commit-reach(repo_get_merge_bases): pass on "missing commits" errors
[git.git] / merge-recursive.c
blob32e9d6665debefd667324abd18f2ba0e9289f2d6
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 "git-compat-util.h"
7 #include "merge-recursive.h"
9 #include "alloc.h"
10 #include "cache-tree.h"
11 #include "commit.h"
12 #include "commit-reach.h"
13 #include "config.h"
14 #include "diff.h"
15 #include "diffcore.h"
16 #include "dir.h"
17 #include "environment.h"
18 #include "gettext.h"
19 #include "hex.h"
20 #include "merge-ll.h"
21 #include "lockfile.h"
22 #include "match-trees.h"
23 #include "name-hash.h"
24 #include "object-file.h"
25 #include "object-name.h"
26 #include "object-store-ll.h"
27 #include "path.h"
28 #include "repository.h"
29 #include "revision.h"
30 #include "sparse-index.h"
31 #include "string-list.h"
32 #include "symlinks.h"
33 #include "tag.h"
34 #include "tree-walk.h"
35 #include "unpack-trees.h"
36 #include "xdiff-interface.h"
38 struct merge_options_internal {
39 int call_depth;
40 int needed_rename_limit;
41 struct hashmap current_file_dir_set;
42 struct string_list df_conflict_file_set;
43 struct unpack_trees_options unpack_opts;
44 struct index_state orig_index;
47 struct path_hashmap_entry {
48 struct hashmap_entry e;
49 char path[FLEX_ARRAY];
52 static int path_hashmap_cmp(const void *cmp_data UNUSED,
53 const struct hashmap_entry *eptr,
54 const struct hashmap_entry *entry_or_key,
55 const void *keydata)
57 const struct path_hashmap_entry *a, *b;
58 const char *key = keydata;
60 a = container_of(eptr, const struct path_hashmap_entry, e);
61 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
63 return fspathcmp(a->path, key ? key : b->path);
67 * For dir_rename_entry, directory names are stored as a full path from the
68 * toplevel of the repository and do not include a trailing '/'. Also:
70 * dir: original name of directory being renamed
71 * non_unique_new_dir: if true, could not determine new_dir
72 * new_dir: final name of directory being renamed
73 * possible_new_dirs: temporary used to help determine new_dir; see comments
74 * in get_directory_renames() for details
76 struct dir_rename_entry {
77 struct hashmap_entry ent;
78 char *dir;
79 unsigned non_unique_new_dir:1;
80 struct strbuf new_dir;
81 struct string_list possible_new_dirs;
84 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
85 char *dir)
87 struct dir_rename_entry key;
89 if (!dir)
90 return NULL;
91 hashmap_entry_init(&key.ent, strhash(dir));
92 key.dir = dir;
93 return hashmap_get_entry(hashmap, &key, ent, NULL);
96 static int dir_rename_cmp(const void *cmp_data UNUSED,
97 const struct hashmap_entry *eptr,
98 const struct hashmap_entry *entry_or_key,
99 const void *keydata UNUSED)
101 const struct dir_rename_entry *e1, *e2;
103 e1 = container_of(eptr, const struct dir_rename_entry, ent);
104 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
106 return strcmp(e1->dir, e2->dir);
109 static void dir_rename_init(struct hashmap *map)
111 hashmap_init(map, dir_rename_cmp, NULL, 0);
114 static void dir_rename_entry_init(struct dir_rename_entry *entry,
115 char *directory)
117 hashmap_entry_init(&entry->ent, strhash(directory));
118 entry->dir = directory;
119 entry->non_unique_new_dir = 0;
120 strbuf_init(&entry->new_dir, 0);
121 string_list_init_nodup(&entry->possible_new_dirs);
124 struct collision_entry {
125 struct hashmap_entry ent;
126 char *target_file;
127 struct string_list source_files;
128 unsigned reported_already:1;
131 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
132 char *target_file)
134 struct collision_entry key;
136 hashmap_entry_init(&key.ent, strhash(target_file));
137 key.target_file = target_file;
138 return hashmap_get_entry(hashmap, &key, ent, NULL);
141 static int collision_cmp(const void *cmp_data UNUSED,
142 const struct hashmap_entry *eptr,
143 const struct hashmap_entry *entry_or_key,
144 const void *keydata UNUSED)
146 const struct collision_entry *e1, *e2;
148 e1 = container_of(eptr, const struct collision_entry, ent);
149 e2 = container_of(entry_or_key, const struct collision_entry, ent);
151 return strcmp(e1->target_file, e2->target_file);
154 static void collision_init(struct hashmap *map)
156 hashmap_init(map, collision_cmp, NULL, 0);
159 static void flush_output(struct merge_options *opt)
161 if (opt->buffer_output < 2 && opt->obuf.len) {
162 fputs(opt->obuf.buf, stdout);
163 strbuf_reset(&opt->obuf);
167 __attribute__((format (printf, 2, 3)))
168 static int err(struct merge_options *opt, const char *err, ...)
170 va_list params;
172 if (opt->buffer_output < 2)
173 flush_output(opt);
174 else {
175 strbuf_complete(&opt->obuf, '\n');
176 strbuf_addstr(&opt->obuf, "error: ");
178 va_start(params, err);
179 strbuf_vaddf(&opt->obuf, err, params);
180 va_end(params);
181 if (opt->buffer_output > 1)
182 strbuf_addch(&opt->obuf, '\n');
183 else {
184 error("%s", opt->obuf.buf);
185 strbuf_reset(&opt->obuf);
188 return -1;
191 static struct tree *shift_tree_object(struct repository *repo,
192 struct tree *one, struct tree *two,
193 const char *subtree_shift)
195 struct object_id shifted;
197 if (!*subtree_shift) {
198 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
199 } else {
200 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
201 subtree_shift);
203 if (oideq(&two->object.oid, &shifted))
204 return two;
205 return lookup_tree(repo, &shifted);
208 static inline void set_commit_tree(struct commit *c, struct tree *t)
210 c->maybe_tree = t;
213 static struct commit *make_virtual_commit(struct repository *repo,
214 struct tree *tree,
215 const char *comment)
217 struct commit *commit = alloc_commit_node(repo);
219 set_merge_remote_desc(commit, comment, (struct object *)commit);
220 set_commit_tree(commit, tree);
221 commit->object.parsed = 1;
222 return commit;
225 enum rename_type {
226 RENAME_NORMAL = 0,
227 RENAME_VIA_DIR,
228 RENAME_ADD,
229 RENAME_DELETE,
230 RENAME_ONE_FILE_TO_ONE,
231 RENAME_ONE_FILE_TO_TWO,
232 RENAME_TWO_FILES_TO_ONE
236 * Since we want to write the index eventually, we cannot reuse the index
237 * for these (temporary) data.
239 struct stage_data {
240 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
241 struct rename_conflict_info *rename_conflict_info;
242 unsigned processed:1;
245 struct rename {
246 unsigned processed:1;
247 struct diff_filepair *pair;
248 const char *branch; /* branch that the rename occurred on */
250 * If directory rename detection affected this rename, what was its
251 * original type ('A' or 'R') and it's original destination before
252 * the directory rename (otherwise, '\0' and NULL for these two vars).
254 char dir_rename_original_type;
255 char *dir_rename_original_dest;
257 * Purpose of src_entry and dst_entry:
259 * If 'before' is renamed to 'after' then src_entry will contain
260 * the versions of 'before' from the merge_base, HEAD, and MERGE in
261 * stages 1, 2, and 3; dst_entry will contain the respective
262 * versions of 'after' in corresponding locations. Thus, we have a
263 * total of six modes and oids, though some will be null. (Stage 0
264 * is ignored; we're interested in handling conflicts.)
266 * Since we don't turn on break-rewrites by default, neither
267 * src_entry nor dst_entry can have all three of their stages have
268 * non-null oids, meaning at most four of the six will be non-null.
269 * Also, since this is a rename, both src_entry and dst_entry will
270 * have at least one non-null oid, meaning at least two will be
271 * non-null. Of the six oids, a typical rename will have three be
272 * non-null. Only two implies a rename/delete, and four implies a
273 * rename/add.
275 struct stage_data *src_entry;
276 struct stage_data *dst_entry;
279 struct rename_conflict_info {
280 enum rename_type rename_type;
281 struct rename *ren1;
282 struct rename *ren2;
285 static inline void setup_rename_conflict_info(enum rename_type rename_type,
286 struct merge_options *opt,
287 struct rename *ren1,
288 struct rename *ren2)
290 struct rename_conflict_info *ci;
293 * When we have two renames involved, it's easiest to get the
294 * correct things into stage 2 and 3, and to make sure that the
295 * content merge puts HEAD before the other branch if we just
296 * ensure that branch1 == opt->branch1. So, simply flip arguments
297 * around if we don't have that.
299 if (ren2 && ren1->branch != opt->branch1) {
300 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
301 return;
304 CALLOC_ARRAY(ci, 1);
305 ci->rename_type = rename_type;
306 ci->ren1 = ren1;
307 ci->ren2 = ren2;
309 ci->ren1->dst_entry->processed = 0;
310 ci->ren1->dst_entry->rename_conflict_info = ci;
311 if (ren2) {
312 ci->ren2->dst_entry->rename_conflict_info = ci;
316 static int show(struct merge_options *opt, int v)
318 return (!opt->priv->call_depth && opt->verbosity >= v) ||
319 opt->verbosity >= 5;
322 __attribute__((format (printf, 3, 4)))
323 static void output(struct merge_options *opt, int v, const char *fmt, ...)
325 va_list ap;
327 if (!show(opt, v))
328 return;
330 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
332 va_start(ap, fmt);
333 strbuf_vaddf(&opt->obuf, fmt, ap);
334 va_end(ap);
336 strbuf_addch(&opt->obuf, '\n');
337 if (!opt->buffer_output)
338 flush_output(opt);
341 static void repo_output_commit_title(struct merge_options *opt,
342 struct repository *repo,
343 struct commit *commit)
345 struct merge_remote_desc *desc;
347 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
348 desc = merge_remote_util(commit);
349 if (desc)
350 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
351 else {
352 strbuf_repo_add_unique_abbrev(&opt->obuf, repo,
353 &commit->object.oid,
354 DEFAULT_ABBREV);
355 strbuf_addch(&opt->obuf, ' ');
356 if (repo_parse_commit(repo, commit) != 0)
357 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
358 else {
359 const char *title;
360 const char *msg = repo_get_commit_buffer(repo, commit, NULL);
361 int len = find_commit_subject(msg, &title);
362 if (len)
363 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
364 repo_unuse_commit_buffer(repo, commit, msg);
367 flush_output(opt);
370 static void output_commit_title(struct merge_options *opt, struct commit *commit)
372 repo_output_commit_title(opt, the_repository, commit);
375 static int add_cacheinfo(struct merge_options *opt,
376 const struct diff_filespec *blob,
377 const char *path, int stage, int refresh, int options)
379 struct index_state *istate = opt->repo->index;
380 struct cache_entry *ce;
381 int ret;
383 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
384 if (!ce)
385 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
387 ret = add_index_entry(istate, ce, options);
388 if (refresh) {
389 struct cache_entry *nce;
391 nce = refresh_cache_entry(istate, ce,
392 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
393 if (!nce)
394 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
395 if (nce != ce)
396 ret = add_index_entry(istate, nce, options);
398 return ret;
401 static inline int merge_detect_rename(struct merge_options *opt)
403 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
406 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
408 parse_tree(tree);
409 init_tree_desc(desc, tree->buffer, tree->size);
412 static int unpack_trees_start(struct merge_options *opt,
413 struct tree *common,
414 struct tree *head,
415 struct tree *merge)
417 int rc;
418 struct tree_desc t[3];
419 struct index_state tmp_index = INDEX_STATE_INIT(opt->repo);
421 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
422 if (opt->priv->call_depth)
423 opt->priv->unpack_opts.index_only = 1;
424 else {
425 opt->priv->unpack_opts.update = 1;
426 /* FIXME: should only do this if !overwrite_ignore */
427 opt->priv->unpack_opts.preserve_ignored = 0;
429 opt->priv->unpack_opts.merge = 1;
430 opt->priv->unpack_opts.head_idx = 2;
431 opt->priv->unpack_opts.fn = threeway_merge;
432 opt->priv->unpack_opts.src_index = opt->repo->index;
433 opt->priv->unpack_opts.dst_index = &tmp_index;
434 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
435 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
437 init_tree_desc_from_tree(t+0, common);
438 init_tree_desc_from_tree(t+1, head);
439 init_tree_desc_from_tree(t+2, merge);
441 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
442 cache_tree_free(&opt->repo->index->cache_tree);
445 * Update opt->repo->index to match the new results, AFTER saving a
446 * copy in opt->priv->orig_index. Update src_index to point to the
447 * saved copy. (verify_uptodate() checks src_index, and the original
448 * index is the one that had the necessary modification timestamps.)
450 opt->priv->orig_index = *opt->repo->index;
451 *opt->repo->index = tmp_index;
452 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
454 return rc;
457 static void unpack_trees_finish(struct merge_options *opt)
459 discard_index(&opt->priv->orig_index);
460 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
463 static int save_files_dirs(const struct object_id *oid UNUSED,
464 struct strbuf *base, const char *path,
465 unsigned int mode, void *context)
467 struct path_hashmap_entry *entry;
468 int baselen = base->len;
469 struct merge_options *opt = context;
471 strbuf_addstr(base, path);
473 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
474 hashmap_entry_init(&entry->e, fspathhash(entry->path));
475 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
477 strbuf_setlen(base, baselen);
478 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
481 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
483 struct pathspec match_all;
484 memset(&match_all, 0, sizeof(match_all));
485 read_tree(opt->repo, tree,
486 &match_all, save_files_dirs, opt);
489 static int get_tree_entry_if_blob(struct repository *r,
490 const struct object_id *tree,
491 const char *path,
492 struct diff_filespec *dfs)
494 int ret;
496 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
497 if (S_ISDIR(dfs->mode)) {
498 oidcpy(&dfs->oid, null_oid());
499 dfs->mode = 0;
501 return ret;
505 * Returns an index_entry instance which doesn't have to correspond to
506 * a real cache entry in Git's index.
508 static struct stage_data *insert_stage_data(struct repository *r,
509 const char *path,
510 struct tree *o, struct tree *a, struct tree *b,
511 struct string_list *entries)
513 struct string_list_item *item;
514 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
515 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
516 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
517 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
518 item = string_list_insert(entries, path);
519 item->util = e;
520 return e;
524 * Create a dictionary mapping file names to stage_data objects. The
525 * dictionary contains one entry for every path with a non-zero stage entry.
527 static struct string_list *get_unmerged(struct index_state *istate)
529 struct string_list *unmerged = xmalloc(sizeof(struct string_list));
530 int i;
532 string_list_init_dup(unmerged);
534 /* TODO: audit for interaction with sparse-index. */
535 ensure_full_index(istate);
536 for (i = 0; i < istate->cache_nr; i++) {
537 struct string_list_item *item;
538 struct stage_data *e;
539 const struct cache_entry *ce = istate->cache[i];
540 if (!ce_stage(ce))
541 continue;
543 item = string_list_lookup(unmerged, ce->name);
544 if (!item) {
545 item = string_list_insert(unmerged, ce->name);
546 item->util = xcalloc(1, sizeof(struct stage_data));
548 e = item->util;
549 e->stages[ce_stage(ce)].mode = ce->ce_mode;
550 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
553 return unmerged;
556 static int string_list_df_name_compare(const char *one, const char *two)
558 int onelen = strlen(one);
559 int twolen = strlen(two);
561 * Here we only care that entries for D/F conflicts are
562 * adjacent, in particular with the file of the D/F conflict
563 * appearing before files below the corresponding directory.
