Sync with 2.33.8
[git/debian.git] / merge-recursive.c
blobd9457797dbb73bfed720ac9ca0b9bcf56575c62a
1 /*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6 #include "cache.h"
7 #include "merge-recursive.h"
9 #include "advice.h"
10 #include "alloc.h"
11 #include "attr.h"
12 #include "blob.h"
13 #include "builtin.h"
14 #include "cache-tree.h"
15 #include "commit.h"
16 #include "commit-reach.h"
17 #include "config.h"
18 #include "diff.h"
19 #include "diffcore.h"
20 #include "dir.h"
21 #include "ll-merge.h"
22 #include "lockfile.h"
23 #include "object-store.h"
24 #include "repository.h"
25 #include "revision.h"
26 #include "string-list.h"
27 #include "submodule-config.h"
28 #include "submodule.h"
29 #include "tag.h"
30 #include "tree-walk.h"
31 #include "unpack-trees.h"
32 #include "xdiff-interface.h"
34 struct merge_options_internal {
35 int call_depth;
36 int needed_rename_limit;
37 struct hashmap current_file_dir_set;
38 struct string_list df_conflict_file_set;
39 struct unpack_trees_options unpack_opts;
40 struct index_state orig_index;
43 struct path_hashmap_entry {
44 struct hashmap_entry e;
45 char path[FLEX_ARRAY];
48 static int path_hashmap_cmp(const void *cmp_data,
49 const struct hashmap_entry *eptr,
50 const struct hashmap_entry *entry_or_key,
51 const void *keydata)
53 const struct path_hashmap_entry *a, *b;
54 const char *key = keydata;
56 a = container_of(eptr, const struct path_hashmap_entry, e);
57 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
59 return fspathcmp(a->path, key ? key : b->path);
63 * For dir_rename_entry, directory names are stored as a full path from the
64 * toplevel of the repository and do not include a trailing '/'. Also:
66 * dir: original name of directory being renamed
67 * non_unique_new_dir: if true, could not determine new_dir
68 * new_dir: final name of directory being renamed
69 * possible_new_dirs: temporary used to help determine new_dir; see comments
70 * in get_directory_renames() for details
72 struct dir_rename_entry {
73 struct hashmap_entry ent;
74 char *dir;
75 unsigned non_unique_new_dir:1;
76 struct strbuf new_dir;
77 struct string_list possible_new_dirs;
80 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
81 char *dir)
83 struct dir_rename_entry key;
85 if (dir == NULL)
86 return NULL;
87 hashmap_entry_init(&key.ent, strhash(dir));
88 key.dir = dir;
89 return hashmap_get_entry(hashmap, &key, ent, NULL);
92 static int dir_rename_cmp(const void *unused_cmp_data,
93 const struct hashmap_entry *eptr,
94 const struct hashmap_entry *entry_or_key,
95 const void *unused_keydata)
97 const struct dir_rename_entry *e1, *e2;
99 e1 = container_of(eptr, const struct dir_rename_entry, ent);
100 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
102 return strcmp(e1->dir, e2->dir);
105 static void dir_rename_init(struct hashmap *map)
107 hashmap_init(map, dir_rename_cmp, NULL, 0);
110 static void dir_rename_entry_init(struct dir_rename_entry *entry,
111 char *directory)
113 hashmap_entry_init(&entry->ent, strhash(directory));
114 entry->dir = directory;
115 entry->non_unique_new_dir = 0;
116 strbuf_init(&entry->new_dir, 0);
117 string_list_init_nodup(&entry->possible_new_dirs);
120 struct collision_entry {
121 struct hashmap_entry ent;
122 char *target_file;
123 struct string_list source_files;
124 unsigned reported_already:1;
127 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
128 char *target_file)
130 struct collision_entry key;
132 hashmap_entry_init(&key.ent, strhash(target_file));
133 key.target_file = target_file;
134 return hashmap_get_entry(hashmap, &key, ent, NULL);
137 static int collision_cmp(const void *unused_cmp_data,
138 const struct hashmap_entry *eptr,
139 const struct hashmap_entry *entry_or_key,
140 const void *unused_keydata)
142 const struct collision_entry *e1, *e2;
144 e1 = container_of(eptr, const struct collision_entry, ent);
145 e2 = container_of(entry_or_key, const struct collision_entry, ent);
147 return strcmp(e1->target_file, e2->target_file);
150 static void collision_init(struct hashmap *map)
152 hashmap_init(map, collision_cmp, NULL, 0);
155 static void flush_output(struct merge_options *opt)
157 if (opt->buffer_output < 2 && opt->obuf.len) {
158 fputs(opt->obuf.buf, stdout);
159 strbuf_reset(&opt->obuf);
163 __attribute__((format (printf, 2, 3)))
164 static int err(struct merge_options *opt, const char *err, ...)
166 va_list params;
168 if (opt->buffer_output < 2)
169 flush_output(opt);
170 else {
171 strbuf_complete(&opt->obuf, '\n');
172 strbuf_addstr(&opt->obuf, "error: ");
174 va_start(params, err);
175 strbuf_vaddf(&opt->obuf, err, params);
176 va_end(params);
177 if (opt->buffer_output > 1)
178 strbuf_addch(&opt->obuf, '\n');
179 else {
180 error("%s", opt->obuf.buf);
181 strbuf_reset(&opt->obuf);
184 return -1;
187 static struct tree *shift_tree_object(struct repository *repo,
188 struct tree *one, struct tree *two,
189 const char *subtree_shift)
191 struct object_id shifted;
193 if (!*subtree_shift) {
194 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
195 } else {
196 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
197 subtree_shift);
199 if (oideq(&two->object.oid, &shifted))
200 return two;
201 return lookup_tree(repo, &shifted);
204 static inline void set_commit_tree(struct commit *c, struct tree *t)
206 c->maybe_tree = t;
209 static struct commit *make_virtual_commit(struct repository *repo,
210 struct tree *tree,
211 const char *comment)
213 struct commit *commit = alloc_commit_node(repo);
215 set_merge_remote_desc(commit, comment, (struct object *)commit);
216 set_commit_tree(commit, tree);
217 commit->object.parsed = 1;
218 return commit;
221 enum rename_type {
222 RENAME_NORMAL = 0,
223 RENAME_VIA_DIR,
224 RENAME_ADD,
225 RENAME_DELETE,
226 RENAME_ONE_FILE_TO_ONE,
227 RENAME_ONE_FILE_TO_TWO,
228 RENAME_TWO_FILES_TO_ONE
232 * Since we want to write the index eventually, we cannot reuse the index
233 * for these (temporary) data.
235 struct stage_data {
236 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
237 struct rename_conflict_info *rename_conflict_info;
238 unsigned processed:1;
241 struct rename {
242 unsigned processed:1;
243 struct diff_filepair *pair;
244 const char *branch; /* branch that the rename occurred on */
246 * If directory rename detection affected this rename, what was its
247 * original type ('A' or 'R') and it's original destination before
248 * the directory rename (otherwise, '\0' and NULL for these two vars).
250 char dir_rename_original_type;
251 char *dir_rename_original_dest;
253 * Purpose of src_entry and dst_entry:
255 * If 'before' is renamed to 'after' then src_entry will contain
256 * the versions of 'before' from the merge_base, HEAD, and MERGE in
257 * stages 1, 2, and 3; dst_entry will contain the respective
258 * versions of 'after' in corresponding locations. Thus, we have a
259 * total of six modes and oids, though some will be null. (Stage 0
260 * is ignored; we're interested in handling conflicts.)
262 * Since we don't turn on break-rewrites by default, neither
263 * src_entry nor dst_entry can have all three of their stages have
264 * non-null oids, meaning at most four of the six will be non-null.
265 * Also, since this is a rename, both src_entry and dst_entry will
266 * have at least one non-null oid, meaning at least two will be
267 * non-null. Of the six oids, a typical rename will have three be
268 * non-null. Only two implies a rename/delete, and four implies a
269 * rename/add.
271 struct stage_data *src_entry;
272 struct stage_data *dst_entry;
275 struct rename_conflict_info {
276 enum rename_type rename_type;
277 struct rename *ren1;
278 struct rename *ren2;
281 static inline void setup_rename_conflict_info(enum rename_type rename_type,
282 struct merge_options *opt,
283 struct rename *ren1,
284 struct rename *ren2)
286 struct rename_conflict_info *ci;
289 * When we have two renames involved, it's easiest to get the
290 * correct things into stage 2 and 3, and to make sure that the
291 * content merge puts HEAD before the other branch if we just
292 * ensure that branch1 == opt->branch1. So, simply flip arguments
293 * around if we don't have that.
295 if (ren2 && ren1->branch != opt->branch1) {
296 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
297 return;
300 CALLOC_ARRAY(ci, 1);
301 ci->rename_type = rename_type;
302 ci->ren1 = ren1;
303 ci->ren2 = ren2;
305 ci->ren1->dst_entry->processed = 0;
306 ci->ren1->dst_entry->rename_conflict_info = ci;
307 if (ren2) {
308 ci->ren2->dst_entry->rename_conflict_info = ci;
312 static int show(struct merge_options *opt, int v)
314 return (!opt->priv->call_depth && opt->verbosity >= v) ||
315 opt->verbosity >= 5;
318 __attribute__((format (printf, 3, 4)))
319 static void output(struct merge_options *opt, int v, const char *fmt, ...)
321 va_list ap;
323 if (!show(opt, v))
324 return;
326 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
328 va_start(ap, fmt);
329 strbuf_vaddf(&opt->obuf, fmt, ap);
330 va_end(ap);
332 strbuf_addch(&opt->obuf, '\n');
333 if (!opt->buffer_output)
334 flush_output(opt);
337 static void repo_output_commit_title(struct merge_options *opt,
338 struct repository *repo,
339 struct commit *commit)
341 struct merge_remote_desc *desc;
343 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
344 desc = merge_remote_util(commit);
345 if (desc)
346 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
347 else {
348 strbuf_repo_add_unique_abbrev(&opt->obuf, repo,
349 &commit->object.oid,
350 DEFAULT_ABBREV);
351 strbuf_addch(&opt->obuf, ' ');
352 if (repo_parse_commit(repo, commit) != 0)
353 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
354 else {
355 const char *title;
356 const char *msg = repo_get_commit_buffer(repo, commit, NULL);
357 int len = find_commit_subject(msg, &title);
358 if (len)
359 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
360 repo_unuse_commit_buffer(repo, commit, msg);
363 flush_output(opt);
366 static void output_commit_title(struct merge_options *opt, struct commit *commit)
368 repo_output_commit_title(opt, the_repository, commit);
371 static int add_cacheinfo(struct merge_options *opt,
372 const struct diff_filespec *blob,
373 const char *path, int stage, int refresh, int options)
375 struct index_state *istate = opt->repo->index;
376 struct cache_entry *ce;
377 int ret;
379 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
380 if (!ce)
381 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
383 ret = add_index_entry(istate, ce, options);
384 if (refresh) {
385 struct cache_entry *nce;
387 nce = refresh_cache_entry(istate, ce,
388 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
389 if (!nce)
390 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
391 if (nce != ce)
392 ret = add_index_entry(istate, nce, options);
394 return ret;
397 static inline int merge_detect_rename(struct merge_options *opt)
399 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
402 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
404 parse_tree(tree);
405 init_tree_desc(desc, tree->buffer, tree->size);
408 static int unpack_trees_start(struct merge_options *opt,
409 struct tree *common,
410 struct tree *head,
411 struct tree *merge)
413 int rc;
414 struct tree_desc t[3];
415 struct index_state tmp_index = { NULL };
417 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
418 if (opt->priv->call_depth)
419 opt->priv->unpack_opts.index_only = 1;
420 else {
421 opt->priv->unpack_opts.update = 1;
422 /* FIXME: should only do this if !overwrite_ignore */
423 opt->priv->unpack_opts.preserve_ignored = 0;
425 opt->priv->unpack_opts.merge = 1;
426 opt->priv->unpack_opts.head_idx = 2;
427 opt->priv->unpack_opts.fn = threeway_merge;
428 opt->priv->unpack_opts.src_index = opt->repo->index;
429 opt->priv->unpack_opts.dst_index = &tmp_index;
430 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
431 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
433 init_tree_desc_from_tree(t+0, common);
434 init_tree_desc_from_tree(t+1, head);
435 init_tree_desc_from_tree(t+2, merge);
437 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
438 cache_tree_free(&opt->repo->index->cache_tree);
441 * Update opt->repo->index to match the new results, AFTER saving a
442 * copy in opt->priv->orig_index. Update src_index to point to the
443 * saved copy. (verify_uptodate() checks src_index, and the original
444 * index is the one that had the necessary modification timestamps.)
446 opt->priv->orig_index = *opt->repo->index;
447 *opt->repo->index = tmp_index;
448 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
450 return rc;
453 static void unpack_trees_finish(struct merge_options *opt)
455 discard_index(&opt->priv->orig_index);
456 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
459 static int save_files_dirs(const struct object_id *oid,
460 struct strbuf *base, const char *path,
461 unsigned int mode, void *context)
463 struct path_hashmap_entry *entry;
464 int baselen = base->len;
465 struct merge_options *opt = context;
467 strbuf_addstr(base, path);
469 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
470 hashmap_entry_init(&entry->e, fspathhash(entry->path));
471 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
473 strbuf_setlen(base, baselen);
474 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
477 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
479 struct pathspec match_all;
480 memset(&match_all, 0, sizeof(match_all));
481 read_tree(opt->repo, tree,
482 &match_all, save_files_dirs, opt);
485 static int get_tree_entry_if_blob(struct repository *r,
486 const struct object_id *tree,
487 const char *path,
488 struct diff_filespec *dfs)
490 int ret;
492 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
493 if (S_ISDIR(dfs->mode)) {
494 oidcpy(&dfs->oid, null_oid());
495 dfs->mode = 0;
497 return ret;
501 * Returns an index_entry instance which doesn't have to correspond to
502 * a real cache entry in Git's index.
504 static struct stage_data *insert_stage_data(struct repository *r,
505 const char *path,
506 struct tree *o, struct tree *a, struct tree *b,
507 struct string_list *entries)
509 struct string_list_item *item;
510 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
511 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
512 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
513 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
514 item = string_list_insert(entries, path);
515 item->util = e;
516 return e;
520 * Create a dictionary mapping file names to stage_data objects. The
521 * dictionary contains one entry for every path with a non-zero stage entry.
523 static struct string_list *get_unmerged(struct index_state *istate)
525 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
526 int i;
528 unmerged->strdup_strings = 1;
530 /* TODO: audit for interaction with sparse-index. */
531 ensure_full_index(istate);
532 for (i = 0; i < istate->cache_nr; i++) {
533 struct string_list_item *item;
534 struct stage_data *e;
535 const struct cache_entry *ce = istate->cache[i];
536 if (!ce_stage(ce))
537 continue;
539 item = string_list_lookup(unmerged, ce->name);
540 if (!item) {
541 item = string_list_insert(unmerged, ce->name);
542 item->util = xcalloc(1, sizeof(struct stage_data));
544 e = item->util;
545 e->stages[ce_stage(ce)].mode = ce->ce_mode;
546 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
549 return unmerged;
552 static int string_list_df_name_compare(const char *one, const char *two)
554 int onelen = strlen(one);
555 int twolen = strlen(two);
557 * Here we only care that entries for D/F conflicts are
558 * adjacent, in particular with the file of the D/F conflict
559 * appearing before files below the corresponding directory.
