compat/fsmonitor: fix socket path in networked SHA256 repos
[git/gitster.git] / merge-recursive.c
blob46ee364af736ac310c215771d315ac10c0c02d5e
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 */
7 #define USE_THE_REPOSITORY_VARIABLE
9 #include "git-compat-util.h"
10 #include "merge-recursive.h"
12 #include "alloc.h"
13 #include "cache-tree.h"
14 #include "commit.h"
15 #include "commit-reach.h"
16 #include "config.h"
17 #include "diff.h"
18 #include "diffcore.h"
19 #include "dir.h"
20 #include "environment.h"
21 #include "gettext.h"
22 #include "hex.h"
23 #include "merge-ll.h"
24 #include "lockfile.h"
25 #include "match-trees.h"
26 #include "name-hash.h"
27 #include "object-file.h"
28 #include "object-name.h"
29 #include "object-store-ll.h"
30 #include "path.h"
31 #include "repository.h"
32 #include "revision.h"
33 #include "sparse-index.h"
34 #include "string-list.h"
35 #include "symlinks.h"
36 #include "tag.h"
37 #include "tree-walk.h"
38 #include "unpack-trees.h"
39 #include "xdiff-interface.h"
41 struct merge_options_internal {
42 int call_depth;
43 int needed_rename_limit;
44 struct hashmap current_file_dir_set;
45 struct string_list df_conflict_file_set;
46 struct unpack_trees_options unpack_opts;
47 struct index_state orig_index;
50 struct path_hashmap_entry {
51 struct hashmap_entry e;
52 char path[FLEX_ARRAY];
55 static int path_hashmap_cmp(const void *cmp_data UNUSED,
56 const struct hashmap_entry *eptr,
57 const struct hashmap_entry *entry_or_key,
58 const void *keydata)
60 const struct path_hashmap_entry *a, *b;
61 const char *key = keydata;
63 a = container_of(eptr, const struct path_hashmap_entry, e);
64 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
66 return fspathcmp(a->path, key ? key : b->path);
70 * For dir_rename_entry, directory names are stored as a full path from the
71 * toplevel of the repository and do not include a trailing '/'. Also:
73 * dir: original name of directory being renamed
74 * non_unique_new_dir: if true, could not determine new_dir
75 * new_dir: final name of directory being renamed
76 * possible_new_dirs: temporary used to help determine new_dir; see comments
77 * in get_directory_renames() for details
79 struct dir_rename_entry {
80 struct hashmap_entry ent;
81 char *dir;
82 unsigned non_unique_new_dir:1;
83 struct strbuf new_dir;
84 struct string_list possible_new_dirs;
87 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
88 char *dir)
90 struct dir_rename_entry key;
92 if (!dir)
93 return NULL;
94 hashmap_entry_init(&key.ent, strhash(dir));
95 key.dir = dir;
96 return hashmap_get_entry(hashmap, &key, ent, NULL);
99 static int dir_rename_cmp(const void *cmp_data UNUSED,
100 const struct hashmap_entry *eptr,
101 const struct hashmap_entry *entry_or_key,
102 const void *keydata UNUSED)
104 const struct dir_rename_entry *e1, *e2;
106 e1 = container_of(eptr, const struct dir_rename_entry, ent);
107 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
109 return strcmp(e1->dir, e2->dir);
112 static void dir_rename_init(struct hashmap *map)
114 hashmap_init(map, dir_rename_cmp, NULL, 0);
117 static void dir_rename_entry_init(struct dir_rename_entry *entry,
118 char *directory)
120 hashmap_entry_init(&entry->ent, strhash(directory));
121 entry->dir = directory;
122 entry->non_unique_new_dir = 0;
123 strbuf_init(&entry->new_dir, 0);
124 string_list_init_nodup(&entry->possible_new_dirs);
127 struct collision_entry {
128 struct hashmap_entry ent;
129 char *target_file;
130 struct string_list source_files;
131 unsigned reported_already:1;
134 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
135 char *target_file)
137 struct collision_entry key;
139 hashmap_entry_init(&key.ent, strhash(target_file));
140 key.target_file = target_file;
141 return hashmap_get_entry(hashmap, &key, ent, NULL);
144 static int collision_cmp(const void *cmp_data UNUSED,
145 const struct hashmap_entry *eptr,
146 const struct hashmap_entry *entry_or_key,
147 const void *keydata UNUSED)
149 const struct collision_entry *e1, *e2;
151 e1 = container_of(eptr, const struct collision_entry, ent);
152 e2 = container_of(entry_or_key, const struct collision_entry, ent);
154 return strcmp(e1->target_file, e2->target_file);
157 static void collision_init(struct hashmap *map)
159 hashmap_init(map, collision_cmp, NULL, 0);
162 static void flush_output(struct merge_options *opt)
164 if (opt->buffer_output < 2 && opt->obuf.len) {
165 fputs(opt->obuf.buf, stdout);
166 strbuf_reset(&opt->obuf);
170 __attribute__((format (printf, 2, 3)))
171 static int err(struct merge_options *opt, const char *err, ...)
173 va_list params;
175 if (opt->buffer_output < 2)
176 flush_output(opt);
177 else {
178 strbuf_complete(&opt->obuf, '\n');
179 strbuf_addstr(&opt->obuf, "error: ");
181 va_start(params, err);
182 strbuf_vaddf(&opt->obuf, err, params);
183 va_end(params);
184 if (opt->buffer_output > 1)
185 strbuf_addch(&opt->obuf, '\n');
186 else {
187 error("%s", opt->obuf.buf);
188 strbuf_reset(&opt->obuf);
191 return -1;
194 static struct tree *shift_tree_object(struct repository *repo,
195 struct tree *one, struct tree *two,
196 const char *subtree_shift)
198 struct object_id shifted;
200 if (!*subtree_shift) {
201 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
202 } else {
203 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
204 subtree_shift);
206 if (oideq(&two->object.oid, &shifted))
207 return two;
208 return lookup_tree(repo, &shifted);
211 static inline void set_commit_tree(struct commit *c, struct tree *t)
213 c->maybe_tree = t;
216 static struct commit *make_virtual_commit(struct repository *repo,
217 struct tree *tree,
218 const char *comment)
220 struct commit *commit = alloc_commit_node(repo);
222 set_merge_remote_desc(commit, comment, (struct object *)commit);
223 set_commit_tree(commit, tree);
224 commit->object.parsed = 1;
225 return commit;
228 enum rename_type {
229 RENAME_NORMAL = 0,
230 RENAME_VIA_DIR,
231 RENAME_ADD,
232 RENAME_DELETE,
233 RENAME_ONE_FILE_TO_ONE,
234 RENAME_ONE_FILE_TO_TWO,
235 RENAME_TWO_FILES_TO_ONE
239 * Since we want to write the index eventually, we cannot reuse the index
240 * for these (temporary) data.
242 struct stage_data {
243 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
244 struct rename_conflict_info *rename_conflict_info;
245 unsigned processed:1;
248 struct rename {
249 unsigned processed:1;
250 struct diff_filepair *pair;
251 const char *branch; /* branch that the rename occurred on */
253 * If directory rename detection affected this rename, what was its
254 * original type ('A' or 'R') and it's original destination before
255 * the directory rename (otherwise, '\0' and NULL for these two vars).
257 char dir_rename_original_type;
258 char *dir_rename_original_dest;
260 * Purpose of src_entry and dst_entry:
262 * If 'before' is renamed to 'after' then src_entry will contain
263 * the versions of 'before' from the merge_base, HEAD, and MERGE in
264 * stages 1, 2, and 3; dst_entry will contain the respective
265 * versions of 'after' in corresponding locations. Thus, we have a
266 * total of six modes and oids, though some will be null. (Stage 0
267 * is ignored; we're interested in handling conflicts.)
269 * Since we don't turn on break-rewrites by default, neither
270 * src_entry nor dst_entry can have all three of their stages have
271 * non-null oids, meaning at most four of the six will be non-null.
272 * Also, since this is a rename, both src_entry and dst_entry will
273 * have at least one non-null oid, meaning at least two will be
274 * non-null. Of the six oids, a typical rename will have three be
275 * non-null. Only two implies a rename/delete, and four implies a
276 * rename/add.
278 struct stage_data *src_entry;
279 struct stage_data *dst_entry;
282 struct rename_conflict_info {
283 enum rename_type rename_type;
284 struct rename *ren1;
285 struct rename *ren2;
288 static inline void setup_rename_conflict_info(enum rename_type rename_type,
289 struct merge_options *opt,
290 struct rename *ren1,
291 struct rename *ren2)
293 struct rename_conflict_info *ci;
296 * When we have two renames involved, it's easiest to get the
297 * correct things into stage 2 and 3, and to make sure that the
298 * content merge puts HEAD before the other branch if we just
299 * ensure that branch1 == opt->branch1. So, simply flip arguments
300 * around if we don't have that.
302 if (ren2 && ren1->branch != opt->branch1) {
303 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
304 return;
307 CALLOC_ARRAY(ci, 1);
308 ci->rename_type = rename_type;
309 ci->ren1 = ren1;
310 ci->ren2 = ren2;
312 ci->ren1->dst_entry->processed = 0;
313 ci->ren1->dst_entry->rename_conflict_info = ci;
314 if (ren2) {
315 ci->ren2->dst_entry->rename_conflict_info = ci;
319 static int show(struct merge_options *opt, int v)
321 return (!opt->priv->call_depth && opt->verbosity >= v) ||
322 opt->verbosity >= 5;
325 __attribute__((format (printf, 3, 4)))
326 static void output(struct merge_options *opt, int v, const char *fmt, ...)
328 va_list ap;
330 if (!show(opt, v))
331 return;
333 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
335 va_start(ap, fmt);
336 strbuf_vaddf(&opt->obuf, fmt, ap);
337 va_end(ap);
339 strbuf_addch(&opt->obuf, '\n');
340 if (!opt->buffer_output)
341 flush_output(opt);
344 static void repo_output_commit_title(struct merge_options *opt,
345 struct repository *repo,
346 struct commit *commit)
348 struct merge_remote_desc *desc;
350 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
351 desc = merge_remote_util(commit);
352 if (desc)
353 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
354 else {
355 strbuf_repo_add_unique_abbrev(&opt->obuf, repo,
356 &commit->object.oid,
357 DEFAULT_ABBREV);
358 strbuf_addch(&opt->obuf, ' ');
359 if (repo_parse_commit(repo, commit) != 0)
360 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
361 else {
362 const char *title;
363 const char *msg = repo_get_commit_buffer(repo, commit, NULL);
364 int len = find_commit_subject(msg, &title);
365 if (len)
366 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
367 repo_unuse_commit_buffer(repo, commit, msg);
370 flush_output(opt);
373 static void output_commit_title(struct merge_options *opt, struct commit *commit)
375 repo_output_commit_title(opt, the_repository, commit);
378 static int add_cacheinfo(struct merge_options *opt,
379 const struct diff_filespec *blob,
380 const char *path, int stage, int refresh, int options)
382 struct index_state *istate = opt->repo->index;
383 struct cache_entry *ce;
384 int ret;
386 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
387 if (!ce)
388 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
390 ret = add_index_entry(istate, ce, options);
391 if (refresh) {
392 struct cache_entry *nce;
394 nce = refresh_cache_entry(istate, ce,
395 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
396 if (!nce)
397 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
398 if (nce != ce)
399 ret = add_index_entry(istate, nce, options);
401 return ret;
404 static inline int merge_detect_rename(struct merge_options *opt)
406 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
409 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
411 if (parse_tree(tree) < 0)
412 exit(128);
413 init_tree_desc(desc, &tree->object.oid, tree->buffer, tree->size);
416 static int unpack_trees_start(struct merge_options *opt,
417 struct tree *common,
418 struct tree *head,
419 struct tree *merge)
421 int rc;
422 struct tree_desc t[3];
423 struct index_state tmp_index = INDEX_STATE_INIT(opt->repo);
425 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
426 if (opt->priv->call_depth)
427 opt->priv->unpack_opts.index_only = 1;
428 else {
429 opt->priv->unpack_opts.update = 1;
430 /* FIXME: should only do this if !overwrite_ignore */
431 opt->priv->unpack_opts.preserve_ignored = 0;
433 opt->priv->unpack_opts.merge = 1;
434 opt->priv->unpack_opts.head_idx = 2;
435 opt->priv->unpack_opts.fn = threeway_merge;
436 opt->priv->unpack_opts.src_index = opt->repo->index;
437 opt->priv->unpack_opts.dst_index = &tmp_index;
438 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
439 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
441 init_tree_desc_from_tree(t+0, common);
442 init_tree_desc_from_tree(t+1, head);
443 init_tree_desc_from_tree(t+2, merge);
445 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
446 cache_tree_free(&opt->repo->index->cache_tree);
449 * Update opt->repo->index to match the new results, AFTER saving a
450 * copy in opt->priv->orig_index. Update src_index to point to the
451 * saved copy. (verify_uptodate() checks src_index, and the original
452 * index is the one that had the necessary modification timestamps.)
454 opt->priv->orig_index = *opt->repo->index;
455 *opt->repo->index = tmp_index;
456 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
458 return rc;
461 static void unpack_trees_finish(struct merge_options *opt)
463 discard_index(&opt->priv->orig_index);
464 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
467 static int save_files_dirs(const struct object_id *oid UNUSED,
468 struct strbuf *base, const char *path,
469 unsigned int mode, void *context)
471 struct path_hashmap_entry *entry;
472 int baselen = base->len;
473 struct merge_options *opt = context;
475 strbuf_addstr(base, path);
477 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
478 hashmap_entry_init(&entry->e, fspathhash(entry->path));
479 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
481 strbuf_setlen(base, baselen);
482 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
485 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
487 struct pathspec match_all;
488 memset(&match_all, 0, sizeof(match_all));
489 read_tree(opt->repo, tree,
490 &match_all, save_files_dirs, opt);
493 static int get_tree_entry_if_blob(struct repository *r,
494 const struct object_id *tree,
495 const char *path,
496 struct diff_filespec *dfs)
498 int ret;
500 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
501 if (S_ISDIR(dfs->mode)) {
502 oidcpy(&dfs->oid, null_oid());
503 dfs->mode = 0;
505 return ret;
509 * Returns an index_entry instance which doesn't have to correspond to
510 * a real cache entry in Git's index.
512 static struct stage_data *insert_stage_data(struct repository *r,
513 const char *path,
514 struct tree *o, struct tree *a, struct tree *b,
515 struct string_list *entries)
517 struct string_list_item *item;
518 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
519 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
520 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
521 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
522 item = string_list_insert(entries, path);
523 item->util = e;
524 return e;
528 * Create a dictionary mapping file names to stage_data objects. The
529 * dictionary contains one entry for every path with a non-zero stage entry.
531 static struct string_list *get_unmerged(struct index_state *istate)
533 struct string_list *unmerged = xmalloc(sizeof(struct string_list));
534 int i;
536 string_list_init_dup(unmerged);
538 /* TODO: audit for interaction with sparse-index. */
539 ensure_full_index(istate);
540 for (i = 0; i < istate->cache_nr; i++) {
541 struct string_list_item *item;
542 struct stage_data *e;
543 const struct cache_entry *ce = istate->cache[i];
544 if (!ce_stage(ce))
545 continue;
547 item = string_list_lookup(unmerged, ce->name);
548 if (!item) {
549 item = string_list_insert(unmerged, ce->name);
550 item->util = xcalloc(1, sizeof(struct stage_data));
552 e = item->util;
553 e->stages[ce_stage(ce)].mode = ce->ce_mode;
554 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
557 return unmerged;
560 static int string_list_df_name_compare(const char *one, const char *two)
562 int onelen = strlen(one);
563 int twolen = strlen(two);
565 * Here we only care that entries for D/F conflicts are
566 * adjacent, in particular with the file of the D/F conflict
567 * appearing before files below the corresponding directory.
