2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
10 #include "cache-tree.h"
11 #include "object-store.h"
12 #include "repository.h"
16 #include "tree-walk.h"
21 #include "unpack-trees.h"
22 #include "string-list.h"
23 #include "xdiff-interface.h"
26 #include "merge-recursive.h"
28 #include "submodule.h"
30 #include "commit-reach.h"
32 struct path_hashmap_entry
{
33 struct hashmap_entry e
;
34 char path
[FLEX_ARRAY
];
37 static int path_hashmap_cmp(const void *cmp_data
,
39 const void *entry_or_key
,
42 const struct path_hashmap_entry
*a
= entry
;
43 const struct path_hashmap_entry
*b
= entry_or_key
;
44 const char *key
= keydata
;
47 return strcasecmp(a
->path
, key
? key
: b
->path
);
49 return strcmp(a
->path
, key
? key
: b
->path
);
52 static unsigned int path_hash(const char *path
)
54 return ignore_case
? strihash(path
) : strhash(path
);
57 static struct dir_rename_entry
*dir_rename_find_entry(struct hashmap
*hashmap
,
60 struct dir_rename_entry key
;
64 hashmap_entry_init(&key
, strhash(dir
));
66 return hashmap_get(hashmap
, &key
, NULL
);
69 static int dir_rename_cmp(const void *unused_cmp_data
,
71 const void *entry_or_key
,
72 const void *unused_keydata
)
74 const struct dir_rename_entry
*e1
= entry
;
75 const struct dir_rename_entry
*e2
= entry_or_key
;
77 return strcmp(e1
->dir
, e2
->dir
);
80 static void dir_rename_init(struct hashmap
*map
)
82 hashmap_init(map
, dir_rename_cmp
, NULL
, 0);
85 static void dir_rename_entry_init(struct dir_rename_entry
*entry
,
88 hashmap_entry_init(entry
, strhash(directory
));
89 entry
->dir
= directory
;
90 entry
->non_unique_new_dir
= 0;
91 strbuf_init(&entry
->new_dir
, 0);
92 string_list_init(&entry
->possible_new_dirs
, 0);
95 static struct collision_entry
*collision_find_entry(struct hashmap
*hashmap
,
98 struct collision_entry key
;
100 hashmap_entry_init(&key
, strhash(target_file
));
101 key
.target_file
= target_file
;
102 return hashmap_get(hashmap
, &key
, NULL
);
105 static int collision_cmp(void *unused_cmp_data
,
106 const struct collision_entry
*e1
,
107 const struct collision_entry
*e2
,
108 const void *unused_keydata
)
110 return strcmp(e1
->target_file
, e2
->target_file
);
113 static void collision_init(struct hashmap
*map
)
115 hashmap_init(map
, (hashmap_cmp_fn
) collision_cmp
, NULL
, 0);
118 static void flush_output(struct merge_options
*o
)
120 if (o
->buffer_output
< 2 && o
->obuf
.len
) {
121 fputs(o
->obuf
.buf
, stdout
);
122 strbuf_reset(&o
->obuf
);
126 static int err(struct merge_options
*o
, const char *err
, ...)
130 if (o
->buffer_output
< 2)
133 strbuf_complete(&o
->obuf
, '\n');
134 strbuf_addstr(&o
->obuf
, "error: ");
136 va_start(params
, err
);
137 strbuf_vaddf(&o
->obuf
, err
, params
);
139 if (o
->buffer_output
> 1)
140 strbuf_addch(&o
->obuf
, '\n');
142 error("%s", o
->obuf
.buf
);
143 strbuf_reset(&o
->obuf
);
149 static struct tree
*shift_tree_object(struct tree
*one
, struct tree
*two
,
150 const char *subtree_shift
)
152 struct object_id shifted
;
154 if (!*subtree_shift
) {
155 shift_tree(&one
->object
.oid
, &two
->object
.oid
, &shifted
, 0);
157 shift_tree_by(&one
->object
.oid
, &two
->object
.oid
, &shifted
,
160 if (oideq(&two
->object
.oid
, &shifted
))
162 return lookup_tree(the_repository
, &shifted
);
165 static struct commit
*make_virtual_commit(struct tree
*tree
, const char *comment
)
167 struct commit
*commit
= alloc_commit_node(the_repository
);
169 set_merge_remote_desc(commit
, comment
, (struct object
*)commit
);
170 commit
->maybe_tree
= tree
;
171 commit
->object
.parsed
= 1;
176 * Since we use get_tree_entry(), which does not put the read object into
177 * the object pool, we cannot rely on a == b.
179 static int oid_eq(const struct object_id
*a
, const struct object_id
*b
)
183 return a
&& b
&& oideq(a
, b
);
191 RENAME_ONE_FILE_TO_ONE
,
192 RENAME_ONE_FILE_TO_TWO
,
193 RENAME_TWO_FILES_TO_ONE
196 struct rename_conflict_info
{
197 enum rename_type rename_type
;
198 struct diff_filepair
*pair1
;
199 struct diff_filepair
*pair2
;
202 struct stage_data
*dst_entry1
;
203 struct stage_data
*dst_entry2
;
204 struct diff_filespec ren1_other
;
205 struct diff_filespec ren2_other
;
209 * Since we want to write the index eventually, we cannot reuse the index
210 * for these (temporary) data.
215 struct object_id oid
;
217 struct rename_conflict_info
*rename_conflict_info
;
218 unsigned processed
:1;
221 static inline void setup_rename_conflict_info(enum rename_type rename_type
,
222 struct diff_filepair
*pair1
,
223 struct diff_filepair
*pair2
,
226 struct stage_data
*dst_entry1
,
227 struct stage_data
*dst_entry2
,
228 struct merge_options
*o
,
229 struct stage_data
*src_entry1
,
230 struct stage_data
*src_entry2
)
232 int ostage1
= 0, ostage2
;
233 struct rename_conflict_info
*ci
;
236 * When we have two renames involved, it's easiest to get the
237 * correct things into stage 2 and 3, and to make sure that the
238 * content merge puts HEAD before the other branch if we just
239 * ensure that branch1 == o->branch1. So, simply flip arguments
240 * around if we don't have that.
242 if (dst_entry2
&& branch1
!= o
->branch1
) {
243 setup_rename_conflict_info(rename_type
,
246 dst_entry2
, dst_entry1
,
248 src_entry2
, src_entry1
);
252 ci
= xcalloc(1, sizeof(struct rename_conflict_info
));
253 ci
->rename_type
= rename_type
;
255 ci
->branch1
= branch1
;
256 ci
->branch2
= branch2
;
258 ci
->dst_entry1
= dst_entry1
;
259 dst_entry1
->rename_conflict_info
= ci
;
260 dst_entry1
->processed
= 0;
262 assert(!pair2
== !dst_entry2
);
264 ci
->dst_entry2
= dst_entry2
;
266 dst_entry2
->rename_conflict_info
= ci
;
270 * For each rename, there could have been
271 * modifications on the side of history where that
272 * file was not renamed.
274 if (rename_type
== RENAME_ADD
||
275 rename_type
== RENAME_TWO_FILES_TO_ONE
) {
276 ostage1
= o
->branch1
== branch1
? 3 : 2;
278 ci
->ren1_other
.path
= pair1
->one
->path
;
279 oidcpy(&ci
->ren1_other
.oid
, &src_entry1
->stages
[ostage1
].oid
);
280 ci
->ren1_other
.mode
= src_entry1
->stages
[ostage1
].mode
;
283 if (rename_type
== RENAME_TWO_FILES_TO_ONE
) {
284 ostage2
= ostage1
^ 1;
286 ci
->ren2_other
.path
= pair2
->one
->path
;
287 oidcpy(&ci
->ren2_other
.oid
, &src_entry2
->stages
[ostage2
].oid
);
288 ci
->ren2_other
.mode
= src_entry2
->stages
[ostage2
].mode
;
292 static int show(struct merge_options
*o
, int v
)
294 return (!o
->call_depth
&& o
->verbosity
>= v
) || o
->verbosity
>= 5;
297 __attribute__((format (printf
, 3, 4)))
298 static void output(struct merge_options
*o
, int v
, const char *fmt
, ...)
305 strbuf_addchars(&o
->obuf
, ' ', o
->call_depth
* 2);
308 strbuf_vaddf(&o
->obuf
, fmt
, ap
);
311 strbuf_addch(&o
->obuf
, '\n');
312 if (!o
->buffer_output
)
316 static void output_commit_title(struct merge_options
*o
, struct commit
*commit
)
318 struct merge_remote_desc
*desc
;
320 strbuf_addchars(&o
->obuf
, ' ', o
->call_depth
* 2);
321 desc
= merge_remote_util(commit
);
323 strbuf_addf(&o
->obuf
, "virtual %s\n", desc
->name
);
325 strbuf_add_unique_abbrev(&o
->obuf
, &commit
->object
.oid
,
327 strbuf_addch(&o
->obuf
, ' ');
328 if (parse_commit(commit
) != 0)
329 strbuf_addstr(&o
->obuf
, _("(bad commit)\n"));
332 const char *msg
= get_commit_buffer(commit
, NULL
);
333 int len
= find_commit_subject(msg
, &title
);
335 strbuf_addf(&o
->obuf
, "%.*s\n", len
, title
);
336 unuse_commit_buffer(commit
, msg
);
342 static int add_cacheinfo(struct merge_options
*o
,
343 unsigned int mode
, const struct object_id
*oid
,
344 const char *path
, int stage
, int refresh
, int options
)
346 struct cache_entry
*ce
;
349 ce
= make_cache_entry(&the_index
, mode
, oid
? oid
: &null_oid
, path
, stage
, 0);
351 return err(o
, _("add_cacheinfo failed for path '%s'; merge aborting."), path
);
353 ret
= add_cache_entry(ce
, options
);
355 struct cache_entry
*nce
;
357 nce
= refresh_cache_entry(&the_index
, ce
, CE_MATCH_REFRESH
| CE_MATCH_IGNORE_MISSING
);
359 return err(o
, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path
);
361 ret
= add_cache_entry(nce
, options
);
366 static void init_tree_desc_from_tree(struct tree_desc
*desc
, struct tree
*tree
)
369 init_tree_desc(desc
, tree
->buffer
, tree
->size
);
372 static int unpack_trees_start(struct merge_options
*o
,
378 struct tree_desc t
[3];
379 struct index_state tmp_index
= { NULL
};
381 memset(&o
->unpack_opts
, 0, sizeof(o
->unpack_opts
));
383 o
->unpack_opts
.index_only
= 1;
385 o
->unpack_opts
.update
= 1;
386 o
->unpack_opts
.merge
= 1;
387 o
->unpack_opts
.head_idx
= 2;
388 o
->unpack_opts
.fn
= threeway_merge
;
389 o
->unpack_opts
.src_index
= &the_index
;
390 o
->unpack_opts
.dst_index
= &tmp_index
;
391 o
->unpack_opts
.aggressive
= !merge_detect_rename(o
);
392 setup_unpack_trees_porcelain(&o
->unpack_opts
, "merge");
394 init_tree_desc_from_tree(t
+0, common
);
395 init_tree_desc_from_tree(t
+1, head
);
396 init_tree_desc_from_tree(t
+2, merge
);
398 rc
= unpack_trees(3, t
, &o
->unpack_opts
);
399 cache_tree_free(&active_cache_tree
);
402 * Update the_index to match the new results, AFTER saving a copy
403 * in o->orig_index. Update src_index to point to the saved copy.