560 * The order of the rest of the list is irrelevant for us.
562 * To achieve this, we sort with df_name_compare and provide
563 * the mode S_IFDIR so that D/F conflicts will sort correctly.
564 * We use the mode S_IFDIR for everything else for simplicity,
565 * since in other cases any changes in their order due to
566 * sorting cause no problems for us.
568 int cmp = df_name_compare(one, onelen, S_IFDIR,
569 two, twolen, S_IFDIR);
571 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
572 * that 'foo' comes before 'foo/bar'.
574 if (cmp)
575 return cmp;
576 return onelen - twolen;
579 static void record_df_conflict_files(struct merge_options *opt,
580 struct string_list *entries)
582 /* If there is a D/F conflict and the file for such a conflict
583 * currently exists in the working tree, we want to allow it to be
584 * removed to make room for the corresponding directory if needed.
585 * The files underneath the directories of such D/F conflicts will
586 * be processed before the corresponding file involved in the D/F
587 * conflict. If the D/F directory ends up being removed by the
588 * merge, then we won't have to touch the D/F file. If the D/F
589 * directory needs to be written to the working copy, then the D/F
590 * file will simply be removed (in make_room_for_path()) to make
591 * room for the necessary paths. Note that if both the directory
592 * and the file need to be present, then the D/F file will be
593 * reinstated with a new unique name at the time it is processed.
595 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
596 const char *last_file = NULL;
597 int last_len = 0;
598 int i;
601 * If we're merging merge-bases, we don't want to bother with
602 * any working directory changes.
604 if (opt->priv->call_depth)
605 return;
607 /* Ensure D/F conflicts are adjacent in the entries list. */
608 for (i = 0; i < entries->nr; i++) {
609 struct string_list_item *next = &entries->items[i];
610 string_list_append(&df_sorted_entries, next->string)->util =
611 next->util;
613 df_sorted_entries.cmp = string_list_df_name_compare;
614 string_list_sort(&df_sorted_entries);
616 string_list_clear(&opt->priv->df_conflict_file_set, 1);
617 for (i = 0; i < df_sorted_entries.nr; i++) {
618 const char *path = df_sorted_entries.items[i].string;
619 int len = strlen(path);
620 struct stage_data *e = df_sorted_entries.items[i].util;
623 * Check if last_file & path correspond to a D/F conflict;
624 * i.e. whether path is last_file+'/'+<something>.
625 * If so, record that it's okay to remove last_file to make
626 * room for path and friends if needed.
628 if (last_file &&
629 len > last_len &&
630 memcmp(path, last_file, last_len) == 0 &&
631 path[last_len] == '/') {
632 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
636 * Determine whether path could exist as a file in the
637 * working directory as a possible D/F conflict. This
638 * will only occur when it exists in stage 2 as a
639 * file.
641 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
642 last_file = path;
643 last_len = len;
644 } else {
645 last_file = NULL;
648 string_list_clear(&df_sorted_entries, 0);
651 static int update_stages(struct merge_options *opt, const char *path,
652 const struct diff_filespec *o,
653 const struct diff_filespec *a,
654 const struct diff_filespec *b)
658 * NOTE: It is usually a bad idea to call update_stages on a path
659 * before calling update_file on that same path, since it can
660 * sometimes lead to spurious "refusing to lose untracked file..."
661 * messages from update_file (via make_room_for path via
662 * would_lose_untracked). Instead, reverse the order of the calls
663 * (executing update_file first and then update_stages).
665 int clear = 1;
666 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
667 if (clear)
668 if (remove_file_from_index(opt->repo->index, path))
669 return -1;
670 if (o)
671 if (add_cacheinfo(opt, o, path, 1, 0, options))
672 return -1;
673 if (a)
674 if (add_cacheinfo(opt, a, path, 2, 0, options))
675 return -1;
676 if (b)
677 if (add_cacheinfo(opt, b, path, 3, 0, options))
678 return -1;
679 return 0;
682 static void update_entry(struct stage_data *entry,
683 struct diff_filespec *o,
684 struct diff_filespec *a,
685 struct diff_filespec *b)
687 entry->processed = 0;
688 entry->stages[1].mode = o->mode;
689 entry->stages[2].mode = a->mode;
690 entry->stages[3].mode = b->mode;
691 oidcpy(&entry->stages[1].oid, &o->oid);
692 oidcpy(&entry->stages[2].oid, &a->oid);
693 oidcpy(&entry->stages[3].oid, &b->oid);
696 static int remove_file(struct merge_options *opt, int clean,
697 const char *path, int no_wd)
699 int update_cache = opt->priv->call_depth || clean;
700 int update_working_directory = !opt->priv->call_depth && !no_wd;
702 if (update_cache) {
703 if (remove_file_from_index(opt->repo->index, path))
704 return -1;
706 if (update_working_directory) {
707 if (ignore_case) {
708 struct cache_entry *ce;
709 ce = index_file_exists(opt->repo->index, path, strlen(path),
710 ignore_case);
711 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
712 return 0;
714 if (remove_path(path))
715 return -1;
717 return 0;
720 /* add a string to a strbuf, but converting "/" to "_" */
721 static void add_flattened_path(struct strbuf *out, const char *s)
723 size_t i = out->len;
724 strbuf_addstr(out, s);
725 for (; i < out->len; i++)
726 if (out->buf[i] == '/')
727 out->buf[i] = '_';
730 static char *unique_path(struct merge_options *opt,
731 const char *path,
732 const char *branch)
734 struct path_hashmap_entry *entry;
735 struct strbuf newpath = STRBUF_INIT;
736 int suffix = 0;
737 size_t base_len;
739 strbuf_addf(&newpath, "%s~", path);
740 add_flattened_path(&newpath, branch);
742 base_len = newpath.len;
743 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
744 fspathhash(newpath.buf), newpath.buf) ||
745 (!opt->priv->call_depth && file_exists(newpath.buf))) {
746 strbuf_setlen(&newpath, base_len);
747 strbuf_addf(&newpath, "_%d", suffix++);
750 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
751 hashmap_entry_init(&entry->e, fspathhash(entry->path));
752 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
753 return strbuf_detach(&newpath, NULL);
757 * Check whether a directory in the index is in the way of an incoming
758 * file. Return 1 if so. If check_working_copy is non-zero, also
759 * check the working directory. If empty_ok is non-zero, also return
760 * 0 in the case where the working-tree dir exists but is empty.
762 static int dir_in_way(struct index_state *istate, const char *path,
763 int check_working_copy, int empty_ok)
765 int pos;
766 struct strbuf dirpath = STRBUF_INIT;
767 struct stat st;
769 strbuf_addstr(&dirpath, path);
770 strbuf_addch(&dirpath, '/');
772 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
774 if (pos < 0)
775 pos = -1 - pos;
776 if (pos < istate->cache_nr &&
777 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
778 strbuf_release(&dirpath);
779 return 1;
782 strbuf_release(&dirpath);
783 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
784 !(empty_ok && is_empty_dir(path)) &&
785 !has_symlink_leading_path(path, strlen(path));
789 * Returns whether path was tracked in the index before the merge started,
790 * and its oid and mode match the specified values
792 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
793 const struct diff_filespec *blob)
795 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
796 struct cache_entry *ce;
798 if (0 > pos)
799 /* we were not tracking this path before the merge */
800 return 0;
802 /* See if the file we were tracking before matches */
803 ce = opt->priv->orig_index.cache[pos];
804 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
808 * Returns whether path was tracked in the index before the merge started
810 static int was_tracked(struct merge_options *opt, const char *path)
812 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
814 if (0 <= pos)
815 /* we were tracking this path before the merge */
816 return 1;
818 return 0;
821 static int would_lose_untracked(struct merge_options *opt, const char *path)
823 struct index_state *istate = opt->repo->index;
826 * This may look like it can be simplified to:
827 * return !was_tracked(opt, path) && file_exists(path)
828 * but it can't. This function needs to know whether path was in
829 * the working tree due to EITHER having been tracked in the index
830 * before the merge OR having been put into the working copy and
831 * index by unpack_trees(). Due to that either-or requirement, we
832 * check the current index instead of the original one.
834 * Note that we do not need to worry about merge-recursive itself
835 * updating the index after unpack_trees() and before calling this
836 * function, because we strictly require all code paths in
837 * merge-recursive to update the working tree first and the index
838 * second. Doing otherwise would break
839 * update_file()/would_lose_untracked(); see every comment in this
840 * file which mentions "update_stages".
842 int pos = index_name_pos(istate, path, strlen(path));
844 if (pos < 0)
845 pos = -1 - pos;
846 while (pos < istate->cache_nr &&
847 !strcmp(path, istate->cache[pos]->name)) {
849 * If stage #0, it is definitely tracked.
850 * If it has stage #2 then it was tracked
851 * before this merge started. All other
852 * cases the path was not tracked.
854 switch (ce_stage(istate->cache[pos])) {
855 case 0:
856 case 2:
857 return 0;
859 pos++;
861 return file_exists(path);
864 static int was_dirty(struct merge_options *opt, const char *path)
866 struct cache_entry *ce;
867 int dirty = 1;
869 if (opt->priv->call_depth || !was_tracked(opt, path))
870 return !dirty;
872 ce = index_file_exists(opt->priv->unpack_opts.src_index,
873 path, strlen(path), ignore_case);
874 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
875 return dirty;
878 static int make_room_for_path(struct merge_options *opt, const char *path)
880 int status, i;
881 const char *msg = _("failed to create path '%s'%s");
883 /* Unlink any D/F conflict files that are in the way */
884 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
885 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
886 size_t pathlen = strlen(path);
887 size_t df_pathlen = strlen(df_path);
888 if (df_pathlen < pathlen &&
889 path[df_pathlen] == '/' &&
890 strncmp(path, df_path, df_pathlen) == 0) {
891 output(opt, 3,
892 _("Removing %s to make room for subdirectory\n"),
893 df_path);
894 unlink(df_path);
895 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
896 i, 0);
897 break;
901 /* Make sure leading directories are created */
902 status = safe_create_leading_directories_const(path);
903 if (status) {
904 if (status == SCLD_EXISTS)
905 /* something else exists */
906 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
907 return err(opt, msg, path, "");
911 * Do not unlink a file in the work tree if we are not
912 * tracking it.
914 if (would_lose_untracked(opt, path))
915 return err(opt, _("refusing to lose untracked file at '%s'"),
916 path);
918 /* Successful unlink is good.. */
919 if (!unlink(path))
920 return 0;
921 /* .. and so is no existing file */
922 if (errno == ENOENT)
923 return 0;
924 /* .. but not some other error (who really cares what?) */
925 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
928 static int update_file_flags(struct merge_options *opt,
929 const struct diff_filespec *contents,
930 const char *path,
931 int update_cache,
932 int update_wd)
934 int ret = 0;
936 if (opt->priv->call_depth)
937 update_wd = 0;
939 if (update_wd) {
940 enum object_type type;
941 void *buf;
942 unsigned long size;
944 if (S_ISGITLINK(contents->mode)) {
946 * We may later decide to recursively descend into
947 * the submodule directory and update its index
948 * and/or work tree, but we do not do that now.
950 update_wd = 0;
951 goto update_index;
954 buf = read_object_file(&contents->oid, &type, &size);
955 if (!buf) {
956 ret = err(opt, _("cannot read object %s '%s'"),
957 oid_to_hex(&contents->oid), path);
958 goto free_buf;
960 if (type != OBJ_BLOB) {
961 ret = err(opt, _("blob expected for %s '%s'"),
962 oid_to_hex(&contents->oid), path);
963 goto free_buf;
965 if (S_ISREG(contents->mode)) {
966 struct strbuf strbuf = STRBUF_INIT;
967 if (convert_to_working_tree(opt->repo->index,
968 path, buf, size, &strbuf, NULL)) {
969 free(buf);
970 size = strbuf.len;
971 buf = strbuf_detach(&strbuf, NULL);
975 if (make_room_for_path(opt, path) < 0) {
976 update_wd = 0;
977 goto free_buf;
979 if (S_ISREG(contents->mode) ||
980 (!has_symlinks && S_ISLNK(contents->mode))) {
981 int fd;
982 int mode = (contents->mode & 0100 ? 0777 : 0666);
984 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
985 if (fd < 0) {
986 ret = err(opt, _("failed to open '%s': %s"),
987 path, strerror(errno));
988 goto free_buf;
990 write_in_full(fd, buf, size);
991 close(fd);
992 } else if (S_ISLNK(contents->mode)) {
993 char *lnk = xmemdupz(buf, size);
994 safe_create_leading_directories_const(path);
995 unlink(path);
996 if (symlink(lnk, path))
997 ret = err(opt, _("failed to symlink '%s': %s"),
998 path, strerror(errno));
999 free(lnk);
1000 } else
1001 ret = err(opt,
1002 _("do not know what to do with %06o %s '%s'"),
1003 contents->mode, oid_to_hex(&contents->oid), path);
1004 free_buf:
1005 free(buf);
1007 update_index:
1008 if (!ret && update_cache) {
1009 int refresh = (!opt->priv->call_depth &&
1010 contents->mode != S_IFGITLINK);
1011 if (add_cacheinfo(opt, contents, path, 0, refresh,
1012 ADD_CACHE_OK_TO_ADD))
1013 return -1;
1015 return ret;
1018 static int update_file(struct merge_options *opt,
1019 int clean,
1020 const struct diff_filespec *contents,
1021 const char *path)
1023 return update_file_flags(opt, contents, path,
1024 opt->priv->call_depth || clean, !opt->priv->call_depth);
1027 /* Low level file merging, update and removal */
1029 struct merge_file_info {
1030 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1031 unsigned clean:1,
1032 merge:1;
1035 static int merge_3way(struct merge_options *opt,
1036 mmbuffer_t *result_buf,
1037 const struct diff_filespec *o,
1038 const struct diff_filespec *a,
1039 const struct diff_filespec *b,
1040 const char *branch1,
1041 const char *branch2,
1042 const int extra_marker_size)
1044 mmfile_t orig, src1, src2;
1045 struct ll_merge_options ll_opts = {0};
1046 char *base, *name1, *name2;
1047 int merge_status;
1049 ll_opts.renormalize = opt->renormalize;
1050 ll_opts.extra_marker_size = extra_marker_size;
1051 ll_opts.xdl_opts = opt->xdl_opts;
1053 if (opt->priv->call_depth) {
1054 ll_opts.virtual_ancestor = 1;
1055 ll_opts.variant = 0;
1056 } else {
1057 switch (opt->recursive_variant) {
1058 case MERGE_VARIANT_OURS:
1059 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1060 break;
1061 case MERGE_VARIANT_THEIRS:
1062 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1063 break;
1064 default:
1065 ll_opts.variant = 0;
1066 break;
1070 assert(a->path && b->path && o->path && opt->ancestor);
1071 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1072 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1073 name1 = mkpathdup("%s:%s", branch1, a->path);
1074 name2 = mkpathdup("%s:%s", branch2, b->path);
1075 } else {
1076 base = mkpathdup("%s", opt->ancestor);
1077 name1 = mkpathdup("%s", branch1);
1078 name2 = mkpathdup("%s", branch2);
1081 read_mmblob(&orig, &o->oid);
1082 read_mmblob(&src1, &a->oid);
1083 read_mmblob(&src2, &b->oid);
1086 * FIXME: Using a->path for normalization rules in ll_merge could be
1087 * wrong if we renamed from a->path to b->path. We should use the
1088 * target path for where the file will be written.