568 * The order of the rest of the list is irrelevant for us.
570 * To achieve this, we sort with df_name_compare and provide
571 * the mode S_IFDIR so that D/F conflicts will sort correctly.
572 * We use the mode S_IFDIR for everything else for simplicity,
573 * since in other cases any changes in their order due to
574 * sorting cause no problems for us.
576 int cmp = df_name_compare(one, onelen, S_IFDIR,
577 two, twolen, S_IFDIR);
579 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
580 * that 'foo' comes before 'foo/bar'.
582 if (cmp)
583 return cmp;
584 return onelen - twolen;
587 static void record_df_conflict_files(struct merge_options *opt,
588 struct string_list *entries)
590 /* If there is a D/F conflict and the file for such a conflict
591 * currently exists in the working tree, we want to allow it to be
592 * removed to make room for the corresponding directory if needed.
593 * The files underneath the directories of such D/F conflicts will
594 * be processed before the corresponding file involved in the D/F
595 * conflict. If the D/F directory ends up being removed by the
596 * merge, then we won't have to touch the D/F file. If the D/F
597 * directory needs to be written to the working copy, then the D/F
598 * file will simply be removed (in make_room_for_path()) to make
599 * room for the necessary paths. Note that if both the directory
600 * and the file need to be present, then the D/F file will be
601 * reinstated with a new unique name at the time it is processed.
603 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
604 const char *last_file = NULL;
605 int last_len = 0;
606 int i;
609 * If we're merging merge-bases, we don't want to bother with
610 * any working directory changes.
612 if (opt->priv->call_depth)
613 return;
615 /* Ensure D/F conflicts are adjacent in the entries list. */
616 for (i = 0; i < entries->nr; i++) {
617 struct string_list_item *next = &entries->items[i];
618 string_list_append(&df_sorted_entries, next->string)->util =
619 next->util;
621 df_sorted_entries.cmp = string_list_df_name_compare;
622 string_list_sort(&df_sorted_entries);
624 string_list_clear(&opt->priv->df_conflict_file_set, 1);
625 for (i = 0; i < df_sorted_entries.nr; i++) {
626 const char *path = df_sorted_entries.items[i].string;
627 int len = strlen(path);
628 struct stage_data *e = df_sorted_entries.items[i].util;
631 * Check if last_file & path correspond to a D/F conflict;
632 * i.e. whether path is last_file+'/'+<something>.
633 * If so, record that it's okay to remove last_file to make
634 * room for path and friends if needed.
636 if (last_file &&
637 len > last_len &&
638 memcmp(path, last_file, last_len) == 0 &&
639 path[last_len] == '/') {
640 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
644 * Determine whether path could exist as a file in the
645 * working directory as a possible D/F conflict. This
646 * will only occur when it exists in stage 2 as a
647 * file.
649 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
650 last_file = path;
651 last_len = len;
652 } else {
653 last_file = NULL;
656 string_list_clear(&df_sorted_entries, 0);
659 static int update_stages(struct merge_options *opt, const char *path,
660 const struct diff_filespec *o,
661 const struct diff_filespec *a,
662 const struct diff_filespec *b)
666 * NOTE: It is usually a bad idea to call update_stages on a path
667 * before calling update_file on that same path, since it can
668 * sometimes lead to spurious "refusing to lose untracked file..."
669 * messages from update_file (via make_room_for path via
670 * would_lose_untracked). Instead, reverse the order of the calls
671 * (executing update_file first and then update_stages).
673 int clear = 1;
674 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
675 if (clear)
676 if (remove_file_from_index(opt->repo->index, path))
677 return -1;
678 if (o)
679 if (add_cacheinfo(opt, o, path, 1, 0, options))
680 return -1;
681 if (a)
682 if (add_cacheinfo(opt, a, path, 2, 0, options))
683 return -1;
684 if (b)
685 if (add_cacheinfo(opt, b, path, 3, 0, options))
686 return -1;
687 return 0;
690 static void update_entry(struct stage_data *entry,
691 struct diff_filespec *o,
692 struct diff_filespec *a,
693 struct diff_filespec *b)
695 entry->processed = 0;
696 entry->stages[1].mode = o->mode;
697 entry->stages[2].mode = a->mode;
698 entry->stages[3].mode = b->mode;
699 oidcpy(&entry->stages[1].oid, &o->oid);
700 oidcpy(&entry->stages[2].oid, &a->oid);
701 oidcpy(&entry->stages[3].oid, &b->oid);
704 static int remove_file(struct merge_options *opt, int clean,
705 const char *path, int no_wd)
707 int update_cache = opt->priv->call_depth || clean;
708 int update_working_directory = !opt->priv->call_depth && !no_wd;
710 if (update_cache) {
711 if (remove_file_from_index(opt->repo->index, path))
712 return -1;
714 if (update_working_directory) {
715 if (ignore_case) {
716 struct cache_entry *ce;
717 ce = index_file_exists(opt->repo->index, path, strlen(path),
718 ignore_case);
719 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
720 return 0;
722 if (remove_path(path))
723 return -1;
725 return 0;
728 /* add a string to a strbuf, but converting "/" to "_" */
729 static void add_flattened_path(struct strbuf *out, const char *s)
731 size_t i = out->len;
732 strbuf_addstr(out, s);
733 for (; i < out->len; i++)
734 if (out->buf[i] == '/')
735 out->buf[i] = '_';
738 static char *unique_path(struct merge_options *opt,
739 const char *path,
740 const char *branch)
742 struct path_hashmap_entry *entry;
743 struct strbuf newpath = STRBUF_INIT;
744 int suffix = 0;
745 size_t base_len;
747 strbuf_addf(&newpath, "%s~", path);
748 add_flattened_path(&newpath, branch);
750 base_len = newpath.len;
751 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
752 fspathhash(newpath.buf), newpath.buf) ||
753 (!opt->priv->call_depth && file_exists(newpath.buf))) {
754 strbuf_setlen(&newpath, base_len);
755 strbuf_addf(&newpath, "_%d", suffix++);
758 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
759 hashmap_entry_init(&entry->e, fspathhash(entry->path));
760 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
761 return strbuf_detach(&newpath, NULL);
765 * Check whether a directory in the index is in the way of an incoming
766 * file. Return 1 if so. If check_working_copy is non-zero, also
767 * check the working directory. If empty_ok is non-zero, also return
768 * 0 in the case where the working-tree dir exists but is empty.
770 static int dir_in_way(struct index_state *istate, const char *path,
771 int check_working_copy, int empty_ok)
773 int pos;
774 struct strbuf dirpath = STRBUF_INIT;
775 struct stat st;
777 strbuf_addstr(&dirpath, path);
778 strbuf_addch(&dirpath, '/');
780 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
782 if (pos < 0)
783 pos = -1 - pos;
784 if (pos < istate->cache_nr &&
785 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
786 strbuf_release(&dirpath);
787 return 1;
790 strbuf_release(&dirpath);
791 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
792 !(empty_ok && is_empty_dir(path)) &&
793 !has_symlink_leading_path(path, strlen(path));
797 * Returns whether path was tracked in the index before the merge started,
798 * and its oid and mode match the specified values
800 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
801 const struct diff_filespec *blob)
803 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
804 struct cache_entry *ce;
806 if (0 > pos)
807 /* we were not tracking this path before the merge */
808 return 0;
810 /* See if the file we were tracking before matches */
811 ce = opt->priv->orig_index.cache[pos];
812 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
816 * Returns whether path was tracked in the index before the merge started
818 static int was_tracked(struct merge_options *opt, const char *path)
820 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
822 if (0 <= pos)
823 /* we were tracking this path before the merge */
824 return 1;
826 return 0;
829 static int would_lose_untracked(struct merge_options *opt, const char *path)
831 struct index_state *istate = opt->repo->index;
834 * This may look like it can be simplified to:
835 * return !was_tracked(opt, path) && file_exists(path)
836 * but it can't. This function needs to know whether path was in
837 * the working tree due to EITHER having been tracked in the index
838 * before the merge OR having been put into the working copy and
839 * index by unpack_trees(). Due to that either-or requirement, we
840 * check the current index instead of the original one.
842 * Note that we do not need to worry about merge-recursive itself
843 * updating the index after unpack_trees() and before calling this
844 * function, because we strictly require all code paths in
845 * merge-recursive to update the working tree first and the index
846 * second. Doing otherwise would break
847 * update_file()/would_lose_untracked(); see every comment in this
848 * file which mentions "update_stages".
850 int pos = index_name_pos(istate, path, strlen(path));
852 if (pos < 0)
853 pos = -1 - pos;
854 while (pos < istate->cache_nr &&
855 !strcmp(path, istate->cache[pos]->name)) {
857 * If stage #0, it is definitely tracked.
858 * If it has stage #2 then it was tracked
859 * before this merge started. All other
860 * cases the path was not tracked.
862 switch (ce_stage(istate->cache[pos])) {
863 case 0:
864 case 2:
865 return 0;
867 pos++;
869 return file_exists(path);
872 static int was_dirty(struct merge_options *opt, const char *path)
874 struct cache_entry *ce;
875 int dirty = 1;
877 if (opt->priv->call_depth || !was_tracked(opt, path))
878 return !dirty;
880 ce = index_file_exists(opt->priv->unpack_opts.src_index,
881 path, strlen(path), ignore_case);
882 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
883 return dirty;
886 static int make_room_for_path(struct merge_options *opt, const char *path)
888 int status, i;
889 const char *msg = _("failed to create path '%s'%s");
891 /* Unlink any D/F conflict files that are in the way */
892 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
893 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
894 size_t pathlen = strlen(path);
895 size_t df_pathlen = strlen(df_path);
896 if (df_pathlen < pathlen &&
897 path[df_pathlen] == '/' &&
898 strncmp(path, df_path, df_pathlen) == 0) {
899 output(opt, 3,
900 _("Removing %s to make room for subdirectory\n"),
901 df_path);
902 unlink(df_path);
903 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
904 i, 0);
905 break;
909 /* Make sure leading directories are created */
910 status = safe_create_leading_directories_const(path);
911 if (status) {
912 if (status == SCLD_EXISTS)
913 /* something else exists */
914 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
915 return err(opt, msg, path, "");
919 * Do not unlink a file in the work tree if we are not
920 * tracking it.
922 if (would_lose_untracked(opt, path))
923 return err(opt, _("refusing to lose untracked file at '%s'"),
924 path);
926 /* Successful unlink is good.. */
927 if (!unlink(path))
928 return 0;
929 /* .. and so is no existing file */
930 if (errno == ENOENT)
931 return 0;
932 /* .. but not some other error (who really cares what?) */
933 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
936 static int update_file_flags(struct merge_options *opt,
937 const struct diff_filespec *contents,
938 const char *path,
939 int update_cache,
940 int update_wd)
942 int ret = 0;
944 if (opt->priv->call_depth)
945 update_wd = 0;
947 if (update_wd) {
948 enum object_type type;
949 void *buf;
950 unsigned long size;
952 if (S_ISGITLINK(contents->mode)) {
954 * We may later decide to recursively descend into
955 * the submodule directory and update its index
956 * and/or work tree, but we do not do that now.
958 update_wd = 0;
959 goto update_index;
962 buf = repo_read_object_file(the_repository, &contents->oid,
963 &type, &size);
964 if (!buf) {
965 ret = err(opt, _("cannot read object %s '%s'"),
966 oid_to_hex(&contents->oid), path);
967 goto free_buf;
969 if (type != OBJ_BLOB) {
970 ret = err(opt, _("blob expected for %s '%s'"),
971 oid_to_hex(&contents->oid), path);
972 goto free_buf;
974 if (S_ISREG(contents->mode)) {
975 struct strbuf strbuf = STRBUF_INIT;
976 if (convert_to_working_tree(opt->repo->index,
977 path, buf, size, &strbuf, NULL)) {
978 free(buf);
979 size = strbuf.len;
980 buf = strbuf_detach(&strbuf, NULL);
984 if (make_room_for_path(opt, path) < 0) {
985 update_wd = 0;
986 goto free_buf;
988 if (S_ISREG(contents->mode) ||
989 (!has_symlinks && S_ISLNK(contents->mode))) {
990 int fd;
991 int mode = (contents->mode & 0100 ? 0777 : 0666);
993 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
994 if (fd < 0) {
995 ret = err(opt, _("failed to open '%s': %s"),
996 path, strerror(errno));
997 goto free_buf;
999 write_in_full(fd, buf, size);
1000 close(fd);
1001 } else if (S_ISLNK(contents->mode)) {
1002 char *lnk = xmemdupz(buf, size);
1003 safe_create_leading_directories_const(path);
1004 unlink(path);
1005 if (symlink(lnk, path))
1006 ret = err(opt, _("failed to symlink '%s': %s"),
1007 path, strerror(errno));
1008 free(lnk);
1009 } else
1010 ret = err(opt,
1011 _("do not know what to do with %06o %s '%s'"),
1012 contents->mode, oid_to_hex(&contents->oid), path);
1013 free_buf:
1014 free(buf);
1016 update_index:
1017 if (!ret && update_cache) {
1018 int refresh = (!opt->priv->call_depth &&
1019 contents->mode != S_IFGITLINK);
1020 if (add_cacheinfo(opt, contents, path, 0, refresh,
1021 ADD_CACHE_OK_TO_ADD))
1022 return -1;
1024 return ret;
1027 static int update_file(struct merge_options *opt,
1028 int clean,
1029 const struct diff_filespec *contents,
1030 const char *path)
1032 return update_file_flags(opt, contents, path,
1033 opt->priv->call_depth || clean, !opt->priv->call_depth);
1036 /* Low level file merging, update and removal */
1038 struct merge_file_info {
1039 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1040 unsigned clean:1,
1041 merge:1;
1044 static int merge_3way(struct merge_options *opt,
1045 mmbuffer_t *result_buf,
1046 const struct diff_filespec *o,
1047 const struct diff_filespec *a,
1048 const struct diff_filespec *b,
1049 const char *branch1,
1050 const char *branch2,
1051 const int extra_marker_size)
1053 mmfile_t orig, src1, src2;
1054 struct ll_merge_options ll_opts = LL_MERGE_OPTIONS_INIT;
1055 char *base, *name1, *name2;
1056 enum ll_merge_result merge_status;
1058 ll_opts.renormalize = opt->renormalize;
1059 ll_opts.extra_marker_size = extra_marker_size;
1060 ll_opts.xdl_opts = opt->xdl_opts;
1061 ll_opts.conflict_style = opt->conflict_style;
1063 if (opt->priv->call_depth) {
1064 ll_opts.virtual_ancestor = 1;
1065 ll_opts.variant = 0;
1066 } else {
1067 switch (opt->recursive_variant) {
1068 case MERGE_VARIANT_OURS:
1069 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1070 break;
1071 case MERGE_VARIANT_THEIRS:
1072 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1073 break;
1074 default:
1075 ll_opts.variant = 0;
1076 break;
1080 assert(a->path && b->path && o->path && opt->ancestor);
1081 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1082 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1083 name1 = mkpathdup("%s:%s", branch1, a->path);
1084 name2 = mkpathdup("%s:%s", branch2, b->path);
1085 } else {
1086 base = mkpathdup("%s", opt->ancestor);
1087 name1 = mkpathdup("%s", branch1);
1088 name2 = mkpathdup("%s", branch2);
1091 read_mmblob(&orig, &o->oid);
1092 read_mmblob(&src1, &a->oid);
1093 read_mmblob(&src2, &b->oid);
1096 * FIXME: Using a->path for normalization rules in ll_merge could be
1097 * wrong if we renamed from a->path to b->path. We should use the
1098 * target path for where the file will be written.
