches
1640 * entry->dir), with entry->new_dir. In perl-speak:
1641 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1642 * NOTE:
1643 * Caller must ensure that old_path starts with entry->dir + '/'.
1645 static char *apply_dir_rename(struct dir_rename_entry *entry,
1646 const char *old_path)
1648 struct strbuf new_path = STRBUF_INIT;
1649 int oldlen, newlen;
1651 if (entry->non_unique_new_dir)
1652 return NULL;
1654 oldlen = strlen(entry->dir);
1655 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1656 strbuf_grow(&new_path, newlen);
1657 strbuf_addbuf(&new_path, &entry->new_dir);
1658 strbuf_addstr(&new_path, &old_path[oldlen]);
1660 return strbuf_detach(&new_path, NULL);
1663 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1664 char **old_dir, char **new_dir)
1666 char *end_of_old, *end_of_new;
1667 int old_len, new_len;
1669 *old_dir = NULL;
1670 *new_dir = NULL;
1673 * For
1674 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1675 * the "e/foo.c" part is the same, we just want to know that
1676 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1677 * so, for this example, this function returns "a/b/c/d" in
1678 * *old_dir and "a/b/some/thing/else" in *new_dir.
1680 * Also, if the basename of the file changed, we don't care. We
1681 * want to know which portion of the directory, if any, changed.
1683 end_of_old = strrchr(old_path, '/');
1684 end_of_new = strrchr(new_path, '/');
1686 if (end_of_old == NULL || end_of_new == NULL)
1687 return;
1688 while (*--end_of_new == *--end_of_old &&
1689 end_of_old != old_path &&
1690 end_of_new != new_path)
1691 ; /* Do nothing; all in the while loop */
1693 * We've found the first non-matching character in the directory
1694 * paths. That means the current directory we were comparing
1695 * represents the rename. Move end_of_old and end_of_new back
1696 * to the full directory name.
1698 if (*end_of_old == '/')
1699 end_of_old++;
1700 if (*end_of_old != '/')
1701 end_of_new++;
1702 end_of_old = strchr(end_of_old, '/');
1703 end_of_new = strchr(end_of_new, '/');
1706 * It may have been the case that old_path and new_path were the same
1707 * directory all along. Don't claim a rename if they're the same.
1709 old_len = end_of_old - old_path;
1710 new_len = end_of_new - new_path;
1712 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1713 *old_dir = xstrndup(old_path, old_len);
1714 *new_dir = xstrndup(new_path, new_len);
1718 static void remove_hashmap_entries(struct hashmap *dir_renames,
1719 struct string_list *items_to_remove)
1721 int i;
1722 struct dir_rename_entry *entry;
1724 for (i = 0; i < items_to_remove->nr; i++) {
1725 entry = items_to_remove->items[i].util;
1726 hashmap_remove(dir_renames, entry, NULL);
1728 string_list_clear(items_to_remove, 0);
1732 * See if there is a directory rename for path, and if there are any file
1733 * level conflicts for the renamed location. If there is a rename and
1734 * there are no conflicts, return the new name. Otherwise, return NULL.
1736 static char *handle_path_level_conflicts(struct merge_options *o,
1737 const char *path,
1738 struct dir_rename_entry *entry,
1739 struct hashmap *collisions,
1740 struct tree *tree)
1742 char *new_path = NULL;
1743 struct collision_entry *collision_ent;
1744 int clean = 1;
1745 struct strbuf collision_paths = STRBUF_INIT;
1748 * entry has the mapping of old directory name to new directory name
1749 * that we want to apply to path.
1751 new_path = apply_dir_rename(entry, path);
1753 if (!new_path) {
1754 /* This should only happen when entry->non_unique_new_dir set */
1755 if (!entry->non_unique_new_dir)
1756 BUG("entry->non_unqiue_dir not set and !new_path");
1757 output(o, 1, _("CONFLICT (directory rename split): "
1758 "Unclear where to place %s because directory "
1759 "%s was renamed to multiple other directories, "
1760 "with no destination getting a majority of the "
1761 "files."),
1762 path, entry->dir);
1763 clean = 0;
1764 return NULL;
1768 * The caller needs to have ensured that it has pre-populated
1769 * collisions with all paths that map to new_path. Do a quick check
1770 * to ensure that's the case.
1772 collision_ent = collision_find_entry(collisions, new_path);
1773 if (collision_ent == NULL)
1774 BUG("collision_ent is NULL");
1777 * Check for one-sided add/add/.../add conflicts, i.e.
1778 * where implicit renames from the other side doing
1779 * directory rename(s) can affect this side of history
1780 * to put multiple paths into the same location. Warn
1781 * and bail on directory renames for such paths.
1783 if (collision_ent->reported_already) {
1784 clean = 0;
1785 } else if (tree_has_path(tree, new_path)) {
1786 collision_ent->reported_already = 1;
1787 strbuf_add_separated_string_list(&collision_paths, ", ",
1788 &collision_ent->source_files);
1789 output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1790 "file/dir at %s in the way of implicit "
1791 "directory rename(s) putting the following "
1792 "path(s) there: %s."),
1793 new_path, collision_paths.buf);
1794 clean = 0;
1795 } else if (collision_ent->source_files.nr > 1) {
1796 collision_ent->reported_already = 1;
1797 strbuf_add_separated_string_list(&collision_paths, ", ",
1798 &collision_ent->source_files);
1799 output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1800 "more than one path to %s; implicit directory "
1801 "renames tried to put these paths there: %s"),
1802 new_path, collision_paths.buf);
1803 clean = 0;
1806 /* Free memory we no longer need */
1807 strbuf_release(&collision_paths);
1808 if (!clean && new_path) {
1809 free(new_path);
1810 return NULL;
1813 return new_path;
1817 * There are a couple things we want to do at the directory level:
1818 * 1. Check for both sides renaming to the same thing, in order to avoid
1819 * implicit renaming of files that should be left in place. (See
1820 * testcase 6b in t6043 for details.)
1821 * 2. Prune directory renames if there are still files left in the
1822 * the original directory. These represent a partial directory rename,
1823 * i.e. a rename where only some of the files within the directory
1824 * were renamed elsewhere. (Technically, this could be done earlier
1825 * in get_directory_renames(), except that would prevent us from
1826 * doing the previous check and thus failing testcase 6b.)
1827 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
1828 * In the future, we could potentially record this info as well and
1829 * omit reporting rename/rename(1to2) conflicts for each path within
1830 * the affected directories, thus cleaning up the merge output.
1831 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1832 * directory level, because merging directories is fine. If it
1833 * causes conflicts for files within those merged directories, then
1834 * that should be detected at the individual path level.