1090 merge_status = ll_merge(result_buf, a->path, &orig, base,
1091 &src1, name1, &src2, name2,
1092 opt->repo->index, &ll_opts);
1094 free(base);
1095 free(name1);
1096 free(name2);
1097 free(orig.ptr);
1098 free(src1.ptr);
1099 free(src2.ptr);
1100 return merge_status;
1103 static int find_first_merges(struct repository *repo,
1104 struct object_array *result, const char *path,
1105 struct commit *a, struct commit *b)
1107 int i, j;
1108 struct object_array merges = OBJECT_ARRAY_INIT;
1109 struct commit *commit;
1110 int contains_another;
1112 char merged_revision[GIT_MAX_HEXSZ + 2];
1113 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1114 "--all", merged_revision, NULL };
1115 struct rev_info revs;
1116 struct setup_revision_opt rev_opts;
1118 memset(result, 0, sizeof(struct object_array));
1119 memset(&rev_opts, 0, sizeof(rev_opts));
1121 /* get all revisions that merge commit a */
1122 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1123 oid_to_hex(&a->object.oid));
1124 repo_init_revisions(repo, &revs, NULL);
1125 /* FIXME: can't handle linked worktrees in submodules yet */
1126 revs.single_worktree = path != NULL;
1127 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1129 /* save all revisions from the above list that contain b */
1130 if (prepare_revision_walk(&revs))
1131 die("revision walk setup failed");
1132 while ((commit = get_revision(&revs)) != NULL) {
1133 struct object *o = &(commit->object);
1134 if (repo_in_merge_bases(repo, b, commit))
1135 add_object_array(o, NULL, &merges);
1137 reset_revision_walk();
1139 /* Now we've got all merges that contain a and b. Prune all
1140 * merges that contain another found merge and save them in
1141 * result.
1143 for (i = 0; i < merges.nr; i++) {
1144 struct commit *m1 = (struct commit *) merges.objects[i].item;
1146 contains_another = 0;
1147 for (j = 0; j < merges.nr; j++) {
1148 struct commit *m2 = (struct commit *) merges.objects[j].item;
1149 if (i != j && repo_in_merge_bases(repo, m2, m1)) {
1150 contains_another = 1;
1151 break;
1155 if (!contains_another)
1156 add_object_array(merges.objects[i].item, NULL, result);
1159 object_array_clear(&merges);
1160 return result->nr;
1163 static void print_commit(struct repository *repo, struct commit *commit)
1165 struct strbuf sb = STRBUF_INIT;
1166 struct pretty_print_context ctx = {0};
1167 ctx.date_mode.type = DATE_NORMAL;
1168 /* FIXME: Merge this with output_commit_title() */
1169 assert(!merge_remote_util(commit));
1170 repo_format_commit_message(repo, commit, " %h: %m %s", &sb, &ctx);
1171 fprintf(stderr, "%s\n", sb.buf);
1172 strbuf_release(&sb);
1175 static int is_valid(const struct diff_filespec *dfs)
1177 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1180 static int merge_submodule(struct merge_options *opt,
1181 struct object_id *result, const char *path,
1182 const struct object_id *base, const struct object_id *a,
1183 const struct object_id *b)
1185 struct repository subrepo;
1186 int ret = 0;
1187 struct commit *commit_base, *commit_a, *commit_b;
1188 int parent_count;
1189 struct object_array merges;
1191 int i;
1192 int search = !opt->priv->call_depth;
1194 /* store a in result in case we fail */
1195 /* FIXME: This is the WRONG resolution for the recursive case when
1196 * we need to be careful to avoid accidentally matching either side.
1197 * Should probably use o instead there, much like we do for merging
1198 * binaries.
1200 oidcpy(result, a);
1202 /* we can not handle deletion conflicts */
1203 if (is_null_oid(base))
1204 return 0;
1205 if (is_null_oid(a))
1206 return 0;
1207 if (is_null_oid(b))
1208 return 0;
1210 if (repo_submodule_init(&subrepo, opt->repo, path, null_oid())) {
1211 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1212 return 0;
1215 if (!(commit_base = lookup_commit_reference(&subrepo, base)) ||
1216 !(commit_a = lookup_commit_reference(&subrepo, a)) ||
1217 !(commit_b = lookup_commit_reference(&subrepo, b))) {
1218 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1219 goto cleanup;
1222 /* check whether both changes are forward */
1223 if (!repo_in_merge_bases(&subrepo, commit_base, commit_a) ||
1224 !repo_in_merge_bases(&subrepo, commit_base, commit_b)) {
1225 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1226 goto cleanup;
1229 /* Case #1: a is contained in b or vice versa */
1230 if (repo_in_merge_bases(&subrepo, commit_a, commit_b)) {
1231 oidcpy(result, b);
1232 if (show(opt, 3)) {
1233 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1234 repo_output_commit_title(opt, &subrepo, commit_b);
1235 } else if (show(opt, 2))
1236 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1237 else
1238 ; /* no output */
1240 ret = 1;
1241 goto cleanup;
1243 if (repo_in_merge_bases(&subrepo, commit_b, commit_a)) {
1244 oidcpy(result, a);
1245 if (show(opt, 3)) {
1246 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1247 repo_output_commit_title(opt, &subrepo, commit_a);
1248 } else if (show(opt, 2))
1249 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1250 else
1251 ; /* no output */
1253 ret = 1;
1254 goto cleanup;
1258 * Case #2: There are one or more merges that contain a and b in
1259 * the submodule. If there is only one, then present it as a
1260 * suggestion to the user, but leave it marked unmerged so the
1261 * user needs to confirm the resolution.
1264 /* Skip the search if makes no sense to the calling context. */
1265 if (!search)
1266 goto cleanup;
1268 /* find commit which merges them */
1269 parent_count = find_first_merges(&subrepo, &merges, path,
1270 commit_a, commit_b);
1271 switch (parent_count) {
1272 case 0:
1273 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1274 break;
1276 case 1:
1277 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1278 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1279 print_commit(&subrepo, (struct commit *) merges.objects[0].item);
1280 output(opt, 2, _(
1281 "If this is correct simply add it to the index "
1282 "for example\n"
1283 "by using:\n\n"
1284 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1285 "which will accept this suggestion.\n"),
1286 oid_to_hex(&merges.objects[0].item->oid), path);
1287 break;
1289 default:
1290 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1291 for (i = 0; i < merges.nr; i++)
1292 print_commit(&subrepo, (struct commit *) merges.objects[i].item);
1295 object_array_clear(&merges);
1296 cleanup:
1297 repo_clear(&subrepo);
1298 return ret;
1301 static int merge_mode_and_contents(struct merge_options *opt,
1302 const struct diff_filespec *o,
1303 const struct diff_filespec *a,
1304 const struct diff_filespec *b,
1305 const char *filename,
1306 const char *branch1,
1307 const char *branch2,
1308 const int extra_marker_size,
1309 struct merge_file_info *result)
1311 if (opt->branch1 != branch1) {
1313 * It's weird getting a reverse merge with HEAD on the bottom
1314 * side of the conflict markers and the other branch on the
1315 * top. Fix that.
1317 return merge_mode_and_contents(opt, o, b, a,
1318 filename,
1319 branch2, branch1,
1320 extra_marker_size, result);
1323 result->merge = 0;
1324 result->clean = 1;
1326 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1327 result->clean = 0;
1329 * FIXME: This is a bad resolution for recursive case; for
1330 * the recursive case we want something that is unlikely to
1331 * accidentally match either side. Also, while it makes
1332 * sense to prefer regular files over symlinks, it doesn't
1333 * make sense to prefer regular files over submodules.
1335 if (S_ISREG(a->mode)) {
1336 result->blob.mode = a->mode;
1337 oidcpy(&result->blob.oid, &a->oid);
1338 } else {
1339 result->blob.mode = b->mode;
1340 oidcpy(&result->blob.oid, &b->oid);
1342 } else {
1343 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1344 result->merge = 1;
1347 * Merge modes
1349 if (a->mode == b->mode || a->mode == o->mode)
1350 result->blob.mode = b->mode;
1351 else {
1352 result->blob.mode = a->mode;
1353 if (b->mode != o->mode) {
1354 result->clean = 0;
1355 result->merge = 1;
1359 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1360 oidcpy(&result->blob.oid, &b->oid);
1361 else if (oideq(&b->oid, &o->oid))
1362 oidcpy(&result->blob.oid, &a->oid);
1363 else if (S_ISREG(a->mode)) {
1364 mmbuffer_t result_buf;
1365 int ret = 0, merge_status;
1367 merge_status = merge_3way(opt, &result_buf, o, a, b,
1368 branch1, branch2,
1369 extra_marker_size);
1371 if ((merge_status < 0) || !result_buf.ptr)
1372 ret = err(opt, _("Failed to execute internal merge"));
1374 if (!ret &&
1375 write_object_file(result_buf.ptr, result_buf.size,
1376 blob_type, &result->blob.oid))
1377 ret = err(opt, _("Unable to add %s to database"),
1378 a->path);
1380 free(result_buf.ptr);
1381 if (ret)
1382 return ret;
1383 /* FIXME: bug, what if modes didn't match? */
1384 result->clean = (merge_status == 0);
1385 } else if (S_ISGITLINK(a->mode)) {
1386 result->clean = merge_submodule(opt, &result->blob.oid,
1387 o->path,
1388 &o->oid,
1389 &a->oid,
1390 &b->oid);
1391 } else if (S_ISLNK(a->mode)) {
1392 switch (opt->recursive_variant) {
1393 case MERGE_VARIANT_NORMAL:
1394 oidcpy(&result->blob.oid, &a->oid);
1395 if (!oideq(&a->oid, &b->oid))
1396 result->clean = 0;
1397 break;
1398 case MERGE_VARIANT_OURS:
1399 oidcpy(&result->blob.oid, &a->oid);
1400 break;
1401 case MERGE_VARIANT_THEIRS:
1402 oidcpy(&result->blob.oid, &b->oid);
1403 break;
1405 } else
1406 BUG("unsupported object type in the tree");
1409 if (result->merge)
1410 output(opt, 2, _("Auto-merging %s"), filename);
1412 return 0;
1415 static int handle_rename_via_dir(struct merge_options *opt,
1416 struct rename_conflict_info *ci)
1419 * Handle file adds that need to be renamed due to directory rename
1420 * detection. This differs from handle_rename_normal, because
1421 * there is no content merge to do; just move the file into the
1422 * desired final location.
1424 const struct rename *ren = ci->ren1;
1425 const struct diff_filespec *dest = ren->pair->two;
1426 char *file_path = dest->path;
1427 int mark_conflicted = (opt->detect_directory_renames ==
1428 MERGE_DIRECTORY_RENAMES_CONFLICT);
1429 assert(ren->dir_rename_original_dest);
1431 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1432 mark_conflicted = 1;
1433 file_path = unique_path(opt, dest->path, ren->branch);
1434 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1435 "writing to %s instead."),
1436 dest->path, file_path);
1439 if (mark_conflicted) {
1441 * Write the file in worktree at file_path. In the index,
1442 * only record the file at dest->path in the appropriate
1443 * higher stage.
1445 if (update_file(opt, 0, dest, file_path))
1446 return -1;
1447 if (file_path != dest->path)
1448 free(file_path);
1449 if (update_stages(opt, dest->path, NULL,
1450 ren->branch == opt->branch1 ? dest : NULL,
1451 ren->branch == opt->branch1 ? NULL : dest))
1452 return -1;
1453 return 0; /* not clean, but conflicted */
1454 } else {
1455 /* Update dest->path both in index and in worktree */
1456 if (update_file(opt, 1, dest, dest->path))
1457 return -1;
1458 return 1; /* clean */
1462 static int handle_change_delete(struct merge_options *opt,
1463 const char *path, const char *old_path,
1464 const struct diff_filespec *o,
1465 const struct diff_filespec *changed,
1466 const char *change_branch,
1467 const char *delete_branch,
1468 const char *change, const char *change_past)
1470 char *alt_path = NULL;
1471 const char *update_path = path;
1472 int ret = 0;
1474 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1475 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1476 update_path = alt_path = unique_path(opt, path, change_branch);
1479 if (opt->priv->call_depth) {
1481 * We cannot arbitrarily accept either a_sha or b_sha as
1482 * correct; since there is no true "middle point" between
1483 * them, simply reuse the base version for virtual merge base.
1485 ret = remove_file_from_index(opt->repo->index, path);
1486 if (!ret)
1487 ret = update_file(opt, 0, o, update_path);
1488 } else {
1490 * Despite the four nearly duplicate messages and argument
1491 * lists below and the ugliness of the nested if-statements,
1492 * having complete messages makes the job easier for
1493 * translators.
1495 * The slight variance among the cases is due to the fact
1496 * that:
1497 * 1) directory/file conflicts (in effect if
1498 * !alt_path) could cause us to need to write the
1499 * file to a different path.
1500 * 2) renames (in effect if !old_path) could mean that
1501 * there are two names for the path that the user
1502 * may know the file by.
1504 if (!alt_path) {
1505 if (!old_path) {
1506 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1507 "and %s in %s. Version %s of %s left in tree."),
1508 change, path, delete_branch, change_past,
1509 change_branch, change_branch, path);
1510 } else {
1511 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1512 "and %s to %s in %s. Version %s of %s left in tree."),
1513 change, old_path, delete_branch, change_past, path,
1514 change_branch, change_branch, path);
1516 } else {
1517 if (!old_path) {
1518 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1519 "and %s in %s. Version %s of %s left in tree at %s."),
1520 change, path, delete_branch, change_past,
1521 change_branch, change_branch, path, alt_path);
1522 } else {
1523 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1524 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1525 change, old_path, delete_branch, change_past, path,
1526 change_branch, change_branch, path, alt_path);
1530 * No need to call update_file() on path when change_branch ==
1531 * opt->branch1 && !alt_path, since that would needlessly touch
1532 * path. We could call update_file_flags() with update_cache=0
1533 * and update_wd=0, but that's a no-op.
1535 if (change_branch != opt->branch1 || alt_path)
1536 ret = update_file(opt, 0, changed, update_path);
1538 free(alt_path);
1540 return ret;
1543 static int handle_rename_delete(struct merge_options *opt,
1544 struct rename_conflict_info *ci)
1546 const struct rename *ren = ci->ren1;
1547 const struct diff_filespec *orig = ren->pair->one;
1548 const struct diff_filespec *dest = ren->pair->two;
1549 const char *rename_branch = ren->branch;
1550 const char *delete_branch = (opt->branch1 == ren->branch ?
1551 opt->branch2 : opt->branch1);
1553 if (handle_change_delete(opt,
1554 opt->priv->call_depth ? orig->path : dest->path,
1555 opt->priv->call_depth ? NULL : orig->path,
1556 orig, dest,
1557 rename_branch, delete_branch,
1558 _("rename"), _("renamed")))
1559 return -1;
1561 if (opt->priv->call_depth)
1562 return remove_file_from_index(opt->repo->index, dest->path);
1563 else
1564 return update_stages(opt, dest->path, NULL,
1565 rename_branch == opt->branch1 ? dest : NULL,
1566 rename_branch == opt->branch1 ? NULL : dest);
1569 static int handle_file_collision(struct merge_options *opt,
1570 const char *collide_path,
1571 const char *prev_path1,
1572 const char *prev_path2,
1573 const char *branch1, const char *branch2,
1574 struct diff_filespec *a,
1575 struct diff_filespec *b)
1577 struct merge_file_info mfi;
1578 struct diff_filespec null;
1579 char *alt_path = NULL;
1580 const char *update_path = collide_path;
1583 * It's easiest to get the correct things into stage 2 and 3, and
1584 * to make sure that the content merge puts HEAD before the other
1585 * branch if we just ensure that branch1 == opt->branch1. So, simply
1586 * flip arguments around if we don't have that.
