Merge branch 'tb/multi-pack-reuse-experiment'
[alt-git.git] / merge-recursive.c
bloba0c3e7a2d9105dd895f0ae2613d0f87f00b8d794
1 /*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6 #include "git-compat-util.h"
7 #include "merge-recursive.h"
9 #include "alloc.h"
10 #include "cache-tree.h"
11 #include "commit.h"
12 #include "commit-reach.h"
13 #include "config.h"
14 #include "diff.h"
15 #include "diffcore.h"
16 #include "dir.h"
17 #include "environment.h"
18 #include "gettext.h"
19 #include "hex.h"
20 #include "merge-ll.h"
21 #include "lockfile.h"
22 #include "match-trees.h"
23 #include "name-hash.h"
24 #include "object-file.h"
25 #include "object-name.h"
26 #include "object-store-ll.h"
27 #include "path.h"
28 #include "repository.h"
29 #include "revision.h"
30 #include "sparse-index.h"
31 #include "string-list.h"
32 #include "symlinks.h"
33 #include "tag.h"
34 #include "tree-walk.h"
35 #include "unpack-trees.h"
36 #include "xdiff-interface.h"
38 struct merge_options_internal {
39 int call_depth;
40 int needed_rename_limit;
41 struct hashmap current_file_dir_set;
42 struct string_list df_conflict_file_set;
43 struct unpack_trees_options unpack_opts;
44 struct index_state orig_index;
47 struct path_hashmap_entry {
48 struct hashmap_entry e;
49 char path[FLEX_ARRAY];
52 static int path_hashmap_cmp(const void *cmp_data UNUSED,
53 const struct hashmap_entry *eptr,
54 const struct hashmap_entry *entry_or_key,
55 const void *keydata)
57 const struct path_hashmap_entry *a, *b;
58 const char *key = keydata;
60 a = container_of(eptr, const struct path_hashmap_entry, e);
61 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
63 return fspathcmp(a->path, key ? key : b->path);
67 * For dir_rename_entry, directory names are stored as a full path from the
68 * toplevel of the repository and do not include a trailing '/'. Also:
70 * dir: original name of directory being renamed
71 * non_unique_new_dir: if true, could not determine new_dir
72 * new_dir: final name of directory being renamed
73 * possible_new_dirs: temporary used to help determine new_dir; see comments
74 * in get_directory_renames() for details
76 struct dir_rename_entry {
77 struct hashmap_entry ent;
78 char *dir;
79 unsigned non_unique_new_dir:1;
80 struct strbuf new_dir;
81 struct string_list possible_new_dirs;
84 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
85 char *dir)
87 struct dir_rename_entry key;
89 if (!dir)
90 return NULL;
91 hashmap_entry_init(&key.ent, strhash(dir));
92 key.dir = dir;
93 return hashmap_get_entry(hashmap, &key, ent, NULL);
96 static int dir_rename_cmp(const void *cmp_data UNUSED,
97 const struct hashmap_entry *eptr,
98 const struct hashmap_entry *entry_or_key,
99 const void *keydata UNUSED)
101 const struct dir_rename_entry *e1, *e2;
103 e1 = container_of(eptr, const struct dir_rename_entry, ent);
104 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
106 return strcmp(e1->dir, e2->dir);
109 static void dir_rename_init(struct hashmap *map)
111 hashmap_init(map, dir_rename_cmp, NULL, 0);
114 static void dir_rename_entry_init(struct dir_rename_entry *entry,
115 char *directory)
117 hashmap_entry_init(&entry->ent, strhash(directory));
118 entry->dir = directory;
119 entry->non_unique_new_dir = 0;
120 strbuf_init(&entry->new_dir, 0);
121 string_list_init_nodup(&entry->possible_new_dirs);
124 struct collision_entry {
125 struct hashmap_entry ent;
126 char *target_file;
127 struct string_list source_files;
128 unsigned reported_already:1;
131 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
132 char *target_file)
134 struct collision_entry key;
136 hashmap_entry_init(&key.ent, strhash(target_file));
137 key.target_file = target_file;
138 return hashmap_get_entry(hashmap, &key, ent, NULL);
141 static int collision_cmp(const void *cmp_data UNUSED,
142 const struct hashmap_entry *eptr,
143 const struct hashmap_entry *entry_or_key,
144 const void *keydata UNUSED)
146 const struct collision_entry *e1, *e2;
148 e1 = container_of(eptr, const struct collision_entry, ent);
149 e2 = container_of(entry_or_key, const struct collision_entry, ent);
151 return strcmp(e1->target_file, e2->target_file);
154 static void collision_init(struct hashmap *map)
156 hashmap_init(map, collision_cmp, NULL, 0);
159 static void flush_output(struct merge_options *opt)
161 if (opt->buffer_output < 2 && opt->obuf.len) {
162 fputs(opt->obuf.buf, stdout);
163 strbuf_reset(&opt->obuf);
167 __attribute__((format (printf, 2, 3)))
168 static int err(struct merge_options *opt, const char *err, ...)
170 va_list params;
172 if (opt->buffer_output < 2)
173 flush_output(opt);
174 else {
175 strbuf_complete(&opt->obuf, '\n');
176 strbuf_addstr(&opt->obuf, "error: ");
178 va_start(params, err);
179 strbuf_vaddf(&opt->obuf, err, params);
180 va_end(params);
181 if (opt->buffer_output > 1)
182 strbuf_addch(&opt->obuf, '\n');
183 else {
184 error("%s", opt->obuf.buf);
185 strbuf_reset(&opt->obuf);
188 return -1;
191 static struct tree *shift_tree_object(struct repository *repo,
192 struct tree *one, struct tree *two,
193 const char *subtree_shift)
195 struct object_id shifted;
197 if (!*subtree_shift) {
198 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
199 } else {
200 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
201 subtree_shift);
203 if (oideq(&two->object.oid, &shifted))
204 return two;
205 return lookup_tree(repo, &shifted);
208 static inline void set_commit_tree(struct commit *c, struct tree *t)
210 c->maybe_tree = t;
213 static struct commit *make_virtual_commit(struct repository *repo,
214 struct tree *tree,
215 const char *comment)
217 struct commit *commit = alloc_commit_node(repo);
219 set_merge_remote_desc(commit, comment, (struct object *)commit);
220 set_commit_tree(commit, tree);
221 commit->object.parsed = 1;
222 return commit;
225 enum rename_type {
226 RENAME_NORMAL = 0,
227 RENAME_VIA_DIR,
228 RENAME_ADD,
229 RENAME_DELETE,
230 RENAME_ONE_FILE_TO_ONE,
231 RENAME_ONE_FILE_TO_TWO,
232 RENAME_TWO_FILES_TO_ONE
236 * Since we want to write the index eventually, we cannot reuse the index
237 * for these (temporary) data.
239 struct stage_data {
240 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
241 struct rename_conflict_info *rename_conflict_info;
242 unsigned processed:1;
245 struct rename {
246 unsigned processed:1;
247 struct diff_filepair *pair;
248 const char *branch; /* branch that the rename occurred on */
250 * If directory rename detection affected this rename, what was its
251 * original type ('A' or 'R') and it's original destination before
252 * the directory rename (otherwise, '\0' and NULL for these two vars).
254 char dir_rename_original_type;
255 char *dir_rename_original_dest;
257 * Purpose of src_entry and dst_entry:
259 * If 'before' is renamed to 'after' then src_entry will contain
260 * the versions of 'before' from the merge_base, HEAD, and MERGE in
261 * stages 1, 2, and 3; dst_entry will contain the respective
262 * versions of 'after' in corresponding locations. Thus, we have a
263 * total of six modes and oids, though some will be null. (Stage 0
264 * is ignored; we're interested in handling conflicts.)
266 * Since we don't turn on break-rewrites by default, neither
267 * src_entry nor dst_entry can have all three of their stages have
268 * non-null oids, meaning at most four of the six will be non-null.
269 * Also, since this is a rename, both src_entry and dst_entry will
270 * have at least one non-null oid, meaning at least two will be
271 * non-null. Of the six oids, a typical rename will have three be
272 * non-null. Only two implies a rename/delete, and four implies a
273 * rename/add.
275 struct stage_data *src_entry;
276 struct stage_data *dst_entry;
279 struct rename_conflict_info {
280 enum rename_type rename_type;
281 struct rename *ren1;
282 struct rename *ren2;
285 static inline void setup_rename_conflict_info(enum rename_type rename_type,
286 struct merge_options *opt,
287 struct rename *ren1,
288 struct rename *ren2)
290 struct rename_conflict_info *ci;
293 * When we have two renames involved, it's easiest to get the
294 * correct things into stage 2 and 3, and to make sure that the
295 * content merge puts HEAD before the other branch if we just
296 * ensure that branch1 == opt->branch1. So, simply flip arguments
297 * around if we don't have that.
299 if (ren2 && ren1->branch != opt->branch1) {
300 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
301 return;
304 CALLOC_ARRAY(ci, 1);
305 ci->rename_type = rename_type;
306 ci->ren1 = ren1;
307 ci->ren2 = ren2;
309 ci->ren1->dst_entry->processed = 0;
310 ci->ren1->dst_entry->rename_conflict_info = ci;
311 if (ren2) {
312 ci->ren2->dst_entry->rename_conflict_info = ci;
316 static int show(struct merge_options *opt, int v)
318 return (!opt->priv->call_depth && opt->verbosity >= v) ||
319 opt->verbosity >= 5;
322 __attribute__((format (printf, 3, 4)))
323 static void output(struct merge_options *opt, int v, const char *fmt, ...)
325 va_list ap;
327 if (!show(opt, v))
328 return;
330 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
332 va_start(ap, fmt);
333 strbuf_vaddf(&opt->obuf, fmt, ap);
334 va_end(ap);
336 strbuf_addch(&opt->obuf, '\n');
337 if (!opt->buffer_output)
338 flush_output(opt);
341 static void repo_output_commit_title(struct merge_options *opt,
342 struct repository *repo,
343 struct commit *commit)
345 struct merge_remote_desc *desc;
347 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
348 desc = merge_remote_util(commit);
349 if (desc)
350 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
351 else {
352 strbuf_repo_add_unique_abbrev(&opt->obuf, repo,
353 &commit->object.oid,
354 DEFAULT_ABBREV);
355 strbuf_addch(&opt->obuf, ' ');
356 if (repo_parse_commit(repo, commit) != 0)
357 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
358 else {
359 const char *title;
360 const char *msg = repo_get_commit_buffer(repo, commit, NULL);
361 int len = find_commit_subject(msg, &title);
362 if (len)
363 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
364 repo_unuse_commit_buffer(repo, commit, msg);
367 flush_output(opt);
370 static void output_commit_title(struct merge_options *opt, struct commit *commit)
372 repo_output_commit_title(opt, the_repository, commit);
375 static int add_cacheinfo(struct merge_options *opt,
376 const struct diff_filespec *blob,
377 const char *path, int stage, int refresh, int options)
379 struct index_state *istate = opt->repo->index;
380 struct cache_entry *ce;
381 int ret;
383 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
384 if (!ce)
385 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
387 ret = add_index_entry(istate, ce, options);
388 if (refresh) {
389 struct cache_entry *nce;
391 nce = refresh_cache_entry(istate, ce,
392 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
393 if (!nce)
394 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
395 if (nce != ce)
396 ret = add_index_entry(istate, nce, options);
398 return ret;
401 static inline int merge_detect_rename(struct merge_options *opt)
403 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
406 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
408 parse_tree(tree);
409 init_tree_desc(desc, tree->buffer, tree->size);
412 static int unpack_trees_start(struct merge_options *opt,
413 struct tree *common,
414 struct tree *head,
415 struct tree *merge)
417 int rc;
418 struct tree_desc t[3];
419 struct index_state tmp_index = INDEX_STATE_INIT(opt->repo);
421 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
422 if (opt->priv->call_depth)
423 opt->priv->unpack_opts.index_only = 1;
424 else {
425 opt->priv->unpack_opts.update = 1;
426 /* FIXME: should only do this if !overwrite_ignore */
427 opt->priv->unpack_opts.preserve_ignored = 0;
429 opt->priv->unpack_opts.merge = 1;
430 opt->priv->unpack_opts.head_idx = 2;
431 opt->priv->unpack_opts.fn = threeway_merge;
432 opt->priv->unpack_opts.src_index = opt->repo->index;
433 opt->priv->unpack_opts.dst_index = &tmp_index;
434 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
435 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
437 init_tree_desc_from_tree(t+0, common);
438 init_tree_desc_from_tree(t+1, head);
439 init_tree_desc_from_tree(t+2, merge);
441 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
442 cache_tree_free(&opt->repo->index->cache_tree);
445 * Update opt->repo->index to match the new results, AFTER saving a
446 * copy in opt->priv->orig_index. Update src_index to point to the
447 * saved copy. (verify_uptodate() checks src_index, and the original
448 * index is the one that had the necessary modification timestamps.)
450 opt->priv->orig_index = *opt->repo->index;
451 *opt->repo->index = tmp_index;
452 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
454 return rc;
457 static void unpack_trees_finish(struct merge_options *opt)
459 discard_index(&opt->priv->orig_index);
460 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
463 static int save_files_dirs(const struct object_id *oid UNUSED,
464 struct strbuf *base, const char *path,
465 unsigned int mode, void *context)
467 struct path_hashmap_entry *entry;
468 int baselen = base->len;
469 struct merge_options *opt = context;
471 strbuf_addstr(base, path);
473 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
474 hashmap_entry_init(&entry->e, fspathhash(entry->path));
475 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
477 strbuf_setlen(base, baselen);
478 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
481 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
483 struct pathspec match_all;
484 memset(&match_all, 0, sizeof(match_all));
485 read_tree(opt->repo, tree,
486 &match_all, save_files_dirs, opt);
489 static int get_tree_entry_if_blob(struct repository *r,
490 const struct object_id *tree,
491 const char *path,
492 struct diff_filespec *dfs)
494 int ret;
496 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
497 if (S_ISDIR(dfs->mode)) {
498 oidcpy(&dfs->oid, null_oid());
499 dfs->mode = 0;
501 return ret;
505 * Returns an index_entry instance which doesn't have to correspond to
506 * a real cache entry in Git's index.
508 static struct stage_data *insert_stage_data(struct repository *r,
509 const char *path,
510 struct tree *o, struct tree *a, struct tree *b,
511 struct string_list *entries)
513 struct string_list_item *item;
514 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
515 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
516 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
517 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
518 item = string_list_insert(entries, path);
519 item->util = e;
520 return e;
524 * Create a dictionary mapping file names to stage_data objects. The
525 * dictionary contains one entry for every path with a non-zero stage entry.
527 static struct string_list *get_unmerged(struct index_state *istate)
529 struct string_list *unmerged = xmalloc(sizeof(struct string_list));
530 int i;
532 string_list_init_dup(unmerged);
534 /* TODO: audit for interaction with sparse-index. */
535 ensure_full_index(istate);
536 for (i = 0; i < istate->cache_nr; i++) {
537 struct string_list_item *item;
538 struct stage_data *e;
539 const struct cache_entry *ce = istate->cache[i];
540 if (!ce_stage(ce))
541 continue;
543 item = string_list_lookup(unmerged, ce->name);
544 if (!item) {
545 item = string_list_insert(unmerged, ce->name);
546 item->util = xcalloc(1, sizeof(struct stage_data));
548 e = item->util;
549 e->stages[ce_stage(ce)].mode = ce->ce_mode;
550 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
553 return unmerged;
556 static int string_list_df_name_compare(const char *one, const char *two)
558 int onelen = strlen(one);
559 int twolen = strlen(two);
561 * Here we only care that entries for D/F conflicts are
562 * adjacent, in particular with the file of the D/F conflict
563 * appearing before files below the corresponding directory.