564 * The order of the rest of the list is irrelevant for us.
566 * To achieve this, we sort with df_name_compare and provide
567 * the mode S_IFDIR so that D/F conflicts will sort correctly.
568 * We use the mode S_IFDIR for everything else for simplicity,
569 * since in other cases any changes in their order due to
570 * sorting cause no problems for us.
572 int cmp = df_name_compare(one, onelen, S_IFDIR,
573 two, twolen, S_IFDIR);
575 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
576 * that 'foo' comes before 'foo/bar'.
578 if (cmp)
579 return cmp;
580 return onelen - twolen;
583 static void record_df_conflict_files(struct merge_options *opt,
584 struct string_list *entries)
586 /* If there is a D/F conflict and the file for such a conflict
587 * currently exists in the working tree, we want to allow it to be
588 * removed to make room for the corresponding directory if needed.
589 * The files underneath the directories of such D/F conflicts will
590 * be processed before the corresponding file involved in the D/F
591 * conflict. If the D/F directory ends up being removed by the
592 * merge, then we won't have to touch the D/F file. If the D/F
593 * directory needs to be written to the working copy, then the D/F
594 * file will simply be removed (in make_room_for_path()) to make
595 * room for the necessary paths. Note that if both the directory
596 * and the file need to be present, then the D/F file will be
597 * reinstated with a new unique name at the time it is processed.
599 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
600 const char *last_file = NULL;
601 int last_len = 0;
602 int i;
605 * If we're merging merge-bases, we don't want to bother with
606 * any working directory changes.
608 if (opt->priv->call_depth)
609 return;
611 /* Ensure D/F conflicts are adjacent in the entries list. */
612 for (i = 0; i < entries->nr; i++) {
613 struct string_list_item *next = &entries->items[i];
614 string_list_append(&df_sorted_entries, next->string)->util =
615 next->util;
617 df_sorted_entries.cmp = string_list_df_name_compare;
618 string_list_sort(&df_sorted_entries);
620 string_list_clear(&opt->priv->df_conflict_file_set, 1);
621 for (i = 0; i < df_sorted_entries.nr; i++) {
622 const char *path = df_sorted_entries.items[i].string;
623 int len = strlen(path);
624 struct stage_data *e = df_sorted_entries.items[i].util;
627 * Check if last_file & path correspond to a D/F conflict;
628 * i.e. whether path is last_file+'/'+<something>.
629 * If so, record that it's okay to remove last_file to make
630 * room for path and friends if needed.
632 if (last_file &&
633 len > last_len &&
634 memcmp(path, last_file, last_len) == 0 &&
635 path[last_len] == '/') {
636 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
640 * Determine whether path could exist as a file in the
641 * working directory as a possible D/F conflict. This
642 * will only occur when it exists in stage 2 as a
643 * file.
645 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
646 last_file = path;
647 last_len = len;
648 } else {
649 last_file = NULL;
652 string_list_clear(&df_sorted_entries, 0);
655 static int update_stages(struct merge_options *opt, const char *path,
656 const struct diff_filespec *o,
657 const struct diff_filespec *a,
658 const struct diff_filespec *b)
662 * NOTE: It is usually a bad idea to call update_stages on a path
663 * before calling update_file on that same path, since it can
664 * sometimes lead to spurious "refusing to lose untracked file..."
665 * messages from update_file (via make_room_for path via
666 * would_lose_untracked). Instead, reverse the order of the calls
667 * (executing update_file first and then update_stages).
669 int clear = 1;
670 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
671 if (clear)
672 if (remove_file_from_index(opt->repo->index, path))
673 return -1;
674 if (o)
675 if (add_cacheinfo(opt, o, path, 1, 0, options))
676 return -1;
677 if (a)
678 if (add_cacheinfo(opt, a, path, 2, 0, options))
679 return -1;
680 if (b)
681 if (add_cacheinfo(opt, b, path, 3, 0, options))
682 return -1;
683 return 0;
686 static void update_entry(struct stage_data *entry,
687 struct diff_filespec *o,
688 struct diff_filespec *a,
689 struct diff_filespec *b)
691 entry->processed = 0;
692 entry->stages[1].mode = o->mode;
693 entry->stages[2].mode = a->mode;
694 entry->stages[3].mode = b->mode;
695 oidcpy(&entry->stages[1].oid, &o->oid);
696 oidcpy(&entry->stages[2].oid, &a->oid);
697 oidcpy(&entry->stages[3].oid, &b->oid);
700 static int remove_file(struct merge_options *opt, int clean,
701 const char *path, int no_wd)
703 int update_cache = opt->priv->call_depth || clean;
704 int update_working_directory = !opt->priv->call_depth && !no_wd;
706 if (update_cache) {
707 if (remove_file_from_index(opt->repo->index, path))
708 return -1;
710 if (update_working_directory) {
711 if (ignore_case) {
712 struct cache_entry *ce;
713 ce = index_file_exists(opt->repo->index, path, strlen(path),
714 ignore_case);
715 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
716 return 0;
718 if (remove_path(path))
719 return -1;
721 return 0;
724 /* add a string to a strbuf, but converting "/" to "_" */
725 static void add_flattened_path(struct strbuf *out, const char *s)
727 size_t i = out->len;
728 strbuf_addstr(out, s);
729 for (; i < out->len; i++)
730 if (out->buf[i] == '/')
731 out->buf[i] = '_';
734 static char *unique_path(struct merge_options *opt,
735 const char *path,
736 const char *branch)
738 struct path_hashmap_entry *entry;
739 struct strbuf newpath = STRBUF_INIT;
740 int suffix = 0;
741 size_t base_len;
743 strbuf_addf(&newpath, "%s~", path);
744 add_flattened_path(&newpath, branch);
746 base_len = newpath.len;
747 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
748 fspathhash(newpath.buf), newpath.buf) ||
749 (!opt->priv->call_depth && file_exists(newpath.buf))) {
750 strbuf_setlen(&newpath, base_len);
751 strbuf_addf(&newpath, "_%d", suffix++);
754 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
755 hashmap_entry_init(&entry->e, fspathhash(entry->path));
756 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
757 return strbuf_detach(&newpath, NULL);
761 * Check whether a directory in the index is in the way of an incoming
762 * file. Return 1 if so. If check_working_copy is non-zero, also
763 * check the working directory. If empty_ok is non-zero, also return
764 * 0 in the case where the working-tree dir exists but is empty.
766 static int dir_in_way(struct index_state *istate, const char *path,
767 int check_working_copy, int empty_ok)
769 int pos;
770 struct strbuf dirpath = STRBUF_INIT;
771 struct stat st;
773 strbuf_addstr(&dirpath, path);
774 strbuf_addch(&dirpath, '/');
776 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
778 if (pos < 0)
779 pos = -1 - pos;
780 if (pos < istate->cache_nr &&
781 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
782 strbuf_release(&dirpath);
783 return 1;
786 strbuf_release(&dirpath);
787 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
788 !(empty_ok && is_empty_dir(path)) &&
789 !has_symlink_leading_path(path, strlen(path));
793 * Returns whether path was tracked in the index before the merge started,
794 * and its oid and mode match the specified values
796 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
797 const struct diff_filespec *blob)
799 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
800 struct cache_entry *ce;
802 if (0 > pos)
803 /* we were not tracking this path before the merge */
804 return 0;
806 /* See if the file we were tracking before matches */
807 ce = opt->priv->orig_index.cache[pos];
808 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
812 * Returns whether path was tracked in the index before the merge started
814 static int was_tracked(struct merge_options *opt, const char *path)
816 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
818 if (0 <= pos)
819 /* we were tracking this path before the merge */
820 return 1;
822 return 0;
825 static int would_lose_untracked(struct merge_options *opt, const char *path)
827 struct index_state *istate = opt->repo->index;
830 * This may look like it can be simplified to:
831 * return !was_tracked(opt, path) && file_exists(path)
832 * but it can't. This function needs to know whether path was in
833 * the working tree due to EITHER having been tracked in the index
834 * before the merge OR having been put into the working copy and
835 * index by unpack_trees(). Due to that either-or requirement, we
836 * check the current index instead of the original one.
838 * Note that we do not need to worry about merge-recursive itself
839 * updating the index after unpack_trees() and before calling this
840 * function, because we strictly require all code paths in
841 * merge-recursive to update the working tree first and the index
842 * second. Doing otherwise would break
843 * update_file()/would_lose_untracked(); see every comment in this
844 * file which mentions "update_stages".
846 int pos = index_name_pos(istate, path, strlen(path));
848 if (pos < 0)
849 pos = -1 - pos;
850 while (pos < istate->cache_nr &&
851 !strcmp(path, istate->cache[pos]->name)) {
853 * If stage #0, it is definitely tracked.
854 * If it has stage #2 then it was tracked
855 * before this merge started. All other
856 * cases the path was not tracked.
858 switch (ce_stage(istate->cache[pos])) {
859 case 0:
860 case 2:
861 return 0;
863 pos++;
865 return file_exists(path);
868 static int was_dirty(struct merge_options *opt, const char *path)
870 struct cache_entry *ce;
871 int dirty = 1;
873 if (opt->priv->call_depth || !was_tracked(opt, path))
874 return !dirty;
876 ce = index_file_exists(opt->priv->unpack_opts.src_index,
877 path, strlen(path), ignore_case);
878 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
879 return dirty;
882 static int make_room_for_path(struct merge_options *opt, const char *path)
884 int status, i;
885 const char *msg = _("failed to create path '%s'%s");
887 /* Unlink any D/F conflict files that are in the way */
888 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
889 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
890 size_t pathlen = strlen(path);
891 size_t df_pathlen = strlen(df_path);
892 if (df_pathlen < pathlen &&
893 path[df_pathlen] == '/' &&
894 strncmp(path, df_path, df_pathlen) == 0) {
895 output(opt, 3,
896 _("Removing %s to make room for subdirectory\n"),
897 df_path);
898 unlink(df_path);
899 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
900 i, 0);
901 break;
905 /* Make sure leading directories are created */
906 status = safe_create_leading_directories_const(path);
907 if (status) {
908 if (status == SCLD_EXISTS)
909 /* something else exists */
910 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
911 return err(opt, msg, path, "");
915 * Do not unlink a file in the work tree if we are not
916 * tracking it.
918 if (would_lose_untracked(opt, path))
919 return err(opt, _("refusing to lose untracked file at '%s'"),
920 path);
922 /* Successful unlink is good.. */
923 if (!unlink(path))
924 return 0;
925 /* .. and so is no existing file */
926 if (errno == ENOENT)
927 return 0;
928 /* .. but not some other error (who really cares what?) */
929 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
932 static int update_file_flags(struct merge_options *opt,
933 const struct diff_filespec *contents,
934 const char *path,
935 int update_cache,
936 int update_wd)
938 int ret = 0;
940 if (opt->priv->call_depth)
941 update_wd = 0;
943 if (update_wd) {
944 enum object_type type;
945 void *buf;
946 unsigned long size;
948 if (S_ISGITLINK(contents->mode)) {
950 * We may later decide to recursively descend into
951 * the submodule directory and update its index
952 * and/or work tree, but we do not do that now.
954 update_wd = 0;
955 goto update_index;
958 buf = repo_read_object_file(the_repository, &contents->oid,
959 &type, &size);
960 if (!buf) {
961 ret = err(opt, _("cannot read object %s '%s'"),
962 oid_to_hex(&contents->oid), path);
963 goto free_buf;
965 if (type != OBJ_BLOB) {
966 ret = err(opt, _("blob expected for %s '%s'"),
967 oid_to_hex(&contents->oid), path);
968 goto free_buf;
970 if (S_ISREG(contents->mode)) {
971 struct strbuf strbuf = STRBUF_INIT;
972 if (convert_to_working_tree(opt->repo->index,
973 path, buf, size, &strbuf, NULL)) {
974 free(buf);
975 size = strbuf.len;
976 buf = strbuf_detach(&strbuf, NULL);
980 if (make_room_for_path(opt, path) < 0) {
981 update_wd = 0;
982 goto free_buf;
984 if (S_ISREG(contents->mode) ||
985 (!has_symlinks && S_ISLNK(contents->mode))) {
986 int fd;
987 int mode = (contents->mode & 0100 ? 0777 : 0666);
989 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
990 if (fd < 0) {
991 ret = err(opt, _("failed to open '%s': %s"),
992 path, strerror(errno));
993 goto free_buf;
995 write_in_full(fd, buf, size);
996 close(fd);
997 } else if (S_ISLNK(contents->mode)) {
998 char *lnk = xmemdupz(buf, size);
999 safe_create_leading_directories_const(path);
1000 unlink(path);
1001 if (symlink(lnk, path))
1002 ret = err(opt, _("failed to symlink '%s': %s"),
1003 path, strerror(errno));
1004 free(lnk);
1005 } else
1006 ret = err(opt,
1007 _("do not know what to do with %06o %s '%s'"),
1008 contents->mode, oid_to_hex(&contents->oid), path);
1009 free_buf:
1010 free(buf);
1012 update_index:
1013 if (!ret && update_cache) {
1014 int refresh = (!opt->priv->call_depth &&
1015 contents->mode != S_IFGITLINK);
1016 if (add_cacheinfo(opt, contents, path, 0, refresh,
1017 ADD_CACHE_OK_TO_ADD))
1018 return -1;
1020 return ret;
1023 static int update_file(struct merge_options *opt,
1024 int clean,
1025 const struct diff_filespec *contents,
1026 const char *path)
1028 return update_file_flags(opt, contents, path,
1029 opt->priv->call_depth || clean, !opt->priv->call_depth);
1032 /* Low level file merging, update and removal */
1034 struct merge_file_info {
1035 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1036 unsigned clean:1,
1037 merge:1;
1040 static int merge_3way(struct merge_options *opt,
1041 mmbuffer_t *result_buf,
1042 const struct diff_filespec *o,
1043 const struct diff_filespec *a,
1044 const struct diff_filespec *b,
1045 const char *branch1,
1046 const char *branch2,
1047 const int extra_marker_size)
1049 mmfile_t orig, src1, src2;
1050 struct ll_merge_options ll_opts = {0};
1051 char *base, *name1, *name2;
1052 enum ll_merge_result merge_status;
1054 ll_opts.renormalize = opt->renormalize;
1055 ll_opts.extra_marker_size = extra_marker_size;
1056 ll_opts.xdl_opts = opt->xdl_opts;
1058 if (opt->priv->call_depth) {
1059 ll_opts.virtual_ancestor = 1;
1060 ll_opts.variant = 0;
1061 } else {
1062 switch (opt->recursive_variant) {
1063 case MERGE_VARIANT_OURS:
1064 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1065 break;
1066 case MERGE_VARIANT_THEIRS:
1067 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1068 break;
1069 default:
1070 ll_opts.variant = 0;
1071 break;
1075 assert(a->path && b->path && o->path && opt->ancestor);
1076 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1077 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1078 name1 = mkpathdup("%s:%s", branch1, a->path);
1079 name2 = mkpathdup("%s:%s", branch2, b->path);
1080 } else {
1081 base = mkpathdup("%s", opt->ancestor);
1082 name1 = mkpathdup("%s", branch1);
1083 name2 = mkpathdup("%s", branch2);
1086 read_mmblob(&orig, &o->oid);
1087 read_mmblob(&src1, &a->oid);
1088 read_mmblob(&src2, &b->oid);
1091 * FIXME: Using a->path for normalization rules in ll_merge could be
1092 * wrong if we renamed from a->path to b->path. We should use the
1093 * target path for where the file will be written.