1588 if (branch1 != opt->branch1) {
1589 return handle_file_collision(opt, collide_path,
1590 prev_path2, prev_path1,
1591 branch2, branch1,
1592 b, a);
1595 /* Remove rename sources if rename/add or rename/rename(2to1) */
1596 if (prev_path1)
1597 remove_file(opt, 1, prev_path1,
1598 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1599 if (prev_path2)
1600 remove_file(opt, 1, prev_path2,
1601 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1604 * Remove the collision path, if it wouldn't cause dirty contents
1605 * or an untracked file to get lost. We'll either overwrite with
1606 * merged contents, or just write out to differently named files.
1608 if (was_dirty(opt, collide_path)) {
1609 output(opt, 1, _("Refusing to lose dirty file at %s"),
1610 collide_path);
1611 update_path = alt_path = unique_path(opt, collide_path, "merged");
1612 } else if (would_lose_untracked(opt, collide_path)) {
1614 * Only way we get here is if both renames were from
1615 * a directory rename AND user had an untracked file
1616 * at the location where both files end up after the
1617 * two directory renames. See testcase 10d of t6043.
1619 output(opt, 1, _("Refusing to lose untracked file at "
1620 "%s, even though it's in the way."),
1621 collide_path);
1622 update_path = alt_path = unique_path(opt, collide_path, "merged");
1623 } else {
1625 * FIXME: It's possible that the two files are identical
1626 * and that the current working copy happens to match, in
1627 * which case we are unnecessarily touching the working
1628 * tree file. It's not a likely enough scenario that I
1629 * want to code up the checks for it and a better fix is
1630 * available if we restructure how unpack_trees() and
1631 * merge-recursive interoperate anyway, so punting for
1632 * now...
1634 remove_file(opt, 0, collide_path, 0);
1637 /* Store things in diff_filespecs for functions that need it */
1638 null.path = (char *)collide_path;
1639 oidcpy(&null.oid, null_oid());
1640 null.mode = 0;
1642 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1643 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1644 return -1;
1645 mfi.clean &= !alt_path;
1646 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1647 return -1;
1648 if (!mfi.clean && !opt->priv->call_depth &&
1649 update_stages(opt, collide_path, NULL, a, b))
1650 return -1;
1651 free(alt_path);
1653 * FIXME: If both a & b both started with conflicts (only possible
1654 * if they came from a rename/rename(2to1)), but had IDENTICAL
1655 * contents including those conflicts, then in the next line we claim
1656 * it was clean. If someone cares about this case, we should have the
1657 * caller notify us if we started with conflicts.
1659 return mfi.clean;
1662 static int handle_rename_add(struct merge_options *opt,
1663 struct rename_conflict_info *ci)
1665 /* a was renamed to c, and a separate c was added. */
1666 struct diff_filespec *a = ci->ren1->pair->one;
1667 struct diff_filespec *c = ci->ren1->pair->two;
1668 char *path = c->path;
1669 char *prev_path_desc;
1670 struct merge_file_info mfi;
1672 const char *rename_branch = ci->ren1->branch;
1673 const char *add_branch = (opt->branch1 == rename_branch ?
1674 opt->branch2 : opt->branch1);
1675 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1677 output(opt, 1, _("CONFLICT (rename/add): "
1678 "Rename %s->%s in %s. Added %s in %s"),
1679 a->path, c->path, rename_branch,
1680 c->path, add_branch);
1682 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1683 ci->ren1->src_entry->stages[other_stage].path = a->path;
1684 if (merge_mode_and_contents(opt, a, c,
1685 &ci->ren1->src_entry->stages[other_stage],
1686 prev_path_desc,
1687 opt->branch1, opt->branch2,
1688 1 + opt->priv->call_depth * 2, &mfi))
1689 return -1;
1690 free(prev_path_desc);
1692 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1693 return handle_file_collision(opt,
1694 c->path, a->path, NULL,
1695 rename_branch, add_branch,
1696 &mfi.blob,
1697 &ci->ren1->dst_entry->stages[other_stage]);
1700 static char *find_path_for_conflict(struct merge_options *opt,
1701 const char *path,
1702 const char *branch1,
1703 const char *branch2)
1705 char *new_path = NULL;
1706 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1707 new_path = unique_path(opt, path, branch1);
1708 output(opt, 1, _("%s is a directory in %s adding "
1709 "as %s instead"),
1710 path, branch2, new_path);
1711 } else if (would_lose_untracked(opt, path)) {
1712 new_path = unique_path(opt, path, branch1);
1713 output(opt, 1, _("Refusing to lose untracked file"
1714 " at %s; adding as %s instead"),
1715 path, new_path);
1718 return new_path;
1722 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1724 static inline int flip_stage(int stage)
1726 return (2 + 3) - stage;
1729 static int handle_rename_rename_1to2(struct merge_options *opt,
1730 struct rename_conflict_info *ci)
1732 /* One file was renamed in both branches, but to different names. */
1733 struct merge_file_info mfi;
1734 struct diff_filespec *add;
1735 struct diff_filespec *o = ci->ren1->pair->one;
1736 struct diff_filespec *a = ci->ren1->pair->two;
1737 struct diff_filespec *b = ci->ren2->pair->two;
1738 char *path_desc;
1740 output(opt, 1, _("CONFLICT (rename/rename): "
1741 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1742 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1743 o->path, a->path, ci->ren1->branch,
1744 o->path, b->path, ci->ren2->branch,
1745 opt->priv->call_depth ? _(" (left unresolved)") : "");
1747 path_desc = xstrfmt("%s and %s, both renamed from %s",
1748 a->path, b->path, o->path);
1749 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1750 ci->ren1->branch, ci->ren2->branch,
1751 opt->priv->call_depth * 2, &mfi))
1752 return -1;
1753 free(path_desc);
1755 if (opt->priv->call_depth)
1756 remove_file_from_index(opt->repo->index, o->path);
1759 * For each destination path, we need to see if there is a
1760 * rename/add collision. If not, we can write the file out
1761 * to the specified location.
1763 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1764 if (is_valid(add)) {
1765 add->path = mfi.blob.path = a->path;
1766 if (handle_file_collision(opt, a->path,
1767 NULL, NULL,
1768 ci->ren1->branch,
1769 ci->ren2->branch,
1770 &mfi.blob, add) < 0)
1771 return -1;
1772 } else {
1773 char *new_path = find_path_for_conflict(opt, a->path,
1774 ci->ren1->branch,
1775 ci->ren2->branch);
1776 if (update_file(opt, 0, &mfi.blob,
1777 new_path ? new_path : a->path))
1778 return -1;
1779 free(new_path);
1780 if (!opt->priv->call_depth &&
1781 update_stages(opt, a->path, NULL, a, NULL))
1782 return -1;
1785 if (!mfi.clean && mfi.blob.mode == a->mode &&
1786 oideq(&mfi.blob.oid, &a->oid)) {
1788 * Getting here means we were attempting to merge a binary
1789 * blob. Since we can't merge binaries, the merge algorithm
1790 * just takes one side. But we don't want to copy the
1791 * contents of one side to both paths; we'd rather use the
1792 * original content at the given path for each path.
1794 oidcpy(&mfi.blob.oid, &b->oid);
1795 mfi.blob.mode = b->mode;
1797 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1798 if (is_valid(add)) {
1799 add->path = mfi.blob.path = b->path;
1800 if (handle_file_collision(opt, b->path,
1801 NULL, NULL,
1802 ci->ren1->branch,
1803 ci->ren2->branch,
1804 add, &mfi.blob) < 0)
1805 return -1;
1806 } else {
1807 char *new_path = find_path_for_conflict(opt, b->path,
1808 ci->ren2->branch,
1809 ci->ren1->branch);
1810 if (update_file(opt, 0, &mfi.blob,
1811 new_path ? new_path : b->path))
1812 return -1;
1813 free(new_path);
1814 if (!opt->priv->call_depth &&
1815 update_stages(opt, b->path, NULL, NULL, b))
1816 return -1;
1819 return 0;
1822 static int handle_rename_rename_2to1(struct merge_options *opt,
1823 struct rename_conflict_info *ci)
1825 /* Two files, a & b, were renamed to the same thing, c. */
1826 struct diff_filespec *a = ci->ren1->pair->one;
1827 struct diff_filespec *b = ci->ren2->pair->one;
1828 struct diff_filespec *c1 = ci->ren1->pair->two;
1829 struct diff_filespec *c2 = ci->ren2->pair->two;
1830 char *path = c1->path; /* == c2->path */
1831 char *path_side_1_desc;
1832 char *path_side_2_desc;
1833 struct merge_file_info mfi_c1;
1834 struct merge_file_info mfi_c2;
1835 int ostage1, ostage2;
1837 output(opt, 1, _("CONFLICT (rename/rename): "
1838 "Rename %s->%s in %s. "
1839 "Rename %s->%s in %s"),
1840 a->path, c1->path, ci->ren1->branch,
1841 b->path, c2->path, ci->ren2->branch);
1843 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1844 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1845 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1846 ostage2 = flip_stage(ostage1);
1847 ci->ren1->src_entry->stages[ostage1].path = a->path;
1848 ci->ren2->src_entry->stages[ostage2].path = b->path;
1849 if (merge_mode_and_contents(opt, a, c1,
1850 &ci->ren1->src_entry->stages[ostage1],
1851 path_side_1_desc,
1852 opt->branch1, opt->branch2,
1853 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1854 merge_mode_and_contents(opt, b,
1855 &ci->ren2->src_entry->stages[ostage2],
1856 c2, path_side_2_desc,
1857 opt->branch1, opt->branch2,
1858 1 + opt->priv->call_depth * 2, &mfi_c2))
1859 return -1;
1860 free(path_side_1_desc);
1861 free(path_side_2_desc);
1862 mfi_c1.blob.path = path;
1863 mfi_c2.blob.path = path;
1865 return handle_file_collision(opt, path, a->path, b->path,
1866 ci->ren1->branch, ci->ren2->branch,
1867 &mfi_c1.blob, &mfi_c2.blob);
1871 * Get the diff_filepairs changed between o_tree and tree.
1873 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1874 struct tree *o_tree,
1875 struct tree *tree)
1877 struct diff_queue_struct *ret;
1878 struct diff_options opts;
1880 repo_diff_setup(opt->repo, &opts);
1881 opts.flags.recursive = 1;
1882 opts.flags.rename_empty = 0;
1883 opts.detect_rename = merge_detect_rename(opt);
1885 * We do not have logic to handle the detection of copies. In
1886 * fact, it may not even make sense to add such logic: would we
1887 * really want a change to a base file to be propagated through
1888 * multiple other files by a merge?
1890 if (opts.detect_rename > DIFF_DETECT_RENAME)
1891 opts.detect_rename = DIFF_DETECT_RENAME;
1892 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 7000;
1893 opts.rename_score = opt->rename_score;
1894 opts.show_rename_progress = opt->show_rename_progress;
1895 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1896 diff_setup_done(&opts);
1897 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1898 diffcore_std(&opts);
1899 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1900 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1902 ret = xmalloc(sizeof(*ret));
1903 *ret = diff_queued_diff;
1905 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1906 diff_queued_diff.nr = 0;
1907 diff_queued_diff.queue = NULL;
1908 diff_flush(&opts);
1909 return ret;
1912 static int tree_has_path(struct repository *r, struct tree *tree,
1913 const char *path)
1915 struct object_id hashy;
1916 unsigned short mode_o;
1918 return !get_tree_entry(r,
1919 &tree->object.oid, path,
1920 &hashy, &mode_o);
1924 * Return a new string that replaces the beginning portion (which matches
1925 * entry->dir), with entry->new_dir. In perl-speak:
1926 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1927 * NOTE:
1928 * Caller must ensure that old_path starts with entry->dir + '/'.
1930 static char *apply_dir_rename(struct dir_rename_entry *entry,
1931 const char *old_path)
1933 struct strbuf new_path = STRBUF_INIT;
1934 int oldlen, newlen;
1936 if (entry->non_unique_new_dir)
1937 return NULL;
1939 oldlen = strlen(entry->dir);
1940 if (entry->new_dir.len == 0)
1942 * If someone renamed/merged a subdirectory into the root
1943 * directory (e.g. 'some/subdir' -> ''), then we want to
1944 * avoid returning
1945 * '' + '/filename'
1946 * as the rename; we need to make old_path + oldlen advance
1947 * past the '/' character.
1949 oldlen++;
1950 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1951 strbuf_grow(&new_path, newlen);
1952 strbuf_addbuf(&new_path, &entry->new_dir);
1953 strbuf_addstr(&new_path, &old_path[oldlen]);
1955 return strbuf_detach(&new_path, NULL);
1958 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1959 char **old_dir, char **new_dir)
1961 char *end_of_old, *end_of_new;
1963 /* Default return values: NULL, meaning no rename */
1964 *old_dir = NULL;
1965 *new_dir = NULL;
1968 * For
1969 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1970 * the "e/foo.c" part is the same, we just want to know that
1971 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1972 * so, for this example, this function returns "a/b/c/d" in
1973 * *old_dir and "a/b/some/thing/else" in *new_dir.
1977 * If the basename of the file changed, we don't care. We want
1978 * to know which portion of the directory, if any, changed.
1980 end_of_old = strrchr(old_path, '/');
1981 end_of_new = strrchr(new_path, '/');
1984 * If end_of_old is NULL, old_path wasn't in a directory, so there
1985 * could not be a directory rename (our rule elsewhere that a
1986 * directory which still exists is not considered to have been
1987 * renamed means the root directory can never be renamed -- because
1988 * the root directory always exists).
1990 if (end_of_old == NULL)
1991 return; /* Note: *old_dir and *new_dir are still NULL */
1994 * If new_path contains no directory (end_of_new is NULL), then we
1995 * have a rename of old_path's directory to the root directory.
1997 if (end_of_new == NULL) {
1998 *old_dir = xstrndup(old_path, end_of_old - old_path);
1999 *new_dir = xstrdup("");
2000 return;
2003 /* Find the first non-matching character traversing backwards */
2004 while (*--end_of_new == *--end_of_old &&
2005 end_of_old != old_path &&
2006 end_of_new != new_path)
2007 ; /* Do nothing; all in the while loop */
2010 * If both got back to the beginning of their strings, then the
2011 * directory didn't change at all, only the basename did.
2013 if (end_of_old == old_path && end_of_new == new_path &&
2014 *end_of_old == *end_of_new)
2015 return; /* Note: *old_dir and *new_dir are still NULL */
2018 * If end_of_new got back to the beginning of its string, and
2019 * end_of_old got back to the beginning of some subdirectory, then
2020 * we have a rename/merge of a subdirectory into the root, which
2021 * needs slightly special handling.