1836 static void handle_directory_level_conflicts(struct merge_options *o,
1837 struct hashmap *dir_re_head,
1838 struct tree *head,
1839 struct hashmap *dir_re_merge,
1840 struct tree *merge)
1842 struct hashmap_iter iter;
1843 struct dir_rename_entry *head_ent;
1844 struct dir_rename_entry *merge_ent;
1846 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1847 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1849 hashmap_iter_init(dir_re_head, &iter);
1850 while ((head_ent = hashmap_iter_next(&iter))) {
1851 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1852 if (merge_ent &&
1853 !head_ent->non_unique_new_dir &&
1854 !merge_ent->non_unique_new_dir &&
1855 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1856 /* 1. Renamed identically; remove it from both sides */
1857 string_list_append(&remove_from_head,
1858 head_ent->dir)->util = head_ent;
1859 strbuf_release(&head_ent->new_dir);
1860 string_list_append(&remove_from_merge,
1861 merge_ent->dir)->util = merge_ent;
1862 strbuf_release(&merge_ent->new_dir);
1863 } else if (tree_has_path(head, head_ent->dir)) {
1864 /* 2. This wasn't a directory rename after all */
1865 string_list_append(&remove_from_head,
1866 head_ent->dir)->util = head_ent;
1867 strbuf_release(&head_ent->new_dir);
1871 remove_hashmap_entries(dir_re_head, &remove_from_head);
1872 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1874 hashmap_iter_init(dir_re_merge, &iter);
1875 while ((merge_ent = hashmap_iter_next(&iter))) {
1876 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1877 if (tree_has_path(merge, merge_ent->dir)) {
1878 /* 2. This wasn't a directory rename after all */
1879 string_list_append(&remove_from_merge,
1880 merge_ent->dir)->util = merge_ent;
1881 } else if (head_ent &&
1882 !head_ent->non_unique_new_dir &&
1883 !merge_ent->non_unique_new_dir) {
1884 /* 3. rename/rename(1to2) */
1886 * We can assume it's not rename/rename(1to1) because
1887 * that was case (1), already checked above. So we
1888 * know that head_ent->new_dir and merge_ent->new_dir
1889 * are different strings.
1891 output(o, 1, _("CONFLICT (rename/rename): "
1892 "Rename directory %s->%s in %s. "
1893 "Rename directory %s->%s in %s"),
1894 head_ent->dir, head_ent->new_dir.buf, o->branch1,
1895 head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1896 string_list_append(&remove_from_head,
1897 head_ent->dir)->util = head_ent;
1898 strbuf_release(&head_ent->new_dir);
1899 string_list_append(&remove_from_merge,
1900 merge_ent->dir)->util = merge_ent;
1901 strbuf_release(&merge_ent->new_dir);
1905 remove_hashmap_entries(dir_re_head, &remove_from_head);
1906 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1909 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1910 struct tree *tree)
1912 struct hashmap *dir_renames;
1913 struct hashmap_iter iter;
1914 struct dir_rename_entry *entry;
1915 int i;
1918 * Typically, we think of a directory rename as all files from a
1919 * certain directory being moved to a target directory. However,
1920 * what if someone first moved two files from the original
1921 * directory in one commit, and then renamed the directory
1922 * somewhere else in a later commit? At merge time, we just know
1923 * that files from the original directory went to two different
1924 * places, and that the bulk of them ended up in the same place.
1925 * We want each directory rename to represent where the bulk of the
1926 * files from that directory end up; this function exists to find
1927 * where the bulk of the files went.
1929 * The first loop below simply iterates through the list of file
1930 * renames, finding out how often each directory rename pair
1931 * possibility occurs.
1933 dir_renames = xmalloc(sizeof(*dir_renames));
1934 dir_rename_init(dir_renames);
1935 for (i = 0; i < pairs->nr; ++i) {
1936 struct string_list_item *item;
1937 int *count;
1938 struct diff_filepair *pair = pairs->queue[i];
1939 char *old_dir, *new_dir;
1941 /* File not part of directory rename if it wasn't renamed */
1942 if (pair->status != 'R')
1943 continue;
1945 get_renamed_dir_portion(pair->one->path, pair->two->path,
1946 &old_dir, &new_dir);
1947 if (!old_dir)
1948 /* Directory didn't change at all; ignore this one. */
1949 continue;
1951 entry = dir_rename_find_entry(dir_renames, old_dir);
1952 if (!entry) {
1953 entry = xmalloc(sizeof(*entry));
1954 dir_rename_entry_init(entry, old_dir);
1955 hashmap_put(dir_renames, entry);
1956 } else {
1957 free(old_dir);
1959 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1960 if (!item) {
1961 item = string_list_insert(&entry->possible_new_dirs,
1962 new_dir);
1963 item->util = xcalloc(1, sizeof(int));
1964 } else {
1965 free(new_dir);
1967 count = item->util;
1968 *count += 1;
1972 * For each directory with files moved out of it, we find out which
1973 * target directory received the most files so we can declare it to
1974 * be the "winning" target location for the directory rename. This
1975 * winner gets recorded in new_dir. If there is no winner
1976 * (multiple target directories received the same number of files),
1977 * we set non_unique_new_dir. Once we've determined the winner (or
1978 * that there is no winner), we no longer need possible_new_dirs.
1980 hashmap_iter_init(dir_renames, &iter);
1981 while ((entry = hashmap_iter_next(&iter))) {
1982 int max = 0;
1983 int bad_max = 0;
1984 char *best = NULL;
1986 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1987 int *count = entry->possible_new_dirs.items[i].util;
1989 if (*count == max)
1990 bad_max = max;
1991 else if (*count > max) {
1992 max = *count;
1993 best = entry->possible_new_dirs.items[i].string;
1996 if (bad_max == max)
1997 entry->non_unique_new_dir = 1;
1998 else {
1999 assert(entry->new_dir.len == 0);
2000 strbuf_addstr(&entry->new_dir, best);
2003 * The relevant directory sub-portion of the original full
2004 * filepaths were xstrndup'ed before inserting into
2005 * possible_new_dirs, and instead of manually iterating the
2006 * list and free'ing each, just lie and tell
2007 * possible_new_dirs that it did the strdup'ing so that it
2008 * will free them for us.
2010 entry->possible_new_dirs.strdup_strings = 1;
2011 string_list_clear(&entry->possible_new_dirs, 1);
2014 return dir_renames;
2017 static struct dir_rename_entry *check_dir_renamed(const char *path,
2018 struct hashmap *dir_renames)
2020 char temp[PATH_MAX];
2021 char *end;
2022 struct dir_rename_entry *entry;
2024 strcpy(temp, path);
2025 while ((end = strrchr(temp, '/'))) {
2026 *end = '\0';
2027 entry = dir_rename_find_entry(dir_renames, temp);
2028 if (entry)
2029 return entry;
2031 return NULL;
2034 static void compute_collisions(struct hashmap *collisions,
2035 struct hashmap *dir_renames,
2036 struct diff_queue_struct *pairs)
2038 int i;
2041 * Multiple files can be mapped to the same path due to directory
2042 * renames done by the other side of history. Since that other
2043 * side of history could have merged multiple directories into one,
2044 * if our side of history added the same file basename to each of
2045 * those directories, then all N of them would get implicitly
2046 * renamed by the directory rename detection into the same path,
2047 * and we'd get an add/add/.../add conflict, and all those adds
2048 * from *this* side of history. This is not representable in the
2049 * index, and users aren't going to easily be able to make sense of
2050 * it. So we need to provide a good warning about what's
2051 * happening, and fall back to no-directory-rename detection
2052 * behavior for those paths.
2054 * See testcases 9e and all of section 5 from t6043 for examples.