564 * The order of the rest of the list is irrelevant for us.
566 * To achieve this, we sort with df_name_compare and provide
567 * the mode S_IFDIR so that D/F conflicts will sort correctly.
568 * We use the mode S_IFDIR for everything else for simplicity,
569 * since in other cases any changes in their order due to
570 * sorting cause no problems for us.
572 int cmp = df_name_compare(one, onelen, S_IFDIR,
573 two, twolen, S_IFDIR);
575 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
576 * that 'foo' comes before 'foo/bar'.
578 if (cmp)
579 return cmp;
580 return onelen - twolen;
583 static void record_df_conflict_files(struct merge_options *opt,
584 struct string_list *entries)
586 /* If there is a D/F conflict and the file for such a conflict
587 * currently exists in the working tree, we want to allow it to be
588 * removed to make room for the corresponding directory if needed.
589 * The files underneath the directories of such D/F conflicts will
590 * be processed before the corresponding file involved in the D/F
591 * conflict. If the D/F directory ends up being removed by the
592 * merge, then we won't have to touch the D/F file. If the D/F
593 * directory needs to be written to the working copy, then the D/F
594 * file will simply be removed (in make_room_for_path()) to make
595 * room for the necessary paths. Note that if both the directory
596 * and the file need to be present, then the D/F file will be
597 * reinstated with a new unique name at the time it is processed.
599 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
600 const char *last_file = NULL;
601 int last_len = 0;
602 int i;
605 * If we're merging merge-bases, we don't want to bother with
606 * any working directory changes.
608 if (opt->priv->call_depth)
609 return;
611 /* Ensure D/F conflicts are adjacent in the entries list. */
612 for (i = 0; i < entries->nr; i++) {
613 struct string_list_item *next = &entries->items[i];
614 string_list_append(&df_sorted_entries, next->string)->util =
615 next->util;
617 df_sorted_entries.cmp = string_list_df_name_compare;
618 string_list_sort(&df_sorted_entries);
620 string_list_clear(&opt->priv->df_conflict_file_set, 1);
621 for (i = 0; i < df_sorted_entries.nr; i++) {
622 const char *path = df_sorted_entries.items[i].string;
623 int len = strlen(path);
624 struct stage_data *e = df_sorted_entries.items[i].util;
627 * Check if last_file & path correspond to a D/F conflict;
628 * i.e. whether path is last_file+'/'+<something>.
629 * If so, record that it's okay to remove last_file to make
630 * room for path and friends if needed.
632 if (last_file &&
633 len > last_len &&
634 memcmp(path, last_file, last_len) == 0 &&
635 path[last_len] == '/') {
636 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
640 * Determine whether path could exist as a file in the
641 * working directory as a possible D/F conflict. This
642 * will only occur when it exists in stage 2 as a
643 * file.
645 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
646 last_file = path;
647 last_len = len;
648 } else {
649 last_file = NULL;
652 string_list_clear(&df_sorted_entries, 0);
655 static int update_stages(struct merge_options *opt, const char *path,
656 const struct diff_filespec *o,
657 const struct diff_filespec *a,
658 const struct diff_filespec *b)
662 * NOTE: It is usually a bad idea to call update_stages on a path
663 * before calling update_file on that same path, since it can
664 * sometimes lead to spurious "refusing to lose untracked file..."
665 * messages from update_file (via make_room_for path via
666 * would_lose_untracked). Instead, reverse the order of the calls
667 * (executing update_file first and then update_stages).
669 int clear = 1;
670 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
671 if (clear)
672 if (remove_file_from_index(opt->repo->index, path))
673 return -1;
674 if (o)
675 if (add_cacheinfo(opt, o, path, 1, 0, options))
676 return -1;
677 if (a)
678 if (add_cacheinfo(opt, a, path, 2, 0, options))
679 return -1;
680 if (b)
681 if (add_cacheinfo(opt, b, path, 3, 0, options))
682 return -1;
683 return 0;
686 static void update_entry(struct stage_data *entry,
687 struct diff_filespec *o,
688 struct diff_filespec *a,
689 struct diff_filespec *b)
691 entry->processed = 0;
692 entry->stages[1].mode = o->mode;
693 entry->stages[2].mode = a->mode;
694 entry->stages[3].mode = b->mode;
695 oidcpy(&entry->stages[1].oid, &o->oid);
696 oidcpy(&entry->stages[2].oid, &a->oid);
697 oidcpy(&entry->stages[3].oid, &b->oid);
700 static int remove_file(struct merge_options *opt, int clean,
701 const char *path, int no_wd)
703 int update_cache = opt->priv->call_depth || clean;
704 int update_working_directory = !opt->priv->call_depth && !no_wd;
706 if (update_cache) {
707 if (remove_file_from_index(opt->repo->index, path))
708 return -1;
710 if (update_working_directory) {
711 if (ignore_case) {
712 struct cache_entry *ce;
713 ce = index_file_exists(opt->repo->index, path, strlen(path),
714 ignore_case);
715 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
716 return 0;
718 if (remove_path(path))
719 return -1;
721 return 0;
724 /* add a string to a strbuf, but converting "/" to "_" */
725 static void add_flattened_path(struct strbuf *out, const char *s)
727 size_t i = out->len;
728 strbuf_addstr(out, s);
729 for (; i < out->len; i++)
730 if (out->buf[i] == '/')
731 out->buf[i] = '_';
734 static char *unique_path(struct merge_options *opt,
735 const char *path,
736 const char *branch)
738 struct path_hashmap_entry *entry;
739 struct strbuf newpath = STRBUF_INIT;
740 int suffix = 0;
741 size_t base_len;
743 strbuf_addf(&newpath, "%s~", path);
744 add_flattened_path(&newpath, branch);
746 base_len = newpath.len;
747 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
748 fspathhash(newpath.buf), newpath.buf) ||
749 (!opt->priv->call_depth && file_exists(newpath.buf))) {
750 strbuf_setlen(&newpath, base_len);
751 strbuf_addf(&newpath, "_%d", suffix++);
754 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
755 hashmap_entry_init(&entry->e, fspathhash(entry->path));
756 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
757 return strbuf_detach(&newpath, NULL);
761 * Check whether a directory in the index is in the way of an incoming
762 * file. Return 1 if so. If check_working_copy is non-zero, also
763 * check the working directory. If empty_ok is non-zero, also return
764 * 0 in the case where the working-tree dir exists but is empty.
766 static int dir_in_way(struct index_state *istate, const char *path,
767 int check_working_copy, int empty_ok)
769 int pos;
770 struct strbuf dirpath = STRBUF_INIT;
771 struct stat st;
773 strbuf_addstr(&dirpath, path);
774 strbuf_addch(&dirpath, '/');
776 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
778 if (pos < 0)
779 pos = -1 - pos;
780 if (pos < istate->cache_nr &&
781 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
782 strbuf_release(&dirpath);
783 return 1;
786 strbuf_release(&dirpath);
787 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
788 !(empty_ok && is_empty_dir(path)) &&
789 !has_symlink_leading_path(path, strlen(path));
793 * Returns whether path was tracked in the index before the merge started,
794 * and its oid and mode match the specified values
796 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
797 const struct diff_filespec *blob)
799 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
800 struct cache_entry *ce;
802 if (0 > pos)
803 /* we were not tracking this path before the merge */
804 return 0;
806 /* See if the file we were tracking before matches */
807 ce = opt->priv->orig_index.cache[pos];
808 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
812 * Returns whether path was tracked in the index before the merge started
814 static int was_tracked(struct merge_options *opt, const char *path)
816 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
818 if (0 <= pos)
819 /* we were tracking this path before the merge */
820 return 1;
822 return 0;
825 static int would_lose_untracked(struct merge_options *opt, const char *path)
827 struct index_state *istate = opt->repo->index;
830 * This may look like it can be simplified to:
831 * return !was_tracked(opt, path) && file_exists(path)
832 * but it can't. This function needs to know whether path was in
833 * the working tree due to EITHER having been tracked in the index
834 * before the merge OR having been put into the working copy and
835 * index by unpack_trees(). Due to that either-or requirement, we
836 * check the current index instead of the original one.
838 * Note that we do not need to worry about merge-recursive itself
839 * updating the index after unpack_trees() and before calling this
840 * function, because we strictly require all code paths in
841 * merge-recursive to update the working tree first and the index
842 * second. Doing otherwise would break
843 * update_file()/would_lose_untracked(); see every comment in this
844 * file which mentions "update_stages".
846 int pos = index_name_pos(istate, path, strlen(path));
848 if (pos < 0)
849 pos = -1 - pos;
850 while (pos < istate->cache_nr &&
851 !strcmp(path, istate->cache[pos]->name)) {
853 * If stage #0, it is definitely tracked.
854 * If it has stage #2 then it was tracked
855 * before this merge started. All other
856 * cases the path was not tracked.
858 switch (ce_stage(istate->cache[pos])) {
859 case 0:
860 case 2:
861 return 0;
863 pos++;
865 return file_exists(path);
868 static int was_dirty(struct merge_options *opt, const char *path)
870 struct cache_entry *ce;
871 int dirty = 1;
873 if (opt->priv->call_depth || !was_tracked(opt, path))
874 return !dirty;
876 ce = index_file_exists(opt->priv->unpack_opts.src_index,
877 path, strlen(path), ignore_case);
878 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
879 return dirty;
882 static int make_room_for_path(struct merge_options *opt, const char *path)
884 int status, i;
885 const char *msg = _("failed to create path '%s'%s");
887 /* Unlink any D/F conflict files that are in the way */
888 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
889 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
890 size_t pathlen = strlen(path);
891 size_t df_pathlen = strlen(df_path);
892 if (df_pathlen < pathlen &&
893 path[df_pathlen] == '/' &&
894 strncmp(path, df_path, df_pathlen) == 0) {
895 output(opt, 3,
896 _("Removing %s to make room for subdirectory\n"),
897 df_path);
898 unlink(df_path);
899 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
900 i, 0);
901 break;
905 /* Make sure leading directories are created */
906 status = safe_create_leading_directories_const(path);
907 if (status) {
908 if (status == SCLD_EXISTS)
909 /* something else exists */
910 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
911 return err(opt, msg, path, "");
915 * Do not unlink a file in the work tree if we are not
916 * tracking it.
918 if (would_lose_untracked(opt, path))
919 return err(opt, _("refusing to lose untracked file at '%s'"),
920 path);
922 /* Successful unlink is good.. */
923 if (!unlink(path))
924 return 0;
925 /* .. and so is no existing file */
926 if (errno == ENOENT)
927 return 0;
928 /* .. but not some other error (who really cares what?) */
929 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
932 static int update_file_flags(struct merge_options *opt,
933 const struct diff_filespec *contents,
934 const char *path,
935 int update_cache,
936 int update_wd)
938 int ret = 0;
940 if (opt->priv->call_depth)
941 update_wd = 0;
943 if (update_wd) {
944 enum object_type type;
945 void *buf;
946 unsigned long size;
948 if (S_ISGITLINK(contents->mode)) {
950 * We may later decide to recursively descend into
951 * the submodule directory and update its index
952 * and/or work tree, but we do not do that now.
954 update_wd = 0;
955 goto update_index;
958 buf = repo_read_object_file(the_repository, &contents->oid,
959 &type, &size);
960 if (!buf) {
961 ret = err(opt, _("cannot read object %s '%s'"),
962 oid_to_hex(&contents->oid), path);
963 goto free_buf;
965 if (type != OBJ_BLOB) {
966 ret = err(opt, _("blob expected for %s '%s'"),
967 oid_to_hex(&contents->oid), path);
968 goto free_buf;
970 if (S_ISREG(contents->mode)) {
971 struct strbuf strbuf = STRBUF_INIT;
972 if (convert_to_working_tree(opt->repo->index,
973 path, buf, size, &strbuf, NULL)) {
974 free(buf);
975 size = strbuf.len;
976 buf = strbuf_detach(&strbuf, NULL);
980 if (make_room_for_path(opt, path) < 0) {
981 update_wd = 0;
982 goto free_buf;
984 if (S_ISREG(contents->mode) ||
985 (!has_symlinks && S_ISLNK(contents->mode))) {
986 int fd;
987 int mode = (contents->mode & 0100 ? 0777 : 0666);
989 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
990 if (fd < 0) {
991 ret = err(opt, _("failed to open '%s': %s"),
992 path, strerror(errno));
993 goto free_buf;
995 write_in_full(fd, buf, size);
996 close(fd);
997 } else if (S_ISLNK(contents->mode)) {
998 char *lnk = xmemdupz(buf, size);
999 safe_create_leading_directories_const(path);
1000 unlink(path);
1001 if (symlink(lnk, path))
1002 ret = err(opt, _("failed to symlink '%s': %s"),
1003 path, strerror(errno));
1004 free(lnk);
1005 } else
1006 ret = err(opt,
1007 _("do not know what to do with %06o %s '%s'"),
1008 contents->mode, oid_to_hex(&contents->oid), path);
1009 free_buf:
1010 free(buf);
1012 update_index:
1013 if (!ret && update_cache) {
1014 int refresh = (!opt->priv->call_depth &&
1015 contents->mode != S_IFGITLINK);
1016 if (add_cacheinfo(opt, contents, path, 0, refresh,
1017 ADD_CACHE_OK_TO_ADD))
1018 return -1;
1020 return ret;
1023 static int update_file(struct merge_options *opt,
1024 int clean,
1025 const struct diff_filespec *contents,
1026 const char *path)
1028 return update_file_flags(opt, contents, path,
1029 opt->priv->call_depth || clean, !opt->priv->call_depth);
1032 /* Low level file merging, update and removal */
1034 struct merge_file_info {
1035 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1036 unsigned clean:1,
1037 merge:1;
1040 static int merge_3way(struct merge_options *opt,
1041 mmbuffer_t *result_buf,
1042 const struct diff_filespec *o,
1043 const struct diff_filespec *a,
1044 const struct diff_filespec *b,
1045 const char *branch1,
1046 const char *branch2,
1047 const int extra_marker_size)
1049 mmfile_t orig, src1, src2;
1050 struct ll_merge_options ll_opts = {0};
1051 char *base, *name1, *name2;
1052 enum ll_merge_result merge_status;
1054 ll_opts.renormalize = opt->renormalize;
1055 ll_opts.extra_marker_size = extra_marker_size;
1056 ll_opts.xdl_opts = opt->xdl_opts;
1058 if (opt->priv->call_depth) {
1059 ll_opts.virtual_ancestor = 1;
1060 ll_opts.variant = 0;
1061 } else {
1062 switch (opt->recursive_variant) {
1063 case MERGE_VARIANT_OURS:
1064 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1065 break;
1066 case MERGE_VARIANT_THEIRS:
1067 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1068 break;
1069 default:
1070 ll_opts.variant = 0;
1071 break;
1075 assert(a->path && b->path && o->path && opt->ancestor);
1076 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1077 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1078 name1 = mkpathdup("%s:%s", branch1, a->path);
1079 name2 = mkpathdup("%s:%s", branch2, b->path);
1080 } else {
1081 base = mkpathdup("%s", opt->ancestor);
1082 name1 = mkpathdup("%s", branch1);
1083 name2 = mkpathdup("%s", branch2);
1086 read_mmblob(&orig, &o->oid);
1087 read_mmblob(&src1, &a->oid);
1088 read_mmblob(&src2, &b->oid);
1091 * FIXME: Using a->path for normalization rules in ll_merge could be
1092 * wrong if we renamed from a->path to b->path. We should use the
1093 * target path for where the file will be written.