404 * (verify_uptodate() checks src_index, and the original index is
405 * the one that had the necessary modification timestamps.)
407 o
->orig_index
= the_index
;
408 the_index
= tmp_index
;
409 o
->unpack_opts
.src_index
= &o
->orig_index
;
414 static void unpack_trees_finish(struct merge_options
*o
)
416 discard_index(&o
->orig_index
);
417 clear_unpack_trees_porcelain(&o
->unpack_opts
);
420 struct tree
*write_tree_from_memory(struct merge_options
*o
)
422 struct tree
*result
= NULL
;
424 if (unmerged_cache()) {
426 fprintf(stderr
, "BUG: There are unmerged index entries:\n");
427 for (i
= 0; i
< active_nr
; i
++) {
428 const struct cache_entry
*ce
= active_cache
[i
];
430 fprintf(stderr
, "BUG: %d %.*s\n", ce_stage(ce
),
431 (int)ce_namelen(ce
), ce
->name
);
433 BUG("unmerged index entries in merge-recursive.c");
436 if (!active_cache_tree
)
437 active_cache_tree
= cache_tree();
439 if (!cache_tree_fully_valid(active_cache_tree
) &&
440 cache_tree_update(&the_index
, 0) < 0) {
441 err(o
, _("error building trees"));
445 result
= lookup_tree(the_repository
, &active_cache_tree
->oid
);
450 static int save_files_dirs(const struct object_id
*oid
,
451 struct strbuf
*base
, const char *path
,
452 unsigned int mode
, int stage
, void *context
)
454 struct path_hashmap_entry
*entry
;
455 int baselen
= base
->len
;
456 struct merge_options
*o
= context
;
458 strbuf_addstr(base
, path
);
460 FLEX_ALLOC_MEM(entry
, path
, base
->buf
, base
->len
);
461 hashmap_entry_init(entry
, path_hash(entry
->path
));
462 hashmap_add(&o
->current_file_dir_set
, entry
);
464 strbuf_setlen(base
, baselen
);
465 return (S_ISDIR(mode
) ? READ_TREE_RECURSIVE
: 0);
468 static void get_files_dirs(struct merge_options
*o
, struct tree
*tree
)
470 struct pathspec match_all
;
471 memset(&match_all
, 0, sizeof(match_all
));
472 read_tree_recursive(tree
, "", 0, 0, &match_all
, save_files_dirs
, o
);
475 static int get_tree_entry_if_blob(const struct object_id
*tree
,
477 struct object_id
*hashy
,
478 unsigned int *mode_o
)
482 ret
= get_tree_entry(tree
, path
, hashy
, mode_o
);
483 if (S_ISDIR(*mode_o
)) {
484 oidcpy(hashy
, &null_oid
);
491 * Returns an index_entry instance which doesn't have to correspond to
492 * a real cache entry in Git's index.
494 static struct stage_data
*insert_stage_data(const char *path
,
495 struct tree
*o
, struct tree
*a
, struct tree
*b
,
496 struct string_list
*entries
)
498 struct string_list_item
*item
;
499 struct stage_data
*e
= xcalloc(1, sizeof(struct stage_data
));
500 get_tree_entry_if_blob(&o
->object
.oid
, path
,
501 &e
->stages
[1].oid
, &e
->stages
[1].mode
);
502 get_tree_entry_if_blob(&a
->object
.oid
, path
,
503 &e
->stages
[2].oid
, &e
->stages
[2].mode
);
504 get_tree_entry_if_blob(&b
->object
.oid
, path
,
505 &e
->stages
[3].oid
, &e
->stages
[3].mode
);
506 item
= string_list_insert(entries
, path
);
512 * Create a dictionary mapping file names to stage_data objects. The
513 * dictionary contains one entry for every path with a non-zero stage entry.
515 static struct string_list
*get_unmerged(void)
517 struct string_list
*unmerged
= xcalloc(1, sizeof(struct string_list
));
520 unmerged
->strdup_strings
= 1;
522 for (i
= 0; i
< active_nr
; i
++) {
523 struct string_list_item
*item
;
524 struct stage_data
*e
;
525 const struct cache_entry
*ce
= active_cache
[i
];
529 item
= string_list_lookup(unmerged
, ce
->name
);
531 item
= string_list_insert(unmerged
, ce
->name
);
532 item
->util
= xcalloc(1, sizeof(struct stage_data
));
535 e
->stages
[ce_stage(ce
)].mode
= ce
->ce_mode
;
536 oidcpy(&e
->stages
[ce_stage(ce
)].oid
, &ce
->oid
);
542 static int string_list_df_name_compare(const char *one
, const char *two
)
544 int onelen
= strlen(one
);
545 int twolen
= strlen(two
);
547 * Here we only care that entries for D/F conflicts are
548 * adjacent, in particular with the file of the D/F conflict
549 * appearing before files below the corresponding directory.
550 * The order of the rest of the list is irrelevant for us.
552 * To achieve this, we sort with df_name_compare and provide
553 * the mode S_IFDIR so that D/F conflicts will sort correctly.
554 * We use the mode S_IFDIR for everything else for simplicity,
555 * since in other cases any changes in their order due to
556 * sorting cause no problems for us.
558 int cmp
= df_name_compare(one
, onelen
, S_IFDIR
,
559 two
, twolen
, S_IFDIR
);
561 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
562 * that 'foo' comes before 'foo/bar'.
566 return onelen
- twolen
;
569 static void record_df_conflict_files(struct merge_options
*o
,
570 struct string_list
*entries
)
572 /* If there is a D/F conflict and the file for such a conflict
573 * currently exists in the working tree, we want to allow it to be
574 * removed to make room for the corresponding directory if needed.
575 * The files underneath the directories of such D/F conflicts will
576 * be processed before the corresponding file involved in the D/F
577 * conflict. If the D/F directory ends up being removed by the
578 * merge, then we won't have to touch the D/F file. If the D/F
579 * directory needs to be written to the working copy, then the D/F
580 * file will simply be removed (in make_room_for_path()) to make
581 * room for the necessary paths. Note that if both the directory
582 * and the file need to be present, then the D/F file will be
583 * reinstated with a new unique name at the time it is processed.
585 struct string_list df_sorted_entries
= STRING_LIST_INIT_NODUP
;
586 const char *last_file
= NULL
;
591 * If we're merging merge-bases, we don't want to bother with
592 * any working directory changes.
597 /* Ensure D/F conflicts are adjacent in the entries list. */
598 for (i
= 0; i
< entries
->nr
; i
++) {
599 struct string_list_item
*next
= &entries
->items
[i
];
600 string_list_append(&df_sorted_entries
, next
->string
)->util
=
603 df_sorted_entries
.cmp
= string_list_df_name_compare
;
604 string_list_sort(&df_sorted_entries
);
606 string_list_clear(&o
->df_conflict_file_set
, 1);
607 for (i
= 0; i
< df_sorted_entries
.nr
; i
++) {
608 const char *path
= df_sorted_entries
.items
[i
].string
;
609 int len
= strlen(path
);
610 struct stage_data
*e
= df_sorted_entries
.items
[i
].util
;
613 * Check if last_file & path correspond to a D/F conflict;
614 * i.e. whether path is last_file+'/'+<something>.
615 * If so, record that it's okay to remove last_file to make
616 * room for path and friends if needed.
620 memcmp(path
, last_file
, last_len
) == 0 &&
621 path
[last_len
] == '/') {
622 string_list_insert(&o
->df_conflict_file_set
, last_file
);
626 * Determine whether path could exist as a file in the
627 * working directory as a possible D/F conflict. This
628 * will only occur when it exists in stage 2 as a
631 if (S_ISREG(e
->stages
[2].mode
) || S_ISLNK(e
->stages
[2].mode
)) {
638 string_list_clear(&df_sorted_entries
, 0);
642 struct diff_filepair
*pair
;
644 * Purpose of src_entry and dst_entry:
646 * If 'before' is renamed to 'after' then src_entry will contain
647 * the versions of 'before' from the merge_base, HEAD, and MERGE in
648 * stages 1, 2, and 3; dst_entry will contain the respective
649 * versions of 'after' in corresponding locations. Thus, we have a
650 * total of six modes and oids, though some will be null. (Stage 0
651 * is ignored; we're interested in handling conflicts.)
653 * Since we don't turn on break-rewrites by default, neither
654 * src_entry nor dst_entry can have all three of their stages have
655 * non-null oids, meaning at most four of the six will be non-null.
656 * Also, since this is a rename, both src_entry and dst_entry will
657 * have at least one non-null oid, meaning at least two will be
658 * non-null. Of the six oids, a typical rename will have three be
659 * non-null. Only two implies a rename/delete, and four implies a
662 struct stage_data
*src_entry
;
663 struct stage_data
*dst_entry
;
664 unsigned add_turned_into_rename
:1;
665 unsigned processed
:1;
668 static int update_stages(struct merge_options
*opt
, const char *path
,
669 const struct diff_filespec
*o
,
670 const struct diff_filespec
*a
,
671 const struct diff_filespec
*b
)
675 * NOTE: It is usually a bad idea to call update_stages on a path
676 * before calling update_file on that same path, since it can
677 * sometimes lead to spurious "refusing to lose untracked file..."
678 * messages from update_file (via make_room_for path via
679 * would_lose_untracked). Instead, reverse the order of the calls
680 * (executing update_file first and then update_stages).