2056 collision_init(collisions);
2058 for (i = 0; i < pairs->nr; ++i) {
2059 struct dir_rename_entry *dir_rename_ent;
2060 struct collision_entry *collision_ent;
2061 char *new_path;
2062 struct diff_filepair *pair = pairs->queue[i];
2064 if (pair->status != 'A' && pair->status != 'R')
2065 continue;
2066 dir_rename_ent = check_dir_renamed(pair->two->path,
2067 dir_renames);
2068 if (!dir_rename_ent)
2069 continue;
2071 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2072 if (!new_path)
2074 * dir_rename_ent->non_unique_new_path is true, which
2075 * means there is no directory rename for us to use,
2076 * which means it won't cause us any additional
2077 * collisions.
2079 continue;
2080 collision_ent = collision_find_entry(collisions, new_path);
2081 if (!collision_ent) {
2082 collision_ent = xcalloc(1,
2083 sizeof(struct collision_entry));
2084 hashmap_entry_init(collision_ent, strhash(new_path));
2085 hashmap_put(collisions, collision_ent);
2086 collision_ent->target_file = new_path;
2087 } else {
2088 free(new_path);
2090 string_list_insert(&collision_ent->source_files,
2091 pair->two->path);
2095 static char *check_for_directory_rename(struct merge_options *o,
2096 const char *path,
2097 struct tree *tree,
2098 struct hashmap *dir_renames,
2099 struct hashmap *dir_rename_exclusions,
2100 struct hashmap *collisions,
2101 int *clean_merge)
2103 char *new_path = NULL;
2104 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2105 struct dir_rename_entry *oentry = NULL;
2107 if (!entry)
2108 return new_path;
2111 * This next part is a little weird. We do not want to do an
2112 * implicit rename into a directory we renamed on our side, because
2113 * that will result in a spurious rename/rename(1to2) conflict. An
2114 * example:
2115 * Base commit: dumbdir/afile, otherdir/bfile
2116 * Side 1: smrtdir/afile, otherdir/bfile
2117 * Side 2: dumbdir/afile, dumbdir/bfile
2118 * Here, while working on Side 1, we could notice that otherdir was
2119 * renamed/merged to dumbdir, and change the diff_filepair for
2120 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2121 * 2 will notice the rename from dumbdir to smrtdir, and do the
2122 * transitive rename to move it from dumbdir/bfile to
2123 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2124 * smrtdir, a rename/rename(1to2) conflict. We really just want
2125 * the file to end up in smrtdir. And the way to achieve that is
2126 * to not let Side1 do the rename to dumbdir, since we know that is
2127 * the source of one of our directory renames.
2129 * That's why oentry and dir_rename_exclusions is here.
2131 * As it turns out, this also prevents N-way transient rename
2132 * confusion; See testcases 9c and 9d of t6043.
2134 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2135 if (oentry) {
2136 output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2137 "to %s, because %s itself was renamed."),
2138 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2139 } else {
2140 new_path = handle_path_level_conflicts(o, path, entry,
2141 collisions, tree);
2142 *clean_merge &= (new_path != NULL);
2145 return new_path;
2148 static void apply_directory_rename_modifications(struct merge_options *o,
2149 struct diff_filepair *pair,
2150 char *new_path,
2151 struct rename *re,
2152 struct tree *tree,
2153 struct tree *o_tree,
2154 struct tree *a_tree,
2155 struct tree *b_tree,
2156 struct string_list *entries,
2157 int *clean)
2159 struct string_list_item *item;
2160 int stage = (tree == a_tree ? 2 : 3);
2161 int update_wd;
2164 * In all cases where we can do directory rename detection,
2165 * unpack_trees() will have read pair->two->path into the
2166 * index and the working copy. We need to remove it so that
2167 * we can instead place it at new_path. It is guaranteed to
2168 * not be untracked (unpack_trees() would have errored out
2169 * saying the file would have been overwritten), but it might
2170 * be dirty, though.
2172 update_wd = !was_dirty(o, pair->two->path);
2173 if (!update_wd)
2174 output(o, 1, _("Refusing to lose dirty file at %s"),
2175 pair->two->path);
2176 remove_file(o, 1, pair->two->path, !update_wd);
2178 /* Find or create a new re->dst_entry */
2179 item = string_list_lookup(entries, new_path);
2180 if (item) {
2182 * Since we're renaming on this side of history, and it's
2183 * due to a directory rename on the other side of history
2184 * (which we only allow when the directory in question no
2185 * longer exists on the other side of history), the
2186 * original entry for re->dst_entry is no longer
2187 * necessary...
2189 re->dst_entry->processed = 1;
2192 * ...because we'll be using this new one.
2194 re->dst_entry = item->util;
2195 } else {
2197 * re->dst_entry is for the before-dir-rename path, and we
2198 * need it to hold information for the after-dir-rename
2199 * path. Before creating a new entry, we need to mark the
2200 * old one as unnecessary (...unless it is shared by
2201 * src_entry, i.e. this didn't use to be a rename, in which
2202 * case we can just allow the normal processing to happen
2203 * for it).
2205 if (pair->status == 'R')
2206 re->dst_entry->processed = 1;
2208 re->dst_entry = insert_stage_data(new_path,
2209 o_tree, a_tree, b_tree,
2210 entries);
2211 item = string_list_insert(entries, new_path);
2212 item->util = re->dst_entry;
2216 * Update the stage_data with the information about the path we are
2217 * moving into place. That slot will be empty and available for us
2218 * to write to because of the collision checks in
2219 * handle_path_level_conflicts(). In other words,
2220 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2221 * open for us to write to.
2223 * It may be tempting to actually update the index at this point as
2224 * well, using update_stages_for_stage_data(), but as per the big
2225 * "NOTE" in update_stages(), doing so will modify the current
2226 * in-memory index which will break calls to would_lose_untracked()
2227 * that we need to make. Instead, we need to just make sure that
2228 * the various conflict_rename_*() functions update the index
2229 * explicitly rather than relying on unpack_trees() to have done it.
2231 get_tree_entry(&tree->object.oid,
2232 pair->two->path,
2233 &re->dst_entry->stages[stage].oid,
2234 &re->dst_entry->stages[stage].mode);
2236 /* Update pair status */
2237 if (pair->status == 'A') {
2239 * Recording rename information for this add makes it look
2240 * like a rename/delete conflict. Make sure we can
2241 * correctly handle this as an add that was moved to a new
2242 * directory instead of reporting a rename/delete conflict.
2244 re->add_turned_into_rename = 1;
2247 * We don't actually look at pair->status again, but it seems
2248 * pedagogically correct to adjust it.
2250 pair->status = 'R';
2253 * Finally, record the new location.
2255 pair->two->path = new_path;
2259 * Get information of all renames which occurred in 'pairs', making use of
2260 * any implicit directory renames inferred from the other side of history.
2261 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2262 * to be able to associate the correct cache entries with the rename
2263 * information; tree is always equal to either a_tree or b_tree.