2023 * Note: There is no need to consider the opposite case, with a
2024 * rename/merge of the root directory into some subdirectory
2025 * because as noted above the root directory always exists so it
2026 * cannot be considered to be renamed.
2028 if (end_of_new == new_path &&
2029 end_of_old != old_path && end_of_old[-1] == '/') {
2030 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2031 *new_dir = xstrdup("");
2032 return;
2036 * We've found the first non-matching character in the directory
2037 * paths. That means the current characters we were looking at
2038 * were part of the first non-matching subdir name going back from
2039 * the end of the strings. Get the whole name by advancing both
2040 * end_of_old and end_of_new to the NEXT '/' character. That will
2041 * represent the entire directory rename.
2043 * The reason for the increment is cases like
2044 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2045 * After dropping the basename and going back to the first
2046 * non-matching character, we're now comparing:
2047 * a/b/s and a/b/
2048 * and we want to be comparing:
2049 * a/b/star/ and a/b/tar/
2050 * but without the pre-increment, the one on the right would stay
2051 * a/b/.
2053 end_of_old = strchr(++end_of_old, '/');
2054 end_of_new = strchr(++end_of_new, '/');
2056 /* Copy the old and new directories into *old_dir and *new_dir. */
2057 *old_dir = xstrndup(old_path, end_of_old - old_path);
2058 *new_dir = xstrndup(new_path, end_of_new - new_path);
2061 static void remove_hashmap_entries(struct hashmap *dir_renames,
2062 struct string_list *items_to_remove)
2064 int i;
2065 struct dir_rename_entry *entry;
2067 for (i = 0; i < items_to_remove->nr; i++) {
2068 entry = items_to_remove->items[i].util;
2069 hashmap_remove(dir_renames, &entry->ent, NULL);
2071 string_list_clear(items_to_remove, 0);
2075 * See if there is a directory rename for path, and if there are any file
2076 * level conflicts for the renamed location. If there is a rename and
2077 * there are no conflicts, return the new name. Otherwise, return NULL.
2079 static char *handle_path_level_conflicts(struct merge_options *opt,
2080 const char *path,
2081 struct dir_rename_entry *entry,
2082 struct hashmap *collisions,
2083 struct tree *tree)
2085 char *new_path = NULL;
2086 struct collision_entry *collision_ent;
2087 int clean = 1;
2088 struct strbuf collision_paths = STRBUF_INIT;
2091 * entry has the mapping of old directory name to new directory name
2092 * that we want to apply to path.
2094 new_path = apply_dir_rename(entry, path);
2096 if (!new_path) {
2097 /* This should only happen when entry->non_unique_new_dir set */
2098 if (!entry->non_unique_new_dir)
2099 BUG("entry->non_unqiue_dir not set and !new_path");
2100 output(opt, 1, _("CONFLICT (directory rename split): "
2101 "Unclear where to place %s because directory "
2102 "%s was renamed to multiple other directories, "
2103 "with no destination getting a majority of the "
2104 "files."),
2105 path, entry->dir);
2106 clean = 0;
2107 return NULL;
2111 * The caller needs to have ensured that it has pre-populated
2112 * collisions with all paths that map to new_path. Do a quick check
2113 * to ensure that's the case.
2115 collision_ent = collision_find_entry(collisions, new_path);
2116 if (collision_ent == NULL)
2117 BUG("collision_ent is NULL");
2120 * Check for one-sided add/add/.../add conflicts, i.e.
2121 * where implicit renames from the other side doing
2122 * directory rename(s) can affect this side of history
2123 * to put multiple paths into the same location. Warn
2124 * and bail on directory renames for such paths.
2126 if (collision_ent->reported_already) {
2127 clean = 0;
2128 } else if (tree_has_path(opt->repo, tree, new_path)) {
2129 collision_ent->reported_already = 1;
2130 strbuf_add_separated_string_list(&collision_paths, ", ",
2131 &collision_ent->source_files);
2132 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2133 "file/dir at %s in the way of implicit "
2134 "directory rename(s) putting the following "
2135 "path(s) there: %s."),
2136 new_path, collision_paths.buf);
2137 clean = 0;
2138 } else if (collision_ent->source_files.nr > 1) {
2139 collision_ent->reported_already = 1;
2140 strbuf_add_separated_string_list(&collision_paths, ", ",
2141 &collision_ent->source_files);
2142 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2143 "more than one path to %s; implicit directory "
2144 "renames tried to put these paths there: %s"),
2145 new_path, collision_paths.buf);
2146 clean = 0;
2149 /* Free memory we no longer need */
2150 strbuf_release(&collision_paths);
2151 if (!clean && new_path) {
2152 free(new_path);
2153 return NULL;
2156 return new_path;
2160 * There are a couple things we want to do at the directory level:
2161 * 1. Check for both sides renaming to the same thing, in order to avoid
2162 * implicit renaming of files that should be left in place. (See
2163 * testcase 6b in t6043 for details.)
2164 * 2. Prune directory renames if there are still files left in the
2165 * original directory. These represent a partial directory rename,
2166 * i.e. a rename where only some of the files within the directory
2167 * were renamed elsewhere. (Technically, this could be done earlier
2168 * in get_directory_renames(), except that would prevent us from
2169 * doing the previous check and thus failing testcase 6b.)
2170 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2171 * In the future, we could potentially record this info as well and
2172 * omit reporting rename/rename(1to2) conflicts for each path within
2173 * the affected directories, thus cleaning up the merge output.
2174 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2175 * directory level, because merging directories is fine. If it
2176 * causes conflicts for files within those merged directories, then
2177 * that should be detected at the individual path level.
2179 static void handle_directory_level_conflicts(struct merge_options *opt,
2180 struct hashmap *dir_re_head,
2181 struct tree *head,
2182 struct hashmap *dir_re_merge,
2183 struct tree *merge)
2185 struct hashmap_iter iter;
2186 struct dir_rename_entry *head_ent;
2187 struct dir_rename_entry *merge_ent;
2189 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2190 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2192 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2193 ent /* member name */) {
2194 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2195 if (merge_ent &&
2196 !head_ent->non_unique_new_dir &&
2197 !merge_ent->non_unique_new_dir &&
2198 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2199 /* 1. Renamed identically; remove it from both sides */
2200 string_list_append(&remove_from_head,
2201 head_ent->dir)->util = head_ent;
2202 strbuf_release(&head_ent->new_dir);
2203 string_list_append(&remove_from_merge,
2204 merge_ent->dir)->util = merge_ent;
2205 strbuf_release(&merge_ent->new_dir);
2206 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2207 /* 2. This wasn't a directory rename after all */
2208 string_list_append(&remove_from_head,
2209 head_ent->dir)->util = head_ent;
2210 strbuf_release(&head_ent->new_dir);
2214 remove_hashmap_entries(dir_re_head, &remove_from_head);
2215 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2217 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2218 ent /* member name */) {
2219 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2220 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2221 /* 2. This wasn't a directory rename after all */
2222 string_list_append(&remove_from_merge,
2223 merge_ent->dir)->util = merge_ent;
2224 } else if (head_ent &&
2225 !head_ent->non_unique_new_dir &&
2226 !merge_ent->non_unique_new_dir) {
2227 /* 3. rename/rename(1to2) */
2229 * We can assume it's not rename/rename(1to1) because
2230 * that was case (1), already checked above. So we
2231 * know that head_ent->new_dir and merge_ent->new_dir
2232 * are different strings.
2234 output(opt, 1, _("CONFLICT (rename/rename): "
2235 "Rename directory %s->%s in %s. "
2236 "Rename directory %s->%s in %s"),
2237 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2238 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2239 string_list_append(&remove_from_head,
2240 head_ent->dir)->util = head_ent;
2241 strbuf_release(&head_ent->new_dir);
2242 string_list_append(&remove_from_merge,
2243 merge_ent->dir)->util = merge_ent;
2244 strbuf_release(&merge_ent->new_dir);
2248 remove_hashmap_entries(dir_re_head, &remove_from_head);
2249 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2252 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2254 struct hashmap *dir_renames;
2255 struct hashmap_iter iter;
2256 struct dir_rename_entry *entry;
2257 int i;
2260 * Typically, we think of a directory rename as all files from a
2261 * certain directory being moved to a target directory. However,
2262 * what if someone first moved two files from the original
2263 * directory in one commit, and then renamed the directory
2264 * somewhere else in a later commit? At merge time, we just know
2265 * that files from the original directory went to two different
2266 * places, and that the bulk of them ended up in the same place.
2267 * We want each directory rename to represent where the bulk of the
2268 * files from that directory end up; this function exists to find
2269 * where the bulk of the files went.
2271 * The first loop below simply iterates through the list of file
2272 * renames, finding out how often each directory rename pair
2273 * possibility occurs.
2275 dir_renames = xmalloc(sizeof(*dir_renames));
2276 dir_rename_init(dir_renames);
2277 for (i = 0; i < pairs->nr; ++i) {
2278 struct string_list_item *item;
2279 int *count;
2280 struct diff_filepair *pair = pairs->queue[i];
2281 char *old_dir, *new_dir;
2283 /* File not part of directory rename if it wasn't renamed */
2284 if (pair->status != 'R')
2285 continue;
2287 get_renamed_dir_portion(pair->one->path, pair->two->path,
2288 &old_dir, &new_dir);
2289 if (!old_dir)
2290 /* Directory didn't change at all; ignore this one. */
2291 continue;
2293 entry = dir_rename_find_entry(dir_renames, old_dir);
2294 if (!entry) {
2295 entry = xmalloc(sizeof(*entry));
2296 dir_rename_entry_init(entry, old_dir);
2297 hashmap_put(dir_renames, &entry->ent);
2298 } else {
2299 free(old_dir);
2301 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2302 if (!item) {
2303 item = string_list_insert(&entry->possible_new_dirs,
2304 new_dir);
2305 item->util = xcalloc(1, sizeof(int));
2306 } else {
2307 free(new_dir);
2309 count = item->util;
2310 *count += 1;
2314 * For each directory with files moved out of it, we find out which
2315 * target directory received the most files so we can declare it to
2316 * be the "winning" target location for the directory rename. This
2317 * winner gets recorded in new_dir. If there is no winner
2318 * (multiple target directories received the same number of files),
2319 * we set non_unique_new_dir. Once we've determined the winner (or
2320 * that there is no winner), we no longer need possible_new_dirs.
2322 hashmap_for_each_entry(dir_renames, &iter, entry,
2323 ent /* member name */) {
2324 int max = 0;
2325 int bad_max = 0;
2326 char *best = NULL;
2328 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2329 int *count = entry->possible_new_dirs.items[i].util;
2331 if (*count == max)
2332 bad_max = max;
2333 else if (*count > max) {
2334 max = *count;
2335 best = entry->possible_new_dirs.items[i].string;
2338 if (bad_max == max)
2339 entry->non_unique_new_dir = 1;
2340 else {
2341 assert(entry->new_dir.len == 0);
2342 strbuf_addstr(&entry->new_dir, best);
2345 * The relevant directory sub-portion of the original full
2346 * filepaths were xstrndup'ed before inserting into
2347 * possible_new_dirs, and instead of manually iterating the
2348 * list and free'ing each, just lie and tell
2349 * possible_new_dirs that it did the strdup'ing so that it
2350 * will free them for us.
2352 entry->possible_new_dirs.strdup_strings = 1;
2353 string_list_clear(&entry->possible_new_dirs, 1);
2356 return dir_renames;
2359 static struct dir_rename_entry *check_dir_renamed(const char *path,
2360 struct hashmap *dir_renames)
2362 char *temp = xstrdup(path);
2363 char *end;
2364 struct dir_rename_entry *entry = NULL;
2366 while ((end = strrchr(temp, '/'))) {
2367 *end = '\0';
2368 entry = dir_rename_find_entry(dir_renames, temp);
2369 if (entry)
2370 break;
2372 free(temp);
2373 return entry;
2376 static void compute_collisions(struct hashmap *collisions,
2377 struct hashmap *dir_renames,
2378 struct diff_queue_struct *pairs)
2380 int i;
2383 * Multiple files can be mapped to the same path due to directory
2384 * renames done by the other side of history. Since that other
2385 * side of history could have merged multiple directories into one,
2386 * if our side of history added the same file basename to each of
2387 * those directories, then all N of them would get implicitly
2388 * renamed by the directory rename detection into the same path,
2389 * and we'd get an add/add/.../add conflict, and all those adds
2390 * from *this* side of history. This is not representable in the
2391 * index, and users aren't going to easily be able to make sense of
2392 * it. So we need to provide a good warning about what's
2393 * happening, and fall back to no-directory-rename detection
2394 * behavior for those paths.
2396 * See testcases 9e and all of section 5 from t6043 for examples.
2398 collision_init(collisions);
2400 for (i = 0; i < pairs->nr; ++i) {
2401 struct dir_rename_entry *dir_rename_ent;
2402 struct collision_entry *collision_ent;
2403 char *new_path;
2404 struct diff_filepair *pair = pairs->queue[i];
2406 if (pair->status != 'A' && pair->status != 'R')
2407 continue;
2408 dir_rename_ent = check_dir_renamed(pair->two->path,
2409 dir_renames);
2410 if (!dir_rename_ent)
2411 continue;
2413 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2414 if (!new_path)
2416 * dir_rename_ent->non_unique_new_path is true, which
2417 * means there is no directory rename for us to use,
2418 * which means it won't cause us any additional
2419 * collisions.
2421 continue;
2422 collision_ent = collision_find_entry(collisions, new_path);
2423 if (!collision_ent) {
2424 CALLOC_ARRAY(collision_ent, 1);
2425 hashmap_entry_init(&collision_ent->ent,
2426 strhash(new_path));
2427 hashmap_put(collisions, &collision_ent->ent);
2428 collision_ent->target_file = new_path;
2429 } else {
2430 free(new_path);
2432 string_list_insert(&collision_ent->source_files,
2433 pair->two->path);
2437 static char *check_for_directory_rename(struct merge_options *opt,
2438 const char *path,
2439 struct tree *tree,
2440 struct hashmap *dir_renames,
2441 struct hashmap *dir_rename_exclusions,
2442 struct hashmap *collisions,
2443 int *clean_merge)
2445 char *new_path = NULL;
2446 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2447 struct dir_rename_entry *oentry = NULL;
2449 if (!entry)
2450 return new_path;
2453 * This next part is a little weird. We do not want to do an
2454 * implicit rename into a directory we renamed on our side, because
2455 * that will result in a spurious rename/rename(1to2) conflict. An
2456 * example:
2457 * Base commit: dumbdir/afile, otherdir/bfile
2458 * Side 1: smrtdir/afile, otherdir/bfile
2459 * Side 2: dumbdir/afile, dumbdir/bfile
2460 * Here, while working on Side 1, we could notice that otherdir was
2461 * renamed/merged to dumbdir, and change the diff_filepair for
2462 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2463 * 2 will notice the rename from dumbdir to smrtdir, and do the
2464 * transitive rename to move it from dumbdir/bfile to
2465 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2466 * smrtdir, a rename/rename(1to2) conflict. We really just want
2467 * the file to end up in smrtdir. And the way to achieve that is
2468 * to not let Side1 do the rename to dumbdir, since we know that is
2469 * the source of one of our directory renames.
2471 * That's why oentry and dir_rename_exclusions is here.
2473 * As it turns out, this also prevents N-way transient rename
2474 * confusion; See testcases 9c and 9d of t6043.