1095 merge_status = ll_merge(result_buf, a->path, &orig, base,
1096 &src1, name1, &src2, name2,
1097 opt->repo->index, &ll_opts);
1098 if (merge_status == LL_MERGE_BINARY_CONFLICT)
1099 warning("Cannot merge binary files: %s (%s vs. %s)",
1100 a->path, name1, name2);
1102 free(base);
1103 free(name1);
1104 free(name2);
1105 free(orig.ptr);
1106 free(src1.ptr);
1107 free(src2.ptr);
1108 return merge_status;
1111 static int find_first_merges(struct repository *repo,
1112 struct object_array *result, const char *path,
1113 struct commit *a, struct commit *b)
1115 int i, j;
1116 struct object_array merges = OBJECT_ARRAY_INIT;
1117 struct commit *commit;
1118 int contains_another;
1120 char merged_revision[GIT_MAX_HEXSZ + 2];
1121 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1122 "--all", merged_revision, NULL };
1123 struct rev_info revs;
1124 struct setup_revision_opt rev_opts;
1126 memset(result, 0, sizeof(struct object_array));
1127 memset(&rev_opts, 0, sizeof(rev_opts));
1129 /* get all revisions that merge commit a */
1130 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1131 oid_to_hex(&a->object.oid));
1132 repo_init_revisions(repo, &revs, NULL);
1133 /* FIXME: can't handle linked worktrees in submodules yet */
1134 revs.single_worktree = path != NULL;
1135 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1137 /* save all revisions from the above list that contain b */
1138 if (prepare_revision_walk(&revs))
1139 die("revision walk setup failed");
1140 while ((commit = get_revision(&revs)) != NULL) {
1141 struct object *o = &(commit->object);
1142 int ret = repo_in_merge_bases(repo, b, commit);
1143 if (ret < 0) {
1144 object_array_clear(&merges);
1145 release_revisions(&revs);
1146 return ret;
1148 if (ret)
1149 add_object_array(o, NULL, &merges);
1151 reset_revision_walk();
1153 /* Now we've got all merges that contain a and b. Prune all
1154 * merges that contain another found merge and save them in
1155 * result.
1157 for (i = 0; i < merges.nr; i++) {
1158 struct commit *m1 = (struct commit *) merges.objects[i].item;
1160 contains_another = 0;
1161 for (j = 0; j < merges.nr; j++) {
1162 struct commit *m2 = (struct commit *) merges.objects[j].item;
1163 if (i != j) {
1164 int ret = repo_in_merge_bases(repo, m2, m1);
1165 if (ret < 0) {
1166 object_array_clear(&merges);
1167 release_revisions(&revs);
1168 return ret;
1170 if (ret > 0) {
1171 contains_another = 1;
1172 break;
1177 if (!contains_another)
1178 add_object_array(merges.objects[i].item, NULL, result);
1181 object_array_clear(&merges);
1182 release_revisions(&revs);
1183 return result->nr;
1186 static void print_commit(struct repository *repo, struct commit *commit)
1188 struct strbuf sb = STRBUF_INIT;
1189 struct pretty_print_context ctx = {0};
1190 ctx.date_mode.type = DATE_NORMAL;
1191 /* FIXME: Merge this with output_commit_title() */
1192 assert(!merge_remote_util(commit));
1193 repo_format_commit_message(repo, commit, " %h: %m %s", &sb, &ctx);
1194 fprintf(stderr, "%s\n", sb.buf);
1195 strbuf_release(&sb);
1198 static int is_valid(const struct diff_filespec *dfs)
1200 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1203 static int merge_submodule(struct merge_options *opt,
1204 struct object_id *result, const char *path,
1205 const struct object_id *base, const struct object_id *a,
1206 const struct object_id *b)
1208 struct repository subrepo;
1209 int ret = 0, ret2;
1210 struct commit *commit_base, *commit_a, *commit_b;
1211 int parent_count;
1212 struct object_array merges;
1214 int i;
1215 int search = !opt->priv->call_depth;
1217 /* store a in result in case we fail */
1218 /* FIXME: This is the WRONG resolution for the recursive case when
1219 * we need to be careful to avoid accidentally matching either side.
1220 * Should probably use o instead there, much like we do for merging
1221 * binaries.
1223 oidcpy(result, a);
1225 /* we can not handle deletion conflicts */
1226 if (is_null_oid(base))
1227 return 0;
1228 if (is_null_oid(a))
1229 return 0;
1230 if (is_null_oid(b))
1231 return 0;
1233 if (repo_submodule_init(&subrepo, opt->repo, path, null_oid())) {
1234 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1235 return 0;
1238 if (!(commit_base = lookup_commit_reference(&subrepo, base)) ||
1239 !(commit_a = lookup_commit_reference(&subrepo, a)) ||
1240 !(commit_b = lookup_commit_reference(&subrepo, b))) {
1241 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1242 goto cleanup;
1245 /* check whether both changes are forward */
1246 ret2 = repo_in_merge_bases(&subrepo, commit_base, commit_a);
1247 if (ret2 < 0) {
1248 output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1249 goto cleanup;
1251 if (ret2 > 0)
1252 ret2 = repo_in_merge_bases(&subrepo, commit_base, commit_b);
1253 if (ret2 < 0) {
1254 output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1255 goto cleanup;
1257 if (!ret2) {
1258 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1259 goto cleanup;
1262 /* Case #1: a is contained in b or vice versa */
1263 ret2 = repo_in_merge_bases(&subrepo, commit_a, commit_b);
1264 if (ret2 < 0) {
1265 output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1266 goto cleanup;
1268 if (ret2) {
1269 oidcpy(result, b);
1270 if (show(opt, 3)) {
1271 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1272 repo_output_commit_title(opt, &subrepo, commit_b);
1273 } else if (show(opt, 2))
1274 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1275 else
1276 ; /* no output */
1278 ret = 1;
1279 goto cleanup;
1281 ret2 = repo_in_merge_bases(&subrepo, commit_b, commit_a);
1282 if (ret2 < 0) {
1283 output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1284 goto cleanup;
1286 if (ret2) {
1287 oidcpy(result, a);
1288 if (show(opt, 3)) {
1289 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1290 repo_output_commit_title(opt, &subrepo, commit_a);
1291 } else if (show(opt, 2))
1292 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1293 else
1294 ; /* no output */
1296 ret = 1;
1297 goto cleanup;
1301 * Case #2: There are one or more merges that contain a and b in
1302 * the submodule. If there is only one, then present it as a
1303 * suggestion to the user, but leave it marked unmerged so the
1304 * user needs to confirm the resolution.
1307 /* Skip the search if makes no sense to the calling context. */
1308 if (!search)
1309 goto cleanup;
1311 /* find commit which merges them */
1312 parent_count = find_first_merges(&subrepo, &merges, path,
1313 commit_a, commit_b);
1314 switch (parent_count) {
1315 case 0:
1316 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1317 break;
1319 case 1:
1320 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1321 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1322 print_commit(&subrepo, (struct commit *) merges.objects[0].item);
1323 output(opt, 2, _(
1324 "If this is correct simply add it to the index "
1325 "for example\n"
1326 "by using:\n\n"
1327 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1328 "which will accept this suggestion.\n"),
1329 oid_to_hex(&merges.objects[0].item->oid), path);
1330 break;
1332 default:
1333 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1334 for (i = 0; i < merges.nr; i++)
1335 print_commit(&subrepo, (struct commit *) merges.objects[i].item);
1338 object_array_clear(&merges);
1339 cleanup:
1340 repo_clear(&subrepo);
1341 return ret;
1344 static int merge_mode_and_contents(struct merge_options *opt,
1345 const struct diff_filespec *o,
1346 const struct diff_filespec *a,
1347 const struct diff_filespec *b,
1348 const char *filename,
1349 const char *branch1,
1350 const char *branch2,
1351 const int extra_marker_size,
1352 struct merge_file_info *result)
1354 if (opt->branch1 != branch1) {
1356 * It's weird getting a reverse merge with HEAD on the bottom
1357 * side of the conflict markers and the other branch on the
1358 * top. Fix that.
1360 return merge_mode_and_contents(opt, o, b, a,
1361 filename,
1362 branch2, branch1,
1363 extra_marker_size, result);
1366 result->merge = 0;
1367 result->clean = 1;
1369 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1370 result->clean = 0;
1372 * FIXME: This is a bad resolution for recursive case; for
1373 * the recursive case we want something that is unlikely to
1374 * accidentally match either side. Also, while it makes
1375 * sense to prefer regular files over symlinks, it doesn't
1376 * make sense to prefer regular files over submodules.
1378 if (S_ISREG(a->mode)) {
1379 result->blob.mode = a->mode;
1380 oidcpy(&result->blob.oid, &a->oid);
1381 } else {
1382 result->blob.mode = b->mode;
1383 oidcpy(&result->blob.oid, &b->oid);
1385 } else {
1386 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1387 result->merge = 1;
1390 * Merge modes
1392 if (a->mode == b->mode || a->mode == o->mode)
1393 result->blob.mode = b->mode;
1394 else {
1395 result->blob.mode = a->mode;
1396 if (b->mode != o->mode) {
1397 result->clean = 0;
1398 result->merge = 1;
1402 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1403 oidcpy(&result->blob.oid, &b->oid);
1404 else if (oideq(&b->oid, &o->oid))
1405 oidcpy(&result->blob.oid, &a->oid);
1406 else if (S_ISREG(a->mode)) {
1407 mmbuffer_t result_buf;
1408 int ret = 0, merge_status;
1410 merge_status = merge_3way(opt, &result_buf, o, a, b,
1411 branch1, branch2,
1412 extra_marker_size);
1414 if ((merge_status < 0) || !result_buf.ptr)
1415 ret = err(opt, _("failed to execute internal merge"));
1417 if (!ret &&
1418 write_object_file(result_buf.ptr, result_buf.size,
1419 OBJ_BLOB, &result->blob.oid))
1420 ret = err(opt, _("unable to add %s to database"),
1421 a->path);
1423 free(result_buf.ptr);
1424 if (ret)
1425 return ret;
1426 /* FIXME: bug, what if modes didn't match? */
1427 result->clean = (merge_status == 0);
1428 } else if (S_ISGITLINK(a->mode)) {
1429 result->clean = merge_submodule(opt, &result->blob.oid,
1430 o->path,
1431 &o->oid,
1432 &a->oid,
1433 &b->oid);
1434 if (result->clean < 0)
1435 return -1;
1436 } else if (S_ISLNK(a->mode)) {
1437 switch (opt->recursive_variant) {
1438 case MERGE_VARIANT_NORMAL:
1439 oidcpy(&result->blob.oid, &a->oid);
1440 if (!oideq(&a->oid, &b->oid))
1441 result->clean = 0;
1442 break;
1443 case MERGE_VARIANT_OURS:
1444 oidcpy(&result->blob.oid, &a->oid);
1445 break;
1446 case MERGE_VARIANT_THEIRS:
1447 oidcpy(&result->blob.oid, &b->oid);
1448 break;
1450 } else
1451 BUG("unsupported object type in the tree");
1454 if (result->merge)
1455 output(opt, 2, _("Auto-merging %s"), filename);
1457 return 0;
1460 static int handle_rename_via_dir(struct merge_options *opt,
1461 struct rename_conflict_info *ci)
1464 * Handle file adds that need to be renamed due to directory rename
1465 * detection. This differs from handle_rename_normal, because
1466 * there is no content merge to do; just move the file into the
1467 * desired final location.
1469 const struct rename *ren = ci->ren1;
1470 const struct diff_filespec *dest = ren->pair->two;
1471 char *file_path = dest->path;
1472 int mark_conflicted = (opt->detect_directory_renames ==
1473 MERGE_DIRECTORY_RENAMES_CONFLICT);
1474 assert(ren->dir_rename_original_dest);
1476 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1477 mark_conflicted = 1;
1478 file_path = unique_path(opt, dest->path, ren->branch);
1479 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1480 "writing to %s instead."),
1481 dest->path, file_path);
1484 if (mark_conflicted) {
1486 * Write the file in worktree at file_path. In the index,
1487 * only record the file at dest->path in the appropriate
1488 * higher stage.
1490 if (update_file(opt, 0, dest, file_path))
1491 return -1;
1492 if (file_path != dest->path)
1493 free(file_path);
1494 if (update_stages(opt, dest->path, NULL,
1495 ren->branch == opt->branch1 ? dest : NULL,
1496 ren->branch == opt->branch1 ? NULL : dest))
1497 return -1;
1498 return 0; /* not clean, but conflicted */
1499 } else {
1500 /* Update dest->path both in index and in worktree */
1501 if (update_file(opt, 1, dest, dest->path))
1502 return -1;
1503 return 1; /* clean */
1507 static int handle_change_delete(struct merge_options *opt,
1508 const char *path, const char *old_path,
1509 const struct diff_filespec *o,
1510 const struct diff_filespec *changed,
1511 const char *change_branch,
1512 const char *delete_branch,
1513 const char *change, const char *change_past)
1515 char *alt_path = NULL;
1516 const char *update_path = path;
1517 int ret = 0;
1519 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1520 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1521 update_path = alt_path = unique_path(opt, path, change_branch);
1524 if (opt->priv->call_depth) {
1526 * We cannot arbitrarily accept either a_sha or b_sha as
1527 * correct; since there is no true "middle point" between
1528 * them, simply reuse the base version for virtual merge base.
1530 ret = remove_file_from_index(opt->repo->index, path);
1531 if (!ret)
1532 ret = update_file(opt, 0, o, update_path);
1533 } else {
1535 * Despite the four nearly duplicate messages and argument
1536 * lists below and the ugliness of the nested if-statements,
1537 * having complete messages makes the job easier for
1538 * translators.
1540 * The slight variance among the cases is due to the fact
1541 * that:
1542 * 1) directory/file conflicts (in effect if
1543 * !alt_path) could cause us to need to write the
1544 * file to a different path.
1545 * 2) renames (in effect if !old_path) could mean that
1546 * there are two names for the path that the user
1547 * may know the file by.
1549 if (!alt_path) {
1550 if (!old_path) {
1551 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1552 "and %s in %s. Version %s of %s left in tree."),
1553 change, path, delete_branch, change_past,
1554 change_branch, change_branch, path);
1555 } else {
1556 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1557 "and %s to %s in %s. Version %s of %s left in tree."),
1558 change, old_path, delete_branch, change_past, path,
1559 change_branch, change_branch, path);
1561 } else {
1562 if (!old_path) {
1563 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1564 "and %s in %s. Version %s of %s left in tree at %s."),
1565 change, path, delete_branch, change_past,
1566 change_branch, change_branch, path, alt_path);
1567 } else {
1568 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1569 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1570 change, old_path, delete_branch, change_past, path,
1571 change_branch, change_branch, path, alt_path);
1575 * No need to call update_file() on path when change_branch ==
1576 * opt->branch1 && !alt_path, since that would needlessly touch
1577 * path. We could call update_file_flags() with update_cache=0
1578 * and update_wd=0, but that's a no-op.
1580 if (change_branch != opt->branch1 || alt_path)
1581 ret = update_file(opt, 0, changed, update_path);
1583 free(alt_path);
1585 return ret;
1588 static int handle_rename_delete(struct merge_options *opt,
1589 struct rename_conflict_info *ci)
1591 const struct rename *ren = ci->ren1;
1592 const struct diff_filespec *orig = ren->pair->one;
1593 const struct diff_filespec *dest = ren->pair->two;
1594 const char *rename_branch = ren->branch;
1595 const char *delete_branch = (opt->branch1 == ren->branch ?