683 int options
= ADD_CACHE_OK_TO_ADD
| ADD_CACHE_SKIP_DFCHECK
;
685 if (remove_file_from_cache(path
))
688 if (add_cacheinfo(opt
, o
->mode
, &o
->oid
, path
, 1, 0, options
))
691 if (add_cacheinfo(opt
, a
->mode
, &a
->oid
, path
, 2, 0, options
))
694 if (add_cacheinfo(opt
, b
->mode
, &b
->oid
, path
, 3, 0, options
))
699 static void update_entry(struct stage_data
*entry
,
700 struct diff_filespec
*o
,
701 struct diff_filespec
*a
,
702 struct diff_filespec
*b
)
704 entry
->processed
= 0;
705 entry
->stages
[1].mode
= o
->mode
;
706 entry
->stages
[2].mode
= a
->mode
;
707 entry
->stages
[3].mode
= b
->mode
;
708 oidcpy(&entry
->stages
[1].oid
, &o
->oid
);
709 oidcpy(&entry
->stages
[2].oid
, &a
->oid
);
710 oidcpy(&entry
->stages
[3].oid
, &b
->oid
);
713 static int remove_file(struct merge_options
*o
, int clean
,
714 const char *path
, int no_wd
)
716 int update_cache
= o
->call_depth
|| clean
;
717 int update_working_directory
= !o
->call_depth
&& !no_wd
;
720 if (remove_file_from_cache(path
))
723 if (update_working_directory
) {
725 struct cache_entry
*ce
;
726 ce
= cache_file_exists(path
, strlen(path
), ignore_case
);
727 if (ce
&& ce_stage(ce
) == 0 && strcmp(path
, ce
->name
))
730 if (remove_path(path
))
736 /* add a string to a strbuf, but converting "/" to "_" */
737 static void add_flattened_path(struct strbuf
*out
, const char *s
)
740 strbuf_addstr(out
, s
);
741 for (; i
< out
->len
; i
++)
742 if (out
->buf
[i
] == '/')
746 static char *unique_path(struct merge_options
*o
, const char *path
, const char *branch
)
748 struct path_hashmap_entry
*entry
;
749 struct strbuf newpath
= STRBUF_INIT
;
753 strbuf_addf(&newpath
, "%s~", path
);
754 add_flattened_path(&newpath
, branch
);
756 base_len
= newpath
.len
;
757 while (hashmap_get_from_hash(&o
->current_file_dir_set
,
758 path_hash(newpath
.buf
), newpath
.buf
) ||
759 (!o
->call_depth
&& file_exists(newpath
.buf
))) {
760 strbuf_setlen(&newpath
, base_len
);
761 strbuf_addf(&newpath
, "_%d", suffix
++);
764 FLEX_ALLOC_MEM(entry
, path
, newpath
.buf
, newpath
.len
);
765 hashmap_entry_init(entry
, path_hash(entry
->path
));
766 hashmap_add(&o
->current_file_dir_set
, entry
);
767 return strbuf_detach(&newpath
, NULL
);
771 * Check whether a directory in the index is in the way of an incoming
772 * file. Return 1 if so. If check_working_copy is non-zero, also
773 * check the working directory. If empty_ok is non-zero, also return
774 * 0 in the case where the working-tree dir exists but is empty.
776 static int dir_in_way(const char *path
, int check_working_copy
, int empty_ok
)
779 struct strbuf dirpath
= STRBUF_INIT
;
782 strbuf_addstr(&dirpath
, path
);
783 strbuf_addch(&dirpath
, '/');
785 pos
= cache_name_pos(dirpath
.buf
, dirpath
.len
);
789 if (pos
< active_nr
&&
790 !strncmp(dirpath
.buf
, active_cache
[pos
]->name
, dirpath
.len
)) {
791 strbuf_release(&dirpath
);
795 strbuf_release(&dirpath
);
796 return check_working_copy
&& !lstat(path
, &st
) && S_ISDIR(st
.st_mode
) &&
797 !(empty_ok
&& is_empty_dir(path
));
801 * Returns whether path was tracked in the index before the merge started,
802 * and its oid and mode match the specified values
804 static int was_tracked_and_matches(struct merge_options
*o
, const char *path
,
805 const struct object_id
*oid
, unsigned mode
)
807 int pos
= index_name_pos(&o
->orig_index
, path
, strlen(path
));
808 struct cache_entry
*ce
;
811 /* we were not tracking this path before the merge */
814 /* See if the file we were tracking before matches */
815 ce
= o
->orig_index
.cache
[pos
];
816 return (oid_eq(&ce
->oid
, oid
) && ce
->ce_mode
== mode
);
820 * Returns whether path was tracked in the index before the merge started
822 static int was_tracked(struct merge_options
*o
, const char *path
)
824 int pos
= index_name_pos(&o
->orig_index
, path
, strlen(path
));
827 /* we were tracking this path before the merge */
833 static int would_lose_untracked(const char *path
)
836 * This may look like it can be simplified to:
837 * return !was_tracked(o, path) && file_exists(path)
838 * but it can't. This function needs to know whether path was in
839 * the working tree due to EITHER having been tracked in the index
840 * before the merge OR having been put into the working copy and
841 * index by unpack_trees(). Due to that either-or requirement, we
842 * check the current index instead of the original one.
844 * Note that we do not need to worry about merge-recursive itself
845 * updating the index after unpack_trees() and before calling this
846 * function, because we strictly require all code paths in
847 * merge-recursive to update the working tree first and the index
848 * second. Doing otherwise would break
849 * update_file()/would_lose_untracked(); see every comment in this
850 * file which mentions "update_stages".
852 int pos
= cache_name_pos(path
, strlen(path
));
856 while (pos
< active_nr
&&
857 !strcmp(path
, active_cache
[pos
]->name
)) {
859 * If stage #0, it is definitely tracked.
860 * If it has stage #2 then it was tracked
861 * before this merge started. All other
862 * cases the path was not tracked.
864 switch (ce_stage(active_cache
[pos
])) {
871 return file_exists(path
);
874 static int was_dirty(struct merge_options
*o
, const char *path
)
876 struct cache_entry
*ce
;
879 if (o
->call_depth
|| !was_tracked(o
, path
))
882 ce
= index_file_exists(o
->unpack_opts
.src_index
,
883 path
, strlen(path
), ignore_case
);
884 dirty
= verify_uptodate(ce
, &o
->unpack_opts
) != 0;
888 static int make_room_for_path(struct merge_options
*o
, const char *path
)
891 const char *msg
= _("failed to create path '%s'%s");
893 /* Unlink any D/F conflict files that are in the way */
894 for (i
= 0; i
< o
->df_conflict_file_set
.nr
; i
++) {
895 const char *df_path
= o
->df_conflict_file_set
.items
[i
].string
;
896 size_t pathlen
= strlen(path
);
897 size_t df_pathlen
= strlen(df_path
);
898 if (df_pathlen
< pathlen
&&
899 path
[df_pathlen
] == '/' &&
900 strncmp(path
, df_path
, df_pathlen
) == 0) {
902 _("Removing %s to make room for subdirectory\n"),
905 unsorted_string_list_delete_item(&o
->df_conflict_file_set
,
911 /* Make sure leading directories are created */
912 status
= safe_create_leading_directories_const(path
);
914 if (status
== SCLD_EXISTS
)
915 /* something else exists */
916 return err(o
, msg
, path
, _(": perhaps a D/F conflict?"));
917 return err(o
, msg
, path
, "");
921 * Do not unlink a file in the work tree if we are not
924 if (would_lose_untracked(path
))
925 return err(o
, _("refusing to lose untracked file at '%s'"),
928 /* Successful unlink is good.. */
931 /* .. and so is no existing file */
934 /* .. but not some other error (who really cares what?) */
935 return err(o
, msg
, path
, _(": perhaps a D/F conflict?"));
938 static int update_file_flags(struct merge_options
*o
,
939 const struct object_id
*oid
,
951 enum object_type type
;
955 if (S_ISGITLINK(mode
)) {
957 * We may later decide to recursively descend into
958 * the submodule directory and update its index
959 * and/or work tree, but we do not do that now.
965 buf
= read_object_file(oid
, &type
, &size
);
967 return err(o
, _("cannot read object %s '%s'"), oid_to_hex(oid
), path
);
968 if (type
!= OBJ_BLOB
) {
969 ret
= err(o
, _("blob expected for %s '%s'"), oid_to_hex(oid
), path
);
973 struct strbuf strbuf
= STRBUF_INIT
;
974 if (convert_to_working_tree(&the_index
, path
, buf
, size
, &strbuf
)) {
977 buf
= strbuf_detach(&strbuf
, NULL
);
981 if (make_room_for_path(o
, path
) < 0) {
985 if (S_ISREG(mode
) || (!has_symlinks
&& S_ISLNK(mode
))) {
991 fd
= open(path
, O_WRONLY
| O_TRUNC
| O_CREAT
, mode
);
993 ret
= err(o
, _("failed to open '%s': %s"),
994 path
, strerror(errno
));
997 write_in_full(fd
, buf
, size
);
999 } else if (S_ISLNK(mode
)) {
1000 char *lnk
= xmemdupz(buf
, size
);
1001 safe_create_leading_directories_const(path
);
1003 if (symlink(lnk
, path
))
1004 ret
= err(o
, _("failed to symlink '%s': %s"),
1005 path
, strerror(errno
));
1009 _("do not know what to do with %06o %s '%s'"),
1010 mode
, oid_to_hex(oid
), path
);
1015 if (!ret
&& update_cache
)
1016 if (add_cacheinfo(o
, mode
, oid
, path
, 0, update_wd
,
1017 ADD_CACHE_OK_TO_ADD
))
1022 static int update_file(struct merge_options
*o
,
1024 const struct object_id
*oid
,
1028 return update_file_flags(o
, oid
, mode
, path
, o
->call_depth
|| clean
, !o
->call_depth
);
1031 /* Low level file merging, update and removal */
1033 struct merge_file_info
{
1034 struct object_id oid
;
1040 static int merge_3way(struct merge_options
*o
,
1041 mmbuffer_t
*result_buf
,
1042 const struct diff_filespec
*one
,
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_name
, *name1
, *name2
;
1054 ll_opts
.renormalize
= o
->renormalize
;
1055 ll_opts
.extra_marker_size
= extra_marker_size
;
1056 ll_opts
.xdl_opts
= o
->xdl_opts
;
1058 if (o
->call_depth
) {
1059 ll_opts
.virtual_ancestor
= 1;
1060 ll_opts
.variant
= 0;
1062 switch (o
->recursive_variant
) {
1063 case MERGE_RECURSIVE_OURS
:
1064 ll_opts
.variant
= XDL_MERGE_FAVOR_OURS
;
1066 case MERGE_RECURSIVE_THEIRS
:
1067 ll_opts
.variant
= XDL_MERGE_FAVOR_THEIRS
;
1070 ll_opts
.variant
= 0;
1075 if (strcmp(a
->path
, b
->path
) ||
1076 (o
->ancestor
!= NULL
&& strcmp(a
->path
, one
->path
) != 0)) {
1077 base_name
= o
->ancestor
== NULL
? NULL
:
1078 mkpathdup("%s:%s", o
->ancestor
, one
->path
);
1079 name1
= mkpathdup("%s:%s", branch1
, a
->path
);
1080 name2
= mkpathdup("%s:%s", branch2
, b
->path
);
1082 base_name
= o
->ancestor
== NULL
? NULL
:
1083 mkpathdup("%s", o
->ancestor
);
1084 name1
= mkpathdup("%s", branch1
);
1085 name2
= mkpathdup("%s", branch2
);
1088 read_mmblob(&orig