1100 merge_status = ll_merge(result_buf, a->path, &orig, base,
1101 &src1, name1, &src2, name2,
1102 opt->repo->index, &ll_opts);
1103 if (merge_status == LL_MERGE_BINARY_CONFLICT)
1104 warning("Cannot merge binary files: %s (%s vs. %s)",
1105 a->path, name1, name2);
1107 free(base);
1108 free(name1);
1109 free(name2);
1110 free(orig.ptr);
1111 free(src1.ptr);
1112 free(src2.ptr);
1113 return merge_status;
1116 static int find_first_merges(struct repository *repo,
1117 struct object_array *result, const char *path,
1118 struct commit *a, struct commit *b)
1120 int i, j;
1121 struct object_array merges = OBJECT_ARRAY_INIT;
1122 struct commit *commit;
1123 int contains_another;
1125 char merged_revision[GIT_MAX_HEXSZ + 2];
1126 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1127 "--all", merged_revision, NULL };
1128 struct rev_info revs;
1129 struct setup_revision_opt rev_opts;
1131 memset(result, 0, sizeof(struct object_array));
1132 memset(&rev_opts, 0, sizeof(rev_opts));
1134 /* get all revisions that merge commit a */
1135 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1136 oid_to_hex(&a->object.oid));
1137 repo_init_revisions(repo, &revs, NULL);
1138 /* FIXME: can't handle linked worktrees in submodules yet */
1139 revs.single_worktree = path != NULL;
1140 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1142 /* save all revisions from the above list that contain b */
1143 if (prepare_revision_walk(&revs))
1144 die("revision walk setup failed");
1145 while ((commit = get_revision(&revs)) != NULL) {
1146 struct object *o = &(commit->object);
1147 int ret = repo_in_merge_bases(repo, b, commit);
1148 if (ret < 0) {
1149 object_array_clear(&merges);
1150 release_revisions(&revs);
1151 return ret;
1153 if (ret)
1154 add_object_array(o, NULL, &merges);
1156 reset_revision_walk();
1158 /* Now we've got all merges that contain a and b. Prune all
1159 * merges that contain another found merge and save them in
1160 * result.
1162 for (i = 0; i < merges.nr; i++) {
1163 struct commit *m1 = (struct commit *) merges.objects[i].item;
1165 contains_another = 0;
1166 for (j = 0; j < merges.nr; j++) {
1167 struct commit *m2 = (struct commit *) merges.objects[j].item;
1168 if (i != j) {
1169 int ret = repo_in_merge_bases(repo, m2, m1);
1170 if (ret < 0) {
1171 object_array_clear(&merges);
1172 release_revisions(&revs);
1173 return ret;
1175 if (ret > 0) {
1176 contains_another = 1;
1177 break;
1182 if (!contains_another)
1183 add_object_array(merges.objects[i].item, NULL, result);
1186 object_array_clear(&merges);
1187 release_revisions(&revs);
1188 return result->nr;
1191 static void print_commit(struct repository *repo, struct commit *commit)
1193 struct strbuf sb = STRBUF_INIT;
1194 struct pretty_print_context ctx = {0};
1195 ctx.date_mode.type = DATE_NORMAL;
1196 /* FIXME: Merge this with output_commit_title() */
1197 assert(!merge_remote_util(commit));
1198 repo_format_commit_message(repo, commit, " %h: %m %s", &sb, &ctx);
1199 fprintf(stderr, "%s\n", sb.buf);
1200 strbuf_release(&sb);
1203 static int is_valid(const struct diff_filespec *dfs)
1205 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1208 static int merge_submodule(struct merge_options *opt,
1209 struct object_id *result, const char *path,
1210 const struct object_id *base, const struct object_id *a,
1211 const struct object_id *b)
1213 struct repository subrepo;
1214 int ret = 0, ret2;
1215 struct commit *commit_base, *commit_a, *commit_b;
1216 int parent_count;
1217 struct object_array merges;
1219 int i;
1220 int search = !opt->priv->call_depth;
1222 /* store a in result in case we fail */
1223 /* FIXME: This is the WRONG resolution for the recursive case when
1224 * we need to be careful to avoid accidentally matching either side.
1225 * Should probably use o instead there, much like we do for merging
1226 * binaries.
1228 oidcpy(result, a);
1230 /* we can not handle deletion conflicts */
1231 if (is_null_oid(base))
1232 return 0;
1233 if (is_null_oid(a))
1234 return 0;
1235 if (is_null_oid(b))
1236 return 0;
1238 if (repo_submodule_init(&subrepo, opt->repo, path, null_oid())) {
1239 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1240 return 0;
1243 if (!(commit_base = lookup_commit_reference(&subrepo, base)) ||
1244 !(commit_a = lookup_commit_reference(&subrepo, a)) ||
1245 !(commit_b = lookup_commit_reference(&subrepo, b))) {
1246 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1247 goto cleanup;
1250 /* check whether both changes are forward */
1251 ret2 = repo_in_merge_bases(&subrepo, commit_base, commit_a);
1252 if (ret2 < 0) {
1253 output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1254 ret = -1;
1255 goto cleanup;
1257 if (ret2 > 0)
1258 ret2 = repo_in_merge_bases(&subrepo, commit_base, commit_b);
1259 if (ret2 < 0) {
1260 output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1261 ret = -1;
1262 goto cleanup;
1264 if (!ret2) {
1265 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1266 goto cleanup;
1269 /* Case #1: a is contained in b or vice versa */
1270 ret2 = repo_in_merge_bases(&subrepo, commit_a, commit_b);
1271 if (ret2 < 0) {
1272 output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1273 ret = -1;
1274 goto cleanup;
1276 if (ret2) {
1277 oidcpy(result, b);
1278 if (show(opt, 3)) {
1279 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1280 repo_output_commit_title(opt, &subrepo, commit_b);
1281 } else if (show(opt, 2))
1282 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1283 else
1284 ; /* no output */
1286 ret = 1;
1287 goto cleanup;
1289 ret2 = repo_in_merge_bases(&subrepo, commit_b, commit_a);
1290 if (ret2 < 0) {
1291 output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1292 ret = -1;
1293 goto cleanup;
1295 if (ret2) {
1296 oidcpy(result, a);
1297 if (show(opt, 3)) {
1298 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1299 repo_output_commit_title(opt, &subrepo, commit_a);
1300 } else if (show(opt, 2))
1301 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1302 else
1303 ; /* no output */
1305 ret = 1;
1306 goto cleanup;
1310 * Case #2: There are one or more merges that contain a and b in
1311 * the submodule. If there is only one, then present it as a
1312 * suggestion to the user, but leave it marked unmerged so the
1313 * user needs to confirm the resolution.
1316 /* Skip the search if makes no sense to the calling context. */
1317 if (!search)
1318 goto cleanup;
1320 /* find commit which merges them */
1321 parent_count = find_first_merges(&subrepo, &merges, path,
1322 commit_a, commit_b);
1323 switch (parent_count) {
1324 case -1:
1325 output(opt, 1,_("Failed to merge submodule %s (repository corrupt)"), path);
1326 ret = -1;
1327 break;
1328 case 0:
1329 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1330 break;
1332 case 1:
1333 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1334 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1335 print_commit(&subrepo, (struct commit *) merges.objects[0].item);
1336 output(opt, 2, _(
1337 "If this is correct simply add it to the index "
1338 "for example\n"
1339 "by using:\n\n"
1340 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1341 "which will accept this suggestion.\n"),
1342 oid_to_hex(&merges.objects[0].item->oid), path);
1343 break;
1345 default:
1346 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1347 for (i = 0; i < merges.nr; i++)
1348 print_commit(&subrepo, (struct commit *) merges.objects[i].item);
1351 object_array_clear(&merges);
1352 cleanup:
1353 repo_clear(&subrepo);
1354 return ret;
1357 static int merge_mode_and_contents(struct merge_options *opt,
1358 const struct diff_filespec *o,
1359 const struct diff_filespec *a,
1360 const struct diff_filespec *b,
1361 const char *filename,
1362 const char *branch1,
1363 const char *branch2,
1364 const int extra_marker_size,
1365 struct merge_file_info *result)
1367 if (opt->branch1 != branch1) {
1369 * It's weird getting a reverse merge with HEAD on the bottom
1370 * side of the conflict markers and the other branch on the
1371 * top. Fix that.
1373 return merge_mode_and_contents(opt, o, b, a,
1374 filename,
1375 branch2, branch1,
1376 extra_marker_size, result);
1379 result->merge = 0;
1380 result->clean = 1;
1382 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1383 result->clean = 0;
1385 * FIXME: This is a bad resolution for recursive case; for
1386 * the recursive case we want something that is unlikely to
1387 * accidentally match either side. Also, while it makes
1388 * sense to prefer regular files over symlinks, it doesn't
1389 * make sense to prefer regular files over submodules.
1391 if (S_ISREG(a->mode)) {
1392 result->blob.mode = a->mode;
1393 oidcpy(&result->blob.oid, &a->oid);
1394 } else {
1395 result->blob.mode = b->mode;
1396 oidcpy(&result->blob.oid, &b->oid);
1398 } else {
1399 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1400 result->merge = 1;
1403 * Merge modes
1405 if (a->mode == b->mode || a->mode == o->mode)
1406 result->blob.mode = b->mode;
1407 else {
1408 result->blob.mode = a->mode;
1409 if (b->mode != o->mode) {
1410 result->clean = 0;
1411 result->merge = 1;
1415 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1416 oidcpy(&result->blob.oid, &b->oid);
1417 else if (oideq(&b->oid, &o->oid))
1418 oidcpy(&result->blob.oid, &a->oid);
1419 else if (S_ISREG(a->mode)) {
1420 mmbuffer_t result_buf;
1421 int ret = 0, merge_status;
1423 merge_status = merge_3way(opt, &result_buf, o, a, b,
1424 branch1, branch2,
1425 extra_marker_size);
1427 if ((merge_status < 0) || !result_buf.ptr)
1428 ret = err(opt, _("failed to execute internal merge"));
1430 if (!ret &&
1431 write_object_file(result_buf.ptr, result_buf.size,
1432 OBJ_BLOB, &result->blob.oid))
1433 ret = err(opt, _("unable to add %s to database"),
1434 a->path);
1436 free(result_buf.ptr);
1437 if (ret)
1438 return ret;
1439 /* FIXME: bug, what if modes didn't match? */
1440 result->clean = (merge_status == 0);
1441 } else if (S_ISGITLINK(a->mode)) {
1442 int clean = merge_submodule(opt, &result->blob.oid,
1443 o->path,
1444 &o->oid,
1445 &a->oid,
1446 &b->oid);
1447 if (clean < 0)
1448 return -1;
1449 result->clean = clean;
1450 } else if (S_ISLNK(a->mode)) {
1451 switch (opt->recursive_variant) {
1452 case MERGE_VARIANT_NORMAL:
1453 oidcpy(&result->blob.oid, &a->oid);
1454 if (!oideq(&a->oid, &b->oid))
1455 result->clean = 0;
1456 break;
1457 case MERGE_VARIANT_OURS:
1458 oidcpy(&result->blob.oid, &a->oid);
1459 break;
1460 case MERGE_VARIANT_THEIRS:
1461 oidcpy(&result->blob.oid, &b->oid);
1462 break;
1464 } else
1465 BUG("unsupported object type in the tree");
1468 if (result->merge)
1469 output(opt, 2, _("Auto-merging %s"), filename);
1471 return 0;
1474 static int handle_rename_via_dir(struct merge_options *opt,
1475 struct rename_conflict_info *ci)
1478 * Handle file adds that need to be renamed due to directory rename
1479 * detection. This differs from handle_rename_normal, because
1480 * there is no content merge to do; just move the file into the
1481 * desired final location.
1483 const struct rename *ren = ci->ren1;
1484 const struct diff_filespec *dest = ren->pair->two;
1485 char *file_path = dest->path;
1486 int mark_conflicted = (opt->detect_directory_renames ==
1487 MERGE_DIRECTORY_RENAMES_CONFLICT);
1488 assert(ren->dir_rename_original_dest);
1490 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1491 mark_conflicted = 1;
1492 file_path = unique_path(opt, dest->path, ren->branch);
1493 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1494 "writing to %s instead."),
1495 dest->path, file_path);
1498 if (mark_conflicted) {
1500 * Write the file in worktree at file_path. In the index,
1501 * only record the file at dest->path in the appropriate
1502 * higher stage.
1504 if (update_file(opt, 0, dest, file_path))
1505 return -1;
1506 if (file_path != dest->path)
1507 free(file_path);
1508 if (update_stages(opt, dest->path, NULL,
1509 ren->branch == opt->branch1 ? dest : NULL,
1510 ren->branch == opt->branch1 ? NULL : dest))
1511 return -1;
1512 return 0; /* not clean, but conflicted */
1513 } else {
1514 /* Update dest->path both in index and in worktree */
1515 if (update_file(opt, 1, dest, dest->path))
1516 return -1;
1517 return 1; /* clean */
1521 static int handle_change_delete(struct merge_options *opt,
1522 const char *path, const char *old_path,
1523 const struct diff_filespec *o,
1524 const struct diff_filespec *changed,
1525 const char *change_branch,
1526 const char *delete_branch,
1527 const char *change, const char *change_past)
1529 char *alt_path = NULL;
1530 const char *update_path = path;
1531 int ret = 0;
1533 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1534 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1535 update_path = alt_path = unique_path(opt, path, change_branch);
1538 if (opt->priv->call_depth) {
1540 * We cannot arbitrarily accept either a_sha or b_sha as
1541 * correct; since there is no true "middle point" between
1542 * them, simply reuse the base version for virtual merge base.
1544 ret = remove_file_from_index(opt->repo->index, path);
1545 if (!ret)
1546 ret = update_file(opt, 0, o, update_path);
1547 } else {
1549 * Despite the four nearly duplicate messages and argument
1550 * lists below and the ugliness of the nested if-statements,
1551 * having complete messages makes the job easier for
1552 * translators.
1554 * The slight variance among the cases is due to the fact
1555 * that:
1556 * 1) directory/file conflicts (in effect if
1557 * !alt_path) could cause us to need to write the
1558 * file to a different path.
1559 * 2) renames (in effect if !old_path) could mean that
1560 * there are two names for the path that the user
1561 * may know the file by.
1563 if (!alt_path) {
1564 if (!old_path) {
1565 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1566 "and %s in %s. Version %s of %s left in tree."),
1567 change, path, delete_branch, change_past,
1568 change_branch, change_branch, path);
1569 } else {
1570 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1571 "and %s to %s in %s. Version %s of %s left in tree."),
1572 change, old_path, delete_branch, change_past, path,
1573 change_branch, change_branch, path);
1575 } else {
1576 if (!old_path) {
1577 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1578 "and %s in %s. Version %s of %s left in tree at %s."),
1579 change, path, delete_branch, change_past,
1580 change_branch, change_branch, path, alt_path);
1581 } else {
1582 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1583 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1584 change, old_path, delete_branch, change_past, path,
1585 change_branch, change_branch, path, alt_path);
1589 * No need to call update_file() on path when change_branch ==
1590 * opt->branch1 && !alt_path, since that would needlessly touch
1591 * path. We could call update_file_flags() with update_cache=0
1592 * and update_wd=0, but that's a no-op.
1594 if (change_branch != opt->branch1 || alt_path)
1595 ret = update_file(opt, 0, changed, update_path);
1597 free(alt_path);
1599 return ret;
1602 static int handle_rename_delete(struct merge_options *opt,
1603 struct rename_conflict_info *ci)
1605 const struct rename *ren = ci->ren1;
1606 const struct diff_filespec *orig = ren->pair->one;
1607 const struct diff_filespec *dest = ren->pair->two;
1608 const char *rename_branch = ren->branch;
1609 const char *delete_branch = (opt->branch1 == ren->branch ?