1095 merge_status = ll_merge(result_buf, a->path, &orig, base,
1096 &src1, name1, &src2, name2,
1097 opt->repo->index, &ll_opts);
1098 if (merge_status == LL_MERGE_BINARY_CONFLICT)
1099 warning("Cannot merge binary files: %s (%s vs. %s)",
1100 a->path, name1, name2);
1102 free(base);
1103 free(name1);
1104 free(name2);
1105 free(orig.ptr);
1106 free(src1.ptr);
1107 free(src2.ptr);
1108 return merge_status;
1111 static int find_first_merges(struct repository *repo,
1112 struct object_array *result, const char *path,
1113 struct commit *a, struct commit *b)
1115 int i, j;
1116 struct object_array merges = OBJECT_ARRAY_INIT;
1117 struct commit *commit;
1118 int contains_another;
1120 char merged_revision[GIT_MAX_HEXSZ + 2];
1121 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1122 "--all", merged_revision, NULL };
1123 struct rev_info revs;
1124 struct setup_revision_opt rev_opts;
1126 memset(result, 0, sizeof(struct object_array));
1127 memset(&rev_opts, 0, sizeof(rev_opts));
1129 /* get all revisions that merge commit a */
1130 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1131 oid_to_hex(&a->object.oid));
1132 repo_init_revisions(repo, &revs, NULL);
1133 /* FIXME: can't handle linked worktrees in submodules yet */
1134 revs.single_worktree = path != NULL;
1135 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1137 /* save all revisions from the above list that contain b */
1138 if (prepare_revision_walk(&revs))
1139 die("revision walk setup failed");
1140 while ((commit = get_revision(&revs)) != NULL) {
1141 struct object *o = &(commit->object);
1142 if (repo_in_merge_bases(repo, b, commit))
1143 add_object_array(o, NULL, &merges);
1145 reset_revision_walk();
1147 /* Now we've got all merges that contain a and b. Prune all
1148 * merges that contain another found merge and save them in
1149 * result.
1151 for (i = 0; i < merges.nr; i++) {
1152 struct commit *m1 = (struct commit *) merges.objects[i].item;
1154 contains_another = 0;
1155 for (j = 0; j < merges.nr; j++) {
1156 struct commit *m2 = (struct commit *) merges.objects[j].item;
1157 if (i != j && repo_in_merge_bases(repo, m2, m1)) {
1158 contains_another = 1;
1159 break;
1163 if (!contains_another)
1164 add_object_array(merges.objects[i].item, NULL, result);
1167 object_array_clear(&merges);
1168 release_revisions(&revs);
1169 return result->nr;
1172 static void print_commit(struct repository *repo, struct commit *commit)
1174 struct strbuf sb = STRBUF_INIT;
1175 struct pretty_print_context ctx = {0};
1176 ctx.date_mode.type = DATE_NORMAL;
1177 /* FIXME: Merge this with output_commit_title() */
1178 assert(!merge_remote_util(commit));
1179 repo_format_commit_message(repo, commit, " %h: %m %s", &sb, &ctx);
1180 fprintf(stderr, "%s\n", sb.buf);
1181 strbuf_release(&sb);
1184 static int is_valid(const struct diff_filespec *dfs)
1186 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1189 static int merge_submodule(struct merge_options *opt,
1190 struct object_id *result, const char *path,
1191 const struct object_id *base, const struct object_id *a,
1192 const struct object_id *b)
1194 struct repository subrepo;
1195 int ret = 0;
1196 struct commit *commit_base, *commit_a, *commit_b;
1197 int parent_count;
1198 struct object_array merges;
1200 int i;
1201 int search = !opt->priv->call_depth;
1203 /* store a in result in case we fail */
1204 /* FIXME: This is the WRONG resolution for the recursive case when
1205 * we need to be careful to avoid accidentally matching either side.
1206 * Should probably use o instead there, much like we do for merging
1207 * binaries.
1209 oidcpy(result, a);
1211 /* we can not handle deletion conflicts */
1212 if (is_null_oid(base))
1213 return 0;
1214 if (is_null_oid(a))
1215 return 0;
1216 if (is_null_oid(b))
1217 return 0;
1219 if (repo_submodule_init(&subrepo, opt->repo, path, null_oid())) {
1220 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1221 return 0;
1224 if (!(commit_base = lookup_commit_reference(&subrepo, base)) ||
1225 !(commit_a = lookup_commit_reference(&subrepo, a)) ||
1226 !(commit_b = lookup_commit_reference(&subrepo, b))) {
1227 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1228 goto cleanup;
1231 /* check whether both changes are forward */
1232 if (!repo_in_merge_bases(&subrepo, commit_base, commit_a) ||
1233 !repo_in_merge_bases(&subrepo, commit_base, commit_b)) {
1234 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1235 goto cleanup;
1238 /* Case #1: a is contained in b or vice versa */
1239 if (repo_in_merge_bases(&subrepo, commit_a, commit_b)) {
1240 oidcpy(result, b);
1241 if (show(opt, 3)) {
1242 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1243 repo_output_commit_title(opt, &subrepo, commit_b);
1244 } else if (show(opt, 2))
1245 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1246 else
1247 ; /* no output */
1249 ret = 1;
1250 goto cleanup;
1252 if (repo_in_merge_bases(&subrepo, commit_b, commit_a)) {
1253 oidcpy(result, a);
1254 if (show(opt, 3)) {
1255 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1256 repo_output_commit_title(opt, &subrepo, commit_a);
1257 } else if (show(opt, 2))
1258 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1259 else
1260 ; /* no output */
1262 ret = 1;
1263 goto cleanup;
1267 * Case #2: There are one or more merges that contain a and b in
1268 * the submodule. If there is only one, then present it as a
1269 * suggestion to the user, but leave it marked unmerged so the
1270 * user needs to confirm the resolution.
1273 /* Skip the search if makes no sense to the calling context. */
1274 if (!search)
1275 goto cleanup;
1277 /* find commit which merges them */
1278 parent_count = find_first_merges(&subrepo, &merges, path,
1279 commit_a, commit_b);
1280 switch (parent_count) {
1281 case 0:
1282 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1283 break;
1285 case 1:
1286 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1287 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1288 print_commit(&subrepo, (struct commit *) merges.objects[0].item);
1289 output(opt, 2, _(
1290 "If this is correct simply add it to the index "
1291 "for example\n"
1292 "by using:\n\n"
1293 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1294 "which will accept this suggestion.\n"),
1295 oid_to_hex(&merges.objects[0].item->oid), path);
1296 break;
1298 default:
1299 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1300 for (i = 0; i < merges.nr; i++)
1301 print_commit(&subrepo, (struct commit *) merges.objects[i].item);
1304 object_array_clear(&merges);
1305 cleanup:
1306 repo_clear(&subrepo);
1307 return ret;
1310 static int merge_mode_and_contents(struct merge_options *opt,
1311 const struct diff_filespec *o,
1312 const struct diff_filespec *a,
1313 const struct diff_filespec *b,
1314 const char *filename,
1315 const char *branch1,
1316 const char *branch2,
1317 const int extra_marker_size,
1318 struct merge_file_info *result)
1320 if (opt->branch1 != branch1) {
1322 * It's weird getting a reverse merge with HEAD on the bottom
1323 * side of the conflict markers and the other branch on the
1324 * top. Fix that.
1326 return merge_mode_and_contents(opt, o, b, a,
1327 filename,
1328 branch2, branch1,
1329 extra_marker_size, result);
1332 result->merge = 0;
1333 result->clean = 1;
1335 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1336 result->clean = 0;
1338 * FIXME: This is a bad resolution for recursive case; for
1339 * the recursive case we want something that is unlikely to
1340 * accidentally match either side. Also, while it makes
1341 * sense to prefer regular files over symlinks, it doesn't
1342 * make sense to prefer regular files over submodules.
1344 if (S_ISREG(a->mode)) {
1345 result->blob.mode = a->mode;
1346 oidcpy(&result->blob.oid, &a->oid);
1347 } else {
1348 result->blob.mode = b->mode;
1349 oidcpy(&result->blob.oid, &b->oid);
1351 } else {
1352 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1353 result->merge = 1;
1356 * Merge modes
1358 if (a->mode == b->mode || a->mode == o->mode)
1359 result->blob.mode = b->mode;
1360 else {
1361 result->blob.mode = a->mode;
1362 if (b->mode != o->mode) {
1363 result->clean = 0;
1364 result->merge = 1;
1368 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1369 oidcpy(&result->blob.oid, &b->oid);
1370 else if (oideq(&b->oid, &o->oid))
1371 oidcpy(&result->blob.oid, &a->oid);
1372 else if (S_ISREG(a->mode)) {
1373 mmbuffer_t result_buf;
1374 int ret = 0, merge_status;
1376 merge_status = merge_3way(opt, &result_buf, o, a, b,
1377 branch1, branch2,
1378 extra_marker_size);
1380 if ((merge_status < 0) || !result_buf.ptr)
1381 ret = err(opt, _("failed to execute internal merge"));
1383 if (!ret &&
1384 write_object_file(result_buf.ptr, result_buf.size,
1385 OBJ_BLOB, &result->blob.oid))
1386 ret = err(opt, _("unable to add %s to database"),
1387 a->path);
1389 free(result_buf.ptr);
1390 if (ret)
1391 return ret;
1392 /* FIXME: bug, what if modes didn't match? */
1393 result->clean = (merge_status == 0);
1394 } else if (S_ISGITLINK(a->mode)) {
1395 result->clean = merge_submodule(opt, &result->blob.oid,
1396 o->path,
1397 &o->oid,
1398 &a->oid,
1399 &b->oid);
1400 } else if (S_ISLNK(a->mode)) {
1401 switch (opt->recursive_variant) {
1402 case MERGE_VARIANT_NORMAL:
1403 oidcpy(&result->blob.oid, &a->oid);
1404 if (!oideq(&a->oid, &b->oid))
1405 result->clean = 0;
1406 break;
1407 case MERGE_VARIANT_OURS:
1408 oidcpy(&result->blob.oid, &a->oid);
1409 break;
1410 case MERGE_VARIANT_THEIRS:
1411 oidcpy(&result->blob.oid, &b->oid);
1412 break;
1414 } else
1415 BUG("unsupported object type in the tree");
1418 if (result->merge)
1419 output(opt, 2, _("Auto-merging %s"), filename);
1421 return 0;
1424 static int handle_rename_via_dir(struct merge_options *opt,
1425 struct rename_conflict_info *ci)
1428 * Handle file adds that need to be renamed due to directory rename
1429 * detection. This differs from handle_rename_normal, because
1430 * there is no content merge to do; just move the file into the
1431 * desired final location.
1433 const struct rename *ren = ci->ren1;
1434 const struct diff_filespec *dest = ren->pair->two;
1435 char *file_path = dest->path;
1436 int mark_conflicted = (opt->detect_directory_renames ==
1437 MERGE_DIRECTORY_RENAMES_CONFLICT);
1438 assert(ren->dir_rename_original_dest);
1440 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1441 mark_conflicted = 1;
1442 file_path = unique_path(opt, dest->path, ren->branch);
1443 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1444 "writing to %s instead."),
1445 dest->path, file_path);
1448 if (mark_conflicted) {
1450 * Write the file in worktree at file_path. In the index,
1451 * only record the file at dest->path in the appropriate
1452 * higher stage.
1454 if (update_file(opt, 0, dest, file_path))
1455 return -1;
1456 if (file_path != dest->path)
1457 free(file_path);
1458 if (update_stages(opt, dest->path, NULL,
1459 ren->branch == opt->branch1 ? dest : NULL,
1460 ren->branch == opt->branch1 ? NULL : dest))
1461 return -1;
1462 return 0; /* not clean, but conflicted */
1463 } else {
1464 /* Update dest->path both in index and in worktree */
1465 if (update_file(opt, 1, dest, dest->path))
1466 return -1;
1467 return 1; /* clean */
1471 static int handle_change_delete(struct merge_options *opt,
1472 const char *path, const char *old_path,
1473 const struct diff_filespec *o,
1474 const struct diff_filespec *changed,
1475 const char *change_branch,
1476 const char *delete_branch,
1477 const char *change, const char *change_past)
1479 char *alt_path = NULL;
1480 const char *update_path = path;
1481 int ret = 0;
1483 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1484 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1485 update_path = alt_path = unique_path(opt, path, change_branch);
1488 if (opt->priv->call_depth) {
1490 * We cannot arbitrarily accept either a_sha or b_sha as
1491 * correct; since there is no true "middle point" between
1492 * them, simply reuse the base version for virtual merge base.
1494 ret = remove_file_from_index(opt->repo->index, path);
1495 if (!ret)
1496 ret = update_file(opt, 0, o, update_path);
1497 } else {
1499 * Despite the four nearly duplicate messages and argument
1500 * lists below and the ugliness of the nested if-statements,
1501 * having complete messages makes the job easier for
1502 * translators.
1504 * The slight variance among the cases is due to the fact
1505 * that:
1506 * 1) directory/file conflicts (in effect if
1507 * !alt_path) could cause us to need to write the
1508 * file to a different path.
1509 * 2) renames (in effect if !old_path) could mean that
1510 * there are two names for the path that the user
1511 * may know the file by.
1513 if (!alt_path) {
1514 if (!old_path) {
1515 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1516 "and %s in %s. Version %s of %s left in tree."),
1517 change, path, delete_branch, change_past,
1518 change_branch, change_branch, path);
1519 } else {
1520 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1521 "and %s to %s in %s. Version %s of %s left in tree."),
1522 change, old_path, delete_branch, change_past, path,
1523 change_branch, change_branch, path);
1525 } else {
1526 if (!old_path) {
1527 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1528 "and %s in %s. Version %s of %s left in tree at %s."),
1529 change, path, delete_branch, change_past,
1530 change_branch, change_branch, path, alt_path);
1531 } else {
1532 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1533 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1534 change, old_path, delete_branch, change_past, path,
1535 change_branch, change_branch, path, alt_path);
1539 * No need to call update_file() on path when change_branch ==
1540 * opt->branch1 && !alt_path, since that would needlessly touch
1541 * path. We could call update_file_flags() with update_cache=0
1542 * and update_wd=0, but that's a no-op.
1544 if (change_branch != opt->branch1 || alt_path)
1545 ret = update_file(opt, 0, changed, update_path);
1547 free(alt_path);
1549 return ret;
1552 static int handle_rename_delete(struct merge_options *opt,
1553 struct rename_conflict_info *ci)
1555 const struct rename *ren = ci->ren1;
1556 const struct diff_filespec *orig = ren->pair->one;
1557 const struct diff_filespec *dest = ren->pair->two;
1558 const char *rename_branch = ren->branch;
1559 const char *delete_branch = (opt->branch1 == ren->branch ?