, &one
->oid
);
1089 read_mmblob(&src1
, &a
->oid
);
1090 read_mmblob(&src2
, &b
->oid
);
1092 merge_status
= ll_merge(result_buf
, a
->path
, &orig
, base_name
,
1093 &src1
, name1
, &src2
, name2
, &ll_opts
);
1101 return merge_status
;
1104 static int find_first_merges(struct object_array
*result
, const char *path
,
1105 struct commit
*a
, struct commit
*b
)
1108 struct object_array merges
= OBJECT_ARRAY_INIT
;
1109 struct commit
*commit
;
1110 int contains_another
;
1112 char merged_revision
[42];
1113 const char *rev_args
[] = { "rev-list", "--merges", "--ancestry-path",
1114 "--all", merged_revision
, NULL
};
1115 struct rev_info revs
;
1116 struct setup_revision_opt rev_opts
;
1118 memset(result
, 0, sizeof(struct object_array
));
1119 memset(&rev_opts
, 0, sizeof(rev_opts
));
1121 /* get all revisions that merge commit a */
1122 xsnprintf(merged_revision
, sizeof(merged_revision
), "^%s",
1123 oid_to_hex(&a
->object
.oid
));
1124 init_revisions(&revs
, NULL
);
1125 rev_opts
.submodule
= path
;
1126 /* FIXME: can't handle linked worktrees in submodules yet */
1127 revs
.single_worktree
= path
!= NULL
;
1128 setup_revisions(ARRAY_SIZE(rev_args
)-1, rev_args
, &revs
, &rev_opts
);
1130 /* save all revisions from the above list that contain b */
1131 if (prepare_revision_walk(&revs
))
1132 die("revision walk setup failed");
1133 while ((commit
= get_revision(&revs
)) != NULL
) {
1134 struct object
*o
= &(commit
->object
);
1135 if (in_merge_bases(b
, commit
))
1136 add_object_array(o
, NULL
, &merges
);
1138 reset_revision_walk();
1140 /* Now we've got all merges that contain a and b. Prune all
1141 * merges that contain another found merge and save them in
1144 for (i
= 0; i
< merges
.nr
; i
++) {
1145 struct commit
*m1
= (struct commit
*) merges
.objects
[i
].item
;
1147 contains_another
= 0;
1148 for (j
= 0; j
< merges
.nr
; j
++) {
1149 struct commit
*m2
= (struct commit
*) merges
.objects
[j
].item
;
1150 if (i
!= j
&& in_merge_bases(m2
, m1
)) {
1151 contains_another
= 1;
1156 if (!contains_another
)
1157 add_object_array(merges
.objects
[i
].item
, NULL
, result
);
1160 object_array_clear(&merges
);
1164 static void print_commit(struct commit
*commit
)
1166 struct strbuf sb
= STRBUF_INIT
;
1167 struct pretty_print_context ctx
= {0};
1168 ctx
.date_mode
.type
= DATE_NORMAL
;
1169 format_commit_message(commit
, " %h: %m %s", &sb
, &ctx
);
1170 fprintf(stderr
, "%s\n", sb
.buf
);
1171 strbuf_release(&sb
);
1174 static int merge_submodule(struct merge_options
*o
,
1175 struct object_id
*result
, const char *path
,
1176 const struct object_id
*base
, const struct object_id
*a
,
1177 const struct object_id
*b
)
1179 struct commit
*commit_base
, *commit_a
, *commit_b
;
1181 struct object_array merges
;
1184 int search
= !o
->call_depth
;
1186 /* store a in result in case we fail */
1189 /* we can not handle deletion conflicts */
1190 if (is_null_oid(base
))
1197 if (add_submodule_odb(path
)) {
1198 output(o
, 1, _("Failed to merge submodule %s (not checked out)"), path
);
1202 if (!(commit_base
= lookup_commit_reference(the_repository
, base
)) ||
1203 !(commit_a
= lookup_commit_reference(the_repository
, a
)) ||
1204 !(commit_b
= lookup_commit_reference(the_repository
, b
))) {
1205 output(o
, 1, _("Failed to merge submodule %s (commits not present)"), path
);
1209 /* check whether both changes are forward */
1210 if (!in_merge_bases(commit_base
, commit_a
) ||
1211 !in_merge_bases(commit_base
, commit_b
)) {
1212 output(o
, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path
);
1216 /* Case #1: a is contained in b or vice versa */
1217 if (in_merge_bases(commit_a
, commit_b
)) {
1220 output(o
, 3, _("Fast-forwarding submodule %s to the following commit:"), path
);
1221 output_commit_title(o
, commit_b
);
1222 } else if (show(o
, 2))
1223 output(o
, 2, _("Fast-forwarding submodule %s"), path
);
1229 if (in_merge_bases(commit_b
, commit_a
)) {
1232 output(o
, 3, _("Fast-forwarding submodule %s to the following commit:"), path
);
1233 output_commit_title(o
, commit_a
);
1234 } else if (show(o
, 2))
1235 output(o
, 2, _("Fast-forwarding submodule %s"), path
);
1243 * Case #2: There are one or more merges that contain a and b in
1244 * the submodule. If there is only one, then present it as a
1245 * suggestion to the user, but leave it marked unmerged so the
1246 * user needs to confirm the resolution.
1249 /* Skip the search if makes no sense to the calling context. */
1253 /* find commit which merges them */
1254 parent_count
= find_first_merges(&merges
, path
, commit_a
, commit_b
);
1255 switch (parent_count
) {
1257 output(o
, 1, _("Failed to merge submodule %s (merge following commits not found)"), path
);
1261 output(o
, 1, _("Failed to merge submodule %s (not fast-forward)"), path
);
1262 output(o
, 2, _("Found a possible merge resolution for the submodule:\n"));
1263 print_commit((struct commit
*) merges
.objects
[0].item
);
1265 "If this is correct simply add it to the index "
1268 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1269 "which will accept this suggestion.\n"),
1270 oid_to_hex(&merges
.objects
[0].item
->oid
), path
);
1274 output(o
, 1, _("Failed to merge submodule %s (multiple merges found)"), path
);
1275 for (i
= 0; i
< merges
.nr
; i
++)
1276 print_commit((struct commit
*) merges
.objects
[i
].item
);
1279 object_array_clear(&merges
);
1283 static int merge_mode_and_contents(struct merge_options
*o
,
1284 const struct diff_filespec
*one
,
1285 const struct diff_filespec
*a
,
1286 const struct diff_filespec
*b
,
1287 const char *filename
,
1288 const char *branch1
,
1289 const char *branch2
,
1290 const int extra_marker_size
,
1291 struct merge_file_info
*result
)
1293 if (o
->branch1
!= branch1
) {
1295 * It's weird getting a reverse merge with HEAD on the bottom
1296 * side of the conflict markers and the other branch on the
1299 return merge_mode_and_contents(o
, one
, b
, a
,
1302 extra_marker_size
, result
);
1308 if ((S_IFMT
& a
->mode
) != (S_IFMT
& b
->mode
)) {
1310 if (S_ISREG(a
->mode
)) {
1311 result
->mode
= a
->mode
;
1312 oidcpy(&result
->oid
, &a
->oid
);
1314 result
->mode
= b
->mode
;
1315 oidcpy(&result
->oid
, &b
->oid
);
1318 if (!oid_eq(&a
->oid
, &one
->oid
) && !oid_eq(&b
->oid
, &one
->oid
))
1324 if (a
->mode
== b
->mode
|| a
->mode
== one
->mode
)
1325 result
->mode
= b
->mode
;
1327 result
->mode
= a
->mode
;
1328 if (b
->mode
!= one
->mode
) {
1334 if (oid_eq(&a
->oid
, &b
->oid
) || oid_eq(&a
->oid
, &one
->oid
))
1335 oidcpy(&result
->oid
, &b
->oid
);
1336 else if (oid_eq(&b
->oid
, &one
->oid
))
1337 oidcpy(&result
->oid
, &a
->oid
);
1338 else if (S_ISREG(a
->mode
)) {
1339 mmbuffer_t result_buf
;
1340 int ret
= 0, merge_status
;
1342 merge_status
= merge_3way(o
, &result_buf
, one
, a
, b
,
1346 if ((merge_status
< 0) || !result_buf
.ptr
)
1347 ret
= err(o
, _("Failed to execute internal merge"));
1350 write_object_file(result_buf
.ptr
, result_buf
.size
,
1351 blob_type
, &result
->oid
))
1352 ret
= err(o
, _("Unable to add %s to database"),
1355 free(result_buf
.ptr
);
1358 result
->clean
= (merge_status
== 0);
1359 } else if (S_ISGITLINK(a
->mode
)) {
1360 result
->clean
= merge_submodule(o
, &result
->oid
,
1365 } else if (S_ISLNK(a
->mode
)) {
1366 switch (o
->recursive_variant
) {
1367 case MERGE_RECURSIVE_NORMAL
:
1368 oidcpy(&result
->oid
, &a
->oid
);
1369 if (!oid_eq(&a
->oid
, &b
->oid
))
1372 case MERGE_RECURSIVE_OURS
:
1373 oidcpy(&result
->oid
, &a
->oid
);
1375 case MERGE_RECURSIVE_THEIRS
:
1376 oidcpy(&result
->oid
, &b
->oid
);
1380 BUG("unsupported object type in the tree");
1384 output(o
, 2, _("Auto-merging %s"), filename
);
1389 static int handle_rename_via_dir(struct merge_options
*o
,
1390 struct diff_filepair
*pair
,
1391 const char *rename_branch
,
1392 const char *other_branch
)
1395 * Handle file adds that need to be renamed due to directory rename
1396 * detection. This differs from handle_rename_normal, because
1397 * there is no content merge to do; just move the file into the
1398 * desired final location.
1400 const struct diff_filespec
*dest
= pair
->two
;
1402 if (!o
->call_depth
&& would_lose_untracked(dest
->path
)) {
1403 char *alt_path
= unique_path(o
, dest
->path
, rename_branch
);
1405 output(o
, 1, _("Error: Refusing to lose untracked file at %s; "
1406 "writing to %s instead."),
1407 dest
->path
, alt_path
);
1409 * Write the file in worktree at alt_path, but not in the
1410 * index. Instead, write to dest->path for the index but
1411 * only at the higher appropriate stage.
1413 if (update_file(o
, 0, &dest
->oid
, dest
->mode
, alt_path
))
1416 return update_stages(o
, dest
->path
, NULL
,
1417 rename_branch
== o
->branch1
? dest
: NULL
,
1418 rename_branch
== o
->branch1
? NULL
: dest
);
1421 /* Update dest->path both in index and in worktree */
1422 if (update_file(o
, 1, &dest
->oid
, dest
->mode
, dest
->path
))
1427 static int handle_change_delete(struct merge_options
*o
,
1428 const char *path
, const char *old_path
,
1429 const struct object_id
*o_oid
, int o_mode
,
1430 const struct object_id
*changed_oid
,
1432 const char *change_branch
,
1433 const char *delete_branch
,
1434 const char *change
, const char *change_past
)
1436 char *alt_path
= NULL
;
1437 const char *update_path
= path
;
1440 if (dir_in_way(path
, !o
->call_depth
, 0) ||
1441 (!o
->call_depth
&& would_lose_untracked(path
))) {
1442 update_path
= alt_path
= unique_path(o
, path
, change_branch
);
1445 if (o
->call_depth
) {
1447 * We cannot arbitrarily accept either a_sha or b_sha as
1448 * correct; since there is no true "middle point" between
1449 * them, simply reuse the base version for virtual merge base.