1610 opt->branch2 : opt->branch1);
1612 if (handle_change_delete(opt,
1613 opt->priv->call_depth ? orig->path : dest->path,
1614 opt->priv->call_depth ? NULL : orig->path,
1615 orig, dest,
1616 rename_branch, delete_branch,
1617 _("rename"), _("renamed")))
1618 return -1;
1620 if (opt->priv->call_depth)
1621 return remove_file_from_index(opt->repo->index, dest->path);
1622 else
1623 return update_stages(opt, dest->path, NULL,
1624 rename_branch == opt->branch1 ? dest : NULL,
1625 rename_branch == opt->branch1 ? NULL : dest);
1628 static int handle_file_collision(struct merge_options *opt,
1629 const char *collide_path,
1630 const char *prev_path1,
1631 const char *prev_path2,
1632 const char *branch1, const char *branch2,
1633 struct diff_filespec *a,
1634 struct diff_filespec *b)
1636 struct merge_file_info mfi;
1637 struct diff_filespec null;
1638 char *alt_path = NULL;
1639 const char *update_path = collide_path;
1642 * It's easiest to get the correct things into stage 2 and 3, and
1643 * to make sure that the content merge puts HEAD before the other
1644 * branch if we just ensure that branch1 == opt->branch1. So, simply
1645 * flip arguments around if we don't have that.
1647 if (branch1 != opt->branch1) {
1648 return handle_file_collision(opt, collide_path,
1649 prev_path2, prev_path1,
1650 branch2, branch1,
1651 b, a);
1654 /* Remove rename sources if rename/add or rename/rename(2to1) */
1655 if (prev_path1)
1656 remove_file(opt, 1, prev_path1,
1657 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1658 if (prev_path2)
1659 remove_file(opt, 1, prev_path2,
1660 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1663 * Remove the collision path, if it wouldn't cause dirty contents
1664 * or an untracked file to get lost. We'll either overwrite with
1665 * merged contents, or just write out to differently named files.
1667 if (was_dirty(opt, collide_path)) {
1668 output(opt, 1, _("Refusing to lose dirty file at %s"),
1669 collide_path);
1670 update_path = alt_path = unique_path(opt, collide_path, "merged");
1671 } else if (would_lose_untracked(opt, collide_path)) {
1673 * Only way we get here is if both renames were from
1674 * a directory rename AND user had an untracked file
1675 * at the location where both files end up after the
1676 * two directory renames. See testcase 10d of t6043.
1678 output(opt, 1, _("Refusing to lose untracked file at "
1679 "%s, even though it's in the way."),
1680 collide_path);
1681 update_path = alt_path = unique_path(opt, collide_path, "merged");
1682 } else {
1684 * FIXME: It's possible that the two files are identical
1685 * and that the current working copy happens to match, in
1686 * which case we are unnecessarily touching the working
1687 * tree file. It's not a likely enough scenario that I
1688 * want to code up the checks for it and a better fix is
1689 * available if we restructure how unpack_trees() and
1690 * merge-recursive interoperate anyway, so punting for
1691 * now...
1693 remove_file(opt, 0, collide_path, 0);
1696 /* Store things in diff_filespecs for functions that need it */
1697 null.path = (char *)collide_path;
1698 oidcpy(&null.oid, null_oid());
1699 null.mode = 0;
1701 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1702 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1703 return -1;
1704 mfi.clean &= !alt_path;
1705 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1706 return -1;
1707 if (!mfi.clean && !opt->priv->call_depth &&
1708 update_stages(opt, collide_path, NULL, a, b))
1709 return -1;
1710 free(alt_path);
1712 * FIXME: If both a & b both started with conflicts (only possible
1713 * if they came from a rename/rename(2to1)), but had IDENTICAL
1714 * contents including those conflicts, then in the next line we claim
1715 * it was clean. If someone cares about this case, we should have the
1716 * caller notify us if we started with conflicts.
1718 return mfi.clean;
1721 static int handle_rename_add(struct merge_options *opt,
1722 struct rename_conflict_info *ci)
1724 /* a was renamed to c, and a separate c was added. */
1725 struct diff_filespec *a = ci->ren1->pair->one;
1726 struct diff_filespec *c = ci->ren1->pair->two;
1727 char *path = c->path;
1728 char *prev_path_desc;
1729 struct merge_file_info mfi;
1731 const char *rename_branch = ci->ren1->branch;
1732 const char *add_branch = (opt->branch1 == rename_branch ?
1733 opt->branch2 : opt->branch1);
1734 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1736 output(opt, 1, _("CONFLICT (rename/add): "
1737 "Rename %s->%s in %s. Added %s in %s"),
1738 a->path, c->path, rename_branch,
1739 c->path, add_branch);
1741 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1742 ci->ren1->src_entry->stages[other_stage].path = a->path;
1743 if (merge_mode_and_contents(opt, a, c,
1744 &ci->ren1->src_entry->stages[other_stage],
1745 prev_path_desc,
1746 opt->branch1, opt->branch2,
1747 1 + opt->priv->call_depth * 2, &mfi))
1748 return -1;
1749 free(prev_path_desc);
1751 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1752 return handle_file_collision(opt,
1753 c->path, a->path, NULL,
1754 rename_branch, add_branch,
1755 &mfi.blob,
1756 &ci->ren1->dst_entry->stages[other_stage]);
1759 static char *find_path_for_conflict(struct merge_options *opt,
1760 const char *path,
1761 const char *branch1,
1762 const char *branch2)
1764 char *new_path = NULL;
1765 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1766 new_path = unique_path(opt, path, branch1);
1767 output(opt, 1, _("%s is a directory in %s adding "
1768 "as %s instead"),
1769 path, branch2, new_path);
1770 } else if (would_lose_untracked(opt, path)) {
1771 new_path = unique_path(opt, path, branch1);
1772 output(opt, 1, _("Refusing to lose untracked file"
1773 " at %s; adding as %s instead"),
1774 path, new_path);
1777 return new_path;
1781 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1783 static inline int flip_stage(int stage)
1785 return (2 + 3) - stage;
1788 static int handle_rename_rename_1to2(struct merge_options *opt,
1789 struct rename_conflict_info *ci)
1791 /* One file was renamed in both branches, but to different names. */
1792 struct merge_file_info mfi;
1793 struct diff_filespec *add;
1794 struct diff_filespec *o = ci->ren1->pair->one;
1795 struct diff_filespec *a = ci->ren1->pair->two;
1796 struct diff_filespec *b = ci->ren2->pair->two;
1797 char *path_desc;
1799 output(opt, 1, _("CONFLICT (rename/rename): "
1800 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1801 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1802 o->path, a->path, ci->ren1->branch,
1803 o->path, b->path, ci->ren2->branch,
1804 opt->priv->call_depth ? _(" (left unresolved)") : "");
1806 path_desc = xstrfmt("%s and %s, both renamed from %s",
1807 a->path, b->path, o->path);
1808 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1809 ci->ren1->branch, ci->ren2->branch,
1810 opt->priv->call_depth * 2, &mfi))
1811 return -1;
1812 free(path_desc);
1814 if (opt->priv->call_depth)
1815 remove_file_from_index(opt->repo->index, o->path);
1818 * For each destination path, we need to see if there is a
1819 * rename/add collision. If not, we can write the file out
1820 * to the specified location.
1822 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1823 if (is_valid(add)) {
1824 add->path = mfi.blob.path = a->path;
1825 if (handle_file_collision(opt, a->path,
1826 NULL, NULL,
1827 ci->ren1->branch,
1828 ci->ren2->branch,
1829 &mfi.blob, add) < 0)
1830 return -1;
1831 } else {
1832 char *new_path = find_path_for_conflict(opt, a->path,
1833 ci->ren1->branch,
1834 ci->ren2->branch);
1835 if (update_file(opt, 0, &mfi.blob,
1836 new_path ? new_path : a->path))
1837 return -1;
1838 free(new_path);
1839 if (!opt->priv->call_depth &&
1840 update_stages(opt, a->path, NULL, a, NULL))
1841 return -1;
1844 if (!mfi.clean && mfi.blob.mode == a->mode &&
1845 oideq(&mfi.blob.oid, &a->oid)) {
1847 * Getting here means we were attempting to merge a binary
1848 * blob. Since we can't merge binaries, the merge algorithm
1849 * just takes one side. But we don't want to copy the
1850 * contents of one side to both paths; we'd rather use the
1851 * original content at the given path for each path.
1853 oidcpy(&mfi.blob.oid, &b->oid);
1854 mfi.blob.mode = b->mode;
1856 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1857 if (is_valid(add)) {
1858 add->path = mfi.blob.path = b->path;
1859 if (handle_file_collision(opt, b->path,
1860 NULL, NULL,
1861 ci->ren1->branch,
1862 ci->ren2->branch,
1863 add, &mfi.blob) < 0)
1864 return -1;
1865 } else {
1866 char *new_path = find_path_for_conflict(opt, b->path,
1867 ci->ren2->branch,
1868 ci->ren1->branch);
1869 if (update_file(opt, 0, &mfi.blob,
1870 new_path ? new_path : b->path))
1871 return -1;
1872 free(new_path);
1873 if (!opt->priv->call_depth &&
1874 update_stages(opt, b->path, NULL, NULL, b))
1875 return -1;
1878 return 0;
1881 static int handle_rename_rename_2to1(struct merge_options *opt,
1882 struct rename_conflict_info *ci)
1884 /* Two files, a & b, were renamed to the same thing, c. */
1885 struct diff_filespec *a = ci->ren1->pair->one;
1886 struct diff_filespec *b = ci->ren2->pair->one;
1887 struct diff_filespec *c1 = ci->ren1->pair->two;
1888 struct diff_filespec *c2 = ci->ren2->pair->two;
1889 char *path = c1->path; /* == c2->path */
1890 char *path_side_1_desc;
1891 char *path_side_2_desc;
1892 struct merge_file_info mfi_c1;
1893 struct merge_file_info mfi_c2;
1894 int ostage1, ostage2;
1896 output(opt, 1, _("CONFLICT (rename/rename): "
1897 "Rename %s->%s in %s. "
1898 "Rename %s->%s in %s"),
1899 a->path, c1->path, ci->ren1->branch,
1900 b->path, c2->path, ci->ren2->branch);
1902 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1903 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1904 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1905 ostage2 = flip_stage(ostage1);
1906 ci->ren1->src_entry->stages[ostage1].path = a->path;
1907 ci->ren2->src_entry->stages[ostage2].path = b->path;
1908 if (merge_mode_and_contents(opt, a, c1,
1909 &ci->ren1->src_entry->stages[ostage1],
1910 path_side_1_desc,
1911 opt->branch1, opt->branch2,
1912 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1913 merge_mode_and_contents(opt, b,
1914 &ci->ren2->src_entry->stages[ostage2],
1915 c2, path_side_2_desc,
1916 opt->branch1, opt->branch2,
1917 1 + opt->priv->call_depth * 2, &mfi_c2))
1918 return -1;
1919 free(path_side_1_desc);
1920 free(path_side_2_desc);
1921 mfi_c1.blob.path = path;
1922 mfi_c2.blob.path = path;
1924 return handle_file_collision(opt, path, a->path, b->path,
1925 ci->ren1->branch, ci->ren2->branch,
1926 &mfi_c1.blob, &mfi_c2.blob);
1930 * Get the diff_filepairs changed between o_tree and tree.
1932 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1933 struct tree *o_tree,
1934 struct tree *tree)
1936 struct diff_queue_struct *ret;
1937 struct diff_options opts;
1939 repo_diff_setup(opt->repo, &opts);
1940 opts.flags.recursive = 1;
1941 opts.flags.rename_empty = 0;
1942 opts.detect_rename = merge_detect_rename(opt);
1944 * We do not have logic to handle the detection of copies. In
1945 * fact, it may not even make sense to add such logic: would we
1946 * really want a change to a base file to be propagated through
1947 * multiple other files by a merge?
1949 if (opts.detect_rename > DIFF_DETECT_RENAME)
1950 opts.detect_rename = DIFF_DETECT_RENAME;
1951 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 7000;
1952 opts.rename_score = opt->rename_score;
1953 opts.show_rename_progress = opt->show_rename_progress;
1954 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1955 diff_setup_done(&opts);
1956 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1957 diffcore_std(&opts);
1958 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1959 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1961 ret = xmalloc(sizeof(*ret));
1962 *ret = diff_queued_diff;
1964 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1965 diff_queued_diff.nr = 0;
1966 diff_queued_diff.queue = NULL;
1967 diff_flush(&opts);
1968 return ret;
1971 static int tree_has_path(struct repository *r, struct tree *tree,
1972 const char *path)
1974 struct object_id hashy;
1975 unsigned short mode_o;
1977 return !get_tree_entry(r,
1978 &tree->object.oid, path,
1979 &hashy, &mode_o);
1983 * Return a new string that replaces the beginning portion (which matches
1984 * entry->dir), with entry->new_dir. In perl-speak:
1985 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1986 * NOTE:
1987 * Caller must ensure that old_path starts with entry->dir + '/'.
1989 static char *apply_dir_rename(struct dir_rename_entry *entry,
1990 const char *old_path)
1992 struct strbuf new_path = STRBUF_INIT;
1993 int oldlen, newlen;
1995 if (entry->non_unique_new_dir)
1996 return NULL;
1998 oldlen = strlen(entry->dir);
1999 if (entry->new_dir.len == 0)
2001 * If someone renamed/merged a subdirectory into the root
2002 * directory (e.g. 'some/subdir' -> ''), then we want to
2003 * avoid returning
2004 * '' + '/filename'
2005 * as the rename; we need to make old_path + oldlen advance
2006 * past the '/' character.
2008 oldlen++;
2009 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
2010 strbuf_grow(&new_path, newlen);
2011 strbuf_addbuf(&new_path, &entry->new_dir);
2012 strbuf_addstr(&new_path, &old_path[oldlen]);
2014 return strbuf_detach(&new_path, NULL);
2017 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
2018 char **old_dir, char **new_dir)
2020 char *end_of_old, *end_of_new;
2022 /* Default return values: NULL, meaning no rename */
2023 *old_dir = NULL;
2024 *new_dir = NULL;
2027 * For
2028 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
2029 * the "e/foo.c" part is the same, we just want to know that
2030 * "a/b/c/d" was renamed to "a/b/some/thing/else"
2031 * so, for this example, this function returns "a/b/c/d" in
2032 * *old_dir and "a/b/some/thing/else" in *new_dir.
2036 * If the basename of the file changed, we don't care. We want
2037 * to know which portion of the directory, if any, changed.
2039 end_of_old = strrchr(old_path, '/');
2040 end_of_new = strrchr(new_path, '/');
2043 * If end_of_old is NULL, old_path wasn't in a directory, so there
2044 * could not be a directory rename (our rule elsewhere that a
2045 * directory which still exists is not considered to have been
2046 * renamed means the root directory can never be renamed -- because
2047 * the root directory always exists).
2049 if (!end_of_old)
2050 return; /* Note: *old_dir and *new_dir are still NULL */
2053 * If new_path contains no directory (end_of_new is NULL), then we
2054 * have a rename of old_path's directory to the root directory.
2056 if (!end_of_new) {
2057 *old_dir = xstrndup(old_path, end_of_old - old_path);
2058 *new_dir = xstrdup("");
2059 return;
2062 /* Find the first non-matching character traversing backwards */
2063 while (*--end_of_new == *--end_of_old &&
2064 end_of_old != old_path &&
2065 end_of_new != new_path)
2066 ; /* Do nothing; all in the while loop */
2069 * If both got back to the beginning of their strings, then the
2070 * directory didn't change at all, only the basename did.