1560 opt->branch2 : opt->branch1);
1562 if (handle_change_delete(opt,
1563 opt->priv->call_depth ? orig->path : dest->path,
1564 opt->priv->call_depth ? NULL : orig->path,
1565 orig, dest,
1566 rename_branch, delete_branch,
1567 _("rename"), _("renamed")))
1568 return -1;
1570 if (opt->priv->call_depth)
1571 return remove_file_from_index(opt->repo->index, dest->path);
1572 else
1573 return update_stages(opt, dest->path, NULL,
1574 rename_branch == opt->branch1 ? dest : NULL,
1575 rename_branch == opt->branch1 ? NULL : dest);
1578 static int handle_file_collision(struct merge_options *opt,
1579 const char *collide_path,
1580 const char *prev_path1,
1581 const char *prev_path2,
1582 const char *branch1, const char *branch2,
1583 struct diff_filespec *a,
1584 struct diff_filespec *b)
1586 struct merge_file_info mfi;
1587 struct diff_filespec null;
1588 char *alt_path = NULL;
1589 const char *update_path = collide_path;
1592 * It's easiest to get the correct things into stage 2 and 3, and
1593 * to make sure that the content merge puts HEAD before the other
1594 * branch if we just ensure that branch1 == opt->branch1. So, simply
1595 * flip arguments around if we don't have that.
1597 if (branch1 != opt->branch1) {
1598 return handle_file_collision(opt, collide_path,
1599 prev_path2, prev_path1,
1600 branch2, branch1,
1601 b, a);
1604 /* Remove rename sources if rename/add or rename/rename(2to1) */
1605 if (prev_path1)
1606 remove_file(opt, 1, prev_path1,
1607 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1608 if (prev_path2)
1609 remove_file(opt, 1, prev_path2,
1610 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1613 * Remove the collision path, if it wouldn't cause dirty contents
1614 * or an untracked file to get lost. We'll either overwrite with
1615 * merged contents, or just write out to differently named files.
1617 if (was_dirty(opt, collide_path)) {
1618 output(opt, 1, _("Refusing to lose dirty file at %s"),
1619 collide_path);
1620 update_path = alt_path = unique_path(opt, collide_path, "merged");
1621 } else if (would_lose_untracked(opt, collide_path)) {
1623 * Only way we get here is if both renames were from
1624 * a directory rename AND user had an untracked file
1625 * at the location where both files end up after the
1626 * two directory renames. See testcase 10d of t6043.
1628 output(opt, 1, _("Refusing to lose untracked file at "
1629 "%s, even though it's in the way."),
1630 collide_path);
1631 update_path = alt_path = unique_path(opt, collide_path, "merged");
1632 } else {
1634 * FIXME: It's possible that the two files are identical
1635 * and that the current working copy happens to match, in
1636 * which case we are unnecessarily touching the working
1637 * tree file. It's not a likely enough scenario that I
1638 * want to code up the checks for it and a better fix is
1639 * available if we restructure how unpack_trees() and
1640 * merge-recursive interoperate anyway, so punting for
1641 * now...
1643 remove_file(opt, 0, collide_path, 0);
1646 /* Store things in diff_filespecs for functions that need it */
1647 null.path = (char *)collide_path;
1648 oidcpy(&null.oid, null_oid());
1649 null.mode = 0;
1651 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1652 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1653 return -1;
1654 mfi.clean &= !alt_path;
1655 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1656 return -1;
1657 if (!mfi.clean && !opt->priv->call_depth &&
1658 update_stages(opt, collide_path, NULL, a, b))
1659 return -1;
1660 free(alt_path);
1662 * FIXME: If both a & b both started with conflicts (only possible
1663 * if they came from a rename/rename(2to1)), but had IDENTICAL
1664 * contents including those conflicts, then in the next line we claim
1665 * it was clean. If someone cares about this case, we should have the
1666 * caller notify us if we started with conflicts.
1668 return mfi.clean;
1671 static int handle_rename_add(struct merge_options *opt,
1672 struct rename_conflict_info *ci)
1674 /* a was renamed to c, and a separate c was added. */
1675 struct diff_filespec *a = ci->ren1->pair->one;
1676 struct diff_filespec *c = ci->ren1->pair->two;
1677 char *path = c->path;
1678 char *prev_path_desc;
1679 struct merge_file_info mfi;
1681 const char *rename_branch = ci->ren1->branch;
1682 const char *add_branch = (opt->branch1 == rename_branch ?
1683 opt->branch2 : opt->branch1);
1684 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1686 output(opt, 1, _("CONFLICT (rename/add): "
1687 "Rename %s->%s in %s. Added %s in %s"),
1688 a->path, c->path, rename_branch,
1689 c->path, add_branch);
1691 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1692 ci->ren1->src_entry->stages[other_stage].path = a->path;
1693 if (merge_mode_and_contents(opt, a, c,
1694 &ci->ren1->src_entry->stages[other_stage],
1695 prev_path_desc,
1696 opt->branch1, opt->branch2,
1697 1 + opt->priv->call_depth * 2, &mfi))
1698 return -1;
1699 free(prev_path_desc);
1701 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1702 return handle_file_collision(opt,
1703 c->path, a->path, NULL,
1704 rename_branch, add_branch,
1705 &mfi.blob,
1706 &ci->ren1->dst_entry->stages[other_stage]);
1709 static char *find_path_for_conflict(struct merge_options *opt,
1710 const char *path,
1711 const char *branch1,
1712 const char *branch2)
1714 char *new_path = NULL;
1715 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1716 new_path = unique_path(opt, path, branch1);
1717 output(opt, 1, _("%s is a directory in %s adding "
1718 "as %s instead"),
1719 path, branch2, new_path);
1720 } else if (would_lose_untracked(opt, path)) {
1721 new_path = unique_path(opt, path, branch1);
1722 output(opt, 1, _("Refusing to lose untracked file"
1723 " at %s; adding as %s instead"),
1724 path, new_path);
1727 return new_path;
1731 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1733 static inline int flip_stage(int stage)
1735 return (2 + 3) - stage;
1738 static int handle_rename_rename_1to2(struct merge_options *opt,
1739 struct rename_conflict_info *ci)
1741 /* One file was renamed in both branches, but to different names. */
1742 struct merge_file_info mfi;
1743 struct diff_filespec *add;
1744 struct diff_filespec *o = ci->ren1->pair->one;
1745 struct diff_filespec *a = ci->ren1->pair->two;
1746 struct diff_filespec *b = ci->ren2->pair->two;
1747 char *path_desc;
1749 output(opt, 1, _("CONFLICT (rename/rename): "
1750 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1751 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1752 o->path, a->path, ci->ren1->branch,
1753 o->path, b->path, ci->ren2->branch,
1754 opt->priv->call_depth ? _(" (left unresolved)") : "");
1756 path_desc = xstrfmt("%s and %s, both renamed from %s",
1757 a->path, b->path, o->path);
1758 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1759 ci->ren1->branch, ci->ren2->branch,
1760 opt->priv->call_depth * 2, &mfi))
1761 return -1;
1762 free(path_desc);
1764 if (opt->priv->call_depth)
1765 remove_file_from_index(opt->repo->index, o->path);
1768 * For each destination path, we need to see if there is a
1769 * rename/add collision. If not, we can write the file out
1770 * to the specified location.
1772 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1773 if (is_valid(add)) {
1774 add->path = mfi.blob.path = a->path;
1775 if (handle_file_collision(opt, a->path,
1776 NULL, NULL,
1777 ci->ren1->branch,
1778 ci->ren2->branch,
1779 &mfi.blob, add) < 0)
1780 return -1;
1781 } else {
1782 char *new_path = find_path_for_conflict(opt, a->path,
1783 ci->ren1->branch,
1784 ci->ren2->branch);
1785 if (update_file(opt, 0, &mfi.blob,
1786 new_path ? new_path : a->path))
1787 return -1;
1788 free(new_path);
1789 if (!opt->priv->call_depth &&
1790 update_stages(opt, a->path, NULL, a, NULL))
1791 return -1;
1794 if (!mfi.clean && mfi.blob.mode == a->mode &&
1795 oideq(&mfi.blob.oid, &a->oid)) {
1797 * Getting here means we were attempting to merge a binary
1798 * blob. Since we can't merge binaries, the merge algorithm
1799 * just takes one side. But we don't want to copy the
1800 * contents of one side to both paths; we'd rather use the
1801 * original content at the given path for each path.
1803 oidcpy(&mfi.blob.oid, &b->oid);
1804 mfi.blob.mode = b->mode;
1806 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1807 if (is_valid(add)) {
1808 add->path = mfi.blob.path = b->path;
1809 if (handle_file_collision(opt, b->path,
1810 NULL, NULL,
1811 ci->ren1->branch,
1812 ci->ren2->branch,
1813 add, &mfi.blob) < 0)
1814 return -1;
1815 } else {
1816 char *new_path = find_path_for_conflict(opt, b->path,
1817 ci->ren2->branch,
1818 ci->ren1->branch);
1819 if (update_file(opt, 0, &mfi.blob,
1820 new_path ? new_path : b->path))
1821 return -1;
1822 free(new_path);
1823 if (!opt->priv->call_depth &&
1824 update_stages(opt, b->path, NULL, NULL, b))
1825 return -1;
1828 return 0;
1831 static int handle_rename_rename_2to1(struct merge_options *opt,
1832 struct rename_conflict_info *ci)
1834 /* Two files, a & b, were renamed to the same thing, c. */
1835 struct diff_filespec *a = ci->ren1->pair->one;
1836 struct diff_filespec *b = ci->ren2->pair->one;
1837 struct diff_filespec *c1 = ci->ren1->pair->two;
1838 struct diff_filespec *c2 = ci->ren2->pair->two;
1839 char *path = c1->path; /* == c2->path */
1840 char *path_side_1_desc;
1841 char *path_side_2_desc;
1842 struct merge_file_info mfi_c1;
1843 struct merge_file_info mfi_c2;
1844 int ostage1, ostage2;
1846 output(opt, 1, _("CONFLICT (rename/rename): "
1847 "Rename %s->%s in %s. "
1848 "Rename %s->%s in %s"),
1849 a->path, c1->path, ci->ren1->branch,
1850 b->path, c2->path, ci->ren2->branch);
1852 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1853 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1854 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1855 ostage2 = flip_stage(ostage1);
1856 ci->ren1->src_entry->stages[ostage1].path = a->path;
1857 ci->ren2->src_entry->stages[ostage2].path = b->path;
1858 if (merge_mode_and_contents(opt, a, c1,
1859 &ci->ren1->src_entry->stages[ostage1],
1860 path_side_1_desc,
1861 opt->branch1, opt->branch2,
1862 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1863 merge_mode_and_contents(opt, b,
1864 &ci->ren2->src_entry->stages[ostage2],
1865 c2, path_side_2_desc,
1866 opt->branch1, opt->branch2,
1867 1 + opt->priv->call_depth * 2, &mfi_c2))
1868 return -1;
1869 free(path_side_1_desc);
1870 free(path_side_2_desc);
1871 mfi_c1.blob.path = path;
1872 mfi_c2.blob.path = path;
1874 return handle_file_collision(opt, path, a->path, b->path,
1875 ci->ren1->branch, ci->ren2->branch,
1876 &mfi_c1.blob, &mfi_c2.blob);
1880 * Get the diff_filepairs changed between o_tree and tree.
1882 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1883 struct tree *o_tree,
1884 struct tree *tree)
1886 struct diff_queue_struct *ret;
1887 struct diff_options opts;
1889 repo_diff_setup(opt->repo, &opts);
1890 opts.flags.recursive = 1;
1891 opts.flags.rename_empty = 0;
1892 opts.detect_rename = merge_detect_rename(opt);
1894 * We do not have logic to handle the detection of copies. In
1895 * fact, it may not even make sense to add such logic: would we
1896 * really want a change to a base file to be propagated through
1897 * multiple other files by a merge?
1899 if (opts.detect_rename > DIFF_DETECT_RENAME)
1900 opts.detect_rename = DIFF_DETECT_RENAME;
1901 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 7000;
1902 opts.rename_score = opt->rename_score;
1903 opts.show_rename_progress = opt->show_rename_progress;
1904 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1905 diff_setup_done(&opts);
1906 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1907 diffcore_std(&opts);
1908 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1909 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1911 ret = xmalloc(sizeof(*ret));
1912 *ret = diff_queued_diff;
1914 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1915 diff_queued_diff.nr = 0;
1916 diff_queued_diff.queue = NULL;
1917 diff_flush(&opts);
1918 return ret;
1921 static int tree_has_path(struct repository *r, struct tree *tree,
1922 const char *path)
1924 struct object_id hashy;
1925 unsigned short mode_o;
1927 return !get_tree_entry(r,
1928 &tree->object.oid, path,
1929 &hashy, &mode_o);
1933 * Return a new string that replaces the beginning portion (which matches
1934 * entry->dir), with entry->new_dir. In perl-speak:
1935 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1936 * NOTE:
1937 * Caller must ensure that old_path starts with entry->dir + '/'.
1939 static char *apply_dir_rename(struct dir_rename_entry *entry,
1940 const char *old_path)
1942 struct strbuf new_path = STRBUF_INIT;
1943 int oldlen, newlen;
1945 if (entry->non_unique_new_dir)
1946 return NULL;
1948 oldlen = strlen(entry->dir);
1949 if (entry->new_dir.len == 0)
1951 * If someone renamed/merged a subdirectory into the root
1952 * directory (e.g. 'some/subdir' -> ''), then we want to
1953 * avoid returning
1954 * '' + '/filename'
1955 * as the rename; we need to make old_path + oldlen advance
1956 * past the '/' character.
1958 oldlen++;
1959 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1960 strbuf_grow(&new_path, newlen);
1961 strbuf_addbuf(&new_path, &entry->new_dir);
1962 strbuf_addstr(&new_path, &old_path[oldlen]);
1964 return strbuf_detach(&new_path, NULL);
1967 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1968 char **old_dir, char **new_dir)
1970 char *end_of_old, *end_of_new;
1972 /* Default return values: NULL, meaning no rename */
1973 *old_dir = NULL;
1974 *new_dir = NULL;
1977 * For
1978 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1979 * the "e/foo.c" part is the same, we just want to know that
1980 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1981 * so, for this example, this function returns "a/b/c/d" in
1982 * *old_dir and "a/b/some/thing/else" in *new_dir.
1986 * If the basename of the file changed, we don't care. We want
1987 * to know which portion of the directory, if any, changed.
1989 end_of_old = strrchr(old_path, '/');
1990 end_of_new = strrchr(new_path, '/');
1993 * If end_of_old is NULL, old_path wasn't in a directory, so there
1994 * could not be a directory rename (our rule elsewhere that a
1995 * directory which still exists is not considered to have been
1996 * renamed means the root directory can never be renamed -- because
1997 * the root directory always exists).
1999 if (!end_of_old)
2000 return; /* Note: *old_dir and *new_dir are still NULL */
2003 * If new_path contains no directory (end_of_new is NULL), then we
2004 * have a rename of old_path's directory to the root directory.
2006 if (!end_of_new) {
2007 *old_dir = xstrndup(old_path, end_of_old - old_path);
2008 *new_dir = xstrdup("");
2009 return;
2012 /* Find the first non-matching character traversing backwards */
2013 while (*--end_of_new == *--end_of_old &&
2014 end_of_old != old_path &&
2015 end_of_new != new_path)
2016 ; /* Do nothing; all in the while loop */
2019 * If both got back to the beginning of their strings, then the
2020 * directory didn't change at all, only the basename did.
2022 if (end_of_old == old_path && end_of_new == new_path &&
2023 *end_of_old == *end_of_new)
2024 return; /* Note: *old_dir and *new_dir are still NULL */
2027 * If end_of_new got back to the beginning of its string, and
2028 * end_of_old got back to the beginning of some subdirectory, then
2029 * we have a rename/merge of a subdirectory into the root, which
2030 * needs slightly special handling.
2032 * Note: There is no need to consider the opposite case, with a
2033 * rename/merge of the root directory into some subdirectory
2034 * because as noted above the root directory always exists so it
2035 * cannot be considered to be renamed.