1451 ret
= remove_file_from_cache(path
);
1453 ret
= update_file(o
, 0, o_oid
, o_mode
, update_path
);
1456 * Despite the four nearly duplicate messages and argument
1457 * lists below and the ugliness of the nested if-statements,
1458 * having complete messages makes the job easier for
1461 * The slight variance among the cases is due to the fact
1463 * 1) directory/file conflicts (in effect if
1464 * !alt_path) could cause us to need to write the
1465 * file to a different path.
1466 * 2) renames (in effect if !old_path) could mean that
1467 * there are two names for the path that the user
1468 * may know the file by.
1472 output(o
, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1473 "and %s in %s. Version %s of %s left in tree."),
1474 change
, path
, delete_branch
, change_past
,
1475 change_branch
, change_branch
, path
);
1477 output(o
, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1478 "and %s to %s in %s. Version %s of %s left in tree."),
1479 change
, old_path
, delete_branch
, change_past
, path
,
1480 change_branch
, change_branch
, path
);
1484 output(o
, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1485 "and %s in %s. Version %s of %s left in tree at %s."),
1486 change
, path
, delete_branch
, change_past
,
1487 change_branch
, change_branch
, path
, alt_path
);
1489 output(o
, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1490 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1491 change
, old_path
, delete_branch
, change_past
, path
,
1492 change_branch
, change_branch
, path
, alt_path
);
1496 * No need to call update_file() on path when change_branch ==
1497 * o->branch1 && !alt_path, since that would needlessly touch
1498 * path. We could call update_file_flags() with update_cache=0
1499 * and update_wd=0, but that's a no-op.
1501 if (change_branch
!= o
->branch1
|| alt_path
)
1502 ret
= update_file(o
, 0, changed_oid
, changed_mode
, update_path
);
1509 static int handle_rename_delete(struct merge_options
*o
,
1510 struct diff_filepair
*pair
,
1511 const char *rename_branch
,
1512 const char *delete_branch
)
1514 const struct diff_filespec
*orig
= pair
->one
;
1515 const struct diff_filespec
*dest
= pair
->two
;
1517 if (handle_change_delete(o
,
1518 o
->call_depth
? orig
->path
: dest
->path
,
1519 o
->call_depth
? NULL
: orig
->path
,
1520 &orig
->oid
, orig
->mode
,
1521 &dest
->oid
, dest
->mode
,
1522 rename_branch
, delete_branch
,
1523 _("rename"), _("renamed")))
1527 return remove_file_from_cache(dest
->path
);
1529 return update_stages(o
, dest
->path
, NULL
,
1530 rename_branch
== o
->branch1
? dest
: NULL
,
1531 rename_branch
== o
->branch1
? NULL
: dest
);
1534 static struct diff_filespec
*filespec_from_entry(struct diff_filespec
*target
,
1535 struct stage_data
*entry
,
1538 struct object_id
*oid
= &entry
->stages
[stage
].oid
;
1539 unsigned mode
= entry
->stages
[stage
].mode
;
1540 if (mode
== 0 || is_null_oid(oid
))
1542 oidcpy(&target
->oid
, oid
);
1543 target
->mode
= mode
;
1547 static int handle_file_collision(struct merge_options
*o
,
1548 const char *collide_path
,
1549 const char *prev_path1
,
1550 const char *prev_path2
,
1551 const char *branch1
, const char *branch2
,
1552 const struct object_id
*a_oid
,
1553 unsigned int a_mode
,
1554 const struct object_id
*b_oid
,
1555 unsigned int b_mode
)
1557 struct merge_file_info mfi
;
1558 struct diff_filespec null
, a
, b
;
1559 char *alt_path
= NULL
;
1560 const char *update_path
= collide_path
;
1563 * It's easiest to get the correct things into stage 2 and 3, and
1564 * to make sure that the content merge puts HEAD before the other
1565 * branch if we just ensure that branch1 == o->branch1. So, simply
1566 * flip arguments around if we don't have that.
1568 if (branch1
!= o
->branch1
) {
1569 return handle_file_collision(o
, collide_path
,
1570 prev_path2
, prev_path1
,
1577 * In the recursive case, we just opt to undo renames
1579 if (o
->call_depth
&& (prev_path1
|| prev_path2
)) {
1580 /* Put first file (a_oid, a_mode) in its original spot */
1582 if (update_file(o
, 1, a_oid
, a_mode
, prev_path1
))
1585 if (update_file(o
, 1, a_oid
, a_mode
, collide_path
))
1589 /* Put second file (b_oid, b_mode) in its original spot */
1591 if (update_file(o
, 1, b_oid
, b_mode
, prev_path2
))
1594 if (update_file(o
, 1, b_oid
, b_mode
, collide_path
))
1598 /* Don't leave something at collision path if unrenaming both */
1599 if (prev_path1
&& prev_path2
)
1600 remove_file(o
, 1, collide_path
, 0);
1605 /* Remove rename sources if rename/add or rename/rename(2to1) */
1607 remove_file(o
, 1, prev_path1
,
1608 o
->call_depth
|| would_lose_untracked(prev_path1
));
1610 remove_file(o
, 1, prev_path2
,
1611 o
->call_depth
|| would_lose_untracked(prev_path2
));
1614 * Remove the collision path, if it wouldn't cause dirty contents
1615 * or an untracked file to get lost. We'll either overwrite with
1616 * merged contents, or just write out to differently named files.
1618 if (was_dirty(o
, collide_path
)) {
1619 output(o
, 1, _("Refusing to lose dirty file at %s"),
1621 update_path
= alt_path
= unique_path(o
, collide_path
, "merged");
1622 } else if (would_lose_untracked(collide_path
)) {
1624 * Only way we get here is if both renames were from
1625 * a directory rename AND user had an untracked file
1626 * at the location where both files end up after the
1627 * two directory renames. See testcase 10d of t6043.
1629 output(o
, 1, _("Refusing to lose untracked file at "
1630 "%s, even though it's in the way."),
1632 update_path
= alt_path
= unique_path(o
, collide_path
, "merged");
1635 * FIXME: It's possible that the two files are identical
1636 * and that the current working copy happens to match, in
1637 * which case we are unnecessarily touching the working
1638 * tree file. It's not a likely enough scenario that I
1639 * want to code up the checks for it and a better fix is
1640 * available if we restructure how unpack_trees() and
1641 * merge-recursive interoperate anyway, so punting for
1644 remove_file(o
, 0, collide_path
, 0);
1647 /* Store things in diff_filespecs for functions that need it */
1648 memset(&a
, 0, sizeof(struct diff_filespec
));
1649 memset(&b
, 0, sizeof(struct diff_filespec
));
1650 null
.path
= a
.path
= b
.path
= (char *)collide_path
;
1651 oidcpy(&null
.oid
, &null_oid
);
1653 oidcpy(&a
.oid
, a_oid
);
1656 oidcpy(&b
.oid
, b_oid
);
1660 if (merge_mode_and_contents(o
, &null
, &a
, &b
, collide_path
,
1661 branch1
, branch2
, o
->call_depth
* 2, &mfi
))
1663 mfi
.clean
&= !alt_path
;
1664 if (update_file(o
, mfi
.clean
, &mfi
.oid
, mfi
.mode
, update_path
))
1666 if (!mfi
.clean
&& !o
->call_depth
&&
1667 update_stages(o
, collide_path
, NULL
, &a
, &b
))
1671 * FIXME: If both a & b both started with conflicts (only possible
1672 * if they came from a rename/rename(2to1)), but had IDENTICAL
1673 * contents including those conflicts, then in the next line we claim
1674 * it was clean. If someone cares about this case, we should have the
1675 * caller notify us if we started with conflicts.
1680 static int handle_rename_add(struct merge_options
*o
,
1681 struct rename_conflict_info
*ci
)
1683 /* a was renamed to c, and a separate c was added. */
1684 struct diff_filespec
*a
= ci
->pair1
->one
;
1685 struct diff_filespec
*c
= ci
->pair1
->two
;
1686 char *path
= c
->path
;
1687 char *prev_path_desc
;
1688 struct merge_file_info mfi
;
1690 int other_stage
= (ci
->branch1
== o
->branch1
? 3 : 2);
1692 output(o
, 1, _("CONFLICT (rename/add): "
1693 "Rename %s->%s in %s. Added %s in %s"),
1694 a
->path
, c
->path
, ci
->branch1
,
1695 c
->path
, ci
->branch2
);
1697 prev_path_desc
= xstrfmt("version of %s from %s", path
, a
->path
);
1698 if (merge_mode_and_contents(o
, a
, c
, &ci
->ren1_other
, prev_path_desc
,
1699 o
->branch1
, o
->branch2
,
1700 1 + o
->call_depth
* 2, &mfi
))
1702 free(prev_path_desc
);
1704 return handle_file_collision(o
,
1705 c
->path
, a
->path
, NULL
,
1706 ci
->branch1
, ci
->branch2
,
1708 &ci
->dst_entry1
->stages
[other_stage
].oid
,
1709 ci
->dst_entry1
->stages
[other_stage
].mode
);
1712 static int handle_rename_rename_1to2(struct merge_options
*o
,
1713 struct rename_conflict_info
*ci
)
1715 /* One file was renamed in both branches, but to different names. */
1716 struct merge_file_info mfi
;
1717 struct diff_filespec other
;
1718 struct diff_filespec
*add
;
1719 struct diff_filespec
*one
= ci
->pair1
->one
;
1720 struct diff_filespec
*a
= ci
->pair1
->two
;
1721 struct diff_filespec
*b
= ci
->pair2
->two
;
1724 output(o
, 1, _("CONFLICT (rename/rename): "
1725 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1726 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1727 one
->path
, a
->path
, ci
->branch1
,
1728 one
->path
, b
->path
, ci
->branch2
,
1729 o
->call_depth
? _(" (left unresolved)") : "");
1731 path_desc
= xstrfmt("%s and %s, both renamed from %s",
1732 a
->path
, b
->path
, one
->path
);
1733 if (merge_mode_and_contents(o
, one
, a
, b
, path_desc
,
1734 ci
->branch1
, ci
->branch2
,
1735 o
->call_depth
* 2, &mfi
))
1739 if (o
->call_depth
) {
1741 * FIXME: For rename/add-source conflicts (if we could detect
1742 * such), this is wrong. We should instead find a unique
1743 * pathname and then either rename the add-source file to that
1744 * unique path, or use that unique path instead of src here.
1746 if (update_file(o
, 0, &mfi
.oid
, mfi
.mode
, one
->path
))
1750 * Above, we put the merged content at the merge-base's
1751 * path. Now we usually need to delete both a->path and
1752 * b->path. However, the rename on each side of the merge
1753 * could also be involved in a rename/add conflict. In
1754 * such cases, we should keep the added file around,
1755 * resolving the conflict at that path in its favor.