1596 opt->branch2 : opt->branch1);
1598 if (handle_change_delete(opt,
1599 opt->priv->call_depth ? orig->path : dest->path,
1600 opt->priv->call_depth ? NULL : orig->path,
1601 orig, dest,
1602 rename_branch, delete_branch,
1603 _("rename"), _("renamed")))
1604 return -1;
1606 if (opt->priv->call_depth)
1607 return remove_file_from_index(opt->repo->index, dest->path);
1608 else
1609 return update_stages(opt, dest->path, NULL,
1610 rename_branch == opt->branch1 ? dest : NULL,
1611 rename_branch == opt->branch1 ? NULL : dest);
1614 static int handle_file_collision(struct merge_options *opt,
1615 const char *collide_path,
1616 const char *prev_path1,
1617 const char *prev_path2,
1618 const char *branch1, const char *branch2,
1619 struct diff_filespec *a,
1620 struct diff_filespec *b)
1622 struct merge_file_info mfi;
1623 struct diff_filespec null;
1624 char *alt_path = NULL;
1625 const char *update_path = collide_path;
1628 * It's easiest to get the correct things into stage 2 and 3, and
1629 * to make sure that the content merge puts HEAD before the other
1630 * branch if we just ensure that branch1 == opt->branch1. So, simply
1631 * flip arguments around if we don't have that.
1633 if (branch1 != opt->branch1) {
1634 return handle_file_collision(opt, collide_path,
1635 prev_path2, prev_path1,
1636 branch2, branch1,
1637 b, a);
1640 /* Remove rename sources if rename/add or rename/rename(2to1) */
1641 if (prev_path1)
1642 remove_file(opt, 1, prev_path1,
1643 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1644 if (prev_path2)
1645 remove_file(opt, 1, prev_path2,
1646 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1649 * Remove the collision path, if it wouldn't cause dirty contents
1650 * or an untracked file to get lost. We'll either overwrite with
1651 * merged contents, or just write out to differently named files.
1653 if (was_dirty(opt, collide_path)) {
1654 output(opt, 1, _("Refusing to lose dirty file at %s"),
1655 collide_path);
1656 update_path = alt_path = unique_path(opt, collide_path, "merged");
1657 } else if (would_lose_untracked(opt, collide_path)) {
1659 * Only way we get here is if both renames were from
1660 * a directory rename AND user had an untracked file
1661 * at the location where both files end up after the
1662 * two directory renames. See testcase 10d of t6043.
1664 output(opt, 1, _("Refusing to lose untracked file at "
1665 "%s, even though it's in the way."),
1666 collide_path);
1667 update_path = alt_path = unique_path(opt, collide_path, "merged");
1668 } else {
1670 * FIXME: It's possible that the two files are identical
1671 * and that the current working copy happens to match, in
1672 * which case we are unnecessarily touching the working
1673 * tree file. It's not a likely enough scenario that I
1674 * want to code up the checks for it and a better fix is
1675 * available if we restructure how unpack_trees() and
1676 * merge-recursive interoperate anyway, so punting for
1677 * now...
1679 remove_file(opt, 0, collide_path, 0);
1682 /* Store things in diff_filespecs for functions that need it */
1683 null.path = (char *)collide_path;
1684 oidcpy(&null.oid, null_oid());
1685 null.mode = 0;
1687 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1688 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1689 return -1;
1690 mfi.clean &= !alt_path;
1691 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1692 return -1;
1693 if (!mfi.clean && !opt->priv->call_depth &&
1694 update_stages(opt, collide_path, NULL, a, b))
1695 return -1;
1696 free(alt_path);
1698 * FIXME: If both a & b both started with conflicts (only possible
1699 * if they came from a rename/rename(2to1)), but had IDENTICAL
1700 * contents including those conflicts, then in the next line we claim
1701 * it was clean. If someone cares about this case, we should have the
1702 * caller notify us if we started with conflicts.
1704 return mfi.clean;
1707 static int handle_rename_add(struct merge_options *opt,
1708 struct rename_conflict_info *ci)
1710 /* a was renamed to c, and a separate c was added. */
1711 struct diff_filespec *a = ci->ren1->pair->one;
1712 struct diff_filespec *c = ci->ren1->pair->two;
1713 char *path = c->path;
1714 char *prev_path_desc;
1715 struct merge_file_info mfi;
1717 const char *rename_branch = ci->ren1->branch;
1718 const char *add_branch = (opt->branch1 == rename_branch ?
1719 opt->branch2 : opt->branch1);
1720 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1722 output(opt, 1, _("CONFLICT (rename/add): "
1723 "Rename %s->%s in %s. Added %s in %s"),
1724 a->path, c->path, rename_branch,
1725 c->path, add_branch);
1727 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1728 ci->ren1->src_entry->stages[other_stage].path = a->path;
1729 if (merge_mode_and_contents(opt, a, c,
1730 &ci->ren1->src_entry->stages[other_stage],
1731 prev_path_desc,
1732 opt->branch1, opt->branch2,
1733 1 + opt->priv->call_depth * 2, &mfi))
1734 return -1;
1735 free(prev_path_desc);
1737 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1738 return handle_file_collision(opt,
1739 c->path, a->path, NULL,
1740 rename_branch, add_branch,
1741 &mfi.blob,
1742 &ci->ren1->dst_entry->stages[other_stage]);
1745 static char *find_path_for_conflict(struct merge_options *opt,
1746 const char *path,
1747 const char *branch1,
1748 const char *branch2)
1750 char *new_path = NULL;
1751 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1752 new_path = unique_path(opt, path, branch1);
1753 output(opt, 1, _("%s is a directory in %s adding "
1754 "as %s instead"),
1755 path, branch2, new_path);
1756 } else if (would_lose_untracked(opt, path)) {
1757 new_path = unique_path(opt, path, branch1);
1758 output(opt, 1, _("Refusing to lose untracked file"
1759 " at %s; adding as %s instead"),
1760 path, new_path);
1763 return new_path;
1767 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1769 static inline int flip_stage(int stage)
1771 return (2 + 3) - stage;
1774 static int handle_rename_rename_1to2(struct merge_options *opt,
1775 struct rename_conflict_info *ci)
1777 /* One file was renamed in both branches, but to different names. */
1778 struct merge_file_info mfi;
1779 struct diff_filespec *add;
1780 struct diff_filespec *o = ci->ren1->pair->one;
1781 struct diff_filespec *a = ci->ren1->pair->two;
1782 struct diff_filespec *b = ci->ren2->pair->two;
1783 char *path_desc;
1785 output(opt, 1, _("CONFLICT (rename/rename): "
1786 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1787 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1788 o->path, a->path, ci->ren1->branch,
1789 o->path, b->path, ci->ren2->branch,
1790 opt->priv->call_depth ? _(" (left unresolved)") : "");
1792 path_desc = xstrfmt("%s and %s, both renamed from %s",
1793 a->path, b->path, o->path);
1794 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1795 ci->ren1->branch, ci->ren2->branch,
1796 opt->priv->call_depth * 2, &mfi))
1797 return -1;
1798 free(path_desc);
1800 if (opt->priv->call_depth)
1801 remove_file_from_index(opt->repo->index, o->path);
1804 * For each destination path, we need to see if there is a
1805 * rename/add collision. If not, we can write the file out
1806 * to the specified location.
1808 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1809 if (is_valid(add)) {
1810 add->path = mfi.blob.path = a->path;
1811 if (handle_file_collision(opt, a->path,
1812 NULL, NULL,
1813 ci->ren1->branch,
1814 ci->ren2->branch,
1815 &mfi.blob, add) < 0)
1816 return -1;
1817 } else {
1818 char *new_path = find_path_for_conflict(opt, a->path,
1819 ci->ren1->branch,
1820 ci->ren2->branch);
1821 if (update_file(opt, 0, &mfi.blob,
1822 new_path ? new_path : a->path))
1823 return -1;
1824 free(new_path);
1825 if (!opt->priv->call_depth &&
1826 update_stages(opt, a->path, NULL, a, NULL))
1827 return -1;
1830 if (!mfi.clean && mfi.blob.mode == a->mode &&
1831 oideq(&mfi.blob.oid, &a->oid)) {
1833 * Getting here means we were attempting to merge a binary
1834 * blob. Since we can't merge binaries, the merge algorithm
1835 * just takes one side. But we don't want to copy the
1836 * contents of one side to both paths; we'd rather use the
1837 * original content at the given path for each path.
1839 oidcpy(&mfi.blob.oid, &b->oid);
1840 mfi.blob.mode = b->mode;
1842 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1843 if (is_valid(add)) {
1844 add->path = mfi.blob.path = b->path;
1845 if (handle_file_collision(opt, b->path,
1846 NULL, NULL,
1847 ci->ren1->branch,
1848 ci->ren2->branch,
1849 add, &mfi.blob) < 0)
1850 return -1;
1851 } else {
1852 char *new_path = find_path_for_conflict(opt, b->path,
1853 ci->ren2->branch,
1854 ci->ren1->branch);
1855 if (update_file(opt, 0, &mfi.blob,
1856 new_path ? new_path : b->path))
1857 return -1;
1858 free(new_path);
1859 if (!opt->priv->call_depth &&
1860 update_stages(opt, b->path, NULL, NULL, b))
1861 return -1;
1864 return 0;
1867 static int handle_rename_rename_2to1(struct merge_options *opt,
1868 struct rename_conflict_info *ci)
1870 /* Two files, a & b, were renamed to the same thing, c. */
1871 struct diff_filespec *a = ci->ren1->pair->one;
1872 struct diff_filespec *b = ci->ren2->pair->one;
1873 struct diff_filespec *c1 = ci->ren1->pair->two;
1874 struct diff_filespec *c2 = ci->ren2->pair->two;
1875 char *path = c1->path; /* == c2->path */
1876 char *path_side_1_desc;
1877 char *path_side_2_desc;
1878 struct merge_file_info mfi_c1;
1879 struct merge_file_info mfi_c2;
1880 int ostage1, ostage2;
1882 output(opt, 1, _("CONFLICT (rename/rename): "
1883 "Rename %s->%s in %s. "
1884 "Rename %s->%s in %s"),
1885 a->path, c1->path, ci->ren1->branch,
1886 b->path, c2->path, ci->ren2->branch);
1888 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1889 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1890 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1891 ostage2 = flip_stage(ostage1);
1892 ci->ren1->src_entry->stages[ostage1].path = a->path;
1893 ci->ren2->src_entry->stages[ostage2].path = b->path;
1894 if (merge_mode_and_contents(opt, a, c1,
1895 &ci->ren1->src_entry->stages[ostage1],
1896 path_side_1_desc,
1897 opt->branch1, opt->branch2,
1898 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1899 merge_mode_and_contents(opt, b,
1900 &ci->ren2->src_entry->stages[ostage2],
1901 c2, path_side_2_desc,
1902 opt->branch1, opt->branch2,
1903 1 + opt->priv->call_depth * 2, &mfi_c2))
1904 return -1;
1905 free(path_side_1_desc);
1906 free(path_side_2_desc);
1907 mfi_c1.blob.path = path;
1908 mfi_c2.blob.path = path;
1910 return handle_file_collision(opt, path, a->path, b->path,
1911 ci->ren1->branch, ci->ren2->branch,
1912 &mfi_c1.blob, &mfi_c2.blob);
1916 * Get the diff_filepairs changed between o_tree and tree.
1918 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1919 struct tree *o_tree,
1920 struct tree *tree)
1922 struct diff_queue_struct *ret;
1923 struct diff_options opts;
1925 repo_diff_setup(opt->repo, &opts);
1926 opts.flags.recursive = 1;
1927 opts.flags.rename_empty = 0;
1928 opts.detect_rename = merge_detect_rename(opt);
1930 * We do not have logic to handle the detection of copies. In
1931 * fact, it may not even make sense to add such logic: would we
1932 * really want a change to a base file to be propagated through
1933 * multiple other files by a merge?
1935 if (opts.detect_rename > DIFF_DETECT_RENAME)
1936 opts.detect_rename = DIFF_DETECT_RENAME;
1937 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 7000;
1938 opts.rename_score = opt->rename_score;
1939 opts.show_rename_progress = opt->show_rename_progress;
1940 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1941 diff_setup_done(&opts);
1942 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1943 diffcore_std(&opts);
1944 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1945 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1947 ret = xmalloc(sizeof(*ret));
1948 *ret = diff_queued_diff;
1950 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1951 diff_queued_diff.nr = 0;
1952 diff_queued_diff.queue = NULL;
1953 diff_flush(&opts);
1954 return ret;
1957 static int tree_has_path(struct repository *r, struct tree *tree,
1958 const char *path)
1960 struct object_id hashy;
1961 unsigned short mode_o;
1963 return !get_tree_entry(r,
1964 &tree->object.oid, path,
1965 &hashy, &mode_o);
1969 * Return a new string that replaces the beginning portion (which matches
1970 * entry->dir), with entry->new_dir. In perl-speak:
1971 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1972 * NOTE:
1973 * Caller must ensure that old_path starts with entry->dir + '/'.
1975 static char *apply_dir_rename(struct dir_rename_entry *entry,
1976 const char *old_path)
1978 struct strbuf new_path = STRBUF_INIT;
1979 int oldlen, newlen;
1981 if (entry->non_unique_new_dir)
1982 return NULL;
1984 oldlen = strlen(entry->dir);
1985 if (entry->new_dir.len == 0)
1987 * If someone renamed/merged a subdirectory into the root
1988 * directory (e.g. 'some/subdir' -> ''), then we want to
1989 * avoid returning
1990 * '' + '/filename'
1991 * as the rename; we need to make old_path + oldlen advance
1992 * past the '/' character.
1994 oldlen++;
1995 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1996 strbuf_grow(&new_path, newlen);
1997 strbuf_addbuf(&new_path, &entry->new_dir);
1998 strbuf_addstr(&new_path, &old_path[oldlen]);
2000 return strbuf_detach(&new_path, NULL);
2003 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
2004 char **old_dir, char **new_dir)
2006 char *end_of_old, *end_of_new;
2008 /* Default return values: NULL, meaning no rename */
2009 *old_dir = NULL;
2010 *new_dir = NULL;
2013 * For
2014 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
2015 * the "e/foo.c" part is the same, we just want to know that
2016 * "a/b/c/d" was renamed to "a/b/some/thing/else"
2017 * so, for this example, this function returns "a/b/c/d" in
2018 * *old_dir and "a/b/some/thing/else" in *new_dir.
2022 * If the basename of the file changed, we don't care. We want
2023 * to know which portion of the directory, if any, changed.
2025 end_of_old = strrchr(old_path, '/');
2026 end_of_new = strrchr(new_path, '/');
2029 * If end_of_old is NULL, old_path wasn't in a directory, so there
2030 * could not be a directory rename (our rule elsewhere that a
2031 * directory which still exists is not considered to have been
2032 * renamed means the root directory can never be renamed -- because
2033 * the root directory always exists).
2035 if (!end_of_old)
2036 return; /* Note: *old_dir and *new_dir are still NULL */
2039 * If new_path contains no directory (end_of_new is NULL), then we
2040 * have a rename of old_path's directory to the root directory.
2042 if (!end_of_new) {
2043 *old_dir = xstrndup(old_path, end_of_old - old_path);
2044 *new_dir = xstrdup("");
2045 return;
2048 /* Find the first non-matching character traversing backwards */
2049 while (*--end_of_new == *--end_of_old &&
2050 end_of_old != old_path &&
2051 end_of_new != new_path)
2052 ; /* Do nothing; all in the while loop */
2055 * If both got back to the beginning of their strings, then the
2056 * directory didn't change at all, only the basename did.