2037 if (end_of_new == new_path &&
2038 end_of_old != old_path && end_of_old[-1] == '/') {
2039 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2040 *new_dir = xstrdup("");
2041 return;
2045 * We've found the first non-matching character in the directory
2046 * paths. That means the current characters we were looking at
2047 * were part of the first non-matching subdir name going back from
2048 * the end of the strings. Get the whole name by advancing both
2049 * end_of_old and end_of_new to the NEXT '/' character. That will
2050 * represent the entire directory rename.
2052 * The reason for the increment is cases like
2053 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2054 * After dropping the basename and going back to the first
2055 * non-matching character, we're now comparing:
2056 * a/b/s and a/b/
2057 * and we want to be comparing:
2058 * a/b/star/ and a/b/tar/
2059 * but without the pre-increment, the one on the right would stay
2060 * a/b/.
2062 end_of_old = strchr(++end_of_old, '/');
2063 end_of_new = strchr(++end_of_new, '/');
2065 /* Copy the old and new directories into *old_dir and *new_dir. */
2066 *old_dir = xstrndup(old_path, end_of_old - old_path);
2067 *new_dir = xstrndup(new_path, end_of_new - new_path);
2070 static void remove_hashmap_entries(struct hashmap *dir_renames,
2071 struct string_list *items_to_remove)
2073 int i;
2074 struct dir_rename_entry *entry;
2076 for (i = 0; i < items_to_remove->nr; i++) {
2077 entry = items_to_remove->items[i].util;
2078 hashmap_remove(dir_renames, &entry->ent, NULL);
2080 string_list_clear(items_to_remove, 0);
2084 * See if there is a directory rename for path, and if there are any file
2085 * level conflicts for the renamed location. If there is a rename and
2086 * there are no conflicts, return the new name. Otherwise, return NULL.
2088 static char *handle_path_level_conflicts(struct merge_options *opt,
2089 const char *path,
2090 struct dir_rename_entry *entry,
2091 struct hashmap *collisions,
2092 struct tree *tree)
2094 char *new_path = NULL;
2095 struct collision_entry *collision_ent;
2096 int clean = 1;
2097 struct strbuf collision_paths = STRBUF_INIT;
2100 * entry has the mapping of old directory name to new directory name
2101 * that we want to apply to path.
2103 new_path = apply_dir_rename(entry, path);
2105 if (!new_path) {
2106 /* This should only happen when entry->non_unique_new_dir set */
2107 if (!entry->non_unique_new_dir)
2108 BUG("entry->non_unique_new_dir not set and !new_path");
2109 output(opt, 1, _("CONFLICT (directory rename split): "
2110 "Unclear where to place %s because directory "
2111 "%s was renamed to multiple other directories, "
2112 "with no destination getting a majority of the "
2113 "files."),
2114 path, entry->dir);
2115 clean = 0;
2116 return NULL;
2120 * The caller needs to have ensured that it has pre-populated
2121 * collisions with all paths that map to new_path. Do a quick check
2122 * to ensure that's the case.
2124 collision_ent = collision_find_entry(collisions, new_path);
2125 if (!collision_ent)
2126 BUG("collision_ent is NULL");
2129 * Check for one-sided add/add/.../add conflicts, i.e.
2130 * where implicit renames from the other side doing
2131 * directory rename(s) can affect this side of history
2132 * to put multiple paths into the same location. Warn
2133 * and bail on directory renames for such paths.
2135 if (collision_ent->reported_already) {
2136 clean = 0;
2137 } else if (tree_has_path(opt->repo, tree, new_path)) {
2138 collision_ent->reported_already = 1;
2139 strbuf_add_separated_string_list(&collision_paths, ", ",
2140 &collision_ent->source_files);
2141 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2142 "file/dir at %s in the way of implicit "
2143 "directory rename(s) putting the following "
2144 "path(s) there: %s."),
2145 new_path, collision_paths.buf);
2146 clean = 0;
2147 } else if (collision_ent->source_files.nr > 1) {
2148 collision_ent->reported_already = 1;
2149 strbuf_add_separated_string_list(&collision_paths, ", ",
2150 &collision_ent->source_files);
2151 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2152 "more than one path to %s; implicit directory "
2153 "renames tried to put these paths there: %s"),
2154 new_path, collision_paths.buf);
2155 clean = 0;
2158 /* Free memory we no longer need */
2159 strbuf_release(&collision_paths);
2160 if (!clean && new_path) {
2161 free(new_path);
2162 return NULL;
2165 return new_path;
2169 * There are a couple things we want to do at the directory level:
2170 * 1. Check for both sides renaming to the same thing, in order to avoid
2171 * implicit renaming of files that should be left in place. (See
2172 * testcase 6b in t6043 for details.)
2173 * 2. Prune directory renames if there are still files left in the
2174 * original directory. These represent a partial directory rename,
2175 * i.e. a rename where only some of the files within the directory
2176 * were renamed elsewhere. (Technically, this could be done earlier
2177 * in get_directory_renames(), except that would prevent us from
2178 * doing the previous check and thus failing testcase 6b.)
2179 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2180 * In the future, we could potentially record this info as well and
2181 * omit reporting rename/rename(1to2) conflicts for each path within
2182 * the affected directories, thus cleaning up the merge output.
2183 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2184 * directory level, because merging directories is fine. If it
2185 * causes conflicts for files within those merged directories, then
2186 * that should be detected at the individual path level.
2188 static void handle_directory_level_conflicts(struct merge_options *opt,
2189 struct hashmap *dir_re_head,
2190 struct tree *head,
2191 struct hashmap *dir_re_merge,
2192 struct tree *merge)
2194 struct hashmap_iter iter;
2195 struct dir_rename_entry *head_ent;
2196 struct dir_rename_entry *merge_ent;
2198 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2199 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2201 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2202 ent /* member name */) {
2203 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2204 if (merge_ent &&
2205 !head_ent->non_unique_new_dir &&
2206 !merge_ent->non_unique_new_dir &&
2207 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2208 /* 1. Renamed identically; remove it from both sides */
2209 string_list_append(&remove_from_head,
2210 head_ent->dir)->util = head_ent;
2211 strbuf_release(&head_ent->new_dir);
2212 string_list_append(&remove_from_merge,
2213 merge_ent->dir)->util = merge_ent;
2214 strbuf_release(&merge_ent->new_dir);
2215 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2216 /* 2. This wasn't a directory rename after all */
2217 string_list_append(&remove_from_head,
2218 head_ent->dir)->util = head_ent;
2219 strbuf_release(&head_ent->new_dir);
2223 remove_hashmap_entries(dir_re_head, &remove_from_head);
2224 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2226 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2227 ent /* member name */) {
2228 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2229 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2230 /* 2. This wasn't a directory rename after all */
2231 string_list_append(&remove_from_merge,
2232 merge_ent->dir)->util = merge_ent;
2233 } else if (head_ent &&
2234 !head_ent->non_unique_new_dir &&
2235 !merge_ent->non_unique_new_dir) {
2236 /* 3. rename/rename(1to2) */
2238 * We can assume it's not rename/rename(1to1) because
2239 * that was case (1), already checked above. So we
2240 * know that head_ent->new_dir and merge_ent->new_dir
2241 * are different strings.
2243 output(opt, 1, _("CONFLICT (rename/rename): "
2244 "Rename directory %s->%s in %s. "
2245 "Rename directory %s->%s in %s"),
2246 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2247 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2248 string_list_append(&remove_from_head,
2249 head_ent->dir)->util = head_ent;
2250 strbuf_release(&head_ent->new_dir);
2251 string_list_append(&remove_from_merge,
2252 merge_ent->dir)->util = merge_ent;
2253 strbuf_release(&merge_ent->new_dir);
2257 remove_hashmap_entries(dir_re_head, &remove_from_head);
2258 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2261 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2263 struct hashmap *dir_renames;
2264 struct hashmap_iter iter;
2265 struct dir_rename_entry *entry;
2266 int i;
2269 * Typically, we think of a directory rename as all files from a
2270 * certain directory being moved to a target directory. However,
2271 * what if someone first moved two files from the original
2272 * directory in one commit, and then renamed the directory
2273 * somewhere else in a later commit? At merge time, we just know
2274 * that files from the original directory went to two different
2275 * places, and that the bulk of them ended up in the same place.
2276 * We want each directory rename to represent where the bulk of the
2277 * files from that directory end up; this function exists to find
2278 * where the bulk of the files went.
2280 * The first loop below simply iterates through the list of file
2281 * renames, finding out how often each directory rename pair
2282 * possibility occurs.
2284 dir_renames = xmalloc(sizeof(*dir_renames));
2285 dir_rename_init(dir_renames);
2286 for (i = 0; i < pairs->nr; ++i) {
2287 struct string_list_item *item;
2288 int *count;
2289 struct diff_filepair *pair = pairs->queue[i];
2290 char *old_dir, *new_dir;
2292 /* File not part of directory rename if it wasn't renamed */
2293 if (pair->status != 'R')
2294 continue;
2296 get_renamed_dir_portion(pair->one->path, pair->two->path,
2297 &old_dir, &new_dir);
2298 if (!old_dir)
2299 /* Directory didn't change at all; ignore this one. */
2300 continue;
2302 entry = dir_rename_find_entry(dir_renames, old_dir);
2303 if (!entry) {
2304 entry = xmalloc(sizeof(*entry));
2305 dir_rename_entry_init(entry, old_dir);
2306 hashmap_put(dir_renames, &entry->ent);
2307 } else {
2308 free(old_dir);
2310 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2311 if (!item) {
2312 item = string_list_insert(&entry->possible_new_dirs,
2313 new_dir);
2314 item->util = xcalloc(1, sizeof(int));
2315 } else {
2316 free(new_dir);
2318 count = item->util;
2319 *count += 1;
2323 * For each directory with files moved out of it, we find out which
2324 * target directory received the most files so we can declare it to
2325 * be the "winning" target location for the directory rename. This
2326 * winner gets recorded in new_dir. If there is no winner
2327 * (multiple target directories received the same number of files),
2328 * we set non_unique_new_dir. Once we've determined the winner (or
2329 * that there is no winner), we no longer need possible_new_dirs.
2331 hashmap_for_each_entry(dir_renames, &iter, entry,
2332 ent /* member name */) {
2333 int max = 0;
2334 int bad_max = 0;
2335 char *best = NULL;
2337 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2338 int *count = entry->possible_new_dirs.items[i].util;
2340 if (*count == max)
2341 bad_max = max;
2342 else if (*count > max) {
2343 max = *count;
2344 best = entry->possible_new_dirs.items[i].string;
2347 if (bad_max == max)
2348 entry->non_unique_new_dir = 1;
2349 else {
2350 assert(entry->new_dir.len == 0);
2351 strbuf_addstr(&entry->new_dir, best);
2354 * The relevant directory sub-portion of the original full
2355 * filepaths were xstrndup'ed before inserting into
2356 * possible_new_dirs, and instead of manually iterating the
2357 * list and free'ing each, just lie and tell
2358 * possible_new_dirs that it did the strdup'ing so that it
2359 * will free them for us.
2361 entry->possible_new_dirs.strdup_strings = 1;
2362 string_list_clear(&entry->possible_new_dirs, 1);
2365 return dir_renames;
2368 static struct dir_rename_entry *check_dir_renamed(const char *path,
2369 struct hashmap *dir_renames)
2371 char *temp = xstrdup(path);
2372 char *end;
2373 struct dir_rename_entry *entry = NULL;
2375 while ((end = strrchr(temp, '/'))) {
2376 *end = '\0';
2377 entry = dir_rename_find_entry(dir_renames, temp);
2378 if (entry)
2379 break;
2381 free(temp);
2382 return entry;
2385 static void compute_collisions(struct hashmap *collisions,
2386 struct hashmap *dir_renames,
2387 struct diff_queue_struct *pairs)
2389 int i;
2392 * Multiple files can be mapped to the same path due to directory
2393 * renames done by the other side of history. Since that other
2394 * side of history could have merged multiple directories into one,
2395 * if our side of history added the same file basename to each of
2396 * those directories, then all N of them would get implicitly
2397 * renamed by the directory rename detection into the same path,
2398 * and we'd get an add/add/.../add conflict, and all those adds
2399 * from *this* side of history. This is not representable in the
2400 * index, and users aren't going to easily be able to make sense of
2401 * it. So we need to provide a good warning about what's
2402 * happening, and fall back to no-directory-rename detection
2403 * behavior for those paths.
2405 * See testcases 9e and all of section 5 from t6043 for examples.
2407 collision_init(collisions);
2409 for (i = 0; i < pairs->nr; ++i) {
2410 struct dir_rename_entry *dir_rename_ent;
2411 struct collision_entry *collision_ent;
2412 char *new_path;
2413 struct diff_filepair *pair = pairs->queue[i];
2415 if (pair->status != 'A' && pair->status != 'R')
2416 continue;
2417 dir_rename_ent = check_dir_renamed(pair->two->path,
2418 dir_renames);
2419 if (!dir_rename_ent)
2420 continue;
2422 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2423 if (!new_path)
2425 * dir_rename_ent->non_unique_new_path is true, which
2426 * means there is no directory rename for us to use,
2427 * which means it won't cause us any additional
2428 * collisions.
2430 continue;
2431 collision_ent = collision_find_entry(collisions, new_path);
2432 if (!collision_ent) {
2433 CALLOC_ARRAY(collision_ent, 1);
2434 hashmap_entry_init(&collision_ent->ent,
2435 strhash(new_path));
2436 hashmap_put(collisions, &collision_ent->ent);
2437 collision_ent->target_file = new_path;
2438 } else {
2439 free(new_path);
2441 string_list_insert(&collision_ent->source_files,
2442 pair->two->path);
2446 static char *check_for_directory_rename(struct merge_options *opt,
2447 const char *path,
2448 struct tree *tree,
2449 struct hashmap *dir_renames,
2450 struct hashmap *dir_rename_exclusions,
2451 struct hashmap *collisions,
2452 int *clean_merge)
2454 char *new_path = NULL;
2455 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2456 struct dir_rename_entry *oentry = NULL;
2458 if (!entry)
2459 return new_path;
2462 * This next part is a little weird. We do not want to do an
2463 * implicit rename into a directory we renamed on our side, because
2464 * that will result in a spurious rename/rename(1to2) conflict. An
2465 * example:
2466 * Base commit: dumbdir/afile, otherdir/bfile
2467 * Side 1: smrtdir/afile, otherdir/bfile
2468 * Side 2: dumbdir/afile, dumbdir/bfile
2469 * Here, while working on Side 1, we could notice that otherdir was
2470 * renamed/merged to dumbdir, and change the diff_filepair for
2471 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2472 * 2 will notice the rename from dumbdir to smrtdir, and do the
2473 * transitive rename to move it from dumbdir/bfile to
2474 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2475 * smrtdir, a rename/rename(1to2) conflict. We really just want
2476 * the file to end up in smrtdir. And the way to achieve that is
2477 * to not let Side1 do the rename to dumbdir, since we know that is
2478 * the source of one of our directory renames.
2480 * That's why oentry and dir_rename_exclusions is here.
2482 * As it turns out, this also prevents N-way transient rename
2483 * confusion; See testcases 9c and 9d of t6043.