1757 add
= filespec_from_entry(&other
, ci
->dst_entry1
, 2 ^ 1);
1759 if (update_file(o
, 0, &add
->oid
, add
->mode
, a
->path
))
1763 remove_file_from_cache(a
->path
);
1764 add
= filespec_from_entry(&other
, ci
->dst_entry2
, 3 ^ 1);
1766 if (update_file(o
, 0, &add
->oid
, add
->mode
, b
->path
))
1770 remove_file_from_cache(b
->path
);
1773 * For each destination path, we need to see if there is a
1774 * rename/add collision. If not, we can write the file out
1775 * to the specified location.
1777 add
= filespec_from_entry(&other
, ci
->dst_entry1
, 2 ^ 1);
1779 if (handle_file_collision(o
, a
->path
,
1781 ci
->branch1
, ci
->branch2
,
1783 &add
->oid
, add
->mode
) < 0)
1786 char *new_path
= NULL
;
1787 if (dir_in_way(a
->path
, !o
->call_depth
, 0)) {
1788 new_path
= unique_path(o
, a
->path
, ci
->branch1
);
1789 output(o
, 1, _("%s is a directory in %s adding "
1791 a
->path
, ci
->branch2
, new_path
);
1792 } else if (would_lose_untracked(a
->path
)) {
1793 new_path
= unique_path(o
, a
->path
, ci
->branch1
);
1794 output(o
, 1, _("Refusing to lose untracked file"
1795 " at %s; adding as %s instead"),
1799 if (update_file(o
, 0, &mfi
.oid
, mfi
.mode
, new_path
? new_path
: a
->path
))
1802 if (update_stages(o
, a
->path
, NULL
, a
, NULL
))
1806 add
= filespec_from_entry(&other
, ci
->dst_entry2
, 3 ^ 1);
1808 if (handle_file_collision(o
, b
->path
,
1810 ci
->branch1
, ci
->branch2
,
1811 &add
->oid
, add
->mode
,
1812 &mfi
.oid
, mfi
.mode
) < 0)
1815 char *new_path
= NULL
;
1816 if (dir_in_way(b
->path
, !o
->call_depth
, 0)) {
1817 new_path
= unique_path(o
, b
->path
, ci
->branch2
);
1818 output(o
, 1, _("%s is a directory in %s adding "
1820 b
->path
, ci
->branch1
, new_path
);
1821 } else if (would_lose_untracked(b
->path
)) {
1822 new_path
= unique_path(o
, b
->path
, ci
->branch2
);
1823 output(o
, 1, _("Refusing to lose untracked file"
1824 " at %s; adding as %s instead"),
1828 if (update_file(o
, 0, &mfi
.oid
, mfi
.mode
, new_path
? new_path
: b
->path
))
1831 if (update_stages(o
, b
->path
, NULL
, NULL
, b
))
1839 static int handle_rename_rename_2to1(struct merge_options
*o
,
1840 struct rename_conflict_info
*ci
)
1842 /* Two files, a & b, were renamed to the same thing, c. */
1843 struct diff_filespec
*a
= ci
->pair1
->one
;
1844 struct diff_filespec
*b
= ci
->pair2
->one
;
1845 struct diff_filespec
*c1
= ci
->pair1
->two
;
1846 struct diff_filespec
*c2
= ci
->pair2
->two
;
1847 char *path
= c1
->path
; /* == c2->path */
1848 char *path_side_1_desc
;
1849 char *path_side_2_desc
;
1850 struct merge_file_info mfi_c1
;
1851 struct merge_file_info mfi_c2
;
1853 output(o
, 1, _("CONFLICT (rename/rename): "
1854 "Rename %s->%s in %s. "
1855 "Rename %s->%s in %s"),
1856 a
->path
, c1
->path
, ci
->branch1
,
1857 b
->path
, c2
->path
, ci
->branch2
);
1859 path_side_1_desc
= xstrfmt("version of %s from %s", path
, a
->path
);
1860 path_side_2_desc
= xstrfmt("version of %s from %s", path
, b
->path
);
1861 if (merge_mode_and_contents(o
, a
, c1
, &ci
->ren1_other
, path_side_1_desc
,
1862 o
->branch1
, o
->branch2
,
1863 1 + o
->call_depth
* 2, &mfi_c1
) ||
1864 merge_mode_and_contents(o
, b
, &ci
->ren2_other
, c2
, path_side_2_desc
,
1865 o
->branch1
, o
->branch2
,
1866 1 + o
->call_depth
* 2, &mfi_c2
))
1868 free(path_side_1_desc
);
1869 free(path_side_2_desc
);
1871 return handle_file_collision(o
, path
, a
->path
, b
->path
,
1872 ci
->branch1
, ci
->branch2
,
1873 &mfi_c1
.oid
, mfi_c1
.mode
,
1874 &mfi_c2
.oid
, mfi_c2
.mode
);
1878 * Get the diff_filepairs changed between o_tree and tree.
1880 static struct diff_queue_struct
*get_diffpairs(struct merge_options
*o
,
1881 struct tree
*o_tree
,
1884 struct diff_queue_struct
*ret
;
1885 struct diff_options opts
;
1888 opts
.flags
.recursive
= 1;
1889 opts
.flags
.rename_empty
= 0;
1890 opts
.detect_rename
= merge_detect_rename(o
);
1892 * We do not have logic to handle the detection of copies. In
1893 * fact, it may not even make sense to add such logic: would we
1894 * really want a change to a base file to be propagated through
1895 * multiple other files by a merge?
1897 if (opts
.detect_rename
> DIFF_DETECT_RENAME
)
1898 opts
.detect_rename
= DIFF_DETECT_RENAME
;
1899 opts
.rename_limit
= o
->merge_rename_limit
>= 0 ? o
->merge_rename_limit
:
1900 o
->diff_rename_limit
>= 0 ? o
->diff_rename_limit
:
1902 opts
.rename_score
= o
->rename_score
;
1903 opts
.show_rename_progress
= o
->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
> o
->needed_rename_limit
)
1909 o
->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
;
1921 static int tree_has_path(struct tree
*tree
, const char *path
)
1923 struct object_id hashy
;
1924 unsigned int mode_o
;
1926 return !get_tree_entry(&tree
->object
.oid
, path
,
1931 * Return a new string that replaces the beginning portion (which matches
1932 * entry->dir), with entry->new_dir. In perl-speak:
1933 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1935 * Caller must ensure that old_path starts with entry->dir + '/'.
1937 static char *apply_dir_rename(struct dir_rename_entry
*entry
,
1938 const char *old_path
)
1940 struct strbuf new_path
= STRBUF_INIT
;
1943 if (entry
->non_unique_new_dir
)
1946 oldlen
= strlen(entry
->dir
);
1947 newlen
= entry
->new_dir
.len
+ (strlen(old_path
) - oldlen
) + 1;
1948 strbuf_grow(&new_path
, newlen
);
1949 strbuf_addbuf(&new_path
, &entry
->new_dir
);
1950 strbuf_addstr(&new_path
, &old_path
[oldlen
]);
1952 return strbuf_detach(&new_path
, NULL
);
1955 static void get_renamed_dir_portion(const char *old_path
, const char *new_path
,
1956 char **old_dir
, char **new_dir
)
1958 char *end_of_old
, *end_of_new
;
1959 int old_len
, new_len
;
1966 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1967 * the "e/foo.c" part is the same, we just want to know that
1968 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1969 * so, for this example, this function returns "a/b/c/d" in
1970 * *old_dir and "a/b/some/thing/else" in *new_dir.
1972 * Also, if the basename of the file changed, we don't care. We
1973 * want to know which portion of the directory, if any, changed.
1975 end_of_old
= strrchr(old_path
, '/');
1976 end_of_new
= strrchr(new_path
, '/');
1978 if (end_of_old
== NULL
|| end_of_new
== NULL
)
1980 while (*--end_of_new
== *--end_of_old
&&
1981 end_of_old
!= old_path
&&
1982 end_of_new
!= new_path
)
1983 ; /* Do nothing; all in the while loop */
1985 * We've found the first non-matching character in the directory
1986 * paths. That means the current directory we were comparing
1987 * represents the rename. Move end_of_old and end_of_new back
1988 * to the full directory name.
1990 if (*end_of_old
== '/')
1992 if (*end_of_old
!= '/')
1994 end_of_old
= strchr(end_of_old
, '/');
1995 end_of_new
= strchr(end_of_new
, '/');
1998 * It may have been the case that old_path and new_path were the same
1999 * directory all along. Don't claim a rename if they're the same.
2001 old_len
= end_of_old
- old_path
;
2002 new_len
= end_of_new
- new_path
;
2004 if (old_len
!= new_len
|| strncmp(old_path
, new_path
, old_len
)) {
2005 *old_dir
= xstrndup(old_path
, old_len
);
2006 *new_dir
= xstrndup(new_path
, new_len
);
2010 static void remove_hashmap_entries(struct hashmap
*dir_renames
,
2011 struct string_list
*items_to_remove
)
2014 struct dir_rename_entry
*entry
;
2016 for (i
= 0; i
< items_to_remove
->nr
; i
++) {
2017 entry
= items_to_remove
->items
[i
].util
;
2018 hashmap_remove(dir_renames
, entry
, NULL
);
2020 string_list_clear(items_to_remove
, 0);
2024 * See if there is a directory rename for path, and if there are any file
2025 * level conflicts for the renamed location. If there is a rename and
2026 * there are no conflicts, return the new name. Otherwise, return NULL.
2028 static char *handle_path_level_conflicts(struct merge_options
*o
,
2030 struct dir_rename_entry
*entry
,
2031 struct hashmap
*collisions
,
2034 char *new_path
= NULL
;
2035 struct collision_entry
*collision_ent
;
2037 struct strbuf collision_paths
= STRBUF_INIT
;
2040 * entry has the mapping of old directory name to new directory name
2041 * that we want to apply to path.
2043 new_path
= apply_dir_rename(entry
, path
);
2046 /* This should only happen when entry->non_unique_new_dir set */
2047 if (!entry
->non_unique_new_dir
)
2048 BUG("entry->non_unqiue_dir not set and !new_path");
2049 output(o
, 1, _("CONFLICT (directory rename split): "
2050 "Unclear where to place %s because directory "
2051 "%s was renamed to multiple other directories, "
2052 "with no destination getting a majority of the "
2060 * The caller needs to have ensured that it has pre-populated
2061 * collisions with all paths that map to new_path. Do a quick check
2062 * to ensure that's the case.
2064 collision_ent
= collision_find_entry(collisions
, new_path
);
2065 if (collision_ent
== NULL
)
2066 BUG("collision_ent is NULL");
2069 * Check for one-sided add/add/.../add conflicts, i.e.
2070 * where implicit renames from the other side doing
2071 * directory rename(s) can affect this side of history
2072 * to put multiple paths into the same location. Warn
2073 * and bail on directory renames for such paths.