2072 if (end_of_old == old_path && end_of_new == new_path &&
2073 *end_of_old == *end_of_new)
2074 return; /* Note: *old_dir and *new_dir are still NULL */
2077 * If end_of_new got back to the beginning of its string, and
2078 * end_of_old got back to the beginning of some subdirectory, then
2079 * we have a rename/merge of a subdirectory into the root, which
2080 * needs slightly special handling.
2082 * Note: There is no need to consider the opposite case, with a
2083 * rename/merge of the root directory into some subdirectory
2084 * because as noted above the root directory always exists so it
2085 * cannot be considered to be renamed.
2087 if (end_of_new == new_path &&
2088 end_of_old != old_path && end_of_old[-1] == '/') {
2089 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2090 *new_dir = xstrdup("");
2091 return;
2095 * We've found the first non-matching character in the directory
2096 * paths. That means the current characters we were looking at
2097 * were part of the first non-matching subdir name going back from
2098 * the end of the strings. Get the whole name by advancing both
2099 * end_of_old and end_of_new to the NEXT '/' character. That will
2100 * represent the entire directory rename.
2102 * The reason for the increment is cases like
2103 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2104 * After dropping the basename and going back to the first
2105 * non-matching character, we're now comparing:
2106 * a/b/s and a/b/
2107 * and we want to be comparing:
2108 * a/b/star/ and a/b/tar/
2109 * but without the pre-increment, the one on the right would stay
2110 * a/b/.
2112 end_of_old = strchr(++end_of_old, '/');
2113 end_of_new = strchr(++end_of_new, '/');
2115 /* Copy the old and new directories into *old_dir and *new_dir. */
2116 *old_dir = xstrndup(old_path, end_of_old - old_path);
2117 *new_dir = xstrndup(new_path, end_of_new - new_path);
2120 static void remove_hashmap_entries(struct hashmap *dir_renames,
2121 struct string_list *items_to_remove)
2123 int i;
2124 struct dir_rename_entry *entry;
2126 for (i = 0; i < items_to_remove->nr; i++) {
2127 entry = items_to_remove->items[i].util;
2128 hashmap_remove(dir_renames, &entry->ent, NULL);
2130 string_list_clear(items_to_remove, 0);
2134 * See if there is a directory rename for path, and if there are any file
2135 * level conflicts for the renamed location. If there is a rename and
2136 * there are no conflicts, return the new name. Otherwise, return NULL.
2138 static char *handle_path_level_conflicts(struct merge_options *opt,
2139 const char *path,
2140 struct dir_rename_entry *entry,
2141 struct hashmap *collisions,
2142 struct tree *tree)
2144 char *new_path = NULL;
2145 struct collision_entry *collision_ent;
2146 int clean = 1;
2147 struct strbuf collision_paths = STRBUF_INIT;
2150 * entry has the mapping of old directory name to new directory name
2151 * that we want to apply to path.
2153 new_path = apply_dir_rename(entry, path);
2155 if (!new_path) {
2156 /* This should only happen when entry->non_unique_new_dir set */
2157 if (!entry->non_unique_new_dir)
2158 BUG("entry->non_unique_new_dir not set and !new_path");
2159 output(opt, 1, _("CONFLICT (directory rename split): "
2160 "Unclear where to place %s because directory "
2161 "%s was renamed to multiple other directories, "
2162 "with no destination getting a majority of the "
2163 "files."),
2164 path, entry->dir);
2165 clean = 0;
2166 return NULL;
2170 * The caller needs to have ensured that it has pre-populated
2171 * collisions with all paths that map to new_path. Do a quick check
2172 * to ensure that's the case.
2174 collision_ent = collision_find_entry(collisions, new_path);
2175 if (!collision_ent)
2176 BUG("collision_ent is NULL");
2179 * Check for one-sided add/add/.../add conflicts, i.e.
2180 * where implicit renames from the other side doing
2181 * directory rename(s) can affect this side of history
2182 * to put multiple paths into the same location. Warn
2183 * and bail on directory renames for such paths.
2185 if (collision_ent->reported_already) {
2186 clean = 0;
2187 } else if (tree_has_path(opt->repo, tree, new_path)) {
2188 collision_ent->reported_already = 1;
2189 strbuf_add_separated_string_list(&collision_paths, ", ",
2190 &collision_ent->source_files);
2191 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2192 "file/dir at %s in the way of implicit "
2193 "directory rename(s) putting the following "
2194 "path(s) there: %s."),
2195 new_path, collision_paths.buf);
2196 clean = 0;
2197 } else if (collision_ent->source_files.nr > 1) {
2198 collision_ent->reported_already = 1;
2199 strbuf_add_separated_string_list(&collision_paths, ", ",
2200 &collision_ent->source_files);
2201 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2202 "more than one path to %s; implicit directory "
2203 "renames tried to put these paths there: %s"),
2204 new_path, collision_paths.buf);
2205 clean = 0;
2208 /* Free memory we no longer need */
2209 strbuf_release(&collision_paths);
2210 if (!clean && new_path) {
2211 free(new_path);
2212 return NULL;
2215 return new_path;
2219 * There are a couple things we want to do at the directory level:
2220 * 1. Check for both sides renaming to the same thing, in order to avoid
2221 * implicit renaming of files that should be left in place. (See
2222 * testcase 6b in t6043 for details.)
2223 * 2. Prune directory renames if there are still files left in the
2224 * original directory. These represent a partial directory rename,
2225 * i.e. a rename where only some of the files within the directory
2226 * were renamed elsewhere. (Technically, this could be done earlier
2227 * in get_directory_renames(), except that would prevent us from
2228 * doing the previous check and thus failing testcase 6b.)
2229 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2230 * In the future, we could potentially record this info as well and
2231 * omit reporting rename/rename(1to2) conflicts for each path within
2232 * the affected directories, thus cleaning up the merge output.
2233 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2234 * directory level, because merging directories is fine. If it
2235 * causes conflicts for files within those merged directories, then
2236 * that should be detected at the individual path level.
2238 static void handle_directory_level_conflicts(struct merge_options *opt,
2239 struct hashmap *dir_re_head,
2240 struct tree *head,
2241 struct hashmap *dir_re_merge,
2242 struct tree *merge)
2244 struct hashmap_iter iter;
2245 struct dir_rename_entry *head_ent;
2246 struct dir_rename_entry *merge_ent;
2248 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2249 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2251 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2252 ent /* member name */) {
2253 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2254 if (merge_ent &&
2255 !head_ent->non_unique_new_dir &&
2256 !merge_ent->non_unique_new_dir &&
2257 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2258 /* 1. Renamed identically; remove it from both sides */
2259 string_list_append(&remove_from_head,
2260 head_ent->dir)->util = head_ent;
2261 strbuf_release(&head_ent->new_dir);
2262 string_list_append(&remove_from_merge,
2263 merge_ent->dir)->util = merge_ent;
2264 strbuf_release(&merge_ent->new_dir);
2265 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2266 /* 2. This wasn't a directory rename after all */
2267 string_list_append(&remove_from_head,
2268 head_ent->dir)->util = head_ent;
2269 strbuf_release(&head_ent->new_dir);
2273 remove_hashmap_entries(dir_re_head, &remove_from_head);
2274 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2276 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2277 ent /* member name */) {
2278 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2279 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2280 /* 2. This wasn't a directory rename after all */
2281 string_list_append(&remove_from_merge,
2282 merge_ent->dir)->util = merge_ent;
2283 } else if (head_ent &&
2284 !head_ent->non_unique_new_dir &&
2285 !merge_ent->non_unique_new_dir) {
2286 /* 3. rename/rename(1to2) */
2288 * We can assume it's not rename/rename(1to1) because
2289 * that was case (1), already checked above. So we
2290 * know that head_ent->new_dir and merge_ent->new_dir
2291 * are different strings.
2293 output(opt, 1, _("CONFLICT (rename/rename): "
2294 "Rename directory %s->%s in %s. "
2295 "Rename directory %s->%s in %s"),
2296 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2297 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2298 string_list_append(&remove_from_head,
2299 head_ent->dir)->util = head_ent;
2300 strbuf_release(&head_ent->new_dir);
2301 string_list_append(&remove_from_merge,
2302 merge_ent->dir)->util = merge_ent;
2303 strbuf_release(&merge_ent->new_dir);
2307 remove_hashmap_entries(dir_re_head, &remove_from_head);
2308 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2311 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2313 struct hashmap *dir_renames;
2314 struct hashmap_iter iter;
2315 struct dir_rename_entry *entry;
2316 int i;
2319 * Typically, we think of a directory rename as all files from a
2320 * certain directory being moved to a target directory. However,
2321 * what if someone first moved two files from the original
2322 * directory in one commit, and then renamed the directory
2323 * somewhere else in a later commit? At merge time, we just know
2324 * that files from the original directory went to two different
2325 * places, and that the bulk of them ended up in the same place.
2326 * We want each directory rename to represent where the bulk of the
2327 * files from that directory end up; this function exists to find
2328 * where the bulk of the files went.
2330 * The first loop below simply iterates through the list of file
2331 * renames, finding out how often each directory rename pair
2332 * possibility occurs.
2334 dir_renames = xmalloc(sizeof(*dir_renames));
2335 dir_rename_init(dir_renames);
2336 for (i = 0; i < pairs->nr; ++i) {
2337 struct string_list_item *item;
2338 int *count;
2339 struct diff_filepair *pair = pairs->queue[i];
2340 char *old_dir, *new_dir;
2342 /* File not part of directory rename if it wasn't renamed */
2343 if (pair->status != 'R')
2344 continue;
2346 get_renamed_dir_portion(pair->one->path, pair->two->path,
2347 &old_dir, &new_dir);
2348 if (!old_dir)
2349 /* Directory didn't change at all; ignore this one. */
2350 continue;
2352 entry = dir_rename_find_entry(dir_renames, old_dir);
2353 if (!entry) {
2354 entry = xmalloc(sizeof(*entry));
2355 dir_rename_entry_init(entry, old_dir);
2356 hashmap_put(dir_renames, &entry->ent);
2357 } else {
2358 free(old_dir);
2360 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2361 if (!item) {
2362 item = string_list_insert(&entry->possible_new_dirs,
2363 new_dir);
2364 item->util = xcalloc(1, sizeof(int));
2365 } else {
2366 free(new_dir);
2368 count = item->util;
2369 *count += 1;
2373 * For each directory with files moved out of it, we find out which
2374 * target directory received the most files so we can declare it to
2375 * be the "winning" target location for the directory rename. This
2376 * winner gets recorded in new_dir. If there is no winner
2377 * (multiple target directories received the same number of files),
2378 * we set non_unique_new_dir. Once we've determined the winner (or
2379 * that there is no winner), we no longer need possible_new_dirs.
2381 hashmap_for_each_entry(dir_renames, &iter, entry,
2382 ent /* member name */) {
2383 int max = 0;
2384 int bad_max = 0;
2385 char *best = NULL;
2387 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2388 int *count = entry->possible_new_dirs.items[i].util;
2390 if (*count == max)
2391 bad_max = max;
2392 else if (*count > max) {
2393 max = *count;
2394 best = entry->possible_new_dirs.items[i].string;
2397 if (bad_max == max)
2398 entry->non_unique_new_dir = 1;
2399 else {
2400 assert(entry->new_dir.len == 0);
2401 strbuf_addstr(&entry->new_dir, best);
2404 * The relevant directory sub-portion of the original full
2405 * filepaths were xstrndup'ed before inserting into
2406 * possible_new_dirs, and instead of manually iterating the
2407 * list and free'ing each, just lie and tell
2408 * possible_new_dirs that it did the strdup'ing so that it
2409 * will free them for us.
2411 entry->possible_new_dirs.strdup_strings = 1;
2412 string_list_clear(&entry->possible_new_dirs, 1);
2415 return dir_renames;
2418 static struct dir_rename_entry *check_dir_renamed(const char *path,
2419 struct hashmap *dir_renames)
2421 char *temp = xstrdup(path);
2422 char *end;
2423 struct dir_rename_entry *entry = NULL;
2425 while ((end = strrchr(temp, '/'))) {
2426 *end = '\0';
2427 entry = dir_rename_find_entry(dir_renames, temp);
2428 if (entry)
2429 break;
2431 free(temp);
2432 return entry;
2435 static void compute_collisions(struct hashmap *collisions,
2436 struct hashmap *dir_renames,
2437 struct diff_queue_struct *pairs)
2439 int i;
2442 * Multiple files can be mapped to the same path due to directory
2443 * renames done by the other side of history. Since that other
2444 * side of history could have merged multiple directories into one,
2445 * if our side of history added the same file basename to each of
2446 * those directories, then all N of them would get implicitly
2447 * renamed by the directory rename detection into the same path,
2448 * and we'd get an add/add/.../add conflict, and all those adds
2449 * from *this* side of history. This is not representable in the
2450 * index, and users aren't going to easily be able to make sense of
2451 * it. So we need to provide a good warning about what's
2452 * happening, and fall back to no-directory-rename detection
2453 * behavior for those paths.
2455 * See testcases 9e and all of section 5 from t6043 for examples.
2457 collision_init(collisions);
2459 for (i = 0; i < pairs->nr; ++i) {
2460 struct dir_rename_entry *dir_rename_ent;
2461 struct collision_entry *collision_ent;
2462 char *new_path;
2463 struct diff_filepair *pair = pairs->queue[i];
2465 if (pair->status != 'A' && pair->status != 'R')
2466 continue;
2467 dir_rename_ent = check_dir_renamed(pair->two->path,
2468 dir_renames);
2469 if (!dir_rename_ent)
2470 continue;
2472 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2473 if (!new_path)
2475 * dir_rename_ent->non_unique_new_path is true, which
2476 * means there is no directory rename for us to use,
2477 * which means it won't cause us any additional
2478 * collisions.
2480 continue;
2481 collision_ent = collision_find_entry(collisions, new_path);
2482 if (!collision_ent) {
2483 CALLOC_ARRAY(collision_ent, 1);
2484 hashmap_entry_init(&collision_ent->ent,
2485 strhash(new_path));
2486 hashmap_put(collisions, &collision_ent->ent);
2487 collision_ent->target_file = new_path;
2488 } else {
2489 free(new_path);
2491 string_list_insert(&collision_ent->source_files,
2492 pair->two->path);
2496 static char *check_for_directory_rename(struct merge_options *opt,
2497 const char *path,
2498 struct tree *tree,
2499 struct hashmap *dir_renames,
2500 struct hashmap *dir_rename_exclusions,
2501 struct hashmap *collisions,
2502 int *clean_merge)
2504 char *new_path = NULL;
2505 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2506 struct dir_rename_entry *oentry = NULL;
2508 if (!entry)
2509 return new_path;
2512 * This next part is a little weird. We do not want to do an
2513 * implicit rename into a directory we renamed on our side, because
2514 * that will result in a spurious rename/rename(1to2) conflict. An
2515 * example:
2516 * Base commit: dumbdir/afile, otherdir/bfile
2517 * Side 1: smrtdir/afile, otherdir/bfile
2518 * Side 2: dumbdir/afile, dumbdir/bfile
2519 * Here, while working on Side 1, we could notice that otherdir was
2520 * renamed/merged to dumbdir, and change the diff_filepair for
2521 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2522 * 2 will notice the rename from dumbdir to smrtdir, and do the
2523 * transitive rename to move it from dumbdir/bfile to
2524 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2525 * smrtdir, a rename/rename(1to2) conflict. We really just want
2526 * the file to end up in smrtdir. And the way to achieve that is
2527 * to not let Side1 do the rename to dumbdir, since we know that is
2528 * the source of one of our directory renames.