2476 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2477 if (oentry) {
2478 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2479 "to %s, because %s itself was renamed."),
2480 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2481 } else {
2482 new_path = handle_path_level_conflicts(opt, path, entry,
2483 collisions, tree);
2484 *clean_merge &= (new_path != NULL);
2487 return new_path;
2490 static void apply_directory_rename_modifications(struct merge_options *opt,
2491 struct diff_filepair *pair,
2492 char *new_path,
2493 struct rename *re,
2494 struct tree *tree,
2495 struct tree *o_tree,
2496 struct tree *a_tree,
2497 struct tree *b_tree,
2498 struct string_list *entries)
2500 struct string_list_item *item;
2501 int stage = (tree == a_tree ? 2 : 3);
2502 int update_wd;
2505 * In all cases where we can do directory rename detection,
2506 * unpack_trees() will have read pair->two->path into the
2507 * index and the working copy. We need to remove it so that
2508 * we can instead place it at new_path. It is guaranteed to
2509 * not be untracked (unpack_trees() would have errored out
2510 * saying the file would have been overwritten), but it might
2511 * be dirty, though.
2513 update_wd = !was_dirty(opt, pair->two->path);
2514 if (!update_wd)
2515 output(opt, 1, _("Refusing to lose dirty file at %s"),
2516 pair->two->path);
2517 remove_file(opt, 1, pair->two->path, !update_wd);
2519 /* Find or create a new re->dst_entry */
2520 item = string_list_lookup(entries, new_path);
2521 if (item) {
2523 * Since we're renaming on this side of history, and it's
2524 * due to a directory rename on the other side of history
2525 * (which we only allow when the directory in question no
2526 * longer exists on the other side of history), the
2527 * original entry for re->dst_entry is no longer
2528 * necessary...
2530 re->dst_entry->processed = 1;
2533 * ...because we'll be using this new one.
2535 re->dst_entry = item->util;
2536 } else {
2538 * re->dst_entry is for the before-dir-rename path, and we
2539 * need it to hold information for the after-dir-rename
2540 * path. Before creating a new entry, we need to mark the
2541 * old one as unnecessary (...unless it is shared by
2542 * src_entry, i.e. this didn't use to be a rename, in which
2543 * case we can just allow the normal processing to happen
2544 * for it).
2546 if (pair->status == 'R')
2547 re->dst_entry->processed = 1;
2549 re->dst_entry = insert_stage_data(opt->repo, new_path,
2550 o_tree, a_tree, b_tree,
2551 entries);
2552 item = string_list_insert(entries, new_path);
2553 item->util = re->dst_entry;
2557 * Update the stage_data with the information about the path we are
2558 * moving into place. That slot will be empty and available for us
2559 * to write to because of the collision checks in
2560 * handle_path_level_conflicts(). In other words,
2561 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2562 * open for us to write to.
2564 * It may be tempting to actually update the index at this point as
2565 * well, using update_stages_for_stage_data(), but as per the big
2566 * "NOTE" in update_stages(), doing so will modify the current
2567 * in-memory index which will break calls to would_lose_untracked()
2568 * that we need to make. Instead, we need to just make sure that
2569 * the various handle_rename_*() functions update the index
2570 * explicitly rather than relying on unpack_trees() to have done it.
2572 get_tree_entry(opt->repo,
2573 &tree->object.oid,
2574 pair->two->path,
2575 &re->dst_entry->stages[stage].oid,
2576 &re->dst_entry->stages[stage].mode);
2579 * Record the original change status (or 'type' of change). If it
2580 * was originally an add ('A'), this lets us differentiate later
2581 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2582 * otherwise look the same). If it was originally a rename ('R'),
2583 * this lets us remember and report accurately about the transitive
2584 * renaming that occurred via the directory rename detection. Also,
2585 * record the original destination name.
2587 re->dir_rename_original_type = pair->status;
2588 re->dir_rename_original_dest = pair->two->path;
2591 * We don't actually look at pair->status again, but it seems
2592 * pedagogically correct to adjust it.
2594 pair->status = 'R';
2597 * Finally, record the new location.
2599 pair->two->path = new_path;
2603 * Get information of all renames which occurred in 'pairs', making use of
2604 * any implicit directory renames inferred from the other side of history.
2605 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2606 * to be able to associate the correct cache entries with the rename
2607 * information; tree is always equal to either a_tree or b_tree.
2609 static struct string_list *get_renames(struct merge_options *opt,
2610 const char *branch,
2611 struct diff_queue_struct *pairs,
2612 struct hashmap *dir_renames,
2613 struct hashmap *dir_rename_exclusions,
2614 struct tree *tree,
2615 struct tree *o_tree,
2616 struct tree *a_tree,
2617 struct tree *b_tree,
2618 struct string_list *entries,
2619 int *clean_merge)
2621 int i;
2622 struct hashmap collisions;
2623 struct hashmap_iter iter;
2624 struct collision_entry *e;
2625 struct string_list *renames;
2627 compute_collisions(&collisions, dir_renames, pairs);
2628 CALLOC_ARRAY(renames, 1);
2630 for (i = 0; i < pairs->nr; ++i) {
2631 struct string_list_item *item;
2632 struct rename *re;
2633 struct diff_filepair *pair = pairs->queue[i];
2634 char *new_path; /* non-NULL only with directory renames */
2636 if (pair->status != 'A' && pair->status != 'R') {
2637 diff_free_filepair(pair);
2638 continue;
2640 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2641 dir_renames,
2642 dir_rename_exclusions,
2643 &collisions,
2644 clean_merge);
2645 if (pair->status != 'R' && !new_path) {
2646 diff_free_filepair(pair);
2647 continue;
2650 re = xmalloc(sizeof(*re));
2651 re->processed = 0;
2652 re->pair = pair;
2653 re->branch = branch;
2654 re->dir_rename_original_type = '\0';
2655 re->dir_rename_original_dest = NULL;
2656 item = string_list_lookup(entries, re->pair->one->path);
2657 if (!item)
2658 re->src_entry = insert_stage_data(opt->repo,
2659 re->pair->one->path,
2660 o_tree, a_tree, b_tree, entries);
2661 else
2662 re->src_entry = item->util;
2664 item = string_list_lookup(entries, re->pair->two->path);
2665 if (!item)
2666 re->dst_entry = insert_stage_data(opt->repo,
2667 re->pair->two->path,
2668 o_tree, a_tree, b_tree, entries);
2669 else
2670 re->dst_entry = item->util;
2671 item = string_list_insert(renames, pair->one->path);
2672 item->util = re;
2673 if (new_path)
2674 apply_directory_rename_modifications(opt, pair, new_path,
2675 re, tree, o_tree,
2676 a_tree, b_tree,
2677 entries);
2680 hashmap_for_each_entry(&collisions, &iter, e,
2681 ent /* member name */) {
2682 free(e->target_file);
2683 string_list_clear(&e->source_files, 0);
2685 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2686 return renames;
2689 static int process_renames(struct merge_options *opt,
2690 struct string_list *a_renames,
2691 struct string_list *b_renames)
2693 int clean_merge = 1, i, j;
2694 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2695 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2696 const struct rename *sre;
2699 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2700 * string_list one-by-one, but O(n log n) to build it unsorted and
2701 * then sort it. Note that as we build the list, we do not need to
2702 * check if the existing destination path is already in the list,
2703 * because the structure of diffcore_rename guarantees we won't
2704 * have duplicates.
2706 for (i = 0; i < a_renames->nr; i++) {
2707 sre = a_renames->items[i].util;
2708 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2709 = (void *)sre;
2711 for (i = 0; i < b_renames->nr; i++) {
2712 sre = b_renames->items[i].util;
2713 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2714 = (void *)sre;
2717 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2718 struct string_list *renames1, *renames2Dst;
2719 struct rename *ren1 = NULL, *ren2 = NULL;
2720 const char *ren1_src, *ren1_dst;
2721 struct string_list_item *lookup;
2723 if (i >= a_renames->nr) {
2724 ren2 = b_renames->items[j++].util;
2725 } else if (j >= b_renames->nr) {
2726 ren1 = a_renames->items[i++].util;
2727 } else {
2728 int compare = strcmp(a_renames->items[i].string,
2729 b_renames->items[j].string);
2730 if (compare <= 0)
2731 ren1 = a_renames->items[i++].util;
2732 if (compare >= 0)
2733 ren2 = b_renames->items[j++].util;
2736 /* TODO: refactor, so that 1/2 are not needed */
2737 if (ren1) {
2738 renames1 = a_renames;
2739 renames2Dst = &b_by_dst;
2740 } else {
2741 renames1 = b_renames;
2742 renames2Dst = &a_by_dst;
2743 SWAP(ren2, ren1);
2746 if (ren1->processed)
2747 continue;
2748 ren1->processed = 1;
2749 ren1->dst_entry->processed = 1;
2750 /* BUG: We should only mark src_entry as processed if we
2751 * are not dealing with a rename + add-source case.
2753 ren1->src_entry->processed = 1;
2755 ren1_src = ren1->pair->one->path;
2756 ren1_dst = ren1->pair->two->path;
2758 if (ren2) {
2759 /* One file renamed on both sides */
2760 const char *ren2_src = ren2->pair->one->path;
2761 const char *ren2_dst = ren2->pair->two->path;
2762 enum rename_type rename_type;
2763 if (strcmp(ren1_src, ren2_src) != 0)
2764 BUG("ren1_src != ren2_src");
2765 ren2->dst_entry->processed = 1;
2766 ren2->processed = 1;
2767 if (strcmp(ren1_dst, ren2_dst) != 0) {
2768 rename_type = RENAME_ONE_FILE_TO_TWO;
2769 clean_merge = 0;
2770 } else {
2771 rename_type = RENAME_ONE_FILE_TO_ONE;
2772 /* BUG: We should only remove ren1_src in
2773 * the base stage (think of rename +
2774 * add-source cases).
2776 remove_file(opt, 1, ren1_src, 1);
2777 update_entry(ren1->dst_entry,
2778 ren1->pair->one,
2779 ren1->pair->two,
2780 ren2->pair->two);
2782 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2783 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2784 /* Two different files renamed to the same thing */
2785 char *ren2_dst;
2786 ren2 = lookup->util;
2787 ren2_dst = ren2->pair->two->path;
2788 if (strcmp(ren1_dst, ren2_dst) != 0)
2789 BUG("ren1_dst != ren2_dst");
2791 clean_merge = 0;
2792 ren2->processed = 1;
2794 * BUG: We should only mark src_entry as processed
2795 * if we are not dealing with a rename + add-source
2796 * case.
2798 ren2->src_entry->processed = 1;
2800 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2801 opt, ren1, ren2);
2802 } else {
2803 /* Renamed in 1, maybe changed in 2 */
2804 /* we only use sha1 and mode of these */
2805 struct diff_filespec src_other, dst_other;
2806 int try_merge;
2809 * unpack_trees loads entries from common-commit
2810 * into stage 1, from head-commit into stage 2, and
2811 * from merge-commit into stage 3. We keep track
2812 * of which side corresponds to the rename.
2814 int renamed_stage = a_renames == renames1 ? 2 : 3;
2815 int other_stage = a_renames == renames1 ? 3 : 2;
2818 * Directory renames have a funny corner case...
2820 int renamed_to_self = !strcmp(ren1_src, ren1_dst);
2822 /* BUG: We should only remove ren1_src in the base
2823 * stage and in other_stage (think of rename +
2824 * add-source case).
2826 if (!renamed_to_self)
2827 remove_file(opt, 1, ren1_src,
2828 renamed_stage == 2 ||
2829 !was_tracked(opt, ren1_src));
2831 oidcpy(&src_other.oid,
2832 &ren1->src_entry->stages[other_stage].oid);
2833 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2834 oidcpy(&dst_other.oid,
2835 &ren1->dst_entry->stages[other_stage].oid);
2836 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2837 try_merge = 0;
2839 if (oideq(&src_other.oid, null_oid()) &&
2840 ren1->dir_rename_original_type == 'A') {
2841 setup_rename_conflict_info(RENAME_VIA_DIR,
2842 opt, ren1, NULL);
2843 } else if (renamed_to_self) {
2844 setup_rename_conflict_info(RENAME_NORMAL,
2845 opt, ren1, NULL);
2846 } else if (oideq(&src_other.oid, null_oid())) {
2847 setup_rename_conflict_info(RENAME_DELETE,
2848 opt, ren1, NULL);
2849 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2850 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2852 * Added file on the other side identical to
2853 * the file being renamed: clean merge.
2854 * Also, there is no need to overwrite the
2855 * file already in the working copy, so call
2856 * update_file_flags() instead of
2857 * update_file().
2859 if (update_file_flags(opt,
2860 ren1->pair->two,
2861 ren1_dst,
2862 1, /* update_cache */
2863 0 /* update_wd */))
2864 clean_merge = -1;
2865 } else if (!oideq(&dst_other.oid, null_oid())) {
2867 * Probably not a clean merge, but it's
2868 * premature to set clean_merge to 0 here,
2869 * because if the rename merges cleanly and
2870 * the merge exactly matches the newly added
2871 * file, then the merge will be clean.
2873 setup_rename_conflict_info(RENAME_ADD,
2874 opt, ren1, NULL);
2875 } else
2876 try_merge = 1;
2878 if (clean_merge < 0)
2879 goto cleanup_and_return;
2880 if (try_merge) {
2881 struct diff_filespec *o, *a, *b;
2882 src_other.path = (char *)ren1_src;
2884 o = ren1->pair->one;
2885 if (a_renames == renames1) {
2886 a = ren1->pair->two;
2887 b = &src_other;
2888 } else {
2889 b = ren1->pair->two;
2890 a = &src_other;
2892 update_entry(ren1->dst_entry, o, a, b);
2893 setup_rename_conflict_info(RENAME_NORMAL,
2894 opt, ren1, NULL);
2898 cleanup_and_return:
2899 string_list_clear(&a_by_dst, 0);
2900 string_list_clear(&b_by_dst, 0);
2902 return clean_merge;
2905 struct rename_info {
2906 struct string_list *head_renames;
2907 struct string_list *merge_renames;
2910 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2911 struct hashmap *dir_renames)
2913 struct hashmap_iter iter;
2914 struct dir_rename_entry *e;
2916 hashmap_for_each_entry(dir_renames, &iter, e,
2917 ent /* member name */) {
2918 free(e->dir);
2919 strbuf_release(&e->new_dir);
2920 /* possible_new_dirs already cleared in get_directory_renames */
2922 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2923 free(dir_renames);
2925 free(pairs->queue);
2926 free(pairs);
2929 static int detect_and_process_renames(struct merge_options *opt,
2930 struct tree *common,
2931 struct tree *head,
2932 struct tree *merge,
2933 struct string_list *entries,
2934 struct rename_info *ri)
2936 struct diff_queue_struct *head_pairs, *merge_pairs;
2937 struct hashmap *dir_re_head, *dir_re_merge;
2938 int clean = 1;
2940 ri->head_renames = NULL;
2941 ri->merge_renames = NULL;
2943 if (!merge_detect_rename(opt))
2944 return 1;
2946 head_pairs = get_diffpairs(opt, common, head);
2947 merge_pairs = get_diffpairs(opt, common, merge);
2949 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2950 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2951 !opt->priv->call_depth)) {
2952 dir_re_head = get_directory_renames(head_pairs);
2953 dir_re_merge = get_directory_renames(merge_pairs);
2955 handle_directory_level_conflicts(opt,
2956 dir_re_head, head,
2957 dir_re_merge, merge);
2958 } else {
2959 dir_re_head = xmalloc(sizeof(*dir_re_head));
2960 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2961 dir_rename_init(dir_re_head);
2962 dir_rename_init(dir_re_merge);
2965 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2966 dir_re_merge, dir_re_head, head,
2967 common, head, merge, entries,
2968 &clean);
2969 if (clean < 0)
2970 goto cleanup;
2971 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2972 dir_re_head, dir_re_merge, merge,
2973 common, head, merge, entries,
2974 &clean);
2975 if (clean < 0)
2976 goto cleanup;
2977 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2979 cleanup:
2981 * Some cleanup is deferred until cleanup_renames() because the
2982 * data structures are still needed and referenced in
2983 * process_entry(). But there are a few things we can free now.