2058 if (end_of_old == old_path && end_of_new == new_path &&
2059 *end_of_old == *end_of_new)
2060 return; /* Note: *old_dir and *new_dir are still NULL */
2063 * If end_of_new got back to the beginning of its string, and
2064 * end_of_old got back to the beginning of some subdirectory, then
2065 * we have a rename/merge of a subdirectory into the root, which
2066 * needs slightly special handling.
2068 * Note: There is no need to consider the opposite case, with a
2069 * rename/merge of the root directory into some subdirectory
2070 * because as noted above the root directory always exists so it
2071 * cannot be considered to be renamed.
2073 if (end_of_new == new_path &&
2074 end_of_old != old_path && end_of_old[-1] == '/') {
2075 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2076 *new_dir = xstrdup("");
2077 return;
2081 * We've found the first non-matching character in the directory
2082 * paths. That means the current characters we were looking at
2083 * were part of the first non-matching subdir name going back from
2084 * the end of the strings. Get the whole name by advancing both
2085 * end_of_old and end_of_new to the NEXT '/' character. That will
2086 * represent the entire directory rename.
2088 * The reason for the increment is cases like
2089 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2090 * After dropping the basename and going back to the first
2091 * non-matching character, we're now comparing:
2092 * a/b/s and a/b/
2093 * and we want to be comparing:
2094 * a/b/star/ and a/b/tar/
2095 * but without the pre-increment, the one on the right would stay
2096 * a/b/.
2098 end_of_old = strchr(++end_of_old, '/');
2099 end_of_new = strchr(++end_of_new, '/');
2101 /* Copy the old and new directories into *old_dir and *new_dir. */
2102 *old_dir = xstrndup(old_path, end_of_old - old_path);
2103 *new_dir = xstrndup(new_path, end_of_new - new_path);
2106 static void remove_hashmap_entries(struct hashmap *dir_renames,
2107 struct string_list *items_to_remove)
2109 int i;
2110 struct dir_rename_entry *entry;
2112 for (i = 0; i < items_to_remove->nr; i++) {
2113 entry = items_to_remove->items[i].util;
2114 hashmap_remove(dir_renames, &entry->ent, NULL);
2116 string_list_clear(items_to_remove, 0);
2120 * See if there is a directory rename for path, and if there are any file
2121 * level conflicts for the renamed location. If there is a rename and
2122 * there are no conflicts, return the new name. Otherwise, return NULL.
2124 static char *handle_path_level_conflicts(struct merge_options *opt,
2125 const char *path,
2126 struct dir_rename_entry *entry,
2127 struct hashmap *collisions,
2128 struct tree *tree)
2130 char *new_path = NULL;
2131 struct collision_entry *collision_ent;
2132 int clean = 1;
2133 struct strbuf collision_paths = STRBUF_INIT;
2136 * entry has the mapping of old directory name to new directory name
2137 * that we want to apply to path.
2139 new_path = apply_dir_rename(entry, path);
2141 if (!new_path) {
2142 /* This should only happen when entry->non_unique_new_dir set */
2143 if (!entry->non_unique_new_dir)
2144 BUG("entry->non_unique_new_dir not set and !new_path");
2145 output(opt, 1, _("CONFLICT (directory rename split): "
2146 "Unclear where to place %s because directory "
2147 "%s was renamed to multiple other directories, "
2148 "with no destination getting a majority of the "
2149 "files."),
2150 path, entry->dir);
2151 clean = 0;
2152 return NULL;
2156 * The caller needs to have ensured that it has pre-populated
2157 * collisions with all paths that map to new_path. Do a quick check
2158 * to ensure that's the case.
2160 collision_ent = collision_find_entry(collisions, new_path);
2161 if (!collision_ent)
2162 BUG("collision_ent is NULL");
2165 * Check for one-sided add/add/.../add conflicts, i.e.
2166 * where implicit renames from the other side doing
2167 * directory rename(s) can affect this side of history
2168 * to put multiple paths into the same location. Warn
2169 * and bail on directory renames for such paths.
2171 if (collision_ent->reported_already) {
2172 clean = 0;
2173 } else if (tree_has_path(opt->repo, tree, new_path)) {
2174 collision_ent->reported_already = 1;
2175 strbuf_add_separated_string_list(&collision_paths, ", ",
2176 &collision_ent->source_files);
2177 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2178 "file/dir at %s in the way of implicit "
2179 "directory rename(s) putting the following "
2180 "path(s) there: %s."),
2181 new_path, collision_paths.buf);
2182 clean = 0;
2183 } else if (collision_ent->source_files.nr > 1) {
2184 collision_ent->reported_already = 1;
2185 strbuf_add_separated_string_list(&collision_paths, ", ",
2186 &collision_ent->source_files);
2187 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2188 "more than one path to %s; implicit directory "
2189 "renames tried to put these paths there: %s"),
2190 new_path, collision_paths.buf);
2191 clean = 0;
2194 /* Free memory we no longer need */
2195 strbuf_release(&collision_paths);
2196 if (!clean && new_path) {
2197 free(new_path);
2198 return NULL;
2201 return new_path;
2205 * There are a couple things we want to do at the directory level:
2206 * 1. Check for both sides renaming to the same thing, in order to avoid
2207 * implicit renaming of files that should be left in place. (See
2208 * testcase 6b in t6043 for details.)
2209 * 2. Prune directory renames if there are still files left in the
2210 * original directory. These represent a partial directory rename,
2211 * i.e. a rename where only some of the files within the directory
2212 * were renamed elsewhere. (Technically, this could be done earlier
2213 * in get_directory_renames(), except that would prevent us from
2214 * doing the previous check and thus failing testcase 6b.)
2215 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2216 * In the future, we could potentially record this info as well and
2217 * omit reporting rename/rename(1to2) conflicts for each path within
2218 * the affected directories, thus cleaning up the merge output.
2219 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2220 * directory level, because merging directories is fine. If it
2221 * causes conflicts for files within those merged directories, then
2222 * that should be detected at the individual path level.
2224 static void handle_directory_level_conflicts(struct merge_options *opt,
2225 struct hashmap *dir_re_head,
2226 struct tree *head,
2227 struct hashmap *dir_re_merge,
2228 struct tree *merge)
2230 struct hashmap_iter iter;
2231 struct dir_rename_entry *head_ent;
2232 struct dir_rename_entry *merge_ent;
2234 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2235 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2237 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2238 ent /* member name */) {
2239 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2240 if (merge_ent &&
2241 !head_ent->non_unique_new_dir &&
2242 !merge_ent->non_unique_new_dir &&
2243 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2244 /* 1. Renamed identically; remove it from both sides */
2245 string_list_append(&remove_from_head,
2246 head_ent->dir)->util = head_ent;
2247 strbuf_release(&head_ent->new_dir);
2248 string_list_append(&remove_from_merge,
2249 merge_ent->dir)->util = merge_ent;
2250 strbuf_release(&merge_ent->new_dir);
2251 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2252 /* 2. This wasn't a directory rename after all */
2253 string_list_append(&remove_from_head,
2254 head_ent->dir)->util = head_ent;
2255 strbuf_release(&head_ent->new_dir);
2259 remove_hashmap_entries(dir_re_head, &remove_from_head);
2260 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2262 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2263 ent /* member name */) {
2264 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2265 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2266 /* 2. This wasn't a directory rename after all */
2267 string_list_append(&remove_from_merge,
2268 merge_ent->dir)->util = merge_ent;
2269 } else if (head_ent &&
2270 !head_ent->non_unique_new_dir &&
2271 !merge_ent->non_unique_new_dir) {
2272 /* 3. rename/rename(1to2) */
2274 * We can assume it's not rename/rename(1to1) because
2275 * that was case (1), already checked above. So we
2276 * know that head_ent->new_dir and merge_ent->new_dir
2277 * are different strings.
2279 output(opt, 1, _("CONFLICT (rename/rename): "
2280 "Rename directory %s->%s in %s. "
2281 "Rename directory %s->%s in %s"),
2282 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2283 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2284 string_list_append(&remove_from_head,
2285 head_ent->dir)->util = head_ent;
2286 strbuf_release(&head_ent->new_dir);
2287 string_list_append(&remove_from_merge,
2288 merge_ent->dir)->util = merge_ent;
2289 strbuf_release(&merge_ent->new_dir);
2293 remove_hashmap_entries(dir_re_head, &remove_from_head);
2294 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2297 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2299 struct hashmap *dir_renames;
2300 struct hashmap_iter iter;
2301 struct dir_rename_entry *entry;
2302 int i;
2305 * Typically, we think of a directory rename as all files from a
2306 * certain directory being moved to a target directory. However,
2307 * what if someone first moved two files from the original
2308 * directory in one commit, and then renamed the directory
2309 * somewhere else in a later commit? At merge time, we just know
2310 * that files from the original directory went to two different
2311 * places, and that the bulk of them ended up in the same place.
2312 * We want each directory rename to represent where the bulk of the
2313 * files from that directory end up; this function exists to find
2314 * where the bulk of the files went.
2316 * The first loop below simply iterates through the list of file
2317 * renames, finding out how often each directory rename pair
2318 * possibility occurs.
2320 dir_renames = xmalloc(sizeof(*dir_renames));
2321 dir_rename_init(dir_renames);
2322 for (i = 0; i < pairs->nr; ++i) {
2323 struct string_list_item *item;
2324 int *count;
2325 struct diff_filepair *pair = pairs->queue[i];
2326 char *old_dir, *new_dir;
2328 /* File not part of directory rename if it wasn't renamed */
2329 if (pair->status != 'R')
2330 continue;
2332 get_renamed_dir_portion(pair->one->path, pair->two->path,
2333 &old_dir, &new_dir);
2334 if (!old_dir)
2335 /* Directory didn't change at all; ignore this one. */
2336 continue;
2338 entry = dir_rename_find_entry(dir_renames, old_dir);
2339 if (!entry) {
2340 entry = xmalloc(sizeof(*entry));
2341 dir_rename_entry_init(entry, old_dir);
2342 hashmap_put(dir_renames, &entry->ent);
2343 } else {
2344 free(old_dir);
2346 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2347 if (!item) {
2348 item = string_list_insert(&entry->possible_new_dirs,
2349 new_dir);
2350 item->util = xcalloc(1, sizeof(int));
2351 } else {
2352 free(new_dir);
2354 count = item->util;
2355 *count += 1;
2359 * For each directory with files moved out of it, we find out which
2360 * target directory received the most files so we can declare it to
2361 * be the "winning" target location for the directory rename. This
2362 * winner gets recorded in new_dir. If there is no winner
2363 * (multiple target directories received the same number of files),
2364 * we set non_unique_new_dir. Once we've determined the winner (or
2365 * that there is no winner), we no longer need possible_new_dirs.
2367 hashmap_for_each_entry(dir_renames, &iter, entry,
2368 ent /* member name */) {
2369 int max = 0;
2370 int bad_max = 0;
2371 char *best = NULL;
2373 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2374 int *count = entry->possible_new_dirs.items[i].util;
2376 if (*count == max)
2377 bad_max = max;
2378 else if (*count > max) {
2379 max = *count;
2380 best = entry->possible_new_dirs.items[i].string;
2383 if (bad_max == max)
2384 entry->non_unique_new_dir = 1;
2385 else {
2386 assert(entry->new_dir.len == 0);
2387 strbuf_addstr(&entry->new_dir, best);
2390 * The relevant directory sub-portion of the original full
2391 * filepaths were xstrndup'ed before inserting into
2392 * possible_new_dirs, and instead of manually iterating the
2393 * list and free'ing each, just lie and tell
2394 * possible_new_dirs that it did the strdup'ing so that it
2395 * will free them for us.
2397 entry->possible_new_dirs.strdup_strings = 1;
2398 string_list_clear(&entry->possible_new_dirs, 1);
2401 return dir_renames;
2404 static struct dir_rename_entry *check_dir_renamed(const char *path,
2405 struct hashmap *dir_renames)
2407 char *temp = xstrdup(path);
2408 char *end;
2409 struct dir_rename_entry *entry = NULL;
2411 while ((end = strrchr(temp, '/'))) {
2412 *end = '\0';
2413 entry = dir_rename_find_entry(dir_renames, temp);
2414 if (entry)
2415 break;
2417 free(temp);
2418 return entry;
2421 static void compute_collisions(struct hashmap *collisions,
2422 struct hashmap *dir_renames,
2423 struct diff_queue_struct *pairs)
2425 int i;
2428 * Multiple files can be mapped to the same path due to directory
2429 * renames done by the other side of history. Since that other
2430 * side of history could have merged multiple directories into one,
2431 * if our side of history added the same file basename to each of
2432 * those directories, then all N of them would get implicitly
2433 * renamed by the directory rename detection into the same path,
2434 * and we'd get an add/add/.../add conflict, and all those adds
2435 * from *this* side of history. This is not representable in the
2436 * index, and users aren't going to easily be able to make sense of
2437 * it. So we need to provide a good warning about what's
2438 * happening, and fall back to no-directory-rename detection
2439 * behavior for those paths.
2441 * See testcases 9e and all of section 5 from t6043 for examples.
2443 collision_init(collisions);
2445 for (i = 0; i < pairs->nr; ++i) {
2446 struct dir_rename_entry *dir_rename_ent;
2447 struct collision_entry *collision_ent;
2448 char *new_path;
2449 struct diff_filepair *pair = pairs->queue[i];
2451 if (pair->status != 'A' && pair->status != 'R')
2452 continue;
2453 dir_rename_ent = check_dir_renamed(pair->two->path,
2454 dir_renames);
2455 if (!dir_rename_ent)
2456 continue;
2458 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2459 if (!new_path)
2461 * dir_rename_ent->non_unique_new_path is true, which
2462 * means there is no directory rename for us to use,
2463 * which means it won't cause us any additional
2464 * collisions.
2466 continue;
2467 collision_ent = collision_find_entry(collisions, new_path);
2468 if (!collision_ent) {
2469 CALLOC_ARRAY(collision_ent, 1);
2470 hashmap_entry_init(&collision_ent->ent,
2471 strhash(new_path));
2472 hashmap_put(collisions, &collision_ent->ent);
2473 collision_ent->target_file = new_path;
2474 } else {
2475 free(new_path);
2477 string_list_insert(&collision_ent->source_files,
2478 pair->two->path);
2482 static char *check_for_directory_rename(struct merge_options *opt,
2483 const char *path,
2484 struct tree *tree,
2485 struct hashmap *dir_renames,
2486 struct hashmap *dir_rename_exclusions,
2487 struct hashmap *collisions,
2488 int *clean_merge)
2490 char *new_path = NULL;
2491 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2492 struct dir_rename_entry *oentry = NULL;
2494 if (!entry)
2495 return new_path;
2498 * This next part is a little weird. We do not want to do an
2499 * implicit rename into a directory we renamed on our side, because
2500 * that will result in a spurious rename/rename(1to2) conflict. An
2501 * example:
2502 * Base commit: dumbdir/afile, otherdir/bfile
2503 * Side 1: smrtdir/afile, otherdir/bfile
2504 * Side 2: dumbdir/afile, dumbdir/bfile
2505 * Here, while working on Side 1, we could notice that otherdir was
2506 * renamed/merged to dumbdir, and change the diff_filepair for
2507 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2508 * 2 will notice the rename from dumbdir to smrtdir, and do the
2509 * transitive rename to move it from dumbdir/bfile to
2510 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2511 * smrtdir, a rename/rename(1to2) conflict. We really just want
2512 * the file to end up in smrtdir. And the way to achieve that is
2513 * to not let Side1 do the rename to dumbdir, since we know that is
2514 * the source of one of our directory renames.
2516 * That's why oentry and dir_rename_exclusions is here.