2265 static struct string_list *get_renames(struct merge_options *o,
2266 struct diff_queue_struct *pairs,
2267 struct hashmap *dir_renames,
2268 struct hashmap *dir_rename_exclusions,
2269 struct tree *tree,
2270 struct tree *o_tree,
2271 struct tree *a_tree,
2272 struct tree *b_tree,
2273 struct string_list *entries,
2274 int *clean_merge)
2276 int i;
2277 struct hashmap collisions;
2278 struct hashmap_iter iter;
2279 struct collision_entry *e;
2280 struct string_list *renames;
2282 compute_collisions(&collisions, dir_renames, pairs);
2283 renames = xcalloc(1, sizeof(struct string_list));
2285 for (i = 0; i < pairs->nr; ++i) {
2286 struct string_list_item *item;
2287 struct rename *re;
2288 struct diff_filepair *pair = pairs->queue[i];
2289 char *new_path; /* non-NULL only with directory renames */
2291 if (pair->status != 'A' && pair->status != 'R') {
2292 diff_free_filepair(pair);
2293 continue;
2295 new_path = check_for_directory_rename(o, pair->two->path, tree,
2296 dir_renames,
2297 dir_rename_exclusions,
2298 &collisions,
2299 clean_merge);
2300 if (pair->status != 'R' && !new_path) {
2301 diff_free_filepair(pair);
2302 continue;
2305 re = xmalloc(sizeof(*re));
2306 re->processed = 0;
2307 re->add_turned_into_rename = 0;
2308 re->pair = pair;
2309 item = string_list_lookup(entries, re->pair->one->path);
2310 if (!item)
2311 re->src_entry = insert_stage_data(re->pair->one->path,
2312 o_tree, a_tree, b_tree, entries);
2313 else
2314 re->src_entry = item->util;
2316 item = string_list_lookup(entries, re->pair->two->path);
2317 if (!item)
2318 re->dst_entry = insert_stage_data(re->pair->two->path,
2319 o_tree, a_tree, b_tree, entries);
2320 else
2321 re->dst_entry = item->util;
2322 item = string_list_insert(renames, pair->one->path);
2323 item->util = re;
2324 if (new_path)
2325 apply_directory_rename_modifications(o, pair, new_path,
2326 re, tree, o_tree,
2327 a_tree, b_tree,
2328 entries,
2329 clean_merge);
2332 hashmap_iter_init(&collisions, &iter);
2333 while ((e = hashmap_iter_next(&iter))) {
2334 free(e->target_file);
2335 string_list_clear(&e->source_files, 0);
2337 hashmap_free(&collisions, 1);
2338 return renames;
2341 static int process_renames(struct merge_options *o,
2342 struct string_list *a_renames,
2343 struct string_list *b_renames)
2345 int clean_merge = 1, i, j;
2346 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2347 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2348 const struct rename *sre;
2350 for (i = 0; i < a_renames->nr; i++) {
2351 sre = a_renames->items[i].util;
2352 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2353 = (void *)sre;
2355 for (i = 0; i < b_renames->nr; i++) {
2356 sre = b_renames->items[i].util;
2357 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2358 = (void *)sre;
2361 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2362 struct string_list *renames1, *renames2Dst;
2363 struct rename *ren1 = NULL, *ren2 = NULL;
2364 const char *branch1, *branch2;
2365 const char *ren1_src, *ren1_dst;
2366 struct string_list_item *lookup;
2368 if (i >= a_renames->nr) {
2369 ren2 = b_renames->items[j++].util;
2370 } else if (j >= b_renames->nr) {
2371 ren1 = a_renames->items[i++].util;
2372 } else {
2373 int compare = strcmp(a_renames->items[i].string,
2374 b_renames->items[j].string);
2375 if (compare <= 0)
2376 ren1 = a_renames->items[i++].util;
2377 if (compare >= 0)
2378 ren2 = b_renames->items[j++].util;
2381 /* TODO: refactor, so that 1/2 are not needed */
2382 if (ren1) {
2383 renames1 = a_renames;
2384 renames2Dst = &b_by_dst;
2385 branch1 = o->branch1;
2386 branch2 = o->branch2;
2387 } else {
2388 renames1 = b_renames;
2389 renames2Dst = &a_by_dst;
2390 branch1 = o->branch2;
2391 branch2 = o->branch1;
2392 SWAP(ren2, ren1);
2395 if (ren1->processed)
2396 continue;
2397 ren1->processed = 1;
2398 ren1->dst_entry->processed = 1;
2399 /* BUG: We should only mark src_entry as processed if we
2400 * are not dealing with a rename + add-source case.
2402 ren1->src_entry->processed = 1;
2404 ren1_src = ren1->pair->one->path;
2405 ren1_dst = ren1->pair->two->path;
2407 if (ren2) {
2408 /* One file renamed on both sides */
2409 const char *ren2_src = ren2->pair->one->path;
2410 const char *ren2_dst = ren2->pair->two->path;
2411 enum rename_type rename_type;
2412 if (strcmp(ren1_src, ren2_src) != 0)
2413 die("BUG: ren1_src != ren2_src");
2414 ren2->dst_entry->processed = 1;
2415 ren2->processed = 1;
2416 if (strcmp(ren1_dst, ren2_dst) != 0) {
2417 rename_type = RENAME_ONE_FILE_TO_TWO;
2418 clean_merge = 0;
2419 } else {
2420 rename_type = RENAME_ONE_FILE_TO_ONE;
2421 /* BUG: We should only remove ren1_src in
2422 * the base stage (think of rename +
2423 * add-source cases).
2425 remove_file(o, 1, ren1_src, 1);
2426 update_entry(ren1->dst_entry,
2427 ren1->pair->one,
2428 ren1->pair->two,
2429 ren2->pair->two);
2431 setup_rename_conflict_info(rename_type,
2432 ren1->pair,
2433 ren2->pair,
2434 branch1,
2435 branch2,
2436 ren1->dst_entry,
2437 ren2->dst_entry,
2439 NULL,
2440 NULL);
2441 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2442 /* Two different files renamed to the same thing */
2443 char *ren2_dst;
2444 ren2 = lookup->util;
2445 ren2_dst = ren2->pair->two->path;
2446 if (strcmp(ren1_dst, ren2_dst) != 0)
2447 die("BUG: ren1_dst != ren2_dst");
2449 clean_merge = 0;
2450 ren2->processed = 1;
2452 * BUG: We should only mark src_entry as processed
2453 * if we are not dealing with a rename + add-source
2454 * case.
2456 ren2->src_entry->processed = 1;
2458 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2459 ren1->pair,
2460 ren2->pair,
2461 branch1,
2462 branch2,
2463 ren1->dst_entry,
2464 ren2->dst_entry,
2466 ren1->src_entry,
2467 ren2->src_entry);
2469 } else {
2470 /* Renamed in 1, maybe changed in 2 */
2471 /* we only use sha1 and mode of these */
2472 struct diff_filespec src_other, dst_other;
2473 int try_merge;
2476 * unpack_trees loads entries from common-commit
2477 * into stage 1, from head-commit into stage 2, and
2478 * from merge-commit into stage 3. We keep track
2479 * of which side corresponds to the rename.
2481 int renamed_stage = a_renames == renames1 ? 2 : 3;
2482 int other_stage = a_renames == renames1 ? 3 : 2;
2484 /* BUG: We should only remove ren1_src in the base
2485 * stage and in other_stage (think of rename +
2486 * add-source case).