2075 if (collision_ent
->reported_already
) {
2077 } else if (tree_has_path(tree
, new_path
)) {
2078 collision_ent
->reported_already
= 1;
2079 strbuf_add_separated_string_list(&collision_paths
, ", ",
2080 &collision_ent
->source_files
);
2081 output(o
, 1, _("CONFLICT (implicit dir rename): Existing "
2082 "file/dir at %s in the way of implicit "
2083 "directory rename(s) putting the following "
2084 "path(s) there: %s."),
2085 new_path
, collision_paths
.buf
);
2087 } else if (collision_ent
->source_files
.nr
> 1) {
2088 collision_ent
->reported_already
= 1;
2089 strbuf_add_separated_string_list(&collision_paths
, ", ",
2090 &collision_ent
->source_files
);
2091 output(o
, 1, _("CONFLICT (implicit dir rename): Cannot map "
2092 "more than one path to %s; implicit directory "
2093 "renames tried to put these paths there: %s"),
2094 new_path
, collision_paths
.buf
);
2098 /* Free memory we no longer need */
2099 strbuf_release(&collision_paths
);
2100 if (!clean
&& new_path
) {
2109 * There are a couple things we want to do at the directory level:
2110 * 1. Check for both sides renaming to the same thing, in order to avoid
2111 * implicit renaming of files that should be left in place. (See
2112 * testcase 6b in t6043 for details.)
2113 * 2. Prune directory renames if there are still files left in the
2114 * the original directory. These represent a partial directory rename,
2115 * i.e. a rename where only some of the files within the directory
2116 * were renamed elsewhere. (Technically, this could be done earlier
2117 * in get_directory_renames(), except that would prevent us from
2118 * doing the previous check and thus failing testcase 6b.)
2119 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2120 * In the future, we could potentially record this info as well and
2121 * omit reporting rename/rename(1to2) conflicts for each path within
2122 * the affected directories, thus cleaning up the merge output.
2123 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2124 * directory level, because merging directories is fine. If it
2125 * causes conflicts for files within those merged directories, then
2126 * that should be detected at the individual path level.
2128 static void handle_directory_level_conflicts(struct merge_options
*o
,
2129 struct hashmap
*dir_re_head
,
2131 struct hashmap
*dir_re_merge
,
2134 struct hashmap_iter iter
;
2135 struct dir_rename_entry
*head_ent
;
2136 struct dir_rename_entry
*merge_ent
;
2138 struct string_list remove_from_head
= STRING_LIST_INIT_NODUP
;
2139 struct string_list remove_from_merge
= STRING_LIST_INIT_NODUP
;
2141 hashmap_iter_init(dir_re_head
, &iter
);
2142 while ((head_ent
= hashmap_iter_next(&iter
))) {
2143 merge_ent
= dir_rename_find_entry(dir_re_merge
, head_ent
->dir
);
2145 !head_ent
->non_unique_new_dir
&&
2146 !merge_ent
->non_unique_new_dir
&&
2147 !strbuf_cmp(&head_ent
->new_dir
, &merge_ent
->new_dir
)) {
2148 /* 1. Renamed identically; remove it from both sides */
2149 string_list_append(&remove_from_head
,
2150 head_ent
->dir
)->util
= head_ent
;
2151 strbuf_release(&head_ent
->new_dir
);
2152 string_list_append(&remove_from_merge
,
2153 merge_ent
->dir
)->util
= merge_ent
;
2154 strbuf_release(&merge_ent
->new_dir
);
2155 } else if (tree_has_path(head
, head_ent
->dir
)) {
2156 /* 2. This wasn't a directory rename after all */
2157 string_list_append(&remove_from_head
,
2158 head_ent
->dir
)->util
= head_ent
;
2159 strbuf_release(&head_ent
->new_dir
);
2163 remove_hashmap_entries(dir_re_head
, &remove_from_head
);
2164 remove_hashmap_entries(dir_re_merge
, &remove_from_merge
);
2166 hashmap_iter_init(dir_re_merge
, &iter
);
2167 while ((merge_ent
= hashmap_iter_next(&iter
))) {
2168 head_ent
= dir_rename_find_entry(dir_re_head
, merge_ent
->dir
);
2169 if (tree_has_path(merge
, merge_ent
->dir
)) {
2170 /* 2. This wasn't a directory rename after all */
2171 string_list_append(&remove_from_merge
,
2172 merge_ent
->dir
)->util
= merge_ent
;
2173 } else if (head_ent
&&
2174 !head_ent
->non_unique_new_dir
&&
2175 !merge_ent
->non_unique_new_dir
) {
2176 /* 3. rename/rename(1to2) */
2178 * We can assume it's not rename/rename(1to1) because
2179 * that was case (1), already checked above. So we
2180 * know that head_ent->new_dir and merge_ent->new_dir
2181 * are different strings.
2183 output(o
, 1, _("CONFLICT (rename/rename): "
2184 "Rename directory %s->%s in %s. "
2185 "Rename directory %s->%s in %s"),
2186 head_ent
->dir
, head_ent
->new_dir
.buf
, o
->branch1
,
2187 head_ent
->dir
, merge_ent
->new_dir
.buf
, o
->branch2
);
2188 string_list_append(&remove_from_head
,
2189 head_ent
->dir
)->util
= head_ent
;
2190 strbuf_release(&head_ent
->new_dir
);
2191 string_list_append(&remove_from_merge
,
2192 merge_ent
->dir
)->util
= merge_ent
;
2193 strbuf_release(&merge_ent
->new_dir
);
2197 remove_hashmap_entries(dir_re_head
, &remove_from_head
);
2198 remove_hashmap_entries(dir_re_merge
, &remove_from_merge
);
2201 static struct hashmap
*get_directory_renames(struct diff_queue_struct
*pairs
,
2204 struct hashmap
*dir_renames
;
2205 struct hashmap_iter iter
;
2206 struct dir_rename_entry
*entry
;
2210 * Typically, we think of a directory rename as all files from a
2211 * certain directory being moved to a target directory. However,
2212 * what if someone first moved two files from the original
2213 * directory in one commit, and then renamed the directory
2214 * somewhere else in a later commit? At merge time, we just know
2215 * that files from the original directory went to two different
2216 * places, and that the bulk of them ended up in the same place.
2217 * We want each directory rename to represent where the bulk of the
2218 * files from that directory end up; this function exists to find
2219 * where the bulk of the files went.
2221 * The first loop below simply iterates through the list of file
2222 * renames, finding out how often each directory rename pair
2223 * possibility occurs.
2225 dir_renames
= xmalloc(sizeof(*dir_renames
));
2226 dir_rename_init(dir_renames
);
2227 for (i
= 0; i
< pairs
->nr
; ++i
) {
2228 struct string_list_item
*item
;
2230 struct diff_filepair
*pair
= pairs
->queue
[i
];
2231 char *old_dir
, *new_dir
;
2233 /* File not part of directory rename if it wasn't renamed */
2234 if (pair
->status
!= 'R')
2237 get_renamed_dir_portion(pair
->one
->path
, pair
->two
->path
,
2238 &old_dir
, &new_dir
);
2240 /* Directory didn't change at all; ignore this one. */
2243 entry
= dir_rename_find_entry(dir_renames
, old_dir
);
2245 entry
= xmalloc(sizeof(*entry
));
2246 dir_rename_entry_init(entry
, old_dir
);
2247 hashmap_put(dir_renames
, entry
);
2251 item
= string_list_lookup(&entry
->possible_new_dirs
, new_dir
);
2253 item
= string_list_insert(&entry
->possible_new_dirs
,
2255 item
->util
= xcalloc(1, sizeof(int));
2264 * For each directory with files moved out of it, we find out which
2265 * target directory received the most files so we can declare it to
2266 * be the "winning" target location for the directory rename. This
2267 * winner gets recorded in new_dir. If there is no winner
2268 * (multiple target directories received the same number of files),
2269 * we set non_unique_new_dir. Once we've determined the winner (or
2270 * that there is no winner), we no longer need possible_new_dirs.
2272 hashmap_iter_init(dir_renames
, &iter
);
2273 while ((entry
= hashmap_iter_next(&iter
))) {
2278 for (i
= 0; i
< entry
->possible_new_dirs
.nr
; i
++) {
2279 int *count
= entry
->possible_new_dirs
.items
[i
].util
;
2283 else if (*count
> max
) {
2285 best
= entry
->possible_new_dirs
.items
[i
].string
;
2289 entry
->non_unique_new_dir
= 1;
2291 assert(entry
->new_dir
.len
== 0);
2292 strbuf_addstr(&entry
->new_dir
, best
);
2295 * The relevant directory sub-portion of the original full
2296 * filepaths were xstrndup'ed before inserting into
2297 * possible_new_dirs, and instead of manually iterating the
2298 * list and free'ing each, just lie and tell
2299 * possible_new_dirs that it did the strdup'ing so that it
2300 * will free them for us.
2302 entry
->possible_new_dirs
.strdup_strings
= 1;
2303 string_list_clear(&entry
->possible_new_dirs
, 1);
2309 static struct dir_rename_entry
*check_dir_renamed(const char *path
,
2310 struct hashmap
*dir_renames
)
2312 char *temp
= xstrdup(path
);
2314 struct dir_rename_entry
*entry
= NULL
;;
2316 while ((end
= strrchr(temp
, '/'))) {
2318 entry
= dir_rename_find_entry(dir_renames
, temp
);
2326 static void compute_collisions(struct hashmap
*collisions
,
2327 struct hashmap
*dir_renames
,
2328 struct diff_queue_struct
*pairs
)
2333 * Multiple files can be mapped to the same path due to directory
2334 * renames done by the other side of history. Since that other
2335 * side of history could have merged multiple directories into one,
2336 * if our side of history added the same file basename to each of
2337 * those directories, then all N of them would get implicitly
2338 * renamed by the directory rename detection into the same path,
2339 * and we'd get an add/add/.../add conflict, and all those adds
2340 * from *this* side of history. This is not representable in the
2341 * index, and users aren't going to easily be able to make sense of
2342 * it. So we need to provide a good warning about what's
2343 * happening, and fall back to no-directory-rename detection
2344 * behavior for those paths.
2346 * See testcases 9e and all of section 5 from t6043 for examples.