2518 * As it turns out, this also prevents N-way transient rename
2519 * confusion; See testcases 9c and 9d of t6043.
2521 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2522 if (oentry) {
2523 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2524 "to %s, because %s itself was renamed."),
2525 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2526 } else {
2527 new_path = handle_path_level_conflicts(opt, path, entry,
2528 collisions, tree);
2529 *clean_merge &= (new_path != NULL);
2532 return new_path;
2535 static void apply_directory_rename_modifications(struct merge_options *opt,
2536 struct diff_filepair *pair,
2537 char *new_path,
2538 struct rename *re,
2539 struct tree *tree,
2540 struct tree *o_tree,
2541 struct tree *a_tree,
2542 struct tree *b_tree,
2543 struct string_list *entries)
2545 struct string_list_item *item;
2546 int stage = (tree == a_tree ? 2 : 3);
2547 int update_wd;
2550 * In all cases where we can do directory rename detection,
2551 * unpack_trees() will have read pair->two->path into the
2552 * index and the working copy. We need to remove it so that
2553 * we can instead place it at new_path. It is guaranteed to
2554 * not be untracked (unpack_trees() would have errored out
2555 * saying the file would have been overwritten), but it might
2556 * be dirty, though.
2558 update_wd = !was_dirty(opt, pair->two->path);
2559 if (!update_wd)
2560 output(opt, 1, _("Refusing to lose dirty file at %s"),
2561 pair->two->path);
2562 remove_file(opt, 1, pair->two->path, !update_wd);
2564 /* Find or create a new re->dst_entry */
2565 item = string_list_lookup(entries, new_path);
2566 if (item) {
2568 * Since we're renaming on this side of history, and it's
2569 * due to a directory rename on the other side of history
2570 * (which we only allow when the directory in question no
2571 * longer exists on the other side of history), the
2572 * original entry for re->dst_entry is no longer
2573 * necessary...
2575 re->dst_entry->processed = 1;
2578 * ...because we'll be using this new one.
2580 re->dst_entry = item->util;
2581 } else {
2583 * re->dst_entry is for the before-dir-rename path, and we
2584 * need it to hold information for the after-dir-rename
2585 * path. Before creating a new entry, we need to mark the
2586 * old one as unnecessary (...unless it is shared by
2587 * src_entry, i.e. this didn't use to be a rename, in which
2588 * case we can just allow the normal processing to happen
2589 * for it).
2591 if (pair->status == 'R')
2592 re->dst_entry->processed = 1;
2594 re->dst_entry = insert_stage_data(opt->repo, new_path,
2595 o_tree, a_tree, b_tree,
2596 entries);
2597 item = string_list_insert(entries, new_path);
2598 item->util = re->dst_entry;
2602 * Update the stage_data with the information about the path we are
2603 * moving into place. That slot will be empty and available for us
2604 * to write to because of the collision checks in
2605 * handle_path_level_conflicts(). In other words,
2606 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2607 * open for us to write to.
2609 * It may be tempting to actually update the index at this point as
2610 * well, using update_stages_for_stage_data(), but as per the big
2611 * "NOTE" in update_stages(), doing so will modify the current
2612 * in-memory index which will break calls to would_lose_untracked()
2613 * that we need to make. Instead, we need to just make sure that
2614 * the various handle_rename_*() functions update the index
2615 * explicitly rather than relying on unpack_trees() to have done it.
2617 get_tree_entry(opt->repo,
2618 &tree->object.oid,
2619 pair->two->path,
2620 &re->dst_entry->stages[stage].oid,
2621 &re->dst_entry->stages[stage].mode);
2624 * Record the original change status (or 'type' of change). If it
2625 * was originally an add ('A'), this lets us differentiate later
2626 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2627 * otherwise look the same). If it was originally a rename ('R'),
2628 * this lets us remember and report accurately about the transitive
2629 * renaming that occurred via the directory rename detection. Also,
2630 * record the original destination name.
2632 re->dir_rename_original_type = pair->status;
2633 re->dir_rename_original_dest = pair->two->path;
2636 * We don't actually look at pair->status again, but it seems
2637 * pedagogically correct to adjust it.
2639 pair->status = 'R';
2642 * Finally, record the new location.
2644 pair->two->path = new_path;
2648 * Get information of all renames which occurred in 'pairs', making use of
2649 * any implicit directory renames inferred from the other side of history.
2650 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2651 * to be able to associate the correct cache entries with the rename
2652 * information; tree is always equal to either a_tree or b_tree.
2654 static struct string_list *get_renames(struct merge_options *opt,
2655 const char *branch,
2656 struct diff_queue_struct *pairs,
2657 struct hashmap *dir_renames,
2658 struct hashmap *dir_rename_exclusions,
2659 struct tree *tree,
2660 struct tree *o_tree,
2661 struct tree *a_tree,
2662 struct tree *b_tree,
2663 struct string_list *entries,
2664 int *clean_merge)
2666 int i;
2667 struct hashmap collisions;
2668 struct hashmap_iter iter;
2669 struct collision_entry *e;
2670 struct string_list *renames;
2672 compute_collisions(&collisions, dir_renames, pairs);
2673 CALLOC_ARRAY(renames, 1);
2675 for (i = 0; i < pairs->nr; ++i) {
2676 struct string_list_item *item;
2677 struct rename *re;
2678 struct diff_filepair *pair = pairs->queue[i];
2679 char *new_path; /* non-NULL only with directory renames */
2681 if (pair->status != 'A' && pair->status != 'R') {
2682 diff_free_filepair(pair);
2683 continue;
2685 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2686 dir_renames,
2687 dir_rename_exclusions,
2688 &collisions,
2689 clean_merge);
2690 if (pair->status != 'R' && !new_path) {
2691 diff_free_filepair(pair);
2692 continue;
2695 re = xmalloc(sizeof(*re));
2696 re->processed = 0;
2697 re->pair = pair;
2698 re->branch = branch;
2699 re->dir_rename_original_type = '\0';
2700 re->dir_rename_original_dest = NULL;
2701 item = string_list_lookup(entries, re->pair->one->path);
2702 if (!item)
2703 re->src_entry = insert_stage_data(opt->repo,
2704 re->pair->one->path,
2705 o_tree, a_tree, b_tree, entries);
2706 else
2707 re->src_entry = item->util;
2709 item = string_list_lookup(entries, re->pair->two->path);
2710 if (!item)
2711 re->dst_entry = insert_stage_data(opt->repo,
2712 re->pair->two->path,
2713 o_tree, a_tree, b_tree, entries);
2714 else
2715 re->dst_entry = item->util;
2716 item = string_list_insert(renames, pair->one->path);
2717 item->util = re;
2718 if (new_path)
2719 apply_directory_rename_modifications(opt, pair, new_path,
2720 re, tree, o_tree,
2721 a_tree, b_tree,
2722 entries);
2725 hashmap_for_each_entry(&collisions, &iter, e,
2726 ent /* member name */) {
2727 free(e->target_file);
2728 string_list_clear(&e->source_files, 0);
2730 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2731 return renames;
2734 static int process_renames(struct merge_options *opt,
2735 struct string_list *a_renames,
2736 struct string_list *b_renames)
2738 int clean_merge = 1, i, j;
2739 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2740 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2741 const struct rename *sre;
2744 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2745 * string_list one-by-one, but O(n log n) to build it unsorted and
2746 * then sort it. Note that as we build the list, we do not need to
2747 * check if the existing destination path is already in the list,
2748 * because the structure of diffcore_rename guarantees we won't
2749 * have duplicates.
2751 for (i = 0; i < a_renames->nr; i++) {
2752 sre = a_renames->items[i].util;
2753 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2754 = (void *)sre;
2756 for (i = 0; i < b_renames->nr; i++) {
2757 sre = b_renames->items[i].util;
2758 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2759 = (void *)sre;
2762 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2763 struct string_list *renames1, *renames2Dst;
2764 struct rename *ren1 = NULL, *ren2 = NULL;
2765 const char *ren1_src, *ren1_dst;
2766 struct string_list_item *lookup;
2768 if (i >= a_renames->nr) {
2769 ren2 = b_renames->items[j++].util;
2770 } else if (j >= b_renames->nr) {
2771 ren1 = a_renames->items[i++].util;
2772 } else {
2773 int compare = strcmp(a_renames->items[i].string,
2774 b_renames->items[j].string);
2775 if (compare <= 0)
2776 ren1 = a_renames->items[i++].util;
2777 if (compare >= 0)
2778 ren2 = b_renames->items[j++].util;
2781 /* TODO: refactor, so that 1/2 are not needed */
2782 if (ren1) {
2783 renames1 = a_renames;
2784 renames2Dst = &b_by_dst;
2785 } else {
2786 renames1 = b_renames;
2787 renames2Dst = &a_by_dst;
2788 SWAP(ren2, ren1);
2791 if (ren1->processed)
2792 continue;
2793 ren1->processed = 1;
2794 ren1->dst_entry->processed = 1;
2795 /* BUG: We should only mark src_entry as processed if we
2796 * are not dealing with a rename + add-source case.
2798 ren1->src_entry->processed = 1;
2800 ren1_src = ren1->pair->one->path;
2801 ren1_dst = ren1->pair->two->path;
2803 if (ren2) {
2804 /* One file renamed on both sides */
2805 const char *ren2_src = ren2->pair->one->path;
2806 const char *ren2_dst = ren2->pair->two->path;
2807 enum rename_type rename_type;
2808 if (strcmp(ren1_src, ren2_src) != 0)
2809 BUG("ren1_src != ren2_src");
2810 ren2->dst_entry->processed = 1;
2811 ren2->processed = 1;
2812 if (strcmp(ren1_dst, ren2_dst) != 0) {
2813 rename_type = RENAME_ONE_FILE_TO_TWO;
2814 clean_merge = 0;
2815 } else {
2816 rename_type = RENAME_ONE_FILE_TO_ONE;
2817 /* BUG: We should only remove ren1_src in
2818 * the base stage (think of rename +
2819 * add-source cases).
2821 remove_file(opt, 1, ren1_src, 1);
2822 update_entry(ren1->dst_entry,
2823 ren1->pair->one,
2824 ren1->pair->two,
2825 ren2->pair->two);
2827 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2828 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2829 /* Two different files renamed to the same thing */
2830 char *ren2_dst;
2831 ren2 = lookup->util;
2832 ren2_dst = ren2->pair->two->path;
2833 if (strcmp(ren1_dst, ren2_dst) != 0)
2834 BUG("ren1_dst != ren2_dst");
2836 clean_merge = 0;
2837 ren2->processed = 1;
2839 * BUG: We should only mark src_entry as processed
2840 * if we are not dealing with a rename + add-source
2841 * case.
2843 ren2->src_entry->processed = 1;
2845 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2846 opt, ren1, ren2);
2847 } else {
2848 /* Renamed in 1, maybe changed in 2 */
2849 /* we only use sha1 and mode of these */
2850 struct diff_filespec src_other, dst_other;
2851 int try_merge;
2854 * unpack_trees loads entries from common-commit
2855 * into stage 1, from head-commit into stage 2, and
2856 * from merge-commit into stage 3. We keep track
2857 * of which side corresponds to the rename.
2859 int renamed_stage = a_renames == renames1 ? 2 : 3;
2860 int other_stage = a_renames == renames1 ? 3 : 2;
2863 * Directory renames have a funny corner case...
2865 int renamed_to_self = !strcmp(ren1_src, ren1_dst);
2867 /* BUG: We should only remove ren1_src in the base
2868 * stage and in other_stage (think of rename +
2869 * add-source case).
2871 if (!renamed_to_self)
2872 remove_file(opt, 1, ren1_src,
2873 renamed_stage == 2 ||
2874 !was_tracked(opt, ren1_src));
2876 oidcpy(&src_other.oid,
2877 &ren1->src_entry->stages[other_stage].oid);
2878 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2879 oidcpy(&dst_other.oid,
2880 &ren1->dst_entry->stages[other_stage].oid);
2881 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2882 try_merge = 0;
2884 if (oideq(&src_other.oid, null_oid()) &&
2885 ren1->dir_rename_original_type == 'A') {
2886 setup_rename_conflict_info(RENAME_VIA_DIR,
2887 opt, ren1, NULL);
2888 } else if (renamed_to_self) {
2889 setup_rename_conflict_info(RENAME_NORMAL,
2890 opt, ren1, NULL);
2891 } else if (oideq(&src_other.oid, null_oid())) {
2892 setup_rename_conflict_info(RENAME_DELETE,
2893 opt, ren1, NULL);
2894 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2895 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2897 * Added file on the other side identical to
2898 * the file being renamed: clean merge.
2899 * Also, there is no need to overwrite the
2900 * file already in the working copy, so call
2901 * update_file_flags() instead of
2902 * update_file().
2904 if (update_file_flags(opt,
2905 ren1->pair->two,
2906 ren1_dst,
2907 1, /* update_cache */
2908 0 /* update_wd */))
2909 clean_merge = -1;
2910 } else if (!oideq(&dst_other.oid, null_oid())) {
2912 * Probably not a clean merge, but it's
2913 * premature to set clean_merge to 0 here,
2914 * because if the rename merges cleanly and
2915 * the merge exactly matches the newly added
2916 * file, then the merge will be clean.
2918 setup_rename_conflict_info(RENAME_ADD,
2919 opt, ren1, NULL);
2920 } else
2921 try_merge = 1;
2923 if (clean_merge < 0)
2924 goto cleanup_and_return;
2925 if (try_merge) {
2926 struct diff_filespec *o, *a, *b;
2927 src_other.path = (char *)ren1_src;
2929 o = ren1->pair->one;
2930 if (a_renames == renames1) {
2931 a = ren1->pair->two;
2932 b = &src_other;
2933 } else {
2934 b = ren1->pair->two;
2935 a = &src_other;
2937 update_entry(ren1->dst_entry, o, a, b);
2938 setup_rename_conflict_info(RENAME_NORMAL,
2939 opt, ren1, NULL);
2943 cleanup_and_return:
2944 string_list_clear(&a_by_dst, 0);
2945 string_list_clear(&b_by_dst, 0);
2947 return clean_merge;
2950 struct rename_info {
2951 struct string_list *head_renames;
2952 struct string_list *merge_renames;
2955 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2956 struct hashmap *dir_renames)
2958 struct hashmap_iter iter;
2959 struct dir_rename_entry *e;
2961 hashmap_for_each_entry(dir_renames, &iter, e,
2962 ent /* member name */) {
2963 free(e->dir);
2964 strbuf_release(&e->new_dir);
2965 /* possible_new_dirs already cleared in get_directory_renames */
2967 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2968 free(dir_renames);
2970 free(pairs->queue);
2971 free(pairs);
2974 static int detect_and_process_renames(struct merge_options *opt,
2975 struct tree *common,
2976 struct tree *head,
2977 struct tree *merge,
2978 struct string_list *entries,
2979 struct rename_info *ri)
2981 struct diff_queue_struct *head_pairs, *merge_pairs;
2982 struct hashmap *dir_re_head, *dir_re_merge;
2983 int clean = 1;
2985 ri->head_renames = NULL;
2986 ri->merge_renames = NULL;
2988 if (!merge_detect_rename(opt))
2989 return 1;
2991 head_pairs = get_diffpairs(opt, common, head);
2992 merge_pairs = get_diffpairs(opt, common, merge);
2994 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2995 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2996 !opt->priv->call_depth)) {
2997 dir_re_head = get_directory_renames(head_pairs);
2998 dir_re_merge = get_directory_renames(merge_pairs);
3000 handle_directory_level_conflicts(opt,
3001 dir_re_head, head,
3002 dir_re_merge, merge);
3003 } else {
3004 dir_re_head = xmalloc(sizeof(*dir_re_head));
3005 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
3006 dir_rename_init(dir_re_head);
3007 dir_rename_init(dir_re_merge);
3010 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
3011 dir_re_merge, dir_re_head, head,
3012 common, head, merge, entries,
3013 &clean);
3014 if (clean < 0)
3015 goto cleanup;
3016 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
3017 dir_re_head, dir_re_merge, merge,
3018 common, head, merge, entries,
3019 &clean);
3020 if (clean < 0)
3021 goto cleanup;
3022 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
3024 cleanup:
3026 * Some cleanup is deferred until cleanup_renames() because the
3027 * data structures are still needed and referenced in
3028 * process_entry(). But there are a few things we can free now.