2485 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2486 if (oentry) {
2487 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2488 "to %s, because %s itself was renamed."),
2489 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2490 } else {
2491 new_path = handle_path_level_conflicts(opt, path, entry,
2492 collisions, tree);
2493 *clean_merge &= (new_path != NULL);
2496 return new_path;
2499 static void apply_directory_rename_modifications(struct merge_options *opt,
2500 struct diff_filepair *pair,
2501 char *new_path,
2502 struct rename *re,
2503 struct tree *tree,
2504 struct tree *o_tree,
2505 struct tree *a_tree,
2506 struct tree *b_tree,
2507 struct string_list *entries)
2509 struct string_list_item *item;
2510 int stage = (tree == a_tree ? 2 : 3);
2511 int update_wd;
2514 * In all cases where we can do directory rename detection,
2515 * unpack_trees() will have read pair->two->path into the
2516 * index and the working copy. We need to remove it so that
2517 * we can instead place it at new_path. It is guaranteed to
2518 * not be untracked (unpack_trees() would have errored out
2519 * saying the file would have been overwritten), but it might
2520 * be dirty, though.
2522 update_wd = !was_dirty(opt, pair->two->path);
2523 if (!update_wd)
2524 output(opt, 1, _("Refusing to lose dirty file at %s"),
2525 pair->two->path);
2526 remove_file(opt, 1, pair->two->path, !update_wd);
2528 /* Find or create a new re->dst_entry */
2529 item = string_list_lookup(entries, new_path);
2530 if (item) {
2532 * Since we're renaming on this side of history, and it's
2533 * due to a directory rename on the other side of history
2534 * (which we only allow when the directory in question no
2535 * longer exists on the other side of history), the
2536 * original entry for re->dst_entry is no longer
2537 * necessary...
2539 re->dst_entry->processed = 1;
2542 * ...because we'll be using this new one.
2544 re->dst_entry = item->util;
2545 } else {
2547 * re->dst_entry is for the before-dir-rename path, and we
2548 * need it to hold information for the after-dir-rename
2549 * path. Before creating a new entry, we need to mark the
2550 * old one as unnecessary (...unless it is shared by
2551 * src_entry, i.e. this didn't use to be a rename, in which
2552 * case we can just allow the normal processing to happen
2553 * for it).
2555 if (pair->status == 'R')
2556 re->dst_entry->processed = 1;
2558 re->dst_entry = insert_stage_data(opt->repo, new_path,
2559 o_tree, a_tree, b_tree,
2560 entries);
2561 item = string_list_insert(entries, new_path);
2562 item->util = re->dst_entry;
2566 * Update the stage_data with the information about the path we are
2567 * moving into place. That slot will be empty and available for us
2568 * to write to because of the collision checks in
2569 * handle_path_level_conflicts(). In other words,
2570 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2571 * open for us to write to.
2573 * It may be tempting to actually update the index at this point as
2574 * well, using update_stages_for_stage_data(), but as per the big
2575 * "NOTE" in update_stages(), doing so will modify the current
2576 * in-memory index which will break calls to would_lose_untracked()
2577 * that we need to make. Instead, we need to just make sure that
2578 * the various handle_rename_*() functions update the index
2579 * explicitly rather than relying on unpack_trees() to have done it.
2581 get_tree_entry(opt->repo,
2582 &tree->object.oid,
2583 pair->two->path,
2584 &re->dst_entry->stages[stage].oid,
2585 &re->dst_entry->stages[stage].mode);
2588 * Record the original change status (or 'type' of change). If it
2589 * was originally an add ('A'), this lets us differentiate later
2590 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2591 * otherwise look the same). If it was originally a rename ('R'),
2592 * this lets us remember and report accurately about the transitive
2593 * renaming that occurred via the directory rename detection. Also,
2594 * record the original destination name.
2596 re->dir_rename_original_type = pair->status;
2597 re->dir_rename_original_dest = pair->two->path;
2600 * We don't actually look at pair->status again, but it seems
2601 * pedagogically correct to adjust it.
2603 pair->status = 'R';
2606 * Finally, record the new location.
2608 pair->two->path = new_path;
2612 * Get information of all renames which occurred in 'pairs', making use of
2613 * any implicit directory renames inferred from the other side of history.
2614 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2615 * to be able to associate the correct cache entries with the rename
2616 * information; tree is always equal to either a_tree or b_tree.
2618 static struct string_list *get_renames(struct merge_options *opt,
2619 const char *branch,
2620 struct diff_queue_struct *pairs,
2621 struct hashmap *dir_renames,
2622 struct hashmap *dir_rename_exclusions,
2623 struct tree *tree,
2624 struct tree *o_tree,
2625 struct tree *a_tree,
2626 struct tree *b_tree,
2627 struct string_list *entries,
2628 int *clean_merge)
2630 int i;
2631 struct hashmap collisions;
2632 struct hashmap_iter iter;
2633 struct collision_entry *e;
2634 struct string_list *renames;
2636 compute_collisions(&collisions, dir_renames, pairs);
2637 CALLOC_ARRAY(renames, 1);
2639 for (i = 0; i < pairs->nr; ++i) {
2640 struct string_list_item *item;
2641 struct rename *re;
2642 struct diff_filepair *pair = pairs->queue[i];
2643 char *new_path; /* non-NULL only with directory renames */
2645 if (pair->status != 'A' && pair->status != 'R') {
2646 diff_free_filepair(pair);
2647 continue;
2649 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2650 dir_renames,
2651 dir_rename_exclusions,
2652 &collisions,
2653 clean_merge);
2654 if (pair->status != 'R' && !new_path) {
2655 diff_free_filepair(pair);
2656 continue;
2659 re = xmalloc(sizeof(*re));
2660 re->processed = 0;
2661 re->pair = pair;
2662 re->branch = branch;
2663 re->dir_rename_original_type = '\0';
2664 re->dir_rename_original_dest = NULL;
2665 item = string_list_lookup(entries, re->pair->one->path);
2666 if (!item)
2667 re->src_entry = insert_stage_data(opt->repo,
2668 re->pair->one->path,
2669 o_tree, a_tree, b_tree, entries);
2670 else
2671 re->src_entry = item->util;
2673 item = string_list_lookup(entries, re->pair->two->path);
2674 if (!item)
2675 re->dst_entry = insert_stage_data(opt->repo,
2676 re->pair->two->path,
2677 o_tree, a_tree, b_tree, entries);
2678 else
2679 re->dst_entry = item->util;
2680 item = string_list_insert(renames, pair->one->path);
2681 item->util = re;
2682 if (new_path)
2683 apply_directory_rename_modifications(opt, pair, new_path,
2684 re, tree, o_tree,
2685 a_tree, b_tree,
2686 entries);
2689 hashmap_for_each_entry(&collisions, &iter, e,
2690 ent /* member name */) {
2691 free(e->target_file);
2692 string_list_clear(&e->source_files, 0);
2694 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2695 return renames;
2698 static int process_renames(struct merge_options *opt,
2699 struct string_list *a_renames,
2700 struct string_list *b_renames)
2702 int clean_merge = 1, i, j;
2703 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2704 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2705 const struct rename *sre;
2708 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2709 * string_list one-by-one, but O(n log n) to build it unsorted and
2710 * then sort it. Note that as we build the list, we do not need to
2711 * check if the existing destination path is already in the list,
2712 * because the structure of diffcore_rename guarantees we won't
2713 * have duplicates.
2715 for (i = 0; i < a_renames->nr; i++) {
2716 sre = a_renames->items[i].util;
2717 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2718 = (void *)sre;
2720 for (i = 0; i < b_renames->nr; i++) {
2721 sre = b_renames->items[i].util;
2722 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2723 = (void *)sre;
2726 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2727 struct string_list *renames1, *renames2Dst;
2728 struct rename *ren1 = NULL, *ren2 = NULL;
2729 const char *ren1_src, *ren1_dst;
2730 struct string_list_item *lookup;
2732 if (i >= a_renames->nr) {
2733 ren2 = b_renames->items[j++].util;
2734 } else if (j >= b_renames->nr) {
2735 ren1 = a_renames->items[i++].util;
2736 } else {
2737 int compare = strcmp(a_renames->items[i].string,
2738 b_renames->items[j].string);
2739 if (compare <= 0)
2740 ren1 = a_renames->items[i++].util;
2741 if (compare >= 0)
2742 ren2 = b_renames->items[j++].util;
2745 /* TODO: refactor, so that 1/2 are not needed */
2746 if (ren1) {
2747 renames1 = a_renames;
2748 renames2Dst = &b_by_dst;
2749 } else {
2750 renames1 = b_renames;
2751 renames2Dst = &a_by_dst;
2752 SWAP(ren2, ren1);
2755 if (ren1->processed)
2756 continue;
2757 ren1->processed = 1;
2758 ren1->dst_entry->processed = 1;
2759 /* BUG: We should only mark src_entry as processed if we
2760 * are not dealing with a rename + add-source case.
2762 ren1->src_entry->processed = 1;
2764 ren1_src = ren1->pair->one->path;
2765 ren1_dst = ren1->pair->two->path;
2767 if (ren2) {
2768 /* One file renamed on both sides */
2769 const char *ren2_src = ren2->pair->one->path;
2770 const char *ren2_dst = ren2->pair->two->path;
2771 enum rename_type rename_type;
2772 if (strcmp(ren1_src, ren2_src) != 0)
2773 BUG("ren1_src != ren2_src");
2774 ren2->dst_entry->processed = 1;
2775 ren2->processed = 1;
2776 if (strcmp(ren1_dst, ren2_dst) != 0) {
2777 rename_type = RENAME_ONE_FILE_TO_TWO;
2778 clean_merge = 0;
2779 } else {
2780 rename_type = RENAME_ONE_FILE_TO_ONE;
2781 /* BUG: We should only remove ren1_src in
2782 * the base stage (think of rename +
2783 * add-source cases).
2785 remove_file(opt, 1, ren1_src, 1);
2786 update_entry(ren1->dst_entry,
2787 ren1->pair->one,
2788 ren1->pair->two,
2789 ren2->pair->two);
2791 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2792 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2793 /* Two different files renamed to the same thing */
2794 char *ren2_dst;
2795 ren2 = lookup->util;
2796 ren2_dst = ren2->pair->two->path;
2797 if (strcmp(ren1_dst, ren2_dst) != 0)
2798 BUG("ren1_dst != ren2_dst");
2800 clean_merge = 0;
2801 ren2->processed = 1;
2803 * BUG: We should only mark src_entry as processed
2804 * if we are not dealing with a rename + add-source
2805 * case.
2807 ren2->src_entry->processed = 1;
2809 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2810 opt, ren1, ren2);
2811 } else {
2812 /* Renamed in 1, maybe changed in 2 */
2813 /* we only use sha1 and mode of these */
2814 struct diff_filespec src_other, dst_other;
2815 int try_merge;
2818 * unpack_trees loads entries from common-commit
2819 * into stage 1, from head-commit into stage 2, and
2820 * from merge-commit into stage 3. We keep track
2821 * of which side corresponds to the rename.
2823 int renamed_stage = a_renames == renames1 ? 2 : 3;
2824 int other_stage = a_renames == renames1 ? 3 : 2;
2827 * Directory renames have a funny corner case...
2829 int renamed_to_self = !strcmp(ren1_src, ren1_dst);
2831 /* BUG: We should only remove ren1_src in the base
2832 * stage and in other_stage (think of rename +
2833 * add-source case).
2835 if (!renamed_to_self)
2836 remove_file(opt, 1, ren1_src,
2837 renamed_stage == 2 ||
2838 !was_tracked(opt, ren1_src));
2840 oidcpy(&src_other.oid,
2841 &ren1->src_entry->stages[other_stage].oid);
2842 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2843 oidcpy(&dst_other.oid,
2844 &ren1->dst_entry->stages[other_stage].oid);
2845 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2846 try_merge = 0;
2848 if (oideq(&src_other.oid, null_oid()) &&
2849 ren1->dir_rename_original_type == 'A') {
2850 setup_rename_conflict_info(RENAME_VIA_DIR,
2851 opt, ren1, NULL);
2852 } else if (renamed_to_self) {
2853 setup_rename_conflict_info(RENAME_NORMAL,
2854 opt, ren1, NULL);
2855 } else if (oideq(&src_other.oid, null_oid())) {
2856 setup_rename_conflict_info(RENAME_DELETE,
2857 opt, ren1, NULL);
2858 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2859 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2861 * Added file on the other side identical to
2862 * the file being renamed: clean merge.
2863 * Also, there is no need to overwrite the
2864 * file already in the working copy, so call
2865 * update_file_flags() instead of
2866 * update_file().
2868 if (update_file_flags(opt,
2869 ren1->pair->two,
2870 ren1_dst,
2871 1, /* update_cache */
2872 0 /* update_wd */))
2873 clean_merge = -1;
2874 } else if (!oideq(&dst_other.oid, null_oid())) {
2876 * Probably not a clean merge, but it's
2877 * premature to set clean_merge to 0 here,
2878 * because if the rename merges cleanly and
2879 * the merge exactly matches the newly added
2880 * file, then the merge will be clean.
2882 setup_rename_conflict_info(RENAME_ADD,
2883 opt, ren1, NULL);
2884 } else
2885 try_merge = 1;
2887 if (clean_merge < 0)
2888 goto cleanup_and_return;
2889 if (try_merge) {
2890 struct diff_filespec *o, *a, *b;
2891 src_other.path = (char *)ren1_src;
2893 o = ren1->pair->one;
2894 if (a_renames == renames1) {
2895 a = ren1->pair->two;
2896 b = &src_other;
2897 } else {
2898 b = ren1->pair->two;
2899 a = &src_other;
2901 update_entry(ren1->dst_entry, o, a, b);
2902 setup_rename_conflict_info(RENAME_NORMAL,
2903 opt, ren1, NULL);
2907 cleanup_and_return:
2908 string_list_clear(&a_by_dst, 0);
2909 string_list_clear(&b_by_dst, 0);
2911 return clean_merge;
2914 struct rename_info {
2915 struct string_list *head_renames;
2916 struct string_list *merge_renames;
2919 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2920 struct hashmap *dir_renames)
2922 struct hashmap_iter iter;
2923 struct dir_rename_entry *e;
2925 hashmap_for_each_entry(dir_renames, &iter, e,
2926 ent /* member name */) {
2927 free(e->dir);
2928 strbuf_release(&e->new_dir);
2929 /* possible_new_dirs already cleared in get_directory_renames */
2931 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2932 free(dir_renames);
2934 free(pairs->queue);
2935 free(pairs);
2938 static int detect_and_process_renames(struct merge_options *opt,
2939 struct tree *common,
2940 struct tree *head,
2941 struct tree *merge,
2942 struct string_list *entries,
2943 struct rename_info *ri)
2945 struct diff_queue_struct *head_pairs, *merge_pairs;
2946 struct hashmap *dir_re_head, *dir_re_merge;
2947 int clean = 1;
2949 ri->head_renames = NULL;
2950 ri->merge_renames = NULL;
2952 if (!merge_detect_rename(opt))
2953 return 1;
2955 head_pairs = get_diffpairs(opt, common, head);
2956 merge_pairs = get_diffpairs(opt, common, merge);
2958 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2959 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2960 !opt->priv->call_depth)) {
2961 dir_re_head = get_directory_renames(head_pairs);
2962 dir_re_merge = get_directory_renames(merge_pairs);
2964 handle_directory_level_conflicts(opt,
2965 dir_re_head, head,
2966 dir_re_merge, merge);
2967 } else {
2968 dir_re_head = xmalloc(sizeof(*dir_re_head));
2969 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2970 dir_rename_init(dir_re_head);
2971 dir_rename_init(dir_re_merge);
2974 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2975 dir_re_merge, dir_re_head, head,
2976 common, head, merge, entries,
2977 &clean);
2978 if (clean < 0)
2979 goto cleanup;
2980 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2981 dir_re_head, dir_re_merge, merge,
2982 common, head, merge, entries,
2983 &clean);
2984 if (clean < 0)
2985 goto cleanup;
2986 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2988 cleanup:
2990 * Some cleanup is deferred until cleanup_renames() because the
2991 * data structures are still needed and referenced in
2992 * process_entry(). But there are a few things we can free now.