2530 * That's why oentry and dir_rename_exclusions is here.
2532 * As it turns out, this also prevents N-way transient rename
2533 * confusion; See testcases 9c and 9d of t6043.
2535 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2536 if (oentry) {
2537 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2538 "to %s, because %s itself was renamed."),
2539 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2540 } else {
2541 new_path = handle_path_level_conflicts(opt, path, entry,
2542 collisions, tree);
2543 *clean_merge &= (new_path != NULL);
2546 return new_path;
2549 static void apply_directory_rename_modifications(struct merge_options *opt,
2550 struct diff_filepair *pair,
2551 char *new_path,
2552 struct rename *re,
2553 struct tree *tree,
2554 struct tree *o_tree,
2555 struct tree *a_tree,
2556 struct tree *b_tree,
2557 struct string_list *entries)
2559 struct string_list_item *item;
2560 int stage = (tree == a_tree ? 2 : 3);
2561 int update_wd;
2564 * In all cases where we can do directory rename detection,
2565 * unpack_trees() will have read pair->two->path into the
2566 * index and the working copy. We need to remove it so that
2567 * we can instead place it at new_path. It is guaranteed to
2568 * not be untracked (unpack_trees() would have errored out
2569 * saying the file would have been overwritten), but it might
2570 * be dirty, though.
2572 update_wd = !was_dirty(opt, pair->two->path);
2573 if (!update_wd)
2574 output(opt, 1, _("Refusing to lose dirty file at %s"),
2575 pair->two->path);
2576 remove_file(opt, 1, pair->two->path, !update_wd);
2578 /* Find or create a new re->dst_entry */
2579 item = string_list_lookup(entries, new_path);
2580 if (item) {
2582 * Since we're renaming on this side of history, and it's
2583 * due to a directory rename on the other side of history
2584 * (which we only allow when the directory in question no
2585 * longer exists on the other side of history), the
2586 * original entry for re->dst_entry is no longer
2587 * necessary...
2589 re->dst_entry->processed = 1;
2592 * ...because we'll be using this new one.
2594 re->dst_entry = item->util;
2595 } else {
2597 * re->dst_entry is for the before-dir-rename path, and we
2598 * need it to hold information for the after-dir-rename
2599 * path. Before creating a new entry, we need to mark the
2600 * old one as unnecessary (...unless it is shared by
2601 * src_entry, i.e. this didn't use to be a rename, in which
2602 * case we can just allow the normal processing to happen
2603 * for it).
2605 if (pair->status == 'R')
2606 re->dst_entry->processed = 1;
2608 re->dst_entry = insert_stage_data(opt->repo, new_path,
2609 o_tree, a_tree, b_tree,
2610 entries);
2611 item = string_list_insert(entries, new_path);
2612 item->util = re->dst_entry;
2616 * Update the stage_data with the information about the path we are
2617 * moving into place. That slot will be empty and available for us
2618 * to write to because of the collision checks in
2619 * handle_path_level_conflicts(). In other words,
2620 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2621 * open for us to write to.
2623 * It may be tempting to actually update the index at this point as
2624 * well, using update_stages_for_stage_data(), but as per the big
2625 * "NOTE" in update_stages(), doing so will modify the current
2626 * in-memory index which will break calls to would_lose_untracked()
2627 * that we need to make. Instead, we need to just make sure that
2628 * the various handle_rename_*() functions update the index
2629 * explicitly rather than relying on unpack_trees() to have done it.
2631 get_tree_entry(opt->repo,
2632 &tree->object.oid,
2633 pair->two->path,
2634 &re->dst_entry->stages[stage].oid,
2635 &re->dst_entry->stages[stage].mode);
2638 * Record the original change status (or 'type' of change). If it
2639 * was originally an add ('A'), this lets us differentiate later
2640 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2641 * otherwise look the same). If it was originally a rename ('R'),
2642 * this lets us remember and report accurately about the transitive
2643 * renaming that occurred via the directory rename detection. Also,
2644 * record the original destination name.
2646 re->dir_rename_original_type = pair->status;
2647 re->dir_rename_original_dest = pair->two->path;
2650 * We don't actually look at pair->status again, but it seems
2651 * pedagogically correct to adjust it.
2653 pair->status = 'R';
2656 * Finally, record the new location.
2658 pair->two->path = new_path;
2662 * Get information of all renames which occurred in 'pairs', making use of
2663 * any implicit directory renames inferred from the other side of history.
2664 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2665 * to be able to associate the correct cache entries with the rename
2666 * information; tree is always equal to either a_tree or b_tree.
2668 static struct string_list *get_renames(struct merge_options *opt,
2669 const char *branch,
2670 struct diff_queue_struct *pairs,
2671 struct hashmap *dir_renames,
2672 struct hashmap *dir_rename_exclusions,
2673 struct tree *tree,
2674 struct tree *o_tree,
2675 struct tree *a_tree,
2676 struct tree *b_tree,
2677 struct string_list *entries,
2678 int *clean_merge)
2680 int i;
2681 struct hashmap collisions;
2682 struct hashmap_iter iter;
2683 struct collision_entry *e;
2684 struct string_list *renames;
2686 compute_collisions(&collisions, dir_renames, pairs);
2687 CALLOC_ARRAY(renames, 1);
2689 for (i = 0; i < pairs->nr; ++i) {
2690 struct string_list_item *item;
2691 struct rename *re;
2692 struct diff_filepair *pair = pairs->queue[i];
2693 char *new_path; /* non-NULL only with directory renames */
2695 if (pair->status != 'A' && pair->status != 'R') {
2696 diff_free_filepair(pair);
2697 continue;
2699 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2700 dir_renames,
2701 dir_rename_exclusions,
2702 &collisions,
2703 clean_merge);
2704 if (pair->status != 'R' && !new_path) {
2705 diff_free_filepair(pair);
2706 continue;
2709 re = xmalloc(sizeof(*re));
2710 re->processed = 0;
2711 re->pair = pair;
2712 re->branch = branch;
2713 re->dir_rename_original_type = '\0';
2714 re->dir_rename_original_dest = NULL;
2715 item = string_list_lookup(entries, re->pair->one->path);
2716 if (!item)
2717 re->src_entry = insert_stage_data(opt->repo,
2718 re->pair->one->path,
2719 o_tree, a_tree, b_tree, entries);
2720 else
2721 re->src_entry = item->util;
2723 item = string_list_lookup(entries, re->pair->two->path);
2724 if (!item)
2725 re->dst_entry = insert_stage_data(opt->repo,
2726 re->pair->two->path,
2727 o_tree, a_tree, b_tree, entries);
2728 else
2729 re->dst_entry = item->util;
2730 item = string_list_insert(renames, pair->one->path);
2731 item->util = re;
2732 if (new_path)
2733 apply_directory_rename_modifications(opt, pair, new_path,
2734 re, tree, o_tree,
2735 a_tree, b_tree,
2736 entries);
2739 hashmap_for_each_entry(&collisions, &iter, e,
2740 ent /* member name */) {
2741 free(e->target_file);
2742 string_list_clear(&e->source_files, 0);
2744 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2745 return renames;
2748 static int process_renames(struct merge_options *opt,
2749 struct string_list *a_renames,
2750 struct string_list *b_renames)
2752 int clean_merge = 1, i, j;
2753 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2754 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2755 const struct rename *sre;
2758 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2759 * string_list one-by-one, but O(n log n) to build it unsorted and
2760 * then sort it. Note that as we build the list, we do not need to
2761 * check if the existing destination path is already in the list,
2762 * because the structure of diffcore_rename guarantees we won't
2763 * have duplicates.
2765 for (i = 0; i < a_renames->nr; i++) {
2766 sre = a_renames->items[i].util;
2767 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2768 = (void *)sre;
2770 for (i = 0; i < b_renames->nr; i++) {
2771 sre = b_renames->items[i].util;
2772 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2773 = (void *)sre;
2776 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2777 struct string_list *renames1, *renames2Dst;
2778 struct rename *ren1 = NULL, *ren2 = NULL;
2779 const char *ren1_src, *ren1_dst;
2780 struct string_list_item *lookup;
2782 if (i >= a_renames->nr) {
2783 ren2 = b_renames->items[j++].util;
2784 } else if (j >= b_renames->nr) {
2785 ren1 = a_renames->items[i++].util;
2786 } else {
2787 int compare = strcmp(a_renames->items[i].string,
2788 b_renames->items[j].string);
2789 if (compare <= 0)
2790 ren1 = a_renames->items[i++].util;
2791 if (compare >= 0)
2792 ren2 = b_renames->items[j++].util;
2795 /* TODO: refactor, so that 1/2 are not needed */
2796 if (ren1) {
2797 renames1 = a_renames;
2798 renames2Dst = &b_by_dst;
2799 } else {
2800 renames1 = b_renames;
2801 renames2Dst = &a_by_dst;
2802 SWAP(ren2, ren1);
2805 if (ren1->processed)
2806 continue;
2807 ren1->processed = 1;
2808 ren1->dst_entry->processed = 1;
2809 /* BUG: We should only mark src_entry as processed if we
2810 * are not dealing with a rename + add-source case.
2812 ren1->src_entry->processed = 1;
2814 ren1_src = ren1->pair->one->path;
2815 ren1_dst = ren1->pair->two->path;
2817 if (ren2) {
2818 /* One file renamed on both sides */
2819 const char *ren2_src = ren2->pair->one->path;
2820 const char *ren2_dst = ren2->pair->two->path;
2821 enum rename_type rename_type;
2822 if (strcmp(ren1_src, ren2_src) != 0)
2823 BUG("ren1_src != ren2_src");
2824 ren2->dst_entry->processed = 1;
2825 ren2->processed = 1;
2826 if (strcmp(ren1_dst, ren2_dst) != 0) {
2827 rename_type = RENAME_ONE_FILE_TO_TWO;
2828 clean_merge = 0;
2829 } else {
2830 rename_type = RENAME_ONE_FILE_TO_ONE;
2831 /* BUG: We should only remove ren1_src in
2832 * the base stage (think of rename +
2833 * add-source cases).
2835 remove_file(opt, 1, ren1_src, 1);
2836 update_entry(ren1->dst_entry,
2837 ren1->pair->one,
2838 ren1->pair->two,
2839 ren2->pair->two);
2841 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2842 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2843 /* Two different files renamed to the same thing */
2844 char *ren2_dst;
2845 ren2 = lookup->util;
2846 ren2_dst = ren2->pair->two->path;
2847 if (strcmp(ren1_dst, ren2_dst) != 0)
2848 BUG("ren1_dst != ren2_dst");
2850 clean_merge = 0;
2851 ren2->processed = 1;
2853 * BUG: We should only mark src_entry as processed
2854 * if we are not dealing with a rename + add-source
2855 * case.
2857 ren2->src_entry->processed = 1;
2859 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2860 opt, ren1, ren2);
2861 } else {
2862 /* Renamed in 1, maybe changed in 2 */
2863 /* we only use sha1 and mode of these */
2864 struct diff_filespec src_other, dst_other;
2865 int try_merge;
2868 * unpack_trees loads entries from common-commit
2869 * into stage 1, from head-commit into stage 2, and
2870 * from merge-commit into stage 3. We keep track
2871 * of which side corresponds to the rename.
2873 int renamed_stage = a_renames == renames1 ? 2 : 3;
2874 int other_stage = a_renames == renames1 ? 3 : 2;
2877 * Directory renames have a funny corner case...
2879 int renamed_to_self = !strcmp(ren1_src, ren1_dst);
2881 /* BUG: We should only remove ren1_src in the base
2882 * stage and in other_stage (think of rename +
2883 * add-source case).
2885 if (!renamed_to_self)
2886 remove_file(opt, 1, ren1_src,
2887 renamed_stage == 2 ||
2888 !was_tracked(opt, ren1_src));
2890 oidcpy(&src_other.oid,
2891 &ren1->src_entry->stages[other_stage].oid);
2892 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2893 oidcpy(&dst_other.oid,
2894 &ren1->dst_entry->stages[other_stage].oid);
2895 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2896 try_merge = 0;
2898 if (oideq(&src_other.oid, null_oid()) &&
2899 ren1->dir_rename_original_type == 'A') {
2900 setup_rename_conflict_info(RENAME_VIA_DIR,
2901 opt, ren1, NULL);
2902 } else if (renamed_to_self) {
2903 setup_rename_conflict_info(RENAME_NORMAL,
2904 opt, ren1, NULL);
2905 } else if (oideq(&src_other.oid, null_oid())) {
2906 setup_rename_conflict_info(RENAME_DELETE,
2907 opt, ren1, NULL);
2908 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2909 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2911 * Added file on the other side identical to
2912 * the file being renamed: clean merge.
2913 * Also, there is no need to overwrite the
2914 * file already in the working copy, so call
2915 * update_file_flags() instead of
2916 * update_file().
2918 if (update_file_flags(opt,
2919 ren1->pair->two,
2920 ren1_dst,
2921 1, /* update_cache */
2922 0 /* update_wd */))
2923 clean_merge = -1;
2924 } else if (!oideq(&dst_other.oid, null_oid())) {
2926 * Probably not a clean merge, but it's
2927 * premature to set clean_merge to 0 here,
2928 * because if the rename merges cleanly and
2929 * the merge exactly matches the newly added
2930 * file, then the merge will be clean.
2932 setup_rename_conflict_info(RENAME_ADD,
2933 opt, ren1, NULL);
2934 } else
2935 try_merge = 1;
2937 if (clean_merge < 0)
2938 goto cleanup_and_return;
2939 if (try_merge) {
2940 struct diff_filespec *o, *a, *b;
2941 src_other.path = (char *)ren1_src;
2943 o = ren1->pair->one;
2944 if (a_renames == renames1) {
2945 a = ren1->pair->two;
2946 b = &src_other;
2947 } else {
2948 b = ren1->pair->two;
2949 a = &src_other;
2951 update_entry(ren1->dst_entry, o, a, b);
2952 setup_rename_conflict_info(RENAME_NORMAL,
2953 opt, ren1, NULL);
2957 cleanup_and_return:
2958 string_list_clear(&a_by_dst, 0);
2959 string_list_clear(&b_by_dst, 0);
2961 return clean_merge;
2964 struct rename_info {
2965 struct string_list *head_renames;
2966 struct string_list *merge_renames;
2969 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2970 struct hashmap *dir_renames)
2972 struct hashmap_iter iter;
2973 struct dir_rename_entry *e;
2975 hashmap_for_each_entry(dir_renames, &iter, e,
2976 ent /* member name */) {
2977 free(e->dir);
2978 strbuf_release(&e->new_dir);
2979 /* possible_new_dirs already cleared in get_directory_renames */
2981 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2982 free(dir_renames);
2984 free(pairs->queue);
2985 free(pairs);
2988 static int detect_and_process_renames(struct merge_options *opt,
2989 struct tree *common,
2990 struct tree *head,
2991 struct tree *merge,
2992 struct string_list *entries,
2993 struct rename_info *ri)
2995 struct diff_queue_struct *head_pairs, *merge_pairs;
2996 struct hashmap *dir_re_head, *dir_re_merge;
2997 int clean = 1;
2999 ri->head_renames = NULL;
3000 ri->merge_renames = NULL;
3002 if (!merge_detect_rename(opt))
3003 return 1;
3005 head_pairs = get_diffpairs(opt, common, head);
3006 merge_pairs = get_diffpairs(opt, common, merge);
3008 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
3009 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3010 !opt->priv->call_depth)) {
3011 dir_re_head = get_directory_renames(head_pairs);
3012 dir_re_merge = get_directory_renames(merge_pairs);
3014 handle_directory_level_conflicts(opt,
3015 dir_re_head, head,
3016 dir_re_merge, merge);
3017 } else {
3018 dir_re_head = xmalloc(sizeof(*dir_re_head));
3019 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
3020 dir_rename_init(dir_re_head);
3021 dir_rename_init(dir_re_merge);
3024 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
3025 dir_re_merge, dir_re_head, head,
3026 common, head, merge, entries,
3027 &clean);
3028 if (clean < 0)
3029 goto cleanup;
3030 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
3031 dir_re_head, dir_re_merge, merge,
3032 common, head, merge, entries,
3033 &clean);
3034 if (clean < 0)
3035 goto cleanup;
3036 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
3038 cleanup:
3040 * Some cleanup is deferred until cleanup_renames() because the
3041 * data structures are still needed and referenced in
3042 * process_entry(). But there are a few things we can free now.