3030 initial_cleanup_rename(head_pairs, dir_re_head);
3031 initial_cleanup_rename(merge_pairs, dir_re_merge);
3033 return clean;
3036 static void final_cleanup_rename(struct string_list *rename)
3038 const struct rename *re;
3039 int i;
3041 if (!rename)
3042 return;
3044 for (i = 0; i < rename->nr; i++) {
3045 re = rename->items[i].util;
3046 diff_free_filepair(re->pair);
3048 string_list_clear(rename, 1);
3049 free(rename);
3052 static void final_cleanup_renames(struct rename_info *re_info)
3054 final_cleanup_rename(re_info->head_renames);
3055 final_cleanup_rename(re_info->merge_renames);
3058 static int read_oid_strbuf(struct merge_options *opt,
3059 const struct object_id *oid,
3060 struct strbuf *dst)
3062 void *buf;
3063 enum object_type type;
3064 unsigned long size;
3065 buf = repo_read_object_file(the_repository, oid, &type, &size);
3066 if (!buf)
3067 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3068 if (type != OBJ_BLOB) {
3069 free(buf);
3070 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3072 strbuf_attach(dst, buf, size, size + 1);
3073 return 0;
3076 static int blob_unchanged(struct merge_options *opt,
3077 const struct diff_filespec *o,
3078 const struct diff_filespec *a,
3079 int renormalize, const char *path)
3081 struct strbuf obuf = STRBUF_INIT;
3082 struct strbuf abuf = STRBUF_INIT;
3083 int ret = 0; /* assume changed for safety */
3084 struct index_state *idx = opt->repo->index;
3086 if (a->mode != o->mode)
3087 return 0;
3088 if (oideq(&o->oid, &a->oid))
3089 return 1;
3090 if (!renormalize)
3091 return 0;
3093 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3094 read_oid_strbuf(opt, &a->oid, &abuf))
3095 goto error_return;
3097 * Note: binary | is used so that both renormalizations are
3098 * performed. Comparison can be skipped if both files are
3099 * unchanged since their sha1s have already been compared.
3101 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3102 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3103 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3105 error_return:
3106 strbuf_release(&obuf);
3107 strbuf_release(&abuf);
3108 return ret;
3111 static int handle_modify_delete(struct merge_options *opt,
3112 const char *path,
3113 const struct diff_filespec *o,
3114 const struct diff_filespec *a,
3115 const struct diff_filespec *b)
3117 const char *modify_branch, *delete_branch;
3118 const struct diff_filespec *changed;
3120 if (is_valid(a)) {
3121 modify_branch = opt->branch1;
3122 delete_branch = opt->branch2;
3123 changed = a;
3124 } else {
3125 modify_branch = opt->branch2;
3126 delete_branch = opt->branch1;
3127 changed = b;
3130 return handle_change_delete(opt,
3131 path, NULL,
3132 o, changed,
3133 modify_branch, delete_branch,
3134 _("modify"), _("modified"));
3137 static int handle_content_merge(struct merge_file_info *mfi,
3138 struct merge_options *opt,
3139 const char *path,
3140 int is_dirty,
3141 const struct diff_filespec *o,
3142 const struct diff_filespec *a,
3143 const struct diff_filespec *b,
3144 struct rename_conflict_info *ci)
3146 const char *reason = _("content");
3147 unsigned df_conflict_remains = 0;
3149 if (!is_valid(o))
3150 reason = _("add/add");
3152 assert(o->path && a->path && b->path);
3153 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3154 S_ISGITLINK(ci->ren1->pair->two->mode)))
3155 df_conflict_remains = 1;
3157 if (merge_mode_and_contents(opt, o, a, b, path,
3158 opt->branch1, opt->branch2,
3159 opt->priv->call_depth * 2, mfi))
3160 return -1;
3163 * We can skip updating the working tree file iff:
3164 * a) The merge is clean
3165 * b) The merge matches what was in HEAD (content, mode, pathname)
3166 * c) The target path is usable (i.e. not involved in D/F conflict)
3168 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3169 !df_conflict_remains) {
3170 int pos;
3171 struct cache_entry *ce;
3173 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3174 if (add_cacheinfo(opt, &mfi->blob, path,
3175 0, (!opt->priv->call_depth && !is_dirty), 0))
3176 return -1;
3178 * However, add_cacheinfo() will delete the old cache entry
3179 * and add a new one. We need to copy over any skip_worktree
3180 * flag to avoid making the file appear as if it were
3181 * deleted by the user.
3183 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3184 ce = opt->priv->orig_index.cache[pos];
3185 if (ce_skip_worktree(ce)) {
3186 pos = index_name_pos(opt->repo->index, path, strlen(path));
3187 ce = opt->repo->index->cache[pos];
3188 ce->ce_flags |= CE_SKIP_WORKTREE;
3190 return mfi->clean;
3193 if (!mfi->clean) {
3194 if (S_ISGITLINK(mfi->blob.mode))
3195 reason = _("submodule");
3196 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3197 reason, path);
3198 if (ci && !df_conflict_remains)
3199 if (update_stages(opt, path, o, a, b))
3200 return -1;
3203 if (df_conflict_remains || is_dirty) {
3204 char *new_path;
3205 if (opt->priv->call_depth) {
3206 remove_file_from_index(opt->repo->index, path);
3207 } else {
3208 if (!mfi->clean) {
3209 if (update_stages(opt, path, o, a, b))
3210 return -1;
3211 } else {
3212 int file_from_stage2 = was_tracked(opt, path);
3214 if (update_stages(opt, path, NULL,
3215 file_from_stage2 ? &mfi->blob : NULL,
3216 file_from_stage2 ? NULL : &mfi->blob))
3217 return -1;
3221 new_path = unique_path(opt, path, ci->ren1->branch);
3222 if (is_dirty) {
3223 output(opt, 1, _("Refusing to lose dirty file at %s"),
3224 path);
3226 output(opt, 1, _("Adding as %s instead"), new_path);
3227 if (update_file(opt, 0, &mfi->blob, new_path)) {
3228 free(new_path);
3229 return -1;
3231 free(new_path);
3232 mfi->clean = 0;
3233 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3234 return -1;
3235 return !is_dirty && mfi->clean;
3238 static int handle_rename_normal(struct merge_options *opt,
3239 const char *path,
3240 const struct diff_filespec *o,
3241 const struct diff_filespec *a,
3242 const struct diff_filespec *b,
3243 struct rename_conflict_info *ci)
3245 struct rename *ren = ci->ren1;
3246 struct merge_file_info mfi;
3247 int clean;
3249 /* Merge the content and write it out */
3250 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3251 o, a, b, ci);
3253 if (clean &&
3254 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3255 ren->dir_rename_original_dest) {
3256 if (update_stages(opt, path,
3257 &mfi.blob, &mfi.blob, &mfi.blob))
3258 return -1;
3259 clean = 0; /* not clean, but conflicted */
3261 return clean;
3264 static void dir_rename_warning(const char *msg,
3265 int is_add,
3266 int clean,
3267 struct merge_options *opt,
3268 struct rename *ren)
3270 const char *other_branch;
3271 other_branch = (ren->branch == opt->branch1 ?
3272 opt->branch2 : opt->branch1);
3273 if (is_add) {
3274 output(opt, clean ? 2 : 1, msg,
3275 ren->pair->one->path, ren->branch,
3276 other_branch, ren->pair->two->path);
3277 return;
3279 output(opt, clean ? 2 : 1, msg,
3280 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3281 other_branch, ren->pair->two->path);
3283 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3284 struct rename *ren)
3286 const char *msg;
3287 int clean = 1, is_add;
3289 if (!ren)
3290 return clean;
3292 /* Return early if ren was not affected/created by a directory rename */
3293 if (!ren->dir_rename_original_dest)
3294 return clean;
3296 /* Sanity checks */
3297 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3298 assert(ren->dir_rename_original_type == 'A' ||
3299 ren->dir_rename_original_type == 'R');
3301 /* Check whether to treat directory renames as a conflict */
3302 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3304 is_add = (ren->dir_rename_original_type == 'A');
3305 if (ren->dir_rename_original_type == 'A' && clean) {
3306 msg = _("Path updated: %s added in %s inside a "
3307 "directory that was renamed in %s; moving it to %s.");
3308 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3309 msg = _("CONFLICT (file location): %s added in %s "
3310 "inside a directory that was renamed in %s, "
3311 "suggesting it should perhaps be moved to %s.");
3312 } else if (ren->dir_rename_original_type == 'R' && clean) {
3313 msg = _("Path updated: %s renamed to %s in %s, inside a "
3314 "directory that was renamed in %s; moving it to %s.");
3315 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3316 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3317 "inside a directory that was renamed in %s, "
3318 "suggesting it should perhaps be moved to %s.");
3319 } else {
3320 BUG("Impossible dir_rename_original_type/clean combination");
3322 dir_rename_warning(msg, is_add, clean, opt, ren);
3324 return clean;
3327 /* Per entry merge function */
3328 static int process_entry(struct merge_options *opt,
3329 const char *path, struct stage_data *entry)
3331 int clean_merge = 1;
3332 int normalize = opt->renormalize;
3334 struct diff_filespec *o = &entry->stages[1];
3335 struct diff_filespec *a = &entry->stages[2];
3336 struct diff_filespec *b = &entry->stages[3];
3337 int o_valid = is_valid(o);
3338 int a_valid = is_valid(a);
3339 int b_valid = is_valid(b);
3340 o->path = a->path = b->path = (char*)path;
3342 entry->processed = 1;
3343 if (entry->rename_conflict_info) {
3344 struct rename_conflict_info *ci = entry->rename_conflict_info;
3345 struct diff_filespec *temp;
3346 int path_clean;
3348 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3349 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3352 * For cases with a single rename, {o,a,b}->path have all been
3353 * set to the rename target path; we need to set two of these
3354 * back to the rename source.
3355 * For rename/rename conflicts, we'll manually fix paths below.
3357 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3358 o->path = temp->path = ci->ren1->pair->one->path;
3359 if (ci->ren2) {
3360 assert(opt->branch1 == ci->ren1->branch);
3363 switch (ci->rename_type) {
3364 case RENAME_NORMAL:
3365 case RENAME_ONE_FILE_TO_ONE:
3366 clean_merge = handle_rename_normal(opt, path, o, a, b,
3367 ci);
3368 break;
3369 case RENAME_VIA_DIR:
3370 clean_merge = handle_rename_via_dir(opt, ci);
3371 break;
3372 case RENAME_ADD:
3374 * Probably unclean merge, but if the renamed file
3375 * merges cleanly and the result can then be
3376 * two-way merged cleanly with the added file, I
3377 * guess it's a clean merge?
3379 clean_merge = handle_rename_add(opt, ci);
3380 break;
3381 case RENAME_DELETE:
3382 clean_merge = 0;
3383 if (handle_rename_delete(opt, ci))
3384 clean_merge = -1;
3385 break;
3386 case RENAME_ONE_FILE_TO_TWO:
3388 * Manually fix up paths; note:
3389 * ren[12]->pair->one->path are equal.
3391 o->path = ci->ren1->pair->one->path;
3392 a->path = ci->ren1->pair->two->path;
3393 b->path = ci->ren2->pair->two->path;
3395 clean_merge = 0;
3396 if (handle_rename_rename_1to2(opt, ci))
3397 clean_merge = -1;
3398 break;
3399 case RENAME_TWO_FILES_TO_ONE:
3401 * Manually fix up paths; note,
3402 * ren[12]->pair->two->path are actually equal.
3404 o->path = NULL;
3405 a->path = ci->ren1->pair->two->path;
3406 b->path = ci->ren2->pair->two->path;
3409 * Probably unclean merge, but if the two renamed
3410 * files merge cleanly and the two resulting files
3411 * can then be two-way merged cleanly, I guess it's
3412 * a clean merge?
3414 clean_merge = handle_rename_rename_2to1(opt, ci);
3415 break;
3416 default:
3417 entry->processed = 0;
3418 break;
3420 if (path_clean < clean_merge)
3421 clean_merge = path_clean;
3422 } else if (o_valid && (!a_valid || !b_valid)) {
3423 /* Case A: Deleted in one */
3424 if ((!a_valid && !b_valid) ||
3425 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3426 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3427 /* Deleted in both or deleted in one and
3428 * unchanged in the other */
3429 if (a_valid)
3430 output(opt, 2, _("Removing %s"), path);
3431 /* do not touch working file if it did not exist */
3432 remove_file(opt, 1, path, !a_valid);
3433 } else {
3434 /* Modify/delete; deleted side may have put a directory in the way */
3435 clean_merge = 0;
3436 if (handle_modify_delete(opt, path, o, a, b))
3437 clean_merge = -1;
3439 } else if ((!o_valid && a_valid && !b_valid) ||
3440 (!o_valid && !a_valid && b_valid)) {
3441 /* Case B: Added in one. */
3442 /* [nothing|directory] -> ([nothing|directory], file) */
3444 const char *add_branch;
3445 const char *other_branch;
3446 const char *conf;
3447 const struct diff_filespec *contents;
3449 if (a_valid) {
3450 add_branch = opt->branch1;
3451 other_branch = opt->branch2;
3452 contents = a;
3453 conf = _("file/directory");
3454 } else {
3455 add_branch = opt->branch2;
3456 other_branch = opt->branch1;
3457 contents = b;
3458 conf = _("directory/file");
3460 if (dir_in_way(opt->repo->index, path,
3461 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3462 0)) {
3463 char *new_path = unique_path(opt, path, add_branch);
3464 clean_merge = 0;
3465 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3466 "Adding %s as %s"),
3467 conf, path, other_branch, path, new_path);
3468 if (update_file(opt, 0, contents, new_path))
3469 clean_merge = -1;
3470 else if (opt->priv->call_depth)
3471 remove_file_from_index(opt->repo->index, path);
3472 free(new_path);
3473 } else {
3474 output(opt, 2, _("Adding %s"), path);
3475 /* do not overwrite file if already present */
3476 if (update_file_flags(opt, contents, path, 1, !a_valid))
3477 clean_merge = -1;
3479 } else if (a_valid && b_valid) {
3480 if (!o_valid) {
3481 /* Case C: Added in both (check for same permissions) */
3482 output(opt, 1,
3483 _("CONFLICT (add/add): Merge conflict in %s"),
3484 path);
3485 clean_merge = handle_file_collision(opt,
3486 path, NULL, NULL,
3487 opt->branch1,
3488 opt->branch2,
3489 a, b);
3490 } else {
3491 /* case D: Modified in both, but differently. */
3492 struct merge_file_info mfi;
3493 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3494 clean_merge = handle_content_merge(&mfi, opt, path,
3495 is_dirty,
3496 o, a, b, NULL);
3498 } else if (!o_valid && !a_valid && !b_valid) {
3500 * this entry was deleted altogether. a_mode == 0 means
3501 * we had that path and want to actively remove it.