2994 initial_cleanup_rename(head_pairs, dir_re_head);
2995 initial_cleanup_rename(merge_pairs, dir_re_merge);
2997 return clean;
3000 static void final_cleanup_rename(struct string_list *rename)
3002 const struct rename *re;
3003 int i;
3005 if (!rename)
3006 return;
3008 for (i = 0; i < rename->nr; i++) {
3009 re = rename->items[i].util;
3010 diff_free_filepair(re->pair);
3012 string_list_clear(rename, 1);
3013 free(rename);
3016 static void final_cleanup_renames(struct rename_info *re_info)
3018 final_cleanup_rename(re_info->head_renames);
3019 final_cleanup_rename(re_info->merge_renames);
3022 static int read_oid_strbuf(struct merge_options *opt,
3023 const struct object_id *oid,
3024 struct strbuf *dst)
3026 void *buf;
3027 enum object_type type;
3028 unsigned long size;
3029 buf = repo_read_object_file(the_repository, oid, &type, &size);
3030 if (!buf)
3031 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3032 if (type != OBJ_BLOB) {
3033 free(buf);
3034 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3036 strbuf_attach(dst, buf, size, size + 1);
3037 return 0;
3040 static int blob_unchanged(struct merge_options *opt,
3041 const struct diff_filespec *o,
3042 const struct diff_filespec *a,
3043 int renormalize, const char *path)
3045 struct strbuf obuf = STRBUF_INIT;
3046 struct strbuf abuf = STRBUF_INIT;
3047 int ret = 0; /* assume changed for safety */
3048 struct index_state *idx = opt->repo->index;
3050 if (a->mode != o->mode)
3051 return 0;
3052 if (oideq(&o->oid, &a->oid))
3053 return 1;
3054 if (!renormalize)
3055 return 0;
3057 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3058 read_oid_strbuf(opt, &a->oid, &abuf))
3059 goto error_return;
3061 * Note: binary | is used so that both renormalizations are
3062 * performed. Comparison can be skipped if both files are
3063 * unchanged since their sha1s have already been compared.
3065 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3066 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3067 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3069 error_return:
3070 strbuf_release(&obuf);
3071 strbuf_release(&abuf);
3072 return ret;
3075 static int handle_modify_delete(struct merge_options *opt,
3076 const char *path,
3077 const struct diff_filespec *o,
3078 const struct diff_filespec *a,
3079 const struct diff_filespec *b)
3081 const char *modify_branch, *delete_branch;
3082 const struct diff_filespec *changed;
3084 if (is_valid(a)) {
3085 modify_branch = opt->branch1;
3086 delete_branch = opt->branch2;
3087 changed = a;
3088 } else {
3089 modify_branch = opt->branch2;
3090 delete_branch = opt->branch1;
3091 changed = b;
3094 return handle_change_delete(opt,
3095 path, NULL,
3096 o, changed,
3097 modify_branch, delete_branch,
3098 _("modify"), _("modified"));
3101 static int handle_content_merge(struct merge_file_info *mfi,
3102 struct merge_options *opt,
3103 const char *path,
3104 int is_dirty,
3105 const struct diff_filespec *o,
3106 const struct diff_filespec *a,
3107 const struct diff_filespec *b,
3108 struct rename_conflict_info *ci)
3110 const char *reason = _("content");
3111 unsigned df_conflict_remains = 0;
3113 if (!is_valid(o))
3114 reason = _("add/add");
3116 assert(o->path && a->path && b->path);
3117 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3118 S_ISGITLINK(ci->ren1->pair->two->mode)))
3119 df_conflict_remains = 1;
3121 if (merge_mode_and_contents(opt, o, a, b, path,
3122 opt->branch1, opt->branch2,
3123 opt->priv->call_depth * 2, mfi))
3124 return -1;
3127 * We can skip updating the working tree file iff:
3128 * a) The merge is clean
3129 * b) The merge matches what was in HEAD (content, mode, pathname)
3130 * c) The target path is usable (i.e. not involved in D/F conflict)
3132 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3133 !df_conflict_remains) {
3134 int pos;
3135 struct cache_entry *ce;
3137 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3138 if (add_cacheinfo(opt, &mfi->blob, path,
3139 0, (!opt->priv->call_depth && !is_dirty), 0))
3140 return -1;
3142 * However, add_cacheinfo() will delete the old cache entry
3143 * and add a new one. We need to copy over any skip_worktree
3144 * flag to avoid making the file appear as if it were
3145 * deleted by the user.
3147 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3148 ce = opt->priv->orig_index.cache[pos];
3149 if (ce_skip_worktree(ce)) {
3150 pos = index_name_pos(opt->repo->index, path, strlen(path));
3151 ce = opt->repo->index->cache[pos];
3152 ce->ce_flags |= CE_SKIP_WORKTREE;
3154 return mfi->clean;
3157 if (!mfi->clean) {
3158 if (S_ISGITLINK(mfi->blob.mode))
3159 reason = _("submodule");
3160 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3161 reason, path);
3162 if (ci && !df_conflict_remains)
3163 if (update_stages(opt, path, o, a, b))
3164 return -1;
3167 if (df_conflict_remains || is_dirty) {
3168 char *new_path;
3169 if (opt->priv->call_depth) {
3170 remove_file_from_index(opt->repo->index, path);
3171 } else {
3172 if (!mfi->clean) {
3173 if (update_stages(opt, path, o, a, b))
3174 return -1;
3175 } else {
3176 int file_from_stage2 = was_tracked(opt, path);
3178 if (update_stages(opt, path, NULL,
3179 file_from_stage2 ? &mfi->blob : NULL,
3180 file_from_stage2 ? NULL : &mfi->blob))
3181 return -1;
3185 new_path = unique_path(opt, path, ci->ren1->branch);
3186 if (is_dirty) {
3187 output(opt, 1, _("Refusing to lose dirty file at %s"),
3188 path);
3190 output(opt, 1, _("Adding as %s instead"), new_path);
3191 if (update_file(opt, 0, &mfi->blob, new_path)) {
3192 free(new_path);
3193 return -1;
3195 free(new_path);
3196 mfi->clean = 0;
3197 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3198 return -1;
3199 return !is_dirty && mfi->clean;
3202 static int handle_rename_normal(struct merge_options *opt,
3203 const char *path,
3204 const struct diff_filespec *o,
3205 const struct diff_filespec *a,
3206 const struct diff_filespec *b,
3207 struct rename_conflict_info *ci)
3209 struct rename *ren = ci->ren1;
3210 struct merge_file_info mfi;
3211 int clean;
3213 /* Merge the content and write it out */
3214 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3215 o, a, b, ci);
3217 if (clean &&
3218 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3219 ren->dir_rename_original_dest) {
3220 if (update_stages(opt, path,
3221 &mfi.blob, &mfi.blob, &mfi.blob))
3222 return -1;
3223 clean = 0; /* not clean, but conflicted */
3225 return clean;
3228 static void dir_rename_warning(const char *msg,
3229 int is_add,
3230 int clean,
3231 struct merge_options *opt,
3232 struct rename *ren)
3234 const char *other_branch;
3235 other_branch = (ren->branch == opt->branch1 ?
3236 opt->branch2 : opt->branch1);
3237 if (is_add) {
3238 output(opt, clean ? 2 : 1, msg,
3239 ren->pair->one->path, ren->branch,
3240 other_branch, ren->pair->two->path);
3241 return;
3243 output(opt, clean ? 2 : 1, msg,
3244 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3245 other_branch, ren->pair->two->path);
3247 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3248 struct rename *ren)
3250 const char *msg;
3251 int clean = 1, is_add;
3253 if (!ren)
3254 return clean;
3256 /* Return early if ren was not affected/created by a directory rename */
3257 if (!ren->dir_rename_original_dest)
3258 return clean;
3260 /* Sanity checks */
3261 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3262 assert(ren->dir_rename_original_type == 'A' ||
3263 ren->dir_rename_original_type == 'R');
3265 /* Check whether to treat directory renames as a conflict */
3266 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3268 is_add = (ren->dir_rename_original_type == 'A');
3269 if (ren->dir_rename_original_type == 'A' && clean) {
3270 msg = _("Path updated: %s added in %s inside a "
3271 "directory that was renamed in %s; moving it to %s.");
3272 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3273 msg = _("CONFLICT (file location): %s added in %s "
3274 "inside a directory that was renamed in %s, "
3275 "suggesting it should perhaps be moved to %s.");
3276 } else if (ren->dir_rename_original_type == 'R' && clean) {
3277 msg = _("Path updated: %s renamed to %s in %s, inside a "
3278 "directory that was renamed in %s; moving it to %s.");
3279 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3280 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3281 "inside a directory that was renamed in %s, "
3282 "suggesting it should perhaps be moved to %s.");
3283 } else {
3284 BUG("Impossible dir_rename_original_type/clean combination");
3286 dir_rename_warning(msg, is_add, clean, opt, ren);
3288 return clean;
3291 /* Per entry merge function */
3292 static int process_entry(struct merge_options *opt,
3293 const char *path, struct stage_data *entry)
3295 int clean_merge = 1;
3296 int normalize = opt->renormalize;
3298 struct diff_filespec *o = &entry->stages[1];
3299 struct diff_filespec *a = &entry->stages[2];
3300 struct diff_filespec *b = &entry->stages[3];
3301 int o_valid = is_valid(o);
3302 int a_valid = is_valid(a);
3303 int b_valid = is_valid(b);
3304 o->path = a->path = b->path = (char*)path;
3306 entry->processed = 1;
3307 if (entry->rename_conflict_info) {
3308 struct rename_conflict_info *ci = entry->rename_conflict_info;
3309 struct diff_filespec *temp;
3310 int path_clean;
3312 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3313 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3316 * For cases with a single rename, {o,a,b}->path have all been
3317 * set to the rename target path; we need to set two of these
3318 * back to the rename source.
3319 * For rename/rename conflicts, we'll manually fix paths below.
3321 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3322 o->path = temp->path = ci->ren1->pair->one->path;
3323 if (ci->ren2) {
3324 assert(opt->branch1 == ci->ren1->branch);
3327 switch (ci->rename_type) {
3328 case RENAME_NORMAL:
3329 case RENAME_ONE_FILE_TO_ONE:
3330 clean_merge = handle_rename_normal(opt, path, o, a, b,
3331 ci);
3332 break;
3333 case RENAME_VIA_DIR:
3334 clean_merge = handle_rename_via_dir(opt, ci);
3335 break;
3336 case RENAME_ADD:
3338 * Probably unclean merge, but if the renamed file
3339 * merges cleanly and the result can then be
3340 * two-way merged cleanly with the added file, I
3341 * guess it's a clean merge?
3343 clean_merge = handle_rename_add(opt, ci);
3344 break;
3345 case RENAME_DELETE:
3346 clean_merge = 0;
3347 if (handle_rename_delete(opt, ci))
3348 clean_merge = -1;
3349 break;
3350 case RENAME_ONE_FILE_TO_TWO:
3352 * Manually fix up paths; note:
3353 * ren[12]->pair->one->path are equal.
3355 o->path = ci->ren1->pair->one->path;
3356 a->path = ci->ren1->pair->two->path;
3357 b->path = ci->ren2->pair->two->path;
3359 clean_merge = 0;
3360 if (handle_rename_rename_1to2(opt, ci))
3361 clean_merge = -1;
3362 break;
3363 case RENAME_TWO_FILES_TO_ONE:
3365 * Manually fix up paths; note,
3366 * ren[12]->pair->two->path are actually equal.
3368 o->path = NULL;
3369 a->path = ci->ren1->pair->two->path;
3370 b->path = ci->ren2->pair->two->path;
3373 * Probably unclean merge, but if the two renamed
3374 * files merge cleanly and the two resulting files
3375 * can then be two-way merged cleanly, I guess it's
3376 * a clean merge?
3378 clean_merge = handle_rename_rename_2to1(opt, ci);
3379 break;
3380 default:
3381 entry->processed = 0;
3382 break;
3384 if (path_clean < clean_merge)
3385 clean_merge = path_clean;
3386 } else if (o_valid && (!a_valid || !b_valid)) {
3387 /* Case A: Deleted in one */
3388 if ((!a_valid && !b_valid) ||
3389 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3390 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3391 /* Deleted in both or deleted in one and
3392 * unchanged in the other */
3393 if (a_valid)
3394 output(opt, 2, _("Removing %s"), path);
3395 /* do not touch working file if it did not exist */
3396 remove_file(opt, 1, path, !a_valid);
3397 } else {
3398 /* Modify/delete; deleted side may have put a directory in the way */
3399 clean_merge = 0;
3400 if (handle_modify_delete(opt, path, o, a, b))
3401 clean_merge = -1;
3403 } else if ((!o_valid && a_valid && !b_valid) ||
3404 (!o_valid && !a_valid && b_valid)) {
3405 /* Case B: Added in one. */
3406 /* [nothing|directory] -> ([nothing|directory], file) */
3408 const char *add_branch;
3409 const char *other_branch;
3410 const char *conf;
3411 const struct diff_filespec *contents;
3413 if (a_valid) {
3414 add_branch = opt->branch1;
3415 other_branch = opt->branch2;
3416 contents = a;
3417 conf = _("file/directory");
3418 } else {
3419 add_branch = opt->branch2;
3420 other_branch = opt->branch1;
3421 contents = b;
3422 conf = _("directory/file");
3424 if (dir_in_way(opt->repo->index, path,
3425 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3426 0)) {
3427 char *new_path = unique_path(opt, path, add_branch);
3428 clean_merge = 0;
3429 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3430 "Adding %s as %s"),
3431 conf, path, other_branch, path, new_path);
3432 if (update_file(opt, 0, contents, new_path))
3433 clean_merge = -1;
3434 else if (opt->priv->call_depth)
3435 remove_file_from_index(opt->repo->index, path);
3436 free(new_path);
3437 } else {
3438 output(opt, 2, _("Adding %s"), path);
3439 /* do not overwrite file if already present */
3440 if (update_file_flags(opt, contents, path, 1, !a_valid))
3441 clean_merge = -1;
3443 } else if (a_valid && b_valid) {
3444 if (!o_valid) {
3445 /* Case C: Added in both (check for same permissions) */
3446 output(opt, 1,
3447 _("CONFLICT (add/add): Merge conflict in %s"),
3448 path);
3449 clean_merge = handle_file_collision(opt,
3450 path, NULL, NULL,
3451 opt->branch1,
3452 opt->branch2,
3453 a, b);
3454 } else {
3455 /* case D: Modified in both, but differently. */
3456 struct merge_file_info mfi;
3457 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3458 clean_merge = handle_content_merge(&mfi, opt, path,
3459 is_dirty,
3460 o, a, b, NULL);
3462 } else if (!o_valid && !a_valid && !b_valid) {
3464 * this entry was deleted altogether. a_mode == 0 means
3465 * we had that path and want to actively remove it.