2985 initial_cleanup_rename(head_pairs, dir_re_head);
2986 initial_cleanup_rename(merge_pairs, dir_re_merge);
2988 return clean;
2991 static void final_cleanup_rename(struct string_list *rename)
2993 const struct rename *re;
2994 int i;
2996 if (rename == NULL)
2997 return;
2999 for (i = 0; i < rename->nr; i++) {
3000 re = rename->items[i].util;
3001 diff_free_filepair(re->pair);
3003 string_list_clear(rename, 1);
3004 free(rename);
3007 static void final_cleanup_renames(struct rename_info *re_info)
3009 final_cleanup_rename(re_info->head_renames);
3010 final_cleanup_rename(re_info->merge_renames);
3013 static int read_oid_strbuf(struct merge_options *opt,
3014 const struct object_id *oid,
3015 struct strbuf *dst)
3017 void *buf;
3018 enum object_type type;
3019 unsigned long size;
3020 buf = read_object_file(oid, &type, &size);
3021 if (!buf)
3022 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3023 if (type != OBJ_BLOB) {
3024 free(buf);
3025 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3027 strbuf_attach(dst, buf, size, size + 1);
3028 return 0;
3031 static int blob_unchanged(struct merge_options *opt,
3032 const struct diff_filespec *o,
3033 const struct diff_filespec *a,
3034 int renormalize, const char *path)
3036 struct strbuf obuf = STRBUF_INIT;
3037 struct strbuf abuf = STRBUF_INIT;
3038 int ret = 0; /* assume changed for safety */
3039 struct index_state *idx = opt->repo->index;
3041 if (a->mode != o->mode)
3042 return 0;
3043 if (oideq(&o->oid, &a->oid))
3044 return 1;
3045 if (!renormalize)
3046 return 0;
3048 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3049 read_oid_strbuf(opt, &a->oid, &abuf))
3050 goto error_return;
3052 * Note: binary | is used so that both renormalizations are
3053 * performed. Comparison can be skipped if both files are
3054 * unchanged since their sha1s have already been compared.
3056 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3057 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3058 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3060 error_return:
3061 strbuf_release(&obuf);
3062 strbuf_release(&abuf);
3063 return ret;
3066 static int handle_modify_delete(struct merge_options *opt,
3067 const char *path,
3068 const struct diff_filespec *o,
3069 const struct diff_filespec *a,
3070 const struct diff_filespec *b)
3072 const char *modify_branch, *delete_branch;
3073 const struct diff_filespec *changed;
3075 if (is_valid(a)) {
3076 modify_branch = opt->branch1;
3077 delete_branch = opt->branch2;
3078 changed = a;
3079 } else {
3080 modify_branch = opt->branch2;
3081 delete_branch = opt->branch1;
3082 changed = b;
3085 return handle_change_delete(opt,
3086 path, NULL,
3087 o, changed,
3088 modify_branch, delete_branch,
3089 _("modify"), _("modified"));
3092 static int handle_content_merge(struct merge_file_info *mfi,
3093 struct merge_options *opt,
3094 const char *path,
3095 int is_dirty,
3096 const struct diff_filespec *o,
3097 const struct diff_filespec *a,
3098 const struct diff_filespec *b,
3099 struct rename_conflict_info *ci)
3101 const char *reason = _("content");
3102 unsigned df_conflict_remains = 0;
3104 if (!is_valid(o))
3105 reason = _("add/add");
3107 assert(o->path && a->path && b->path);
3108 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3109 S_ISGITLINK(ci->ren1->pair->two->mode)))
3110 df_conflict_remains = 1;
3112 if (merge_mode_and_contents(opt, o, a, b, path,
3113 opt->branch1, opt->branch2,
3114 opt->priv->call_depth * 2, mfi))
3115 return -1;
3118 * We can skip updating the working tree file iff:
3119 * a) The merge is clean
3120 * b) The merge matches what was in HEAD (content, mode, pathname)
3121 * c) The target path is usable (i.e. not involved in D/F conflict)
3123 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3124 !df_conflict_remains) {
3125 int pos;
3126 struct cache_entry *ce;
3128 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3129 if (add_cacheinfo(opt, &mfi->blob, path,
3130 0, (!opt->priv->call_depth && !is_dirty), 0))
3131 return -1;
3133 * However, add_cacheinfo() will delete the old cache entry
3134 * and add a new one. We need to copy over any skip_worktree
3135 * flag to avoid making the file appear as if it were
3136 * deleted by the user.
3138 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3139 ce = opt->priv->orig_index.cache[pos];
3140 if (ce_skip_worktree(ce)) {
3141 pos = index_name_pos(opt->repo->index, path, strlen(path));
3142 ce = opt->repo->index->cache[pos];
3143 ce->ce_flags |= CE_SKIP_WORKTREE;
3145 return mfi->clean;
3148 if (!mfi->clean) {
3149 if (S_ISGITLINK(mfi->blob.mode))
3150 reason = _("submodule");
3151 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3152 reason, path);
3153 if (ci && !df_conflict_remains)
3154 if (update_stages(opt, path, o, a, b))
3155 return -1;
3158 if (df_conflict_remains || is_dirty) {
3159 char *new_path;
3160 if (opt->priv->call_depth) {
3161 remove_file_from_index(opt->repo->index, path);
3162 } else {
3163 if (!mfi->clean) {
3164 if (update_stages(opt, path, o, a, b))
3165 return -1;
3166 } else {
3167 int file_from_stage2 = was_tracked(opt, path);
3169 if (update_stages(opt, path, NULL,
3170 file_from_stage2 ? &mfi->blob : NULL,
3171 file_from_stage2 ? NULL : &mfi->blob))
3172 return -1;
3176 new_path = unique_path(opt, path, ci->ren1->branch);
3177 if (is_dirty) {
3178 output(opt, 1, _("Refusing to lose dirty file at %s"),
3179 path);
3181 output(opt, 1, _("Adding as %s instead"), new_path);
3182 if (update_file(opt, 0, &mfi->blob, new_path)) {
3183 free(new_path);
3184 return -1;
3186 free(new_path);
3187 mfi->clean = 0;
3188 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3189 return -1;
3190 return !is_dirty && mfi->clean;
3193 static int handle_rename_normal(struct merge_options *opt,
3194 const char *path,
3195 const struct diff_filespec *o,
3196 const struct diff_filespec *a,
3197 const struct diff_filespec *b,
3198 struct rename_conflict_info *ci)
3200 struct rename *ren = ci->ren1;
3201 struct merge_file_info mfi;
3202 int clean;
3204 /* Merge the content and write it out */
3205 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3206 o, a, b, ci);
3208 if (clean &&
3209 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3210 ren->dir_rename_original_dest) {
3211 if (update_stages(opt, path,
3212 &mfi.blob, &mfi.blob, &mfi.blob))
3213 return -1;
3214 clean = 0; /* not clean, but conflicted */
3216 return clean;
3219 static void dir_rename_warning(const char *msg,
3220 int is_add,
3221 int clean,
3222 struct merge_options *opt,
3223 struct rename *ren)
3225 const char *other_branch;
3226 other_branch = (ren->branch == opt->branch1 ?
3227 opt->branch2 : opt->branch1);
3228 if (is_add) {
3229 output(opt, clean ? 2 : 1, msg,
3230 ren->pair->one->path, ren->branch,
3231 other_branch, ren->pair->two->path);
3232 return;
3234 output(opt, clean ? 2 : 1, msg,
3235 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3236 other_branch, ren->pair->two->path);
3238 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3239 struct rename *ren)
3241 const char *msg;
3242 int clean = 1, is_add;
3244 if (!ren)
3245 return clean;
3247 /* Return early if ren was not affected/created by a directory rename */
3248 if (!ren->dir_rename_original_dest)
3249 return clean;
3251 /* Sanity checks */
3252 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3253 assert(ren->dir_rename_original_type == 'A' ||
3254 ren->dir_rename_original_type == 'R');
3256 /* Check whether to treat directory renames as a conflict */
3257 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3259 is_add = (ren->dir_rename_original_type == 'A');
3260 if (ren->dir_rename_original_type == 'A' && clean) {
3261 msg = _("Path updated: %s added in %s inside a "
3262 "directory that was renamed in %s; moving it to %s.");
3263 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3264 msg = _("CONFLICT (file location): %s added in %s "
3265 "inside a directory that was renamed in %s, "
3266 "suggesting it should perhaps be moved to %s.");
3267 } else if (ren->dir_rename_original_type == 'R' && clean) {
3268 msg = _("Path updated: %s renamed to %s in %s, inside a "
3269 "directory that was renamed in %s; moving it to %s.");
3270 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3271 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3272 "inside a directory that was renamed in %s, "
3273 "suggesting it should perhaps be moved to %s.");
3274 } else {
3275 BUG("Impossible dir_rename_original_type/clean combination");
3277 dir_rename_warning(msg, is_add, clean, opt, ren);
3279 return clean;
3282 /* Per entry merge function */
3283 static int process_entry(struct merge_options *opt,
3284 const char *path, struct stage_data *entry)
3286 int clean_merge = 1;
3287 int normalize = opt->renormalize;
3289 struct diff_filespec *o = &entry->stages[1];
3290 struct diff_filespec *a = &entry->stages[2];
3291 struct diff_filespec *b = &entry->stages[3];
3292 int o_valid = is_valid(o);
3293 int a_valid = is_valid(a);
3294 int b_valid = is_valid(b);
3295 o->path = a->path = b->path = (char*)path;
3297 entry->processed = 1;
3298 if (entry->rename_conflict_info) {
3299 struct rename_conflict_info *ci = entry->rename_conflict_info;
3300 struct diff_filespec *temp;
3301 int path_clean;
3303 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3304 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3307 * For cases with a single rename, {o,a,b}->path have all been
3308 * set to the rename target path; we need to set two of these
3309 * back to the rename source.
3310 * For rename/rename conflicts, we'll manually fix paths below.
3312 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3313 o->path = temp->path = ci->ren1->pair->one->path;
3314 if (ci->ren2) {
3315 assert(opt->branch1 == ci->ren1->branch);
3318 switch (ci->rename_type) {
3319 case RENAME_NORMAL:
3320 case RENAME_ONE_FILE_TO_ONE:
3321 clean_merge = handle_rename_normal(opt, path, o, a, b,
3322 ci);
3323 break;
3324 case RENAME_VIA_DIR:
3325 clean_merge = handle_rename_via_dir(opt, ci);
3326 break;
3327 case RENAME_ADD:
3329 * Probably unclean merge, but if the renamed file
3330 * merges cleanly and the result can then be
3331 * two-way merged cleanly with the added file, I
3332 * guess it's a clean merge?
3334 clean_merge = handle_rename_add(opt, ci);
3335 break;
3336 case RENAME_DELETE:
3337 clean_merge = 0;
3338 if (handle_rename_delete(opt, ci))
3339 clean_merge = -1;
3340 break;
3341 case RENAME_ONE_FILE_TO_TWO:
3343 * Manually fix up paths; note:
3344 * ren[12]->pair->one->path are equal.
3346 o->path = ci->ren1->pair->one->path;
3347 a->path = ci->ren1->pair->two->path;
3348 b->path = ci->ren2->pair->two->path;
3350 clean_merge = 0;
3351 if (handle_rename_rename_1to2(opt, ci))
3352 clean_merge = -1;
3353 break;
3354 case RENAME_TWO_FILES_TO_ONE:
3356 * Manually fix up paths; note,
3357 * ren[12]->pair->two->path are actually equal.
3359 o->path = NULL;
3360 a->path = ci->ren1->pair->two->path;
3361 b->path = ci->ren2->pair->two->path;
3364 * Probably unclean merge, but if the two renamed
3365 * files merge cleanly and the two resulting files
3366 * can then be two-way merged cleanly, I guess it's
3367 * a clean merge?
3369 clean_merge = handle_rename_rename_2to1(opt, ci);
3370 break;
3371 default:
3372 entry->processed = 0;
3373 break;
3375 if (path_clean < clean_merge)
3376 clean_merge = path_clean;
3377 } else if (o_valid && (!a_valid || !b_valid)) {
3378 /* Case A: Deleted in one */
3379 if ((!a_valid && !b_valid) ||
3380 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3381 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3382 /* Deleted in both or deleted in one and
3383 * unchanged in the other */
3384 if (a_valid)
3385 output(opt, 2, _("Removing %s"), path);
3386 /* do not touch working file if it did not exist */
3387 remove_file(opt, 1, path, !a_valid);
3388 } else {
3389 /* Modify/delete; deleted side may have put a directory in the way */
3390 clean_merge = 0;
3391 if (handle_modify_delete(opt, path, o, a, b))
3392 clean_merge = -1;
3394 } else if ((!o_valid && a_valid && !b_valid) ||
3395 (!o_valid && !a_valid && b_valid)) {
3396 /* Case B: Added in one. */
3397 /* [nothing|directory] -> ([nothing|directory], file) */
3399 const char *add_branch;
3400 const char *other_branch;
3401 const char *conf;
3402 const struct diff_filespec *contents;
3404 if (a_valid) {
3405 add_branch = opt->branch1;
3406 other_branch = opt->branch2;
3407 contents = a;
3408 conf = _("file/directory");
3409 } else {
3410 add_branch = opt->branch2;
3411 other_branch = opt->branch1;
3412 contents = b;
3413 conf = _("directory/file");
3415 if (dir_in_way(opt->repo->index, path,
3416 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3417 0)) {
3418 char *new_path = unique_path(opt, path, add_branch);
3419 clean_merge = 0;
3420 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3421 "Adding %s as %s"),
3422 conf, path, other_branch, path, new_path);
3423 if (update_file(opt, 0, contents, new_path))
3424 clean_merge = -1;
3425 else if (opt->priv->call_depth)
3426 remove_file_from_index(opt->repo->index, path);
3427 free(new_path);
3428 } else {
3429 output(opt, 2, _("Adding %s"), path);
3430 /* do not overwrite file if already present */
3431 if (update_file_flags(opt, contents, path, 1, !a_valid))
3432 clean_merge = -1;
3434 } else if (a_valid && b_valid) {
3435 if (!o_valid) {
3436 /* Case C: Added in both (check for same permissions) */
3437 output(opt, 1,
3438 _("CONFLICT (add/add): Merge conflict in %s"),
3439 path);
3440 clean_merge = handle_file_collision(opt,
3441 path, NULL, NULL,
3442 opt->branch1,
3443 opt->branch2,
3444 a, b);
3445 } else {
3446 /* case D: Modified in both, but differently. */
3447 struct merge_file_info mfi;
3448 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3449 clean_merge = handle_content_merge(&mfi, opt, path,
3450 is_dirty,
3451 o, a, b, NULL);
3453 } else if (!o_valid && !a_valid && !b_valid) {
3455 * this entry was deleted altogether. a_mode == 0 means
3456 * we had that path and want to actively remove it.