3503 remove_file(opt, 1, path, !a->mode);
3504 } else
3505 BUG("fatal merge failure, shouldn't happen.");
3507 return clean_merge;
3510 static int merge_trees_internal(struct merge_options *opt,
3511 struct tree *head,
3512 struct tree *merge,
3513 struct tree *merge_base,
3514 struct tree **result)
3516 struct index_state *istate = opt->repo->index;
3517 int code, clean;
3519 if (opt->subtree_shift) {
3520 merge = shift_tree_object(opt->repo, head, merge,
3521 opt->subtree_shift);
3522 merge_base = shift_tree_object(opt->repo, head, merge_base,
3523 opt->subtree_shift);
3526 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3527 output(opt, 0, _("Already up to date."));
3528 *result = head;
3529 return 1;
3532 code = unpack_trees_start(opt, merge_base, head, merge);
3534 if (code != 0) {
3535 if (show(opt, 4) || opt->priv->call_depth)
3536 err(opt, _("merging of trees %s and %s failed"),
3537 oid_to_hex(&head->object.oid),
3538 oid_to_hex(&merge->object.oid));
3539 unpack_trees_finish(opt);
3540 return -1;
3543 if (unmerged_index(istate)) {
3544 struct string_list *entries;
3545 struct rename_info re_info;
3546 int i;
3548 * Only need the hashmap while processing entries, so
3549 * initialize it here and free it when we are done running
3550 * through the entries. Keeping it in the merge_options as
3551 * opposed to decaring a local hashmap is for convenience
3552 * so that we don't have to pass it to around.
3554 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3555 NULL, 512);
3556 get_files_dirs(opt, head);
3557 get_files_dirs(opt, merge);
3559 entries = get_unmerged(opt->repo->index);
3560 clean = detect_and_process_renames(opt, merge_base, head, merge,
3561 entries, &re_info);
3562 record_df_conflict_files(opt, entries);
3563 if (clean < 0)
3564 goto cleanup;
3565 for (i = entries->nr-1; 0 <= i; i--) {
3566 const char *path = entries->items[i].string;
3567 struct stage_data *e = entries->items[i].util;
3568 if (!e->processed) {
3569 int ret = process_entry(opt, path, e);
3570 if (!ret)
3571 clean = 0;
3572 else if (ret < 0) {
3573 clean = ret;
3574 goto cleanup;
3578 for (i = 0; i < entries->nr; i++) {
3579 struct stage_data *e = entries->items[i].util;
3580 if (!e->processed)
3581 BUG("unprocessed path??? %s",
3582 entries->items[i].string);
3585 cleanup:
3586 final_cleanup_renames(&re_info);
3588 string_list_clear(entries, 1);
3589 free(entries);
3591 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3592 struct path_hashmap_entry, e);
3594 if (clean < 0) {
3595 unpack_trees_finish(opt);
3596 return clean;
3599 else
3600 clean = 1;
3602 unpack_trees_finish(opt);
3604 if (opt->priv->call_depth &&
3605 !(*result = write_in_core_index_as_tree(opt->repo)))
3606 return -1;
3608 return clean;
3612 * Merge the commits h1 and h2, returning a flag (int) indicating the
3613 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3614 * virtual commit and write its location to *result.
3616 static int merge_recursive_internal(struct merge_options *opt,
3617 struct commit *h1,
3618 struct commit *h2,
3619 struct commit_list *merge_bases,
3620 struct commit **result)
3622 struct commit_list *iter;
3623 struct commit *merged_merge_bases;
3624 struct tree *result_tree;
3625 int clean;
3626 const char *ancestor_name;
3627 struct strbuf merge_base_abbrev = STRBUF_INIT;
3629 if (show(opt, 4)) {
3630 output(opt, 4, _("Merging:"));
3631 output_commit_title(opt, h1);
3632 output_commit_title(opt, h2);
3635 if (!merge_bases) {
3636 if (repo_get_merge_bases(the_repository, h1, h2,
3637 &merge_bases) < 0)
3638 return -1;
3639 merge_bases = reverse_commit_list(merge_bases);
3642 if (show(opt, 5)) {
3643 unsigned cnt = commit_list_count(merge_bases);
3645 output(opt, 5, Q_("found %u common ancestor:",
3646 "found %u common ancestors:", cnt), cnt);
3647 for (iter = merge_bases; iter; iter = iter->next)
3648 output_commit_title(opt, iter->item);
3651 merged_merge_bases = pop_commit(&merge_bases);
3652 if (!merged_merge_bases) {
3653 /* if there is no common ancestor, use an empty tree */
3654 struct tree *tree;
3656 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3657 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3658 "ancestor");
3659 ancestor_name = "empty tree";
3660 } else if (opt->ancestor && !opt->priv->call_depth) {
3661 ancestor_name = opt->ancestor;
3662 } else if (merge_bases) {
3663 ancestor_name = "merged common ancestors";
3664 } else {
3665 strbuf_add_unique_abbrev(&merge_base_abbrev,
3666 &merged_merge_bases->object.oid,
3667 DEFAULT_ABBREV);
3668 ancestor_name = merge_base_abbrev.buf;
3671 for (iter = merge_bases; iter; iter = iter->next) {
3672 const char *saved_b1, *saved_b2;
3673 opt->priv->call_depth++;
3675 * When the merge fails, the result contains files
3676 * with conflict markers. The cleanness flag is
3677 * ignored (unless indicating an error), it was never
3678 * actually used, as result of merge_trees has always
3679 * overwritten it: the committed "conflicts" were
3680 * already resolved.
3682 discard_index(opt->repo->index);
3683 saved_b1 = opt->branch1;
3684 saved_b2 = opt->branch2;
3685 opt->branch1 = "Temporary merge branch 1";
3686 opt->branch2 = "Temporary merge branch 2";
3687 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3688 NULL, &merged_merge_bases) < 0)
3689 return -1;
3690 opt->branch1 = saved_b1;
3691 opt->branch2 = saved_b2;
3692 opt->priv->call_depth--;
3694 if (!merged_merge_bases)
3695 return err(opt, _("merge returned no commit"));
3699 * FIXME: Since merge_recursive_internal() is only ever called by
3700 * places that ensure the index is loaded first
3701 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3702 * case where the merge base was unique that means when we get here
3703 * we immediately discard the index and re-read it, which is a
3704 * complete waste of time. We should only be discarding and
3705 * re-reading if we were forced to recurse.
3707 discard_index(opt->repo->index);
3708 if (!opt->priv->call_depth)
3709 repo_read_index(opt->repo);
3711 opt->ancestor = ancestor_name;
3712 clean = merge_trees_internal(opt,
3713 repo_get_commit_tree(opt->repo, h1),
3714 repo_get_commit_tree(opt->repo, h2),
3715 repo_get_commit_tree(opt->repo,
3716 merged_merge_bases),
3717 &result_tree);
3718 strbuf_release(&merge_base_abbrev);
3719 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3720 if (clean < 0) {
3721 flush_output(opt);
3722 return clean;
3725 if (opt->priv->call_depth) {
3726 *result = make_virtual_commit(opt->repo, result_tree,
3727 "merged tree");
3728 commit_list_insert(h1, &(*result)->parents);
3729 commit_list_insert(h2, &(*result)->parents->next);
3731 return clean;
3734 static int merge_start(struct merge_options *opt, struct tree *head)
3736 struct strbuf sb = STRBUF_INIT;
3738 /* Sanity checks on opt */
3739 assert(opt->repo);
3741 assert(opt->branch1 && opt->branch2);
3743 assert(opt->detect_renames >= -1 &&
3744 opt->detect_renames <= DIFF_DETECT_COPY);
3745 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3746 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3747 assert(opt->rename_limit >= -1);
3748 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3749 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3751 assert(opt->xdl_opts >= 0);
3752 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3753 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3755 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3756 assert(opt->buffer_output <= 2);
3757 assert(opt->obuf.len == 0);
3759 assert(opt->priv == NULL);
3761 /* Not supported; option specific to merge-ort */
3762 assert(!opt->record_conflict_msgs_as_headers);
3763 assert(!opt->msg_header_prefix);
3765 /* Sanity check on repo state; index must match head */
3766 if (repo_index_has_changes(opt->repo, head, &sb)) {
3767 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3768 sb.buf);
3769 strbuf_release(&sb);
3770 return -1;
3773 CALLOC_ARRAY(opt->priv, 1);
3774 string_list_init_dup(&opt->priv->df_conflict_file_set);
3775 return 0;
3778 static void merge_finalize(struct merge_options *opt)
3780 flush_output(opt);
3781 if (!opt->priv->call_depth && opt->buffer_output < 2)
3782 strbuf_release(&opt->obuf);
3783 if (show(opt, 2))
3784 diff_warn_rename_limit("merge.renamelimit",
3785 opt->priv->needed_rename_limit, 0);
3786 FREE_AND_NULL(opt->priv);
3789 int merge_trees(struct merge_options *opt,
3790 struct tree *head,
3791 struct tree *merge,
3792 struct tree *merge_base)
3794 int clean;
3795 struct tree *ignored;
3797 assert(opt->ancestor != NULL);
3799 if (merge_start(opt, head))
3800 return -1;
3801 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3802 merge_finalize(opt);
3804 return clean;
3807 int merge_recursive(struct merge_options *opt,
3808 struct commit *h1,
3809 struct commit *h2,
3810 struct commit_list *merge_bases,
3811 struct commit **result)
3813 int clean;
3815 assert(opt->ancestor == NULL ||
3816 !strcmp(opt->ancestor, "constructed merge base"));
3818 prepare_repo_settings(opt->repo);
3819 opt->repo->settings.command_requires_full_index = 1;
3821 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3822 return -1;
3823 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3824 merge_finalize(opt);
3826 return clean;
3829 static struct commit *get_ref(struct repository *repo,
3830 const struct object_id *oid,
3831 const char *name)
3833 struct object *object;
3835 object = deref_tag(repo, parse_object(repo, oid),
3836 name, strlen(name));
3837 if (!object)
3838 return NULL;
3839 if (object->type == OBJ_TREE)
3840 return make_virtual_commit(repo, (struct tree*)object, name);
3841 if (object->type != OBJ_COMMIT)
3842 return NULL;
3843 if (repo_parse_commit(repo, (struct commit *)object))
3844 return NULL;
3845 return (struct commit *)object;
3848 int merge_recursive_generic(struct merge_options *opt,
3849 const struct object_id *head,
3850 const struct object_id *merge,
3851 int num_merge_bases,
3852 const struct object_id **merge_bases,
3853 struct commit **result)
3855 int clean;
3856 struct lock_file lock = LOCK_INIT;
3857 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3858 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3859 struct commit_list *ca = NULL;
3861 if (merge_bases) {
3862 int i;
3863 for (i = 0; i < num_merge_bases; ++i) {
3864 struct commit *base;
3865 if (!(base = get_ref(opt->repo, merge_bases[i],
3866 oid_to_hex(merge_bases[i]))))
3867 return err(opt, _("Could not parse object '%s'"),
3868 oid_to_hex(merge_bases[i]));
3869 commit_list_insert(base, &ca);
3871 if (num_merge_bases == 1)
3872 opt->ancestor = "constructed merge base";
3875 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3876 clean = merge_recursive(opt, head_commit, next_commit, ca,
3877 result);
3878 if (clean < 0) {
3879 rollback_lock_file(&lock);
3880 return clean;
3883 if (write_locked_index(opt->repo->index, &lock,
3884 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3885 return err(opt, _("Unable to write index."));
3887 return clean ? 0 : 1;
3890 static void merge_recursive_config(struct merge_options *opt)
3892 char *value = NULL;
3893 int renormalize = 0;
3894 git_config_get_int("merge.verbosity", &opt->verbosity);
3895 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3896 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3897 git_config_get_bool("merge.renormalize", &renormalize);
3898 opt->renormalize = renormalize;
3899 if (!git_config_get_string("diff.renames", &value)) {
3900 opt->detect_renames = git_config_rename("diff.renames", value);
3901 free(value);
3903 if (!git_config_get_string("merge.renames", &value)) {
3904 opt->detect_renames = git_config_rename("merge.renames", value);
3905 free(value);
3907 if (!git_config_get_string("merge.directoryrenames", &value)) {
3908 int boolval = git_parse_maybe_bool(value);
3909 if (0 <= boolval) {
3910 opt->detect_directory_renames = boolval ?
3911 MERGE_DIRECTORY_RENAMES_TRUE :
3912 MERGE_DIRECTORY_RENAMES_NONE;
3913 } else if (!strcasecmp(value, "conflict")) {
3914 opt->detect_directory_renames =
3915 MERGE_DIRECTORY_RENAMES_CONFLICT;
3916 } /* avoid erroring on values from future versions of git */
3917 free(value);
3919 git_config(git_xmerge_config, NULL);
3922 void init_merge_options(struct merge_options *opt,
3923 struct repository *repo)
3925 const char *merge_verbosity;
3926 memset(opt, 0, sizeof(struct merge_options));
3928 opt->repo = repo;
3930 opt->detect_renames = -1;
3931 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3932 opt->rename_limit = -1;
3934 opt->verbosity = 2;
3935 opt->buffer_output = 1;
3936 strbuf_init(&opt->obuf, 0);
3938 opt->renormalize = 0;
3940 merge_recursive_config(opt);
3941 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3942 if (merge_verbosity)
3943 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3944 if (opt->verbosity >= 5)
3945 opt->buffer_output = 0;
3949 * For now, members of merge_options do not need deep copying, but
3950 * it may change in the future, in which case we would need to update
3951 * this, and also make a matching change to clear_merge_options() to
3952 * release the resources held by a copied instance.
3954 void copy_merge_options(struct merge_options *dst, struct merge_options *src)
3956 *dst = *src;
3959 void clear_merge_options(struct merge_options *opt UNUSED)
3961 ; /* no-op as our copy is shallow right now */
3964 int parse_merge_opt(struct merge_options *opt, const char *s)
3966 const char *arg;
3968 if (!s || !*s)
3969 return -1;
3970 if (!strcmp(s, "ours"))
3971 opt->recursive_variant = MERGE_VARIANT_OURS;
3972 else if (!strcmp(s, "theirs"))
3973 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3974 else if (!strcmp(s, "subtree"))
3975 opt->subtree_shift = "";
3976 else if (skip_prefix(s, "subtree=", &arg))
3977 opt->subtree_shift = arg;
3978 else if (!strcmp(s, "patience"))
3979 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3980 else if (!strcmp(s, "histogram"))
3981 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3982 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3983 long value = parse_algorithm_value(arg);
3984 if (value < 0)
3985 return -1;
3986 /* clear out previous settings */
3987 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3988 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3989 opt->xdl_opts |= value;
3991 else if (!strcmp(s, "ignore-space-change"))
3992 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3993 else if (!strcmp(s, "ignore-all-space"))
3994 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3995 else if (!strcmp(s, "ignore-space-at-eol"))
3996 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3997 else if (!strcmp(s, "ignore-cr-at-eol"))
3998 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3999 else if (!strcmp(s, "renormalize"))
4000 opt->renormalize = 1;
4001 else if (!strcmp(s, "no-renormalize"))
4002 opt->renormalize = 0;
4003 else if (!strcmp(s, "no-renames"))
4004 opt->detect_renames = 0;
4005 else if (!strcmp(s, "find-renames")) {
4006 opt->detect_renames = 1;
4007 opt->rename_score = 0;
4009 else if (skip_prefix(s, "find-renames=", &arg) ||
4010 skip_prefix(s, "rename-threshold=", &arg)) {
4011 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
4012 return -1;
4013 opt->detect_renames = 1;
4016 * Please update $__git_merge_strategy_options in
4017 * git-completion.bash when you add new options
4019 else
4020 return -1;
4021 return 0;