3467 remove_file(opt, 1, path, !a->mode);
3468 } else
3469 BUG("fatal merge failure, shouldn't happen.");
3471 return clean_merge;
3474 static int merge_trees_internal(struct merge_options *opt,
3475 struct tree *head,
3476 struct tree *merge,
3477 struct tree *merge_base,
3478 struct tree **result)
3480 struct index_state *istate = opt->repo->index;
3481 int code, clean;
3483 if (opt->subtree_shift) {
3484 merge = shift_tree_object(opt->repo, head, merge,
3485 opt->subtree_shift);
3486 merge_base = shift_tree_object(opt->repo, head, merge_base,
3487 opt->subtree_shift);
3490 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3491 output(opt, 0, _("Already up to date."));
3492 *result = head;
3493 return 1;
3496 code = unpack_trees_start(opt, merge_base, head, merge);
3498 if (code != 0) {
3499 if (show(opt, 4) || opt->priv->call_depth)
3500 err(opt, _("merging of trees %s and %s failed"),
3501 oid_to_hex(&head->object.oid),
3502 oid_to_hex(&merge->object.oid));
3503 unpack_trees_finish(opt);
3504 return -1;
3507 if (unmerged_index(istate)) {
3508 struct string_list *entries;
3509 struct rename_info re_info;
3510 int i;
3512 * Only need the hashmap while processing entries, so
3513 * initialize it here and free it when we are done running
3514 * through the entries. Keeping it in the merge_options as
3515 * opposed to decaring a local hashmap is for convenience
3516 * so that we don't have to pass it to around.
3518 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3519 NULL, 512);
3520 get_files_dirs(opt, head);
3521 get_files_dirs(opt, merge);
3523 entries = get_unmerged(opt->repo->index);
3524 clean = detect_and_process_renames(opt, merge_base, head, merge,
3525 entries, &re_info);
3526 record_df_conflict_files(opt, entries);
3527 if (clean < 0)
3528 goto cleanup;
3529 for (i = entries->nr-1; 0 <= i; i--) {
3530 const char *path = entries->items[i].string;
3531 struct stage_data *e = entries->items[i].util;
3532 if (!e->processed) {
3533 int ret = process_entry(opt, path, e);
3534 if (!ret)
3535 clean = 0;
3536 else if (ret < 0) {
3537 clean = ret;
3538 goto cleanup;
3542 for (i = 0; i < entries->nr; i++) {
3543 struct stage_data *e = entries->items[i].util;
3544 if (!e->processed)
3545 BUG("unprocessed path??? %s",
3546 entries->items[i].string);
3549 cleanup:
3550 final_cleanup_renames(&re_info);
3552 string_list_clear(entries, 1);
3553 free(entries);
3555 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3556 struct path_hashmap_entry, e);
3558 if (clean < 0) {
3559 unpack_trees_finish(opt);
3560 return clean;
3563 else
3564 clean = 1;
3566 unpack_trees_finish(opt);
3568 if (opt->priv->call_depth &&
3569 !(*result = write_in_core_index_as_tree(opt->repo)))
3570 return -1;
3572 return clean;
3576 * Merge the commits h1 and h2, returning a flag (int) indicating the
3577 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3578 * virtual commit and write its location to *result.
3580 static int merge_recursive_internal(struct merge_options *opt,
3581 struct commit *h1,
3582 struct commit *h2,
3583 struct commit_list *merge_bases,
3584 struct commit **result)
3586 struct commit_list *iter;
3587 struct commit *merged_merge_bases;
3588 struct tree *result_tree;
3589 int clean;
3590 const char *ancestor_name;
3591 struct strbuf merge_base_abbrev = STRBUF_INIT;
3593 if (show(opt, 4)) {
3594 output(opt, 4, _("Merging:"));
3595 output_commit_title(opt, h1);
3596 output_commit_title(opt, h2);
3599 if (!merge_bases) {
3600 merge_bases = repo_get_merge_bases(the_repository, h1, h2);
3601 merge_bases = reverse_commit_list(merge_bases);
3604 if (show(opt, 5)) {
3605 unsigned cnt = commit_list_count(merge_bases);
3607 output(opt, 5, Q_("found %u common ancestor:",
3608 "found %u common ancestors:", cnt), cnt);
3609 for (iter = merge_bases; iter; iter = iter->next)
3610 output_commit_title(opt, iter->item);
3613 merged_merge_bases = pop_commit(&merge_bases);
3614 if (!merged_merge_bases) {
3615 /* if there is no common ancestor, use an empty tree */
3616 struct tree *tree;
3618 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3619 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3620 "ancestor");
3621 ancestor_name = "empty tree";
3622 } else if (opt->ancestor && !opt->priv->call_depth) {
3623 ancestor_name = opt->ancestor;
3624 } else if (merge_bases) {
3625 ancestor_name = "merged common ancestors";
3626 } else {
3627 strbuf_add_unique_abbrev(&merge_base_abbrev,
3628 &merged_merge_bases->object.oid,
3629 DEFAULT_ABBREV);
3630 ancestor_name = merge_base_abbrev.buf;
3633 for (iter = merge_bases; iter; iter = iter->next) {
3634 const char *saved_b1, *saved_b2;
3635 opt->priv->call_depth++;
3637 * When the merge fails, the result contains files
3638 * with conflict markers. The cleanness flag is
3639 * ignored (unless indicating an error), it was never
3640 * actually used, as result of merge_trees has always
3641 * overwritten it: the committed "conflicts" were
3642 * already resolved.
3644 discard_index(opt->repo->index);
3645 saved_b1 = opt->branch1;
3646 saved_b2 = opt->branch2;
3647 opt->branch1 = "Temporary merge branch 1";
3648 opt->branch2 = "Temporary merge branch 2";
3649 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3650 NULL, &merged_merge_bases) < 0)
3651 return -1;
3652 opt->branch1 = saved_b1;
3653 opt->branch2 = saved_b2;
3654 opt->priv->call_depth--;
3656 if (!merged_merge_bases)
3657 return err(opt, _("merge returned no commit"));
3661 * FIXME: Since merge_recursive_internal() is only ever called by
3662 * places that ensure the index is loaded first
3663 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3664 * case where the merge base was unique that means when we get here
3665 * we immediately discard the index and re-read it, which is a
3666 * complete waste of time. We should only be discarding and
3667 * re-reading if we were forced to recurse.
3669 discard_index(opt->repo->index);
3670 if (!opt->priv->call_depth)
3671 repo_read_index(opt->repo);
3673 opt->ancestor = ancestor_name;
3674 clean = merge_trees_internal(opt,
3675 repo_get_commit_tree(opt->repo, h1),
3676 repo_get_commit_tree(opt->repo, h2),
3677 repo_get_commit_tree(opt->repo,
3678 merged_merge_bases),
3679 &result_tree);
3680 strbuf_release(&merge_base_abbrev);
3681 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3682 if (clean < 0) {
3683 flush_output(opt);
3684 return clean;
3687 if (opt->priv->call_depth) {
3688 *result = make_virtual_commit(opt->repo, result_tree,
3689 "merged tree");
3690 commit_list_insert(h1, &(*result)->parents);
3691 commit_list_insert(h2, &(*result)->parents->next);
3693 return clean;
3696 static int merge_start(struct merge_options *opt, struct tree *head)
3698 struct strbuf sb = STRBUF_INIT;
3700 /* Sanity checks on opt */
3701 assert(opt->repo);
3703 assert(opt->branch1 && opt->branch2);
3705 assert(opt->detect_renames >= -1 &&
3706 opt->detect_renames <= DIFF_DETECT_COPY);
3707 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3708 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3709 assert(opt->rename_limit >= -1);
3710 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3711 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3713 assert(opt->xdl_opts >= 0);
3714 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3715 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3717 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3718 assert(opt->buffer_output <= 2);
3719 assert(opt->obuf.len == 0);
3721 assert(opt->priv == NULL);
3723 /* Not supported; option specific to merge-ort */
3724 assert(!opt->record_conflict_msgs_as_headers);
3725 assert(!opt->msg_header_prefix);
3727 /* Sanity check on repo state; index must match head */
3728 if (repo_index_has_changes(opt->repo, head, &sb)) {
3729 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3730 sb.buf);
3731 strbuf_release(&sb);
3732 return -1;
3735 CALLOC_ARRAY(opt->priv, 1);
3736 string_list_init_dup(&opt->priv->df_conflict_file_set);
3737 return 0;
3740 static void merge_finalize(struct merge_options *opt)
3742 flush_output(opt);
3743 if (!opt->priv->call_depth && opt->buffer_output < 2)
3744 strbuf_release(&opt->obuf);
3745 if (show(opt, 2))
3746 diff_warn_rename_limit("merge.renamelimit",
3747 opt->priv->needed_rename_limit, 0);
3748 FREE_AND_NULL(opt->priv);
3751 int merge_trees(struct merge_options *opt,
3752 struct tree *head,
3753 struct tree *merge,
3754 struct tree *merge_base)
3756 int clean;
3757 struct tree *ignored;
3759 assert(opt->ancestor != NULL);
3761 if (merge_start(opt, head))
3762 return -1;
3763 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3764 merge_finalize(opt);
3766 return clean;
3769 int merge_recursive(struct merge_options *opt,
3770 struct commit *h1,
3771 struct commit *h2,
3772 struct commit_list *merge_bases,
3773 struct commit **result)
3775 int clean;
3777 assert(opt->ancestor == NULL ||
3778 !strcmp(opt->ancestor, "constructed merge base"));
3780 prepare_repo_settings(opt->repo);
3781 opt->repo->settings.command_requires_full_index = 1;
3783 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3784 return -1;
3785 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3786 merge_finalize(opt);
3788 return clean;
3791 static struct commit *get_ref(struct repository *repo,
3792 const struct object_id *oid,
3793 const char *name)
3795 struct object *object;
3797 object = deref_tag(repo, parse_object(repo, oid),
3798 name, strlen(name));
3799 if (!object)
3800 return NULL;
3801 if (object->type == OBJ_TREE)
3802 return make_virtual_commit(repo, (struct tree*)object, name);
3803 if (object->type != OBJ_COMMIT)
3804 return NULL;
3805 if (repo_parse_commit(repo, (struct commit *)object))
3806 return NULL;
3807 return (struct commit *)object;
3810 int merge_recursive_generic(struct merge_options *opt,
3811 const struct object_id *head,
3812 const struct object_id *merge,
3813 int num_merge_bases,
3814 const struct object_id **merge_bases,
3815 struct commit **result)
3817 int clean;
3818 struct lock_file lock = LOCK_INIT;
3819 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3820 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3821 struct commit_list *ca = NULL;
3823 if (merge_bases) {
3824 int i;
3825 for (i = 0; i < num_merge_bases; ++i) {
3826 struct commit *base;
3827 if (!(base = get_ref(opt->repo, merge_bases[i],
3828 oid_to_hex(merge_bases[i]))))
3829 return err(opt, _("Could not parse object '%s'"),
3830 oid_to_hex(merge_bases[i]));
3831 commit_list_insert(base, &ca);
3833 if (num_merge_bases == 1)
3834 opt->ancestor = "constructed merge base";
3837 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3838 clean = merge_recursive(opt, head_commit, next_commit, ca,
3839 result);
3840 if (clean < 0) {
3841 rollback_lock_file(&lock);
3842 return clean;
3845 if (write_locked_index(opt->repo->index, &lock,
3846 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3847 return err(opt, _("Unable to write index."));
3849 return clean ? 0 : 1;
3852 static void merge_recursive_config(struct merge_options *opt)
3854 char *value = NULL;
3855 int renormalize = 0;
3856 git_config_get_int("merge.verbosity", &opt->verbosity);
3857 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3858 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3859 git_config_get_bool("merge.renormalize", &renormalize);
3860 opt->renormalize = renormalize;
3861 if (!git_config_get_string("diff.renames", &value)) {
3862 opt->detect_renames = git_config_rename("diff.renames", value);
3863 free(value);
3865 if (!git_config_get_string("merge.renames", &value)) {
3866 opt->detect_renames = git_config_rename("merge.renames", value);
3867 free(value);
3869 if (!git_config_get_string("merge.directoryrenames", &value)) {
3870 int boolval = git_parse_maybe_bool(value);
3871 if (0 <= boolval) {
3872 opt->detect_directory_renames = boolval ?
3873 MERGE_DIRECTORY_RENAMES_TRUE :
3874 MERGE_DIRECTORY_RENAMES_NONE;
3875 } else if (!strcasecmp(value, "conflict")) {
3876 opt->detect_directory_renames =
3877 MERGE_DIRECTORY_RENAMES_CONFLICT;
3878 } /* avoid erroring on values from future versions of git */
3879 free(value);
3881 git_config(git_xmerge_config, NULL);
3884 void init_merge_options(struct merge_options *opt,
3885 struct repository *repo)
3887 const char *merge_verbosity;
3888 memset(opt, 0, sizeof(struct merge_options));
3890 opt->repo = repo;
3892 opt->detect_renames = -1;
3893 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3894 opt->rename_limit = -1;
3896 opt->verbosity = 2;
3897 opt->buffer_output = 1;
3898 strbuf_init(&opt->obuf, 0);
3900 opt->renormalize = 0;
3902 merge_recursive_config(opt);
3903 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3904 if (merge_verbosity)
3905 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3906 if (opt->verbosity >= 5)
3907 opt->buffer_output = 0;
3911 * For now, members of merge_options do not need deep copying, but
3912 * it may change in the future, in which case we would need to update
3913 * this, and also make a matching change to clear_merge_options() to
3914 * release the resources held by a copied instance.
3916 void copy_merge_options(struct merge_options *dst, struct merge_options *src)
3918 *dst = *src;
3921 void clear_merge_options(struct merge_options *opt UNUSED)
3923 ; /* no-op as our copy is shallow right now */
3926 int parse_merge_opt(struct merge_options *opt, const char *s)
3928 const char *arg;
3930 if (!s || !*s)
3931 return -1;
3932 if (!strcmp(s, "ours"))
3933 opt->recursive_variant = MERGE_VARIANT_OURS;
3934 else if (!strcmp(s, "theirs"))
3935 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3936 else if (!strcmp(s, "subtree"))
3937 opt->subtree_shift = "";
3938 else if (skip_prefix(s, "subtree=", &arg))
3939 opt->subtree_shift = arg;
3940 else if (!strcmp(s, "patience"))
3941 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3942 else if (!strcmp(s, "histogram"))
3943 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3944 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3945 long value = parse_algorithm_value(arg);
3946 if (value < 0)
3947 return -1;
3948 /* clear out previous settings */
3949 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3950 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3951 opt->xdl_opts |= value;
3953 else if (!strcmp(s, "ignore-space-change"))
3954 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3955 else if (!strcmp(s, "ignore-all-space"))
3956 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3957 else if (!strcmp(s, "ignore-space-at-eol"))
3958 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3959 else if (!strcmp(s, "ignore-cr-at-eol"))
3960 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3961 else if (!strcmp(s, "renormalize"))
3962 opt->renormalize = 1;
3963 else if (!strcmp(s, "no-renormalize"))
3964 opt->renormalize = 0;
3965 else if (!strcmp(s, "no-renames"))
3966 opt->detect_renames = 0;
3967 else if (!strcmp(s, "find-renames")) {
3968 opt->detect_renames = 1;
3969 opt->rename_score = 0;
3971 else if (skip_prefix(s, "find-renames=", &arg) ||
3972 skip_prefix(s, "rename-threshold=", &arg)) {
3973 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3974 return -1;
3975 opt->detect_renames = 1;
3978 * Please update $__git_merge_strategy_options in
3979 * git-completion.bash when you add new options
3981 else
3982 return -1;
3983 return 0;