3458 remove_file(opt, 1, path, !a->mode);
3459 } else
3460 BUG("fatal merge failure, shouldn't happen.");
3462 return clean_merge;
3465 static int merge_trees_internal(struct merge_options *opt,
3466 struct tree *head,
3467 struct tree *merge,
3468 struct tree *merge_base,
3469 struct tree **result)
3471 struct index_state *istate = opt->repo->index;
3472 int code, clean;
3474 if (opt->subtree_shift) {
3475 merge = shift_tree_object(opt->repo, head, merge,
3476 opt->subtree_shift);
3477 merge_base = shift_tree_object(opt->repo, head, merge_base,
3478 opt->subtree_shift);
3481 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3482 output(opt, 0, _("Already up to date."));
3483 *result = head;
3484 return 1;
3487 code = unpack_trees_start(opt, merge_base, head, merge);
3489 if (code != 0) {
3490 if (show(opt, 4) || opt->priv->call_depth)
3491 err(opt, _("merging of trees %s and %s failed"),
3492 oid_to_hex(&head->object.oid),
3493 oid_to_hex(&merge->object.oid));
3494 unpack_trees_finish(opt);
3495 return -1;
3498 if (unmerged_index(istate)) {
3499 struct string_list *entries;
3500 struct rename_info re_info;
3501 int i;
3503 * Only need the hashmap while processing entries, so
3504 * initialize it here and free it when we are done running
3505 * through the entries. Keeping it in the merge_options as
3506 * opposed to decaring a local hashmap is for convenience
3507 * so that we don't have to pass it to around.
3509 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3510 NULL, 512);
3511 get_files_dirs(opt, head);
3512 get_files_dirs(opt, merge);
3514 entries = get_unmerged(opt->repo->index);
3515 clean = detect_and_process_renames(opt, merge_base, head, merge,
3516 entries, &re_info);
3517 record_df_conflict_files(opt, entries);
3518 if (clean < 0)
3519 goto cleanup;
3520 for (i = entries->nr-1; 0 <= i; i--) {
3521 const char *path = entries->items[i].string;
3522 struct stage_data *e = entries->items[i].util;
3523 if (!e->processed) {
3524 int ret = process_entry(opt, path, e);
3525 if (!ret)
3526 clean = 0;
3527 else if (ret < 0) {
3528 clean = ret;
3529 goto cleanup;
3533 for (i = 0; i < entries->nr; i++) {
3534 struct stage_data *e = entries->items[i].util;
3535 if (!e->processed)
3536 BUG("unprocessed path??? %s",
3537 entries->items[i].string);
3540 cleanup:
3541 final_cleanup_renames(&re_info);
3543 string_list_clear(entries, 1);
3544 free(entries);
3546 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3547 struct path_hashmap_entry, e);
3549 if (clean < 0) {
3550 unpack_trees_finish(opt);
3551 return clean;
3554 else
3555 clean = 1;
3557 unpack_trees_finish(opt);
3559 if (opt->priv->call_depth &&
3560 !(*result = write_in_core_index_as_tree(opt->repo)))
3561 return -1;
3563 return clean;
3567 * Merge the commits h1 and h2, returning a flag (int) indicating the
3568 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3569 * virtual commit and write its location to *result.
3571 static int merge_recursive_internal(struct merge_options *opt,
3572 struct commit *h1,
3573 struct commit *h2,
3574 struct commit_list *merge_bases,
3575 struct commit **result)
3577 struct commit_list *iter;
3578 struct commit *merged_merge_bases;
3579 struct tree *result_tree;
3580 int clean;
3581 const char *ancestor_name;
3582 struct strbuf merge_base_abbrev = STRBUF_INIT;
3584 if (show(opt, 4)) {
3585 output(opt, 4, _("Merging:"));
3586 output_commit_title(opt, h1);
3587 output_commit_title(opt, h2);
3590 if (!merge_bases) {
3591 merge_bases = get_merge_bases(h1, h2);
3592 merge_bases = reverse_commit_list(merge_bases);
3595 if (show(opt, 5)) {
3596 unsigned cnt = commit_list_count(merge_bases);
3598 output(opt, 5, Q_("found %u common ancestor:",
3599 "found %u common ancestors:", cnt), cnt);
3600 for (iter = merge_bases; iter; iter = iter->next)
3601 output_commit_title(opt, iter->item);
3604 merged_merge_bases = pop_commit(&merge_bases);
3605 if (merged_merge_bases == NULL) {
3606 /* if there is no common ancestor, use an empty tree */
3607 struct tree *tree;
3609 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3610 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3611 "ancestor");
3612 ancestor_name = "empty tree";
3613 } else if (opt->ancestor && !opt->priv->call_depth) {
3614 ancestor_name = opt->ancestor;
3615 } else if (merge_bases) {
3616 ancestor_name = "merged common ancestors";
3617 } else {
3618 strbuf_add_unique_abbrev(&merge_base_abbrev,
3619 &merged_merge_bases->object.oid,
3620 DEFAULT_ABBREV);
3621 ancestor_name = merge_base_abbrev.buf;
3624 for (iter = merge_bases; iter; iter = iter->next) {
3625 const char *saved_b1, *saved_b2;
3626 opt->priv->call_depth++;
3628 * When the merge fails, the result contains files
3629 * with conflict markers. The cleanness flag is
3630 * ignored (unless indicating an error), it was never
3631 * actually used, as result of merge_trees has always
3632 * overwritten it: the committed "conflicts" were
3633 * already resolved.
3635 discard_index(opt->repo->index);
3636 saved_b1 = opt->branch1;
3637 saved_b2 = opt->branch2;
3638 opt->branch1 = "Temporary merge branch 1";
3639 opt->branch2 = "Temporary merge branch 2";
3640 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3641 NULL, &merged_merge_bases) < 0)
3642 return -1;
3643 opt->branch1 = saved_b1;
3644 opt->branch2 = saved_b2;
3645 opt->priv->call_depth--;
3647 if (!merged_merge_bases)
3648 return err(opt, _("merge returned no commit"));
3652 * FIXME: Since merge_recursive_internal() is only ever called by
3653 * places that ensure the index is loaded first
3654 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3655 * case where the merge base was unique that means when we get here
3656 * we immediately discard the index and re-read it, which is a
3657 * complete waste of time. We should only be discarding and
3658 * re-reading if we were forced to recurse.
3660 discard_index(opt->repo->index);
3661 if (!opt->priv->call_depth)
3662 repo_read_index(opt->repo);
3664 opt->ancestor = ancestor_name;
3665 clean = merge_trees_internal(opt,
3666 repo_get_commit_tree(opt->repo, h1),
3667 repo_get_commit_tree(opt->repo, h2),
3668 repo_get_commit_tree(opt->repo,
3669 merged_merge_bases),
3670 &result_tree);
3671 strbuf_release(&merge_base_abbrev);
3672 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3673 if (clean < 0) {
3674 flush_output(opt);
3675 return clean;
3678 if (opt->priv->call_depth) {
3679 *result = make_virtual_commit(opt->repo, result_tree,
3680 "merged tree");
3681 commit_list_insert(h1, &(*result)->parents);
3682 commit_list_insert(h2, &(*result)->parents->next);
3684 return clean;
3687 static int merge_start(struct merge_options *opt, struct tree *head)
3689 struct strbuf sb = STRBUF_INIT;
3691 /* Sanity checks on opt */
3692 assert(opt->repo);
3694 assert(opt->branch1 && opt->branch2);
3696 assert(opt->detect_renames >= -1 &&
3697 opt->detect_renames <= DIFF_DETECT_COPY);
3698 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3699 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3700 assert(opt->rename_limit >= -1);
3701 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3702 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3704 assert(opt->xdl_opts >= 0);
3705 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3706 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3708 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3709 assert(opt->buffer_output <= 2);
3710 assert(opt->obuf.len == 0);
3712 assert(opt->priv == NULL);
3714 /* Sanity check on repo state; index must match head */
3715 if (repo_index_has_changes(opt->repo, head, &sb)) {
3716 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3717 sb.buf);
3718 strbuf_release(&sb);
3719 return -1;
3722 CALLOC_ARRAY(opt->priv, 1);
3723 string_list_init_dup(&opt->priv->df_conflict_file_set);
3724 return 0;
3727 static void merge_finalize(struct merge_options *opt)
3729 flush_output(opt);
3730 if (!opt->priv->call_depth && opt->buffer_output < 2)
3731 strbuf_release(&opt->obuf);
3732 if (show(opt, 2))
3733 diff_warn_rename_limit("merge.renamelimit",
3734 opt->priv->needed_rename_limit, 0);
3735 FREE_AND_NULL(opt->priv);
3738 int merge_trees(struct merge_options *opt,
3739 struct tree *head,
3740 struct tree *merge,
3741 struct tree *merge_base)
3743 int clean;
3744 struct tree *ignored;
3746 assert(opt->ancestor != NULL);
3748 if (merge_start(opt, head))
3749 return -1;
3750 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3751 merge_finalize(opt);
3753 return clean;
3756 int merge_recursive(struct merge_options *opt,
3757 struct commit *h1,
3758 struct commit *h2,
3759 struct commit_list *merge_bases,
3760 struct commit **result)
3762 int clean;
3764 assert(opt->ancestor == NULL ||
3765 !strcmp(opt->ancestor, "constructed merge base"));
3767 prepare_repo_settings(opt->repo);
3768 opt->repo->settings.command_requires_full_index = 1;
3770 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3771 return -1;
3772 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3773 merge_finalize(opt);
3775 return clean;
3778 static struct commit *get_ref(struct repository *repo,
3779 const struct object_id *oid,
3780 const char *name)
3782 struct object *object;
3784 object = deref_tag(repo, parse_object(repo, oid),
3785 name, strlen(name));
3786 if (!object)
3787 return NULL;
3788 if (object->type == OBJ_TREE)
3789 return make_virtual_commit(repo, (struct tree*)object, name);
3790 if (object->type != OBJ_COMMIT)
3791 return NULL;
3792 if (parse_commit((struct commit *)object))
3793 return NULL;
3794 return (struct commit *)object;
3797 int merge_recursive_generic(struct merge_options *opt,
3798 const struct object_id *head,
3799 const struct object_id *merge,
3800 int num_merge_bases,
3801 const struct object_id **merge_bases,
3802 struct commit **result)
3804 int clean;
3805 struct lock_file lock = LOCK_INIT;
3806 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3807 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3808 struct commit_list *ca = NULL;
3810 if (merge_bases) {
3811 int i;
3812 for (i = 0; i < num_merge_bases; ++i) {
3813 struct commit *base;
3814 if (!(base = get_ref(opt->repo, merge_bases[i],
3815 oid_to_hex(merge_bases[i]))))
3816 return err(opt, _("Could not parse object '%s'"),
3817 oid_to_hex(merge_bases[i]));
3818 commit_list_insert(base, &ca);
3820 if (num_merge_bases == 1)
3821 opt->ancestor = "constructed merge base";
3824 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3825 clean = merge_recursive(opt, head_commit, next_commit, ca,
3826 result);
3827 if (clean < 0) {
3828 rollback_lock_file(&lock);
3829 return clean;
3832 if (write_locked_index(opt->repo->index, &lock,
3833 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3834 return err(opt, _("Unable to write index."));
3836 return clean ? 0 : 1;
3839 static void merge_recursive_config(struct merge_options *opt)
3841 char *value = NULL;
3842 int renormalize = 0;
3843 git_config_get_int("merge.verbosity", &opt->verbosity);
3844 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3845 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3846 git_config_get_bool("merge.renormalize", &renormalize);
3847 opt->renormalize = renormalize;
3848 if (!git_config_get_string("diff.renames", &value)) {
3849 opt->detect_renames = git_config_rename("diff.renames", value);
3850 free(value);
3852 if (!git_config_get_string("merge.renames", &value)) {
3853 opt->detect_renames = git_config_rename("merge.renames", value);
3854 free(value);
3856 if (!git_config_get_string("merge.directoryrenames", &value)) {
3857 int boolval = git_parse_maybe_bool(value);
3858 if (0 <= boolval) {
3859 opt->detect_directory_renames = boolval ?
3860 MERGE_DIRECTORY_RENAMES_TRUE :
3861 MERGE_DIRECTORY_RENAMES_NONE;
3862 } else if (!strcasecmp(value, "conflict")) {
3863 opt->detect_directory_renames =
3864 MERGE_DIRECTORY_RENAMES_CONFLICT;
3865 } /* avoid erroring on values from future versions of git */
3866 free(value);
3868 git_config(git_xmerge_config, NULL);
3871 void init_merge_options(struct merge_options *opt,
3872 struct repository *repo)
3874 const char *merge_verbosity;
3875 memset(opt, 0, sizeof(struct merge_options));
3877 opt->repo = repo;
3879 opt->detect_renames = -1;
3880 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3881 opt->rename_limit = -1;
3883 opt->verbosity = 2;
3884 opt->buffer_output = 1;
3885 strbuf_init(&opt->obuf, 0);
3887 opt->renormalize = 0;
3889 merge_recursive_config(opt);
3890 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3891 if (merge_verbosity)
3892 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3893 if (opt->verbosity >= 5)
3894 opt->buffer_output = 0;
3897 int parse_merge_opt(struct merge_options *opt, const char *s)
3899 const char *arg;
3901 if (!s || !*s)
3902 return -1;
3903 if (!strcmp(s, "ours"))
3904 opt->recursive_variant = MERGE_VARIANT_OURS;
3905 else if (!strcmp(s, "theirs"))
3906 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3907 else if (!strcmp(s, "subtree"))
3908 opt->subtree_shift = "";
3909 else if (skip_prefix(s, "subtree=", &arg))
3910 opt->subtree_shift = arg;
3911 else if (!strcmp(s, "patience"))
3912 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3913 else if (!strcmp(s, "histogram"))
3914 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3915 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3916 long value = parse_algorithm_value(arg);
3917 if (value < 0)
3918 return -1;
3919 /* clear out previous settings */
3920 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3921 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3922 opt->xdl_opts |= value;
3924 else if (!strcmp(s, "ignore-space-change"))
3925 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3926 else if (!strcmp(s, "ignore-all-space"))
3927 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3928 else if (!strcmp(s, "ignore-space-at-eol"))
3929 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3930 else if (!strcmp(s, "ignore-cr-at-eol"))
3931 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3932 else if (!strcmp(s, "renormalize"))
3933 opt->renormalize = 1;
3934 else if (!strcmp(s, "no-renormalize"))
3935 opt->renormalize = 0;
3936 else if (!strcmp(s, "no-renames"))
3937 opt->detect_renames = 0;
3938 else if (!strcmp(s, "find-renames")) {
3939 opt->detect_renames = 1;
3940 opt->rename_score = 0;
3942 else if (skip_prefix(s, "find-renames=", &arg) ||
3943 skip_prefix(s, "rename-threshold=", &arg)) {
3944 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3945 return -1;
3946 opt->detect_renames = 1;
3949 * Please update $__git_merge_strategy_options in
3950 * git-completion.bash when you add new options
3952 else
3953 return -1;
3954 return 0;