3044 initial_cleanup_rename(head_pairs, dir_re_head);
3045 initial_cleanup_rename(merge_pairs, dir_re_merge);
3047 return clean;
3050 static void final_cleanup_rename(struct string_list *rename)
3052 const struct rename *re;
3053 int i;
3055 if (!rename)
3056 return;
3058 for (i = 0; i < rename->nr; i++) {
3059 re = rename->items[i].util;
3060 diff_free_filepair(re->pair);
3062 string_list_clear(rename, 1);
3063 free(rename);
3066 static void final_cleanup_renames(struct rename_info *re_info)
3068 final_cleanup_rename(re_info->head_renames);
3069 final_cleanup_rename(re_info->merge_renames);
3072 static int read_oid_strbuf(struct merge_options *opt,
3073 const struct object_id *oid,
3074 struct strbuf *dst)
3076 void *buf;
3077 enum object_type type;
3078 unsigned long size;
3079 buf = repo_read_object_file(the_repository, oid, &type, &size);
3080 if (!buf)
3081 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3082 if (type != OBJ_BLOB) {
3083 free(buf);
3084 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3086 strbuf_attach(dst, buf, size, size + 1);
3087 return 0;
3090 static int blob_unchanged(struct merge_options *opt,
3091 const struct diff_filespec *o,
3092 const struct diff_filespec *a,
3093 int renormalize, const char *path)
3095 struct strbuf obuf = STRBUF_INIT;
3096 struct strbuf abuf = STRBUF_INIT;
3097 int ret = 0; /* assume changed for safety */
3098 struct index_state *idx = opt->repo->index;
3100 if (a->mode != o->mode)
3101 return 0;
3102 if (oideq(&o->oid, &a->oid))
3103 return 1;
3104 if (!renormalize)
3105 return 0;
3107 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3108 read_oid_strbuf(opt, &a->oid, &abuf))
3109 goto error_return;
3111 * Note: binary | is used so that both renormalizations are
3112 * performed. Comparison can be skipped if both files are
3113 * unchanged since their sha1s have already been compared.
3115 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3116 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3117 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3119 error_return:
3120 strbuf_release(&obuf);
3121 strbuf_release(&abuf);
3122 return ret;
3125 static int handle_modify_delete(struct merge_options *opt,
3126 const char *path,
3127 const struct diff_filespec *o,
3128 const struct diff_filespec *a,
3129 const struct diff_filespec *b)
3131 const char *modify_branch, *delete_branch;
3132 const struct diff_filespec *changed;
3134 if (is_valid(a)) {
3135 modify_branch = opt->branch1;
3136 delete_branch = opt->branch2;
3137 changed = a;
3138 } else {
3139 modify_branch = opt->branch2;
3140 delete_branch = opt->branch1;
3141 changed = b;
3144 return handle_change_delete(opt,
3145 path, NULL,
3146 o, changed,
3147 modify_branch, delete_branch,
3148 _("modify"), _("modified"));
3151 static int handle_content_merge(struct merge_file_info *mfi,
3152 struct merge_options *opt,
3153 const char *path,
3154 int is_dirty,
3155 const struct diff_filespec *o,
3156 const struct diff_filespec *a,
3157 const struct diff_filespec *b,
3158 struct rename_conflict_info *ci)
3160 const char *reason = _("content");
3161 unsigned df_conflict_remains = 0;
3163 if (!is_valid(o))
3164 reason = _("add/add");
3166 assert(o->path && a->path && b->path);
3167 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3168 S_ISGITLINK(ci->ren1->pair->two->mode)))
3169 df_conflict_remains = 1;
3171 if (merge_mode_and_contents(opt, o, a, b, path,
3172 opt->branch1, opt->branch2,
3173 opt->priv->call_depth * 2, mfi))
3174 return -1;
3177 * We can skip updating the working tree file iff:
3178 * a) The merge is clean
3179 * b) The merge matches what was in HEAD (content, mode, pathname)
3180 * c) The target path is usable (i.e. not involved in D/F conflict)
3182 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3183 !df_conflict_remains) {
3184 int pos;
3185 struct cache_entry *ce;
3187 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3188 if (add_cacheinfo(opt, &mfi->blob, path,
3189 0, (!opt->priv->call_depth && !is_dirty), 0))
3190 return -1;
3192 * However, add_cacheinfo() will delete the old cache entry
3193 * and add a new one. We need to copy over any skip_worktree
3194 * flag to avoid making the file appear as if it were
3195 * deleted by the user.
3197 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3198 ce = opt->priv->orig_index.cache[pos];
3199 if (ce_skip_worktree(ce)) {
3200 pos = index_name_pos(opt->repo->index, path, strlen(path));
3201 ce = opt->repo->index->cache[pos];
3202 ce->ce_flags |= CE_SKIP_WORKTREE;
3204 return mfi->clean;
3207 if (!mfi->clean) {
3208 if (S_ISGITLINK(mfi->blob.mode))
3209 reason = _("submodule");
3210 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3211 reason, path);
3212 if (ci && !df_conflict_remains)
3213 if (update_stages(opt, path, o, a, b))
3214 return -1;
3217 if (df_conflict_remains || is_dirty) {
3218 char *new_path;
3219 if (opt->priv->call_depth) {
3220 remove_file_from_index(opt->repo->index, path);
3221 } else {
3222 if (!mfi->clean) {
3223 if (update_stages(opt, path, o, a, b))
3224 return -1;
3225 } else {
3226 int file_from_stage2 = was_tracked(opt, path);
3228 if (update_stages(opt, path, NULL,
3229 file_from_stage2 ? &mfi->blob : NULL,
3230 file_from_stage2 ? NULL : &mfi->blob))
3231 return -1;
3235 new_path = unique_path(opt, path, ci->ren1->branch);
3236 if (is_dirty) {
3237 output(opt, 1, _("Refusing to lose dirty file at %s"),
3238 path);
3240 output(opt, 1, _("Adding as %s instead"), new_path);
3241 if (update_file(opt, 0, &mfi->blob, new_path)) {
3242 free(new_path);
3243 return -1;
3245 free(new_path);
3246 mfi->clean = 0;
3247 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3248 return -1;
3249 return !is_dirty && mfi->clean;
3252 static int handle_rename_normal(struct merge_options *opt,
3253 const char *path,
3254 const struct diff_filespec *o,
3255 const struct diff_filespec *a,
3256 const struct diff_filespec *b,
3257 struct rename_conflict_info *ci)
3259 struct rename *ren = ci->ren1;
3260 struct merge_file_info mfi;
3261 int clean;
3263 /* Merge the content and write it out */
3264 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3265 o, a, b, ci);
3267 if (clean &&
3268 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3269 ren->dir_rename_original_dest) {
3270 if (update_stages(opt, path,
3271 &mfi.blob, &mfi.blob, &mfi.blob))
3272 return -1;
3273 clean = 0; /* not clean, but conflicted */
3275 return clean;
3278 static void dir_rename_warning(const char *msg,
3279 int is_add,
3280 int clean,
3281 struct merge_options *opt,
3282 struct rename *ren)
3284 const char *other_branch;
3285 other_branch = (ren->branch == opt->branch1 ?
3286 opt->branch2 : opt->branch1);
3287 if (is_add) {
3288 output(opt, clean ? 2 : 1, msg,
3289 ren->pair->one->path, ren->branch,
3290 other_branch, ren->pair->two->path);
3291 return;
3293 output(opt, clean ? 2 : 1, msg,
3294 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3295 other_branch, ren->pair->two->path);
3297 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3298 struct rename *ren)
3300 const char *msg;
3301 int clean = 1, is_add;
3303 if (!ren)
3304 return clean;
3306 /* Return early if ren was not affected/created by a directory rename */
3307 if (!ren->dir_rename_original_dest)
3308 return clean;
3310 /* Sanity checks */
3311 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3312 assert(ren->dir_rename_original_type == 'A' ||
3313 ren->dir_rename_original_type == 'R');
3315 /* Check whether to treat directory renames as a conflict */
3316 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3318 is_add = (ren->dir_rename_original_type == 'A');
3319 if (ren->dir_rename_original_type == 'A' && clean) {
3320 msg = _("Path updated: %s added in %s inside a "
3321 "directory that was renamed in %s; moving it to %s.");
3322 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3323 msg = _("CONFLICT (file location): %s added in %s "
3324 "inside a directory that was renamed in %s, "
3325 "suggesting it should perhaps be moved to %s.");
3326 } else if (ren->dir_rename_original_type == 'R' && clean) {
3327 msg = _("Path updated: %s renamed to %s in %s, inside a "
3328 "directory that was renamed in %s; moving it to %s.");
3329 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3330 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3331 "inside a directory that was renamed in %s, "
3332 "suggesting it should perhaps be moved to %s.");
3333 } else {
3334 BUG("Impossible dir_rename_original_type/clean combination");
3336 dir_rename_warning(msg, is_add, clean, opt, ren);
3338 return clean;
3341 /* Per entry merge function */
3342 static int process_entry(struct merge_options *opt,
3343 const char *path, struct stage_data *entry)
3345 int clean_merge = 1;
3346 int normalize = opt->renormalize;
3348 struct diff_filespec *o = &entry->stages[1];
3349 struct diff_filespec *a = &entry->stages[2];
3350 struct diff_filespec *b = &entry->stages[3];
3351 int o_valid = is_valid(o);
3352 int a_valid = is_valid(a);
3353 int b_valid = is_valid(b);
3354 o->path = a->path = b->path = (char*)path;
3356 entry->processed = 1;
3357 if (entry->rename_conflict_info) {
3358 struct rename_conflict_info *ci = entry->rename_conflict_info;
3359 struct diff_filespec *temp;
3360 int path_clean;
3362 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3363 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3366 * For cases with a single rename, {o,a,b}->path have all been
3367 * set to the rename target path; we need to set two of these
3368 * back to the rename source.
3369 * For rename/rename conflicts, we'll manually fix paths below.
3371 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3372 o->path = temp->path = ci->ren1->pair->one->path;
3373 if (ci->ren2) {
3374 assert(opt->branch1 == ci->ren1->branch);
3377 switch (ci->rename_type) {
3378 case RENAME_NORMAL:
3379 case RENAME_ONE_FILE_TO_ONE:
3380 clean_merge = handle_rename_normal(opt, path, o, a, b,
3381 ci);
3382 break;
3383 case RENAME_VIA_DIR:
3384 clean_merge = handle_rename_via_dir(opt, ci);
3385 break;
3386 case RENAME_ADD:
3388 * Probably unclean merge, but if the renamed file
3389 * merges cleanly and the result can then be
3390 * two-way merged cleanly with the added file, I
3391 * guess it's a clean merge?
3393 clean_merge = handle_rename_add(opt, ci);
3394 break;
3395 case RENAME_DELETE:
3396 clean_merge = 0;
3397 if (handle_rename_delete(opt, ci))
3398 clean_merge = -1;
3399 break;
3400 case RENAME_ONE_FILE_TO_TWO:
3402 * Manually fix up paths; note:
3403 * ren[12]->pair->one->path are equal.
3405 o->path = ci->ren1->pair->one->path;
3406 a->path = ci->ren1->pair->two->path;
3407 b->path = ci->ren2->pair->two->path;
3409 clean_merge = 0;
3410 if (handle_rename_rename_1to2(opt, ci))
3411 clean_merge = -1;
3412 break;
3413 case RENAME_TWO_FILES_TO_ONE:
3415 * Manually fix up paths; note,
3416 * ren[12]->pair->two->path are actually equal.
3418 o->path = NULL;
3419 a->path = ci->ren1->pair->two->path;
3420 b->path = ci->ren2->pair->two->path;
3423 * Probably unclean merge, but if the two renamed
3424 * files merge cleanly and the two resulting files
3425 * can then be two-way merged cleanly, I guess it's
3426 * a clean merge?
3428 clean_merge = handle_rename_rename_2to1(opt, ci);
3429 break;
3430 default:
3431 entry->processed = 0;
3432 break;
3434 if (path_clean < clean_merge)
3435 clean_merge = path_clean;
3436 } else if (o_valid && (!a_valid || !b_valid)) {
3437 /* Case A: Deleted in one */
3438 if ((!a_valid && !b_valid) ||
3439 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3440 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3441 /* Deleted in both or deleted in one and
3442 * unchanged in the other */
3443 if (a_valid)
3444 output(opt, 2, _("Removing %s"), path);
3445 /* do not touch working file if it did not exist */
3446 remove_file(opt, 1, path, !a_valid);
3447 } else {
3448 /* Modify/delete; deleted side may have put a directory in the way */
3449 clean_merge = 0;
3450 if (handle_modify_delete(opt, path, o, a, b))
3451 clean_merge = -1;
3453 } else if ((!o_valid && a_valid && !b_valid) ||
3454 (!o_valid && !a_valid && b_valid)) {
3455 /* Case B: Added in one. */
3456 /* [nothing|directory] -> ([nothing|directory], file) */
3458 const char *add_branch;
3459 const char *other_branch;
3460 const char *conf;
3461 const struct diff_filespec *contents;
3463 if (a_valid) {
3464 add_branch = opt->branch1;
3465 other_branch = opt->branch2;
3466 contents = a;
3467 conf = _("file/directory");
3468 } else {
3469 add_branch = opt->branch2;
3470 other_branch = opt->branch1;
3471 contents = b;
3472 conf = _("directory/file");
3474 if (dir_in_way(opt->repo->index, path,
3475 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3476 0)) {
3477 char *new_path = unique_path(opt, path, add_branch);
3478 clean_merge = 0;
3479 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3480 "Adding %s as %s"),
3481 conf, path, other_branch, path, new_path);
3482 if (update_file(opt, 0, contents, new_path))
3483 clean_merge = -1;
3484 else if (opt->priv->call_depth)
3485 remove_file_from_index(opt->repo->index, path);
3486 free(new_path);
3487 } else {
3488 output(opt, 2, _("Adding %s"), path);
3489 /* do not overwrite file if already present */
3490 if (update_file_flags(opt, contents, path, 1, !a_valid))
3491 clean_merge = -1;
3493 } else if (a_valid && b_valid) {
3494 if (!o_valid) {
3495 /* Case C: Added in both (check for same permissions) */
3496 output(opt, 1,
3497 _("CONFLICT (add/add): Merge conflict in %s"),
3498 path);
3499 clean_merge = handle_file_collision(opt,
3500 path, NULL, NULL,
3501 opt->branch1,
3502 opt->branch2,
3503 a, b);
3504 } else {
3505 /* case D: Modified in both, but differently. */
3506 struct merge_file_info mfi;
3507 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3508 clean_merge = handle_content_merge(&mfi, opt, path,
3509 is_dirty,
3510 o, a, b, NULL);
3512 } else if (!o_valid && !a_valid && !b_valid) {
3514 * this entry was deleted altogether. a_mode == 0 means
3515 * we had that path and want to actively remove it.