2348 collision_init(collisions
);
2350 for (i
= 0; i
< pairs
->nr
; ++i
) {
2351 struct dir_rename_entry
*dir_rename_ent
;
2352 struct collision_entry
*collision_ent
;
2354 struct diff_filepair
*pair
= pairs
->queue
[i
];
2356 if (pair
->status
!= 'A' && pair
->status
!= 'R')
2358 dir_rename_ent
= check_dir_renamed(pair
->two
->path
,
2360 if (!dir_rename_ent
)
2363 new_path
= apply_dir_rename(dir_rename_ent
, pair
->two
->path
);
2366 * dir_rename_ent->non_unique_new_path is true, which
2367 * means there is no directory rename for us to use,
2368 * which means it won't cause us any additional
2372 collision_ent
= collision_find_entry(collisions
, new_path
);
2373 if (!collision_ent
) {
2374 collision_ent
= xcalloc(1,
2375 sizeof(struct collision_entry
));
2376 hashmap_entry_init(collision_ent
, strhash(new_path
));
2377 hashmap_put(collisions
, collision_ent
);
2378 collision_ent
->target_file
= new_path
;
2382 string_list_insert(&collision_ent
->source_files
,
2387 static char *check_for_directory_rename(struct merge_options
*o
,
2390 struct hashmap
*dir_renames
,
2391 struct hashmap
*dir_rename_exclusions
,
2392 struct hashmap
*collisions
,
2395 char *new_path
= NULL
;
2396 struct dir_rename_entry
*entry
= check_dir_renamed(path
, dir_renames
);
2397 struct dir_rename_entry
*oentry
= NULL
;
2403 * This next part is a little weird. We do not want to do an
2404 * implicit rename into a directory we renamed on our side, because
2405 * that will result in a spurious rename/rename(1to2) conflict. An
2407 * Base commit: dumbdir/afile, otherdir/bfile
2408 * Side 1: smrtdir/afile, otherdir/bfile
2409 * Side 2: dumbdir/afile, dumbdir/bfile
2410 * Here, while working on Side 1, we could notice that otherdir was
2411 * renamed/merged to dumbdir, and change the diff_filepair for
2412 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2413 * 2 will notice the rename from dumbdir to smrtdir, and do the
2414 * transitive rename to move it from dumbdir/bfile to
2415 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2416 * smrtdir, a rename/rename(1to2) conflict. We really just want
2417 * the file to end up in smrtdir. And the way to achieve that is
2418 * to not let Side1 do the rename to dumbdir, since we know that is
2419 * the source of one of our directory renames.
2421 * That's why oentry and dir_rename_exclusions is here.
2423 * As it turns out, this also prevents N-way transient rename
2424 * confusion; See testcases 9c and 9d of t6043.
2426 oentry
= dir_rename_find_entry(dir_rename_exclusions
, entry
->new_dir
.buf
);
2428 output(o
, 1, _("WARNING: Avoiding applying %s -> %s rename "
2429 "to %s, because %s itself was renamed."),
2430 entry
->dir
, entry
->new_dir
.buf
, path
, entry
->new_dir
.buf
);
2432 new_path
= handle_path_level_conflicts(o
, path
, entry
,
2434 *clean_merge
&= (new_path
!= NULL
);
2440 static void apply_directory_rename_modifications(struct merge_options
*o
,
2441 struct diff_filepair
*pair
,
2445 struct tree
*o_tree
,
2446 struct tree
*a_tree
,
2447 struct tree
*b_tree
,
2448 struct string_list
*entries
,
2451 struct string_list_item
*item
;
2452 int stage
= (tree
== a_tree
? 2 : 3);
2456 * In all cases where we can do directory rename detection,
2457 * unpack_trees() will have read pair->two->path into the
2458 * index and the working copy. We need to remove it so that
2459 * we can instead place it at new_path. It is guaranteed to
2460 * not be untracked (unpack_trees() would have errored out
2461 * saying the file would have been overwritten), but it might
2464 update_wd
= !was_dirty(o
, pair
->two
->path
);
2466 output(o
, 1, _("Refusing to lose dirty file at %s"),
2468 remove_file(o
, 1, pair
->two
->path
, !update_wd
);
2470 /* Find or create a new re->dst_entry */
2471 item
= string_list_lookup(entries
, new_path
);
2474 * Since we're renaming on this side of history, and it's
2475 * due to a directory rename on the other side of history
2476 * (which we only allow when the directory in question no
2477 * longer exists on the other side of history), the
2478 * original entry for re->dst_entry is no longer
2481 re
->dst_entry
->processed
= 1;
2484 * ...because we'll be using this new one.
2486 re
->dst_entry
= item
->util
;
2489 * re->dst_entry is for the before-dir-rename path, and we
2490 * need it to hold information for the after-dir-rename
2491 * path. Before creating a new entry, we need to mark the
2492 * old one as unnecessary (...unless it is shared by
2493 * src_entry, i.e. this didn't use to be a rename, in which
2494 * case we can just allow the normal processing to happen
2497 if (pair
->status
== 'R')
2498 re
->dst_entry
->processed
= 1;
2500 re
->dst_entry
= insert_stage_data(new_path
,
2501 o_tree
, a_tree
, b_tree
,
2503 item
= string_list_insert(entries
, new_path
);
2504 item
->util
= re
->dst_entry
;
2508 * Update the stage_data with the information about the path we are
2509 * moving into place. That slot will be empty and available for us
2510 * to write to because of the collision checks in
2511 * handle_path_level_conflicts(). In other words,
2512 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2513 * open for us to write to.
2515 * It may be tempting to actually update the index at this point as
2516 * well, using update_stages_for_stage_data(), but as per the big
2517 * "NOTE" in update_stages(), doing so will modify the current
2518 * in-memory index which will break calls to would_lose_untracked()
2519 * that we need to make. Instead, we need to just make sure that
2520 * the various handle_rename_*() functions update the index
2521 * explicitly rather than relying on unpack_trees() to have done it.
2523 get_tree_entry(&tree
->object
.oid
,
2525 &re
->dst_entry
->stages
[stage
].oid
,
2526 &re
->dst_entry
->stages
[stage
].mode
);
2528 /* Update pair status */
2529 if (pair
->status
== 'A') {
2531 * Recording rename information for this add makes it look
2532 * like a rename/delete conflict. Make sure we can
2533 * correctly handle this as an add that was moved to a new
2534 * directory instead of reporting a rename/delete conflict.
2536 re
->add_turned_into_rename
= 1;
2539 * We don't actually look at pair->status again, but it seems
2540 * pedagogically correct to adjust it.
2545 * Finally, record the new location.
2547 pair
->two
->path
= new_path
;
2551 * Get information of all renames which occurred in 'pairs', making use of
2552 * any implicit directory renames inferred from the other side of history.
2553 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2554 * to be able to associate the correct cache entries with the rename
2555 * information; tree is always equal to either a_tree or b_tree.
2557 static struct string_list
*get_renames(struct merge_options
*o
,
2558 struct diff_queue_struct
*pairs
,
2559 struct hashmap
*dir_renames
,
2560 struct hashmap
*dir_rename_exclusions
,
2562 struct tree
*o_tree
,
2563 struct tree
*a_tree
,
2564 struct tree
*b_tree
,
2565 struct string_list
*entries
,
2569 struct hashmap collisions
;
2570 struct hashmap_iter iter
;
2571 struct collision_entry
*e
;
2572 struct string_list
*renames
;
2574 compute_collisions(&collisions
, dir_renames
, pairs
);
2575 renames
= xcalloc(1, sizeof(struct string_list
));
2577 for (i
= 0; i
< pairs
->nr
; ++i
) {
2578 struct string_list_item
*item
;
2580 struct diff_filepair
*pair
= pairs
->queue
[i
];
2581 char *new_path
; /* non-NULL only with directory renames */
2583 if (pair
->status
!= 'A' && pair
->status
!= 'R') {
2584 diff_free_filepair(pair
);
2587 new_path
= check_for_directory_rename(o
, pair
->two
->path
, tree
,
2589 dir_rename_exclusions
,
2592 if (pair
->status
!= 'R' && !new_path
) {
2593 diff_free_filepair(pair
);
2597 re
= xmalloc(sizeof(*re
));
2599 re
->add_turned_into_rename
= 0;
2601 item
= string_list_lookup(entries
, re
->pair
->one
->path
);
2603 re
->src_entry
= insert_stage_data(re
->pair
->one
->path
,
2604 o_tree
, a_tree
, b_tree
, entries
);
2606 re
->src_entry
= item
->util
;
2608 item
= string_list_lookup(entries
, re
->pair
->two
->path
);
2610 re
->dst_entry
= insert_stage_data(re
->pair
->two
->path
,
2611 o_tree
, a_tree
, b_tree
, entries
);
2613 re
->dst_entry
= item
->util
;
2614 item
= string_list_insert(renames
, pair
->one
->path
);
2617 apply_directory_rename_modifications(o
, pair
, new_path
,
2624 hashmap_iter_init(&collisions
, &iter
);
2625 while ((e
= hashmap_iter_next(&iter
))) {
2626 free(e
->target_file
);
2627 string_list_clear(&e
->source_files
, 0);
2629 hashmap_free(&collisions
, 1);
2633 static int process_renames(struct merge_options
*o
,
2634 struct string_list
*a_renames
,
2635 struct string_list
*b_renames
)
2637 int clean_merge
= 1, i
, j
;
2638 struct string_list a_by_dst
= STRING_LIST_INIT_NODUP
;
2639 struct string_list b_by_dst
= STRING_LIST_INIT_NODUP
;
2640 const struct rename
*sre
;
2642 for (i
= 0; i
< a_renames
->nr
; i
++) {
2643 sre
= a_renames
->items
[i
].util
;
2644 string_list_insert(&a_by_dst
, sre
->pair
->two
->path
)->util
2647 for (i
= 0; i
< b_renames
->nr
; i
++) {
2648 sre
= b_renames
->items
[i
].util
;
2649 string_list_insert(&b_by_dst
, sre
->pair
->two
->path
)->util
2653 for (i
= 0, j
= 0; i
< a_renames
->nr
|| j
< b_renames
->nr
;) {
2654 struct string_list
*renames1
, *renames2Dst
;
2655 struct rename
*ren1
= NULL
, *ren2
= NULL
;
2656 const char *branch1
, *branch2
;
2657 const char *ren1_src
, *ren1_dst
;
2658 struct string_list_item
*lookup
;
2660 if (i
>= a_renames
->nr
) {
2661 ren2
= b_renames
->items
[j
++].util
;
2662 } else if (j
>= b_renames
->nr
) {
2663 ren1
= a_renames
->items
[i
++].util
;
2665 int compare
= strcmp(a_renames
->items
[i
].string
,
2666 b_renames
->items
[j
].string
);
2668 ren1
= a_renames
->items
[i
++].util
;
2670 ren2
= b_renames
->items
[j
++].util
;
2673 /* TODO: refactor, so that 1/2 are not needed */
2675 renames1
= a_renames
;
2676 renames2Dst
= &b_by_dst
;
2677 branch1
= o
->branch1
;
2678 branch2
= o
->branch2
;
2680 renames1
= b_renames
;
2681 renames2Dst
= &a_by_dst
;
2682 branch1
= o
->branch2
;
2683 branch2
= o
->branch1
;
2687 if (ren1
->processed
)
2689 ren1
->processed
= 1;
2690 ren1
->dst_entry
->processed
= 1;
2691 /* BUG: We should only mark src_entry as processed if we
2692 * are not dealing with a rename + add-source case.
2694 ren1
->src_entry
->processed
= 1;
2696 ren1_src
= ren1
->pair
->one
->path
;
2697 ren1_dst
= ren1
->pair
->two
->path
;
2700 /* One file renamed on both sides