2488 remove_file(o, 1, ren1_src,
2489 renamed_stage == 2 || !was_tracked(o, ren1_src));
2491 oidcpy(&src_other.oid,
2492 &ren1->src_entry->stages[other_stage].oid);
2493 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2494 oidcpy(&dst_other.oid,
2495 &ren1->dst_entry->stages[other_stage].oid);
2496 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2497 try_merge = 0;
2499 if (oid_eq(&src_other.oid, &null_oid) &&
2500 ren1->add_turned_into_rename) {
2501 setup_rename_conflict_info(RENAME_DIR,
2502 ren1->pair,
2503 NULL,
2504 branch1,
2505 branch2,
2506 ren1->dst_entry,
2507 NULL,
2509 NULL,
2510 NULL);
2511 } else if (oid_eq(&src_other.oid, &null_oid)) {
2512 setup_rename_conflict_info(RENAME_DELETE,
2513 ren1->pair,
2514 NULL,
2515 branch1,
2516 branch2,
2517 ren1->dst_entry,
2518 NULL,
2520 NULL,
2521 NULL);
2522 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2523 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2525 * Added file on the other side identical to
2526 * the file being renamed: clean merge.
2527 * Also, there is no need to overwrite the
2528 * file already in the working copy, so call
2529 * update_file_flags() instead of
2530 * update_file().
2532 if (update_file_flags(o,
2533 &ren1->pair->two->oid,
2534 ren1->pair->two->mode,
2535 ren1_dst,
2536 1, /* update_cache */
2537 0 /* update_wd */))
2538 clean_merge = -1;
2539 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2540 clean_merge = 0;
2541 try_merge = 1;
2542 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2543 "%s added in %s"),
2544 ren1_src, ren1_dst, branch1,
2545 ren1_dst, branch2);
2546 if (o->call_depth) {
2547 struct merge_file_info mfi;
2548 if (merge_file_one(o, ren1_dst, &null_oid, 0,
2549 &ren1->pair->two->oid,
2550 ren1->pair->two->mode,
2551 &dst_other.oid,
2552 dst_other.mode,
2553 branch1, branch2, &mfi)) {
2554 clean_merge = -1;
2555 goto cleanup_and_return;
2557 output(o, 1, _("Adding merged %s"), ren1_dst);
2558 if (update_file(o, 0, &mfi.oid,
2559 mfi.mode, ren1_dst))
2560 clean_merge = -1;
2561 try_merge = 0;
2562 } else {
2563 char *new_path = unique_path(o, ren1_dst, branch2);
2564 output(o, 1, _("Adding as %s instead"), new_path);
2565 if (update_file(o, 0, &dst_other.oid,
2566 dst_other.mode, new_path))
2567 clean_merge = -1;
2568 free(new_path);
2570 } else
2571 try_merge = 1;
2573 if (clean_merge < 0)
2574 goto cleanup_and_return;
2575 if (try_merge) {
2576 struct diff_filespec *one, *a, *b;
2577 src_other.path = (char *)ren1_src;
2579 one = ren1->pair->one;
2580 if (a_renames == renames1) {
2581 a = ren1->pair->two;
2582 b = &src_other;
2583 } else {
2584 b = ren1->pair->two;
2585 a = &src_other;
2587 update_entry(ren1->dst_entry, one, a, b);
2588 setup_rename_conflict_info(RENAME_NORMAL,
2589 ren1->pair,
2590 NULL,
2591 branch1,
2592 NULL,
2593 ren1->dst_entry,
2594 NULL,
2596 NULL,
2597 NULL);
2601 cleanup_and_return:
2602 string_list_clear(&a_by_dst, 0);
2603 string_list_clear(&b_by_dst, 0);
2605 return clean_merge;
2608 struct rename_info {
2609 struct string_list *head_renames;
2610 struct string_list *merge_renames;
2613 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2614 struct hashmap *dir_renames)
2616 struct hashmap_iter iter;
2617 struct dir_rename_entry *e;
2619 hashmap_iter_init(dir_renames, &iter);
2620 while ((e = hashmap_iter_next(&iter))) {
2621 free(e->dir);
2622 strbuf_release(&e->new_dir);
2623 /* possible_new_dirs already cleared in get_directory_renames */
2625 hashmap_free(dir_renames, 1);
2626 free(dir_renames);
2628 free(pairs->queue);
2629 free(pairs);
2632 static int handle_renames(struct merge_options *o,
2633 struct tree *common,
2634 struct tree *head,
2635 struct tree *merge,
2636 struct string_list *entries,
2637 struct rename_info *ri)
2639 struct diff_queue_struct *head_pairs, *merge_pairs;
2640 struct hashmap *dir_re_head, *dir_re_merge;
2641 int clean = 1;
2643 ri->head_renames = NULL;
2644 ri->merge_renames = NULL;
2646 if (!o->detect_rename)
2647 return 1;
2649 head_pairs = get_diffpairs(o, common, head);
2650 merge_pairs = get_diffpairs(o, common, merge);
2652 dir_re_head = get_directory_renames(head_pairs, head);
2653 dir_re_merge = get_directory_renames(merge_pairs, merge);
2655 handle_directory_level_conflicts(o,
2656 dir_re_head, head,
2657 dir_re_merge, merge);
2659 ri->head_renames = get_renames(o, head_pairs,
2660 dir_re_merge, dir_re_head, head,
2661 common, head, merge, entries,
2662 &clean);
2663 if (clean < 0)
2664 goto cleanup;
2665 ri->merge_renames = get_renames(o, merge_pairs,
2666 dir_re_head, dir_re_merge, merge,
2667 common, head, merge, entries,
2668 &clean);
2669 if (clean < 0)
2670 goto cleanup;
2671 clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2673 cleanup:
2675 * Some cleanup is deferred until cleanup_renames() because the
2676 * data structures are still needed and referenced in
2677 * process_entry(). But there are a few things we can free now.
2679 initial_cleanup_rename(head_pairs, dir_re_head);
2680 initial_cleanup_rename(merge_pairs, dir_re_merge);
2682 return clean;
2685 static void final_cleanup_rename(struct string_list *rename)
2687 const struct rename *re;
2688 int i;
2690 if (rename == NULL)
2691 return;
2693 for (i = 0; i < rename->nr; i++) {
2694 re = rename->items[i].util;
2695 diff_free_filepair(re->pair);
2697 string_list_clear(rename, 1);
2698 free(rename);
2701 static void final_cleanup_renames(struct rename_info *re_info)
2703 final_cleanup_rename(re_info->head_renames);
2704 final_cleanup_rename(re_info->merge_renames);
2707 static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2709 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2712 static int read_oid_strbuf(struct merge_options *o,
2713 const struct object_id *oid, struct strbuf *dst)
2715 void *buf;
2716 enum object_type type;
2717 unsigned long size;
2718 buf = read_object_file(oid, &type, &size);
2719 if (!buf)
2720 return err(o, _("cannot read object %s"), oid_to_hex(oid));
2721 if (type != OBJ_BLOB) {
2722 free(buf);
2723 return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2725 strbuf_attach(dst, buf, size, size + 1);
2726 return 0;
2729 static int blob_unchanged(struct merge_options *opt,
2730 const struct object_id *o_oid,
2731 unsigned o_mode,
2732 const struct object_id *a_oid,
2733 unsigned a_mode,
2734 int renormalize, const char *path)
2736 struct strbuf o = STRBUF_INIT;
2737 struct strbuf a = STRBUF_INIT;
2738 int ret = 0; /* assume changed for safety */
2740 if (a_mode != o_mode)
2741 return 0;
2742 if (oid_eq(o_oid, a_oid))
2743 return 1;
2744 if (!renormalize)
2745 return 0;
2747 assert(o_oid && a_oid);
2748 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2749 goto error_return;
2751 * Note: binary | is used so that both renormalizations are
2752 * performed. Comparison can be skipped if both files are
2753 * unchanged since their sha1s have already been compared.