3517 remove_file(opt, 1, path, !a->mode);
3518 } else
3519 BUG("fatal merge failure, shouldn't happen.");
3521 return clean_merge;
3524 static int merge_trees_internal(struct merge_options *opt,
3525 struct tree *head,
3526 struct tree *merge,
3527 struct tree *merge_base,
3528 struct tree **result)
3530 struct index_state *istate = opt->repo->index;
3531 int code, clean;
3533 if (opt->subtree_shift) {
3534 merge = shift_tree_object(opt->repo, head, merge,
3535 opt->subtree_shift);
3536 merge_base = shift_tree_object(opt->repo, head, merge_base,
3537 opt->subtree_shift);
3540 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3541 output(opt, 0, _("Already up to date."));
3542 *result = head;
3543 return 1;
3546 code = unpack_trees_start(opt, merge_base, head, merge);
3548 if (code != 0) {
3549 if (show(opt, 4) || opt->priv->call_depth)
3550 err(opt, _("merging of trees %s and %s failed"),
3551 oid_to_hex(&head->object.oid),
3552 oid_to_hex(&merge->object.oid));
3553 unpack_trees_finish(opt);
3554 return -1;
3557 if (unmerged_index(istate)) {
3558 struct string_list *entries;
3559 struct rename_info re_info;
3560 int i;
3562 * Only need the hashmap while processing entries, so
3563 * initialize it here and free it when we are done running
3564 * through the entries. Keeping it in the merge_options as
3565 * opposed to decaring a local hashmap is for convenience
3566 * so that we don't have to pass it to around.
3568 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3569 NULL, 512);
3570 get_files_dirs(opt, head);
3571 get_files_dirs(opt, merge);
3573 entries = get_unmerged(opt->repo->index);
3574 clean = detect_and_process_renames(opt, merge_base, head, merge,
3575 entries, &re_info);
3576 record_df_conflict_files(opt, entries);
3577 if (clean < 0)
3578 goto cleanup;
3579 for (i = entries->nr-1; 0 <= i; i--) {
3580 const char *path = entries->items[i].string;
3581 struct stage_data *e = entries->items[i].util;
3582 if (!e->processed) {
3583 int ret = process_entry(opt, path, e);
3584 if (!ret)
3585 clean = 0;
3586 else if (ret < 0) {
3587 clean = ret;
3588 goto cleanup;
3592 for (i = 0; i < entries->nr; i++) {
3593 struct stage_data *e = entries->items[i].util;
3594 if (!e->processed)
3595 BUG("unprocessed path??? %s",
3596 entries->items[i].string);
3599 cleanup:
3600 final_cleanup_renames(&re_info);
3602 string_list_clear(entries, 1);
3603 free(entries);
3605 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3606 struct path_hashmap_entry, e);
3608 if (clean < 0) {
3609 unpack_trees_finish(opt);
3610 return clean;
3613 else
3614 clean = 1;
3616 unpack_trees_finish(opt);
3618 if (opt->priv->call_depth &&
3619 !(*result = write_in_core_index_as_tree(opt->repo)))
3620 return -1;
3622 return clean;
3626 * Merge the commits h1 and h2, returning a flag (int) indicating the
3627 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3628 * virtual commit and write its location to *result.
3630 static int merge_recursive_internal(struct merge_options *opt,
3631 struct commit *h1,
3632 struct commit *h2,
3633 struct commit_list *merge_bases,
3634 struct commit **result)
3636 struct commit_list *iter;
3637 struct commit *merged_merge_bases;
3638 struct tree *result_tree;
3639 int clean;
3640 const char *ancestor_name;
3641 struct strbuf merge_base_abbrev = STRBUF_INIT;
3643 if (show(opt, 4)) {
3644 output(opt, 4, _("Merging:"));
3645 output_commit_title(opt, h1);
3646 output_commit_title(opt, h2);
3649 if (!merge_bases) {
3650 if (repo_get_merge_bases(the_repository, h1, h2,
3651 &merge_bases) < 0)
3652 return -1;
3653 merge_bases = reverse_commit_list(merge_bases);
3656 if (show(opt, 5)) {
3657 unsigned cnt = commit_list_count(merge_bases);
3659 output(opt, 5, Q_("found %u common ancestor:",
3660 "found %u common ancestors:", cnt), cnt);
3661 for (iter = merge_bases; iter; iter = iter->next)
3662 output_commit_title(opt, iter->item);
3665 merged_merge_bases = pop_commit(&merge_bases);
3666 if (!merged_merge_bases) {
3667 /* if there is no common ancestor, use an empty tree */
3668 struct tree *tree;
3670 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3671 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3672 "ancestor");
3673 ancestor_name = "empty tree";
3674 } else if (opt->ancestor && !opt->priv->call_depth) {
3675 ancestor_name = opt->ancestor;
3676 } else if (merge_bases) {
3677 ancestor_name = "merged common ancestors";
3678 } else {
3679 strbuf_add_unique_abbrev(&merge_base_abbrev,
3680 &merged_merge_bases->object.oid,
3681 DEFAULT_ABBREV);
3682 ancestor_name = merge_base_abbrev.buf;
3685 for (iter = merge_bases; iter; iter = iter->next) {
3686 const char *saved_b1, *saved_b2;
3687 opt->priv->call_depth++;
3689 * When the merge fails, the result contains files
3690 * with conflict markers. The cleanness flag is
3691 * ignored (unless indicating an error), it was never
3692 * actually used, as result of merge_trees has always
3693 * overwritten it: the committed "conflicts" were
3694 * already resolved.
3696 discard_index(opt->repo->index);
3697 saved_b1 = opt->branch1;
3698 saved_b2 = opt->branch2;
3699 opt->branch1 = "Temporary merge branch 1";
3700 opt->branch2 = "Temporary merge branch 2";
3701 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3702 NULL, &merged_merge_bases) < 0)
3703 return -1;
3704 opt->branch1 = saved_b1;
3705 opt->branch2 = saved_b2;
3706 opt->priv->call_depth--;
3708 if (!merged_merge_bases)
3709 return err(opt, _("merge returned no commit"));
3713 * FIXME: Since merge_recursive_internal() is only ever called by
3714 * places that ensure the index is loaded first
3715 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3716 * case where the merge base was unique that means when we get here
3717 * we immediately discard the index and re-read it, which is a
3718 * complete waste of time. We should only be discarding and
3719 * re-reading if we were forced to recurse.
3721 discard_index(opt->repo->index);
3722 if (!opt->priv->call_depth)
3723 repo_read_index(opt->repo);
3725 opt->ancestor = ancestor_name;
3726 clean = merge_trees_internal(opt,
3727 repo_get_commit_tree(opt->repo, h1),
3728 repo_get_commit_tree(opt->repo, h2),
3729 repo_get_commit_tree(opt->repo,
3730 merged_merge_bases),
3731 &result_tree);
3732 strbuf_release(&merge_base_abbrev);
3733 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3734 if (clean < 0) {
3735 flush_output(opt);
3736 return clean;
3739 if (opt->priv->call_depth) {
3740 *result = make_virtual_commit(opt->repo, result_tree,
3741 "merged tree");
3742 commit_list_insert(h1, &(*result)->parents);
3743 commit_list_insert(h2, &(*result)->parents->next);
3745 return clean;
3748 static int merge_start(struct merge_options *opt, struct tree *head)
3750 struct strbuf sb = STRBUF_INIT;
3752 /* Sanity checks on opt */
3753 assert(opt->repo);
3755 assert(opt->branch1 && opt->branch2);
3757 assert(opt->detect_renames >= -1 &&
3758 opt->detect_renames <= DIFF_DETECT_COPY);
3759 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3760 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3761 assert(opt->rename_limit >= -1);
3762 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3763 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3765 assert(opt->xdl_opts >= 0);
3766 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3767 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3769 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3770 assert(opt->buffer_output <= 2);
3771 assert(opt->obuf.len == 0);
3773 assert(opt->priv == NULL);
3775 /* Not supported; option specific to merge-ort */
3776 assert(!opt->record_conflict_msgs_as_headers);
3777 assert(!opt->msg_header_prefix);
3779 /* Sanity check on repo state; index must match head */
3780 if (repo_index_has_changes(opt->repo, head, &sb)) {
3781 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3782 sb.buf);
3783 strbuf_release(&sb);
3784 return -1;
3787 CALLOC_ARRAY(opt->priv, 1);
3788 string_list_init_dup(&opt->priv->df_conflict_file_set);
3789 return 0;
3792 static void merge_finalize(struct merge_options *opt)
3794 flush_output(opt);
3795 if (!opt->priv->call_depth && opt->buffer_output < 2)
3796 strbuf_release(&opt->obuf);
3797 if (show(opt, 2))
3798 diff_warn_rename_limit("merge.renamelimit",
3799 opt->priv->needed_rename_limit, 0);
3800 FREE_AND_NULL(opt->priv);
3803 int merge_trees(struct merge_options *opt,
3804 struct tree *head,
3805 struct tree *merge,
3806 struct tree *merge_base)
3808 int clean;
3809 struct tree *ignored;
3811 assert(opt->ancestor != NULL);
3813 if (merge_start(opt, head))
3814 return -1;
3815 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3816 merge_finalize(opt);
3818 return clean;
3821 int merge_recursive(struct merge_options *opt,
3822 struct commit *h1,
3823 struct commit *h2,
3824 struct commit_list *merge_bases,
3825 struct commit **result)
3827 int clean;
3829 assert(opt->ancestor == NULL ||
3830 !strcmp(opt->ancestor, "constructed merge base"));
3832 prepare_repo_settings(opt->repo);
3833 opt->repo->settings.command_requires_full_index = 1;
3835 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3836 return -1;
3837 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3838 merge_finalize(opt);
3840 return clean;
3843 static struct commit *get_ref(struct repository *repo,
3844 const struct object_id *oid,
3845 const char *name)
3847 struct object *object;
3849 object = deref_tag(repo, parse_object(repo, oid),
3850 name, strlen(name));
3851 if (!object)
3852 return NULL;
3853 if (object->type == OBJ_TREE)
3854 return make_virtual_commit(repo, (struct tree*)object, name);
3855 if (object->type != OBJ_COMMIT)
3856 return NULL;
3857 if (repo_parse_commit(repo, (struct commit *)object))
3858 return NULL;
3859 return (struct commit *)object;
3862 int merge_recursive_generic(struct merge_options *opt,
3863 const struct object_id *head,
3864 const struct object_id *merge,
3865 int num_merge_bases,
3866 const struct object_id **merge_bases,
3867 struct commit **result)
3869 int clean;
3870 struct lock_file lock = LOCK_INIT;
3871 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3872 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3873 struct commit_list *ca = NULL;
3875 if (merge_bases) {
3876 int i;
3877 for (i = 0; i < num_merge_bases; ++i) {
3878 struct commit *base;
3879 if (!(base = get_ref(opt->repo, merge_bases[i],
3880 oid_to_hex(merge_bases[i]))))
3881 return err(opt, _("Could not parse object '%s'"),
3882 oid_to_hex(merge_bases[i]));
3883 commit_list_insert(base, &ca);
3885 if (num_merge_bases == 1)
3886 opt->ancestor = "constructed merge base";
3889 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3890 clean = merge_recursive(opt, head_commit, next_commit, ca,
3891 result);
3892 if (clean < 0) {
3893 rollback_lock_file(&lock);
3894 return clean;
3897 if (write_locked_index(opt->repo->index, &lock,
3898 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3899 return err(opt, _("Unable to write index."));
3901 return clean ? 0 : 1;
3904 static void merge_recursive_config(struct merge_options *opt)
3906 char *value = NULL;
3907 int renormalize = 0;
3908 git_config_get_int("merge.verbosity", &opt->verbosity);
3909 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3910 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3911 git_config_get_bool("merge.renormalize", &renormalize);
3912 opt->renormalize = renormalize;
3913 if (!git_config_get_string("diff.renames", &value)) {
3914 opt->detect_renames = git_config_rename("diff.renames", value);
3915 free(value);
3917 if (!git_config_get_string("merge.renames", &value)) {
3918 opt->detect_renames = git_config_rename("merge.renames", value);
3919 free(value);
3921 if (!git_config_get_string("merge.directoryrenames", &value)) {
3922 int boolval = git_parse_maybe_bool(value);
3923 if (0 <= boolval) {
3924 opt->detect_directory_renames = boolval ?
3925 MERGE_DIRECTORY_RENAMES_TRUE :
3926 MERGE_DIRECTORY_RENAMES_NONE;
3927 } else if (!strcasecmp(value, "conflict")) {
3928 opt->detect_directory_renames =
3929 MERGE_DIRECTORY_RENAMES_CONFLICT;
3930 } /* avoid erroring on values from future versions of git */
3931 free(value);
3933 git_config(git_xmerge_config, NULL);
3936 void init_merge_options(struct merge_options *opt,
3937 struct repository *repo)
3939 const char *merge_verbosity;
3940 memset(opt, 0, sizeof(struct merge_options));
3942 opt->repo = repo;
3944 opt->detect_renames = -1;
3945 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3946 opt->rename_limit = -1;
3948 opt->verbosity = 2;
3949 opt->buffer_output = 1;
3950 strbuf_init(&opt->obuf, 0);
3952 opt->renormalize = 0;
3954 opt->conflict_style = -1;
3956 merge_recursive_config(opt);
3957 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3958 if (merge_verbosity)
3959 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3960 if (opt->verbosity >= 5)
3961 opt->buffer_output = 0;
3965 * For now, members of merge_options do not need deep copying, but
3966 * it may change in the future, in which case we would need to update
3967 * this, and also make a matching change to clear_merge_options() to
3968 * release the resources held by a copied instance.
3970 void copy_merge_options(struct merge_options *dst, struct merge_options *src)
3972 *dst = *src;
3975 void clear_merge_options(struct merge_options *opt UNUSED)
3977 ; /* no-op as our copy is shallow right now */
3980 int parse_merge_opt(struct merge_options *opt, const char *s)
3982 const char *arg;
3984 if (!s || !*s)
3985 return -1;
3986 if (!strcmp(s, "ours"))
3987 opt->recursive_variant = MERGE_VARIANT_OURS;
3988 else if (!strcmp(s, "theirs"))
3989 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3990 else if (!strcmp(s, "subtree"))
3991 opt->subtree_shift = "";
3992 else if (skip_prefix(s, "subtree=", &arg))
3993 opt->subtree_shift = arg;
3994 else if (!strcmp(s, "patience"))
3995 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3996 else if (!strcmp(s, "histogram"))
3997 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3998 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3999 long value = parse_algorithm_value(arg);
4000 if (value < 0)
4001 return -1;
4002 /* clear out previous settings */
4003 DIFF_XDL_CLR(opt, NEED_MINIMAL);
4004 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
4005 opt->xdl_opts |= value;
4007 else if (!strcmp(s, "ignore-space-change"))
4008 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
4009 else if (!strcmp(s, "ignore-all-space"))
4010 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
4011 else if (!strcmp(s, "ignore-space-at-eol"))
4012 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
4013 else if (!strcmp(s, "ignore-cr-at-eol"))
4014 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
4015 else if (!strcmp(s, "renormalize"))
4016 opt->renormalize = 1;
4017 else if (!strcmp(s, "no-renormalize"))
4018 opt->renormalize = 0;
4019 else if (!strcmp(s, "no-renames"))
4020 opt->detect_renames = 0;
4021 else if (!strcmp(s, "find-renames")) {
4022 opt->detect_renames = 1;
4023 opt->rename_score = 0;
4025 else if (skip_prefix(s, "find-renames=", &arg) ||
4026 skip_prefix(s, "rename-threshold=", &arg)) {
4027 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
4028 return -1;
4029 opt->detect_renames = 1;
4032 * Please update $__git_merge_strategy_options in
4033 * git-completion.bash when you add new options
4035 else
4036 return -1;
4037 return 0;