2755 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2756 renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2757 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2759 error_return:
2760 strbuf_release(&o);
2761 strbuf_release(&a);
2762 return ret;
2765 static int handle_modify_delete(struct merge_options *o,
2766 const char *path,
2767 struct object_id *o_oid, int o_mode,
2768 struct object_id *a_oid, int a_mode,
2769 struct object_id *b_oid, int b_mode)
2771 const char *modify_branch, *delete_branch;
2772 struct object_id *changed_oid;
2773 int changed_mode;
2775 if (a_oid) {
2776 modify_branch = o->branch1;
2777 delete_branch = o->branch2;
2778 changed_oid = a_oid;
2779 changed_mode = a_mode;
2780 } else {
2781 modify_branch = o->branch2;
2782 delete_branch = o->branch1;
2783 changed_oid = b_oid;
2784 changed_mode = b_mode;
2787 return handle_change_delete(o,
2788 path, NULL,
2789 o_oid, o_mode,
2790 changed_oid, changed_mode,
2791 modify_branch, delete_branch,
2792 _("modify"), _("modified"));
2795 static int merge_content(struct merge_options *o,
2796 const char *path,
2797 int is_dirty,
2798 struct object_id *o_oid, int o_mode,
2799 struct object_id *a_oid, int a_mode,
2800 struct object_id *b_oid, int b_mode,
2801 struct rename_conflict_info *rename_conflict_info)
2803 const char *reason = _("content");
2804 const char *path1 = NULL, *path2 = NULL;
2805 struct merge_file_info mfi;
2806 struct diff_filespec one, a, b;
2807 unsigned df_conflict_remains = 0;
2809 if (!o_oid) {
2810 reason = _("add/add");
2811 o_oid = (struct object_id *)&null_oid;
2813 one.path = a.path = b.path = (char *)path;
2814 oidcpy(&one.oid, o_oid);
2815 one.mode = o_mode;
2816 oidcpy(&a.oid, a_oid);
2817 a.mode = a_mode;
2818 oidcpy(&b.oid, b_oid);
2819 b.mode = b_mode;
2821 if (rename_conflict_info) {
2822 struct diff_filepair *pair1 = rename_conflict_info->pair1;
2824 path1 = (o->branch1 == rename_conflict_info->branch1) ?
2825 pair1->two->path : pair1->one->path;
2826 /* If rename_conflict_info->pair2 != NULL, we are in
2827 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
2828 * normal rename.
2830 path2 = (rename_conflict_info->pair2 ||
2831 o->branch2 == rename_conflict_info->branch1) ?
2832 pair1->two->path : pair1->one->path;
2834 if (dir_in_way(path, !o->call_depth,
2835 S_ISGITLINK(pair1->two->mode)))
2836 df_conflict_remains = 1;
2838 if (merge_file_special_markers(o, &one, &a, &b, path,
2839 o->branch1, path1,
2840 o->branch2, path2, &mfi))
2841 return -1;
2844 * We can skip updating the working tree file iff:
2845 * a) The merge is clean
2846 * b) The merge matches what was in HEAD (content, mode, pathname)
2847 * c) The target path is usable (i.e. not involved in D/F conflict)
2849 if (mfi.clean &&
2850 was_tracked_and_matches(o, path, &mfi.oid, mfi.mode) &&
2851 !df_conflict_remains) {
2852 output(o, 3, _("Skipped %s (merged same as existing)"), path);
2853 if (add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2854 0, (!o->call_depth && !is_dirty), 0))
2855 return -1;
2856 return mfi.clean;
2859 if (!mfi.clean) {
2860 if (S_ISGITLINK(mfi.mode))
2861 reason = _("submodule");
2862 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2863 reason, path);
2864 if (rename_conflict_info && !df_conflict_remains)
2865 if (update_stages(o, path, &one, &a, &b))
2866 return -1;
2869 if (df_conflict_remains || is_dirty) {
2870 char *new_path;
2871 if (o->call_depth) {
2872 remove_file_from_cache(path);
2873 } else {
2874 if (!mfi.clean) {
2875 if (update_stages(o, path, &one, &a, &b))
2876 return -1;
2877 } else {
2878 int file_from_stage2 = was_tracked(o, path);
2879 struct diff_filespec merged;
2880 oidcpy(&merged.oid, &mfi.oid);
2881 merged.mode = mfi.mode;
2883 if (update_stages(o, path, NULL,
2884 file_from_stage2 ? &merged : NULL,
2885 file_from_stage2 ? NULL : &merged))
2886 return -1;
2890 new_path = unique_path(o, path, rename_conflict_info->branch1);
2891 if (is_dirty) {
2892 output(o, 1, _("Refusing to lose dirty file at %s"),
2893 path);
2895 output(o, 1, _("Adding as %s instead"), new_path);
2896 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2897 free(new_path);
2898 return -1;
2900 free(new_path);
2901 mfi.clean = 0;
2902 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2903 return -1;
2904 return !is_dirty && mfi.clean;
2907 static int conflict_rename_normal(struct merge_options *o,
2908 const char *path,
2909 struct object_id *o_oid, unsigned int o_mode,
2910 struct object_id *a_oid, unsigned int a_mode,
2911 struct object_id *b_oid, unsigned int b_mode,
2912 struct rename_conflict_info *ci)
2914 /* Merge the content and write it out */
2915 return merge_content(o, path, was_dirty(o, path),
2916 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2917 ci);
2920 /* Per entry merge function */
2921 static int process_entry(struct merge_options *o,
2922 const char *path, struct stage_data *entry)
2924 int clean_merge = 1;
2925 int normalize = o->renormalize;
2926 unsigned o_mode = entry->stages[1].mode;
2927 unsigned a_mode = entry->stages[2].mode;
2928 unsigned b_mode = entry->stages[3].mode;
2929 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2930 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2931 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2933 entry->processed = 1;
2934 if (entry->rename_conflict_info) {
2935 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2936 switch (conflict_info->rename_type) {
2937 case RENAME_NORMAL:
2938 case RENAME_ONE_FILE_TO_ONE:
2939 clean_merge = conflict_rename_normal(o,
2940 path,
2941 o_oid, o_mode,
2942 a_oid, a_mode,
2943 b_oid, b_mode,
2944 conflict_info);
2945 break;
2946 case RENAME_DIR:
2947 clean_merge = 1;
2948 if (conflict_rename_dir(o,
2949 conflict_info->pair1,
2950 conflict_info->branch1,
2951 conflict_info->branch2))
2952 clean_merge = -1;
2953 break;
2954 case RENAME_DELETE:
2955 clean_merge = 0;
2956 if (conflict_rename_delete(o,
2957 conflict_info->pair1,
2958 conflict_info->branch1,
2959 conflict_info->branch2))
2960 clean_merge = -1;
2961 break;
2962 case RENAME_ONE_FILE_TO_TWO:
2963 clean_merge = 0;
2964 if (conflict_rename_rename_1to2(o, conflict_info))
2965 clean_merge = -1;
2966 break;
2967 case RENAME_TWO_FILES_TO_ONE:
2968 clean_merge = 0;
2969 if (conflict_rename_rename_2to1(o, conflict_info))
2970 clean_merge = -1;
2971 break;
2972 default:
2973 entry->processed = 0;
2974 break;
2976 } else if (o_oid && (!a_oid || !b_oid)) {
2977 /* Case A: Deleted in one */
2978 if ((!a_oid && !b_oid) ||
2979 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2980 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2981 /* Deleted in both or deleted in one and
2982 * unchanged in the other */
2983 if (a_oid)
2984 output(o, 2, _("Removing %s"), path);
2985 /* do not touch working file if it did not exist */
2986 remove_file(o, 1, path, !a_oid);
2987 } else {
2988 /* Modify/delete; deleted side may have put a directory in the way */
2989 clean_merge = 0;
2990 if (handle_modify_delete(o, path, o_oid, o_mode,
2991 a_oid, a_mode, b_oid, b_mode))
2992 clean_merge = -1;
2994 } else if ((!o_oid && a_oid && !b_oid) ||
2995 (!o_oid && !a_oid && b_oid)) {
2996 /* Case B: Added in one. */
2997 /* [nothing|directory] -> ([nothing|directory], file) */
2999 const char *add_branch;
3000 const char *other_branch;
3001 unsigned mode;
3002 const struct object_id *oid;
3003 const char *conf;
3005 if (a_oid) {
3006 add_branch = o->branch1;
3007 other_branch = o->branch2;
3008 mode = a_mode;
3009 oid = a_oid;
3010 conf = _("file/directory");
3011 } else {
3012 add_branch = o->branch2;
3013 other_branch = o->branch1;
3014 mode = b_mode;
3015 oid = b_oid;
3016 conf = _("directory/file");
3018 if (dir_in_way(path,
3019 !o->call_depth && !S_ISGITLINK(a_mode),
3020 0)) {
3021 char *new_path = unique_path(o, path, add_branch);
3022 clean_merge = 0;
3023 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3024 "Adding %s as %s"),
3025 conf, path, other_branch, path, new_path);
3026 if (update_file(o, 0, oid, mode, new_path))
3027 clean_merge = -1;
3028 else if (o->call_depth)
3029 remove_file_from_cache(path);
3030 free(new_path);
3031 } else {
3032 output(o, 2, _("Adding %s"), path);
3033 /* do not overwrite file if already present */
3034 if (update_file_flags(o, oid, mode, path, 1, !a_oid))
3035 clean_merge = -1;
3037 } else if (a_oid && b_oid) {
3038 /* Case C: Added in both (check for same permissions) and */
3039 /* case D: Modified in both, but differently. */
3040 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3041 clean_merge = merge_content(o, path, is_dirty,
3042 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
3043 NULL);
3044 } else if (!o_oid && !a_oid && !b_oid) {
3046 * this entry was deleted altogether. a_mode == 0 means
3047 * we had that path and want to actively remove it.
3049 remove_file(o, 1, path, !a_mode);
3050 } else
3051 die("BUG: fatal merge failure, shouldn't happen.");
3053 return clean_merge;
3056 int merge_trees(struct merge_options *o,
3057 struct tree *head,
3058 struct tree *merge,
3059 struct tree *common,
3060 struct tree **result)
3062 int code, clean;
3064 if (o->subtree_shift) {
3065 merge = shift_tree_object(head, merge, o->subtree_shift);
3066 common = shift_tree_object(head, common, o->subtree_shift);
3069 if (oid_eq(&common->object.oid, &merge->object.oid)) {
3070 struct strbuf sb = STRBUF_INIT;
3072 if (!o->call_depth && index_has_changes(&sb)) {
3073 err(o, _("Dirty index: cannot merge (dirty: %s)"),
3074 sb.buf);
3075 return 0;
3077 output(o, 0, _("Already up to date!"));
3078 *result = head;
3079 return 1;
3082 code = git_merge_trees(o, common, head, merge);
3084 if (code != 0) {
3085 if (show(o, 4) || o->call_depth)
3086 err(o, _("merging of trees %s and %s failed"),
3087 oid_to_hex(&head->object.oid),
3088 oid_to_hex(&merge->object.oid));
3089 return -1;
3092 if (unmerged_cache()) {
3093 struct string_list *entries;
3094 struct rename_info re_info;
3095 int i;
3097 * Only need the hashmap while processing entries, so
3098 * initialize it here and free it when we are done running
3099 * through the entries. Keeping it in the merge_options as
3100 * opposed to decaring a local hashmap is for convenience
3101 * so that we don't have to pass it to around.
3103 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3104 get_files_dirs(o, head);
3105 get_files_dirs(o, merge);
3107 entries = get_unmerged();
3108 clean = handle_renames(o, common, head, merge, entries,
3109 &re_info);
3110 record_df_conflict_files(o, entries);
3111 if (clean < 0)
3112 goto cleanup;
3113 for (i = entries->nr-1; 0 <= i; i--) {
3114 const char *path = entries->items[i].string;
3115 struct stage_data *e = entries->items[i].util;
3116 if (!e->processed) {
3117 int ret = process_entry(o, path, e);
3118 if (!ret)
3119 clean = 0;
3120 else if (ret < 0) {
3121 clean = ret;
3122 goto cleanup;
3126 for (i = 0; i < entries->nr; i++) {
3127 struct stage_data *e = entries->items[i].util;
3128 if (!e->processed)
3129 die("BUG: unprocessed path??? %s",
3130 entries->items[i].string);
3133 cleanup:
3134 final_cleanup_renames(&re_info);
3136 string_list_clear(entries, 1);
3137 free(entries);
3139 hashmap_free(&o->current_file_dir_set, 1);
3141 if (clean < 0)
3142 return clean;
3144 else
3145 clean = 1;
3147 /* Free the extra index left from git_merge_trees() */
3149 * FIXME: Need to also free data allocated by
3150 * setup_unpack_trees_porcelain() tucked away in o->unpack_opts.msgs,
3151 * but the problem is that only half of it refers to dynamically
3152 * allocated data, while the other half points at static strings.
3154 discard_index(&o->orig_index);
3156 if (o->call_depth && !(*result = write_tree_from_memory(o)))
3157 return -1;
3159 return clean;
3162 static struct commit_list *reverse_commit_list(struct commit_list *list)
3164 struct commit_list *next = NULL, *current, *backup;
3165 for (current = list; current; current = backup) {
3166 backup = current->next;
3167 current->next = next;
3168 next = current;
3170 return next;
3174 * Merge the commits h1 and h2, return the resulting virtual
3175 * commit object and a flag indicating the cleanness of the merge.
3177 int merge_recursive(struct merge_options *o,
3178 struct commit *h1,
3179 struct commit *h2,
3180 struct commit_list *ca,
3181 struct commit **result)
3183 struct commit_list *iter;
3184 struct commit *merged_common_ancestors;
3185 struct tree *mrtree;
3186 int clean;
3188 if (show(o, 4)) {
3189 output(o, 4, _("Merging:"));
3190 output_commit_title(o, h1);
3191 output_commit_title(o, h2);
3194 if (!ca) {
3195 ca = get_merge_bases(h1, h2);
3196 ca = reverse_commit_list(ca);
3199 if (show(o, 5)) {
3200 unsigned cnt = commit_list_count(ca);
3202 output(o, 5, Q_("found %u common ancestor:",
3203 "found %u common ancestors:", cnt), cnt);
3204 for (iter = ca; iter; iter = iter->next)
3205 output_commit_title(o, iter->item);
3208 merged_common_ancestors = pop_commit(&ca);
3209 if (merged_common_ancestors == NULL) {
3210 /* if there is no common ancestor, use an empty tree */
3211 struct tree *tree;
3213 tree = lookup_tree(the_hash_algo->empty_tree);
3214 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3217 for (iter = ca; iter; iter = iter->next) {
3218 const char *saved_b1, *saved_b2;
3219 o->call_depth++;
3221 * When the merge fails, the result contains files
3222 * with conflict markers. The cleanness flag is
3223 * ignored (unless indicating an error), it was never
3224 * actually used, as result of merge_trees has always
3225 * overwritten it: the committed "conflicts" were
3226 * already resolved.
3228 discard_cache();
3229 saved_b1 = o->branch1;
3230 saved_b2 = o->branch2;
3231 o->branch1 = "Temporary merge branch 1";
3232 o->branch2 = "Temporary merge branch 2";
3233 if (merge_recursive(o, merged_common_ancestors, iter->item,
3234 NULL, &merged_common_ancestors) < 0)
3235 return -1;
3236 o->branch1 = saved_b1;
3237 o->branch2 = saved_b2;
3238 o->call_depth--;
3240 if (!merged_common_ancestors)
3241 return err(o, _("merge returned no commit"));
3244 discard_cache();
3245 if (!o->call_depth)
3246 read_cache();
3248 o->ancestor = "merged common ancestors";
3249 clean = merge_trees(o, get_commit_tree(h1), get_commit_tree(h2),
3250 get_commit_tree(merged_common_ancestors),
3251 &mrtree);
3252 if (clean < 0) {
3253 flush_output(o);
3254 return clean;
3257 if (o->call_depth) {
3258 *result = make_virtual_commit(mrtree, "merged tree");
3259 commit_list_insert(h1, &(*result)->parents);
3260 commit_list_insert(h2, &(*result)->parents->next);
3262 flush_output(o);
3263 if (!o->call_depth && o->buffer_output < 2)
3264 strbuf_release(&o->obuf);
3265 if (show(o, 2))
3266 diff_warn_rename_limit("merge.renamelimit",
3267 o->needed_rename_limit, 0);
3268 return clean;
3271 static struct commit *get_ref(const struct object_id *oid, const char *name)
3273 struct object *object;
3275 object = deref_tag(parse_object(oid), name, strlen(name));
3276 if (!object)
3277 return NULL;
3278 if (object->type == OBJ_TREE)
3279 return make_virtual_commit((struct tree*)object, name);
3280 if (object->type != OBJ_COMMIT)
3281 return NULL;
3282 if (parse_commit((struct commit *)object))
3283 return NULL;
3284 return (struct commit *)object;
3287 int merge_recursive_generic(struct merge_options *o,
3288 const struct object_id *head,
3289 const struct object_id *merge,
3290 int num_base_list,
3291 const struct object_id **base_list,
3292 struct commit **result)
3294 int clean;
3295 struct lock_file lock = LOCK_INIT;
3296 struct commit *head_commit = get_ref(head, o->branch1);
3297 struct commit *next_commit = get_ref(merge, o->branch2);
3298 struct commit_list *ca = NULL;
3300 if (base_list) {
3301 int i;
3302 for (i = 0; i < num_base_list; ++i) {
3303 struct commit *base;
3304 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3305 return err(o, _("Could not parse object '%s'"),
3306 oid_to_hex(base_list[i]));
3307 commit_list_insert(base, &ca);
3311 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3312 clean = merge_recursive(o, head_commit, next_commit, ca,
3313 result);
3314 if (clean < 0) {
3315 rollback_lock_file(&lock);
3316 return clean;
3319 if (write_locked_index(&the_index, &lock,
3320 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3321 return err(o, _("Unable to write index."));
3323 return clean ? 0 : 1;
3326 static void merge_recursive_config(struct merge_options *o)
3328 git_config_get_int("merge.verbosity", &o->verbosity);
3329 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3330 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3331 git_config(git_xmerge_config, NULL);
3334 void init_merge_options(struct merge_options *o)
3336 const char *merge_verbosity;
3337 memset(o, 0, sizeof(struct merge_options));
3338 o->verbosity = 2;
3339 o->buffer_output = 1;
3340 o->diff_rename_limit = -1;
3341 o->merge_rename_limit = -1;
3342 o->renormalize = 0;
3343 o->detect_rename = 1;
3344 merge_recursive_config(o);
3345 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3346 if (merge_verbosity)
3347 o->verbosity = strtol(merge_verbosity, NULL, 10);
3348 if (o->verbosity >= 5)
3349 o->buffer_output = 0;
3350 strbuf_init(&o->obuf, 0);
3351 string_list_init(&o->df_conflict_file_set, 1);
3354 int parse_merge_opt(struct merge_options *o, const char *s)
3356 const char *arg;
3358 if (!s || !*s)
3359 return -1;
3360 if (!strcmp(s, "ours"))
3361 o->recursive_variant = MERGE_RECURSIVE_OURS;
3362 else if (!strcmp(s, "theirs"))
3363 o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3364 else if (!strcmp(s, "subtree"))
3365 o->subtree_shift = "";
3366 else if (skip_prefix(s, "subtree=", &arg))
3367 o->subtree_shift = arg;
3368 else if (!strcmp(s, "patience"))
3369 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3370 else if (!strcmp(s, "histogram"))
3371 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3372 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3373 long value = parse_algorithm_value(arg);
3374 if (value < 0)
3375 return -1;
3376 /* clear out previous settings */
3377 DIFF_XDL_CLR(o, NEED_MINIMAL);
3378 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3379 o->xdl_opts |= value;
3381 else if (!strcmp(s, "ignore-space-change"))
3382 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3383 else if (!strcmp(s, "ignore-all-space"))
3384 DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3385 else if (!strcmp(s, "ignore-space-at-eol"))
3386 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3387 else if (!strcmp(s, "ignore-cr-at-eol"))
3388 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3389 else if (!strcmp(s, "renormalize"))
3390 o->renormalize = 1;
3391 else if (!strcmp(s, "no-renormalize"))
3392 o->renormalize = 0;
3393 else if (!strcmp(s, "no-renames"))
3394 o->detect_rename = 0;
3395 else if (!strcmp(s, "find-renames")) {
3396 o->detect_rename = 1;
3397 o->rename_score = 0;
3399 else if (skip_prefix(s, "find-renames=", &arg) ||
3400 skip_prefix(s, "rename-threshold=", &arg)) {
3401 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3402 return -1;
3403 o->detect_rename = 1;
3405 else
3406 return -1;
3407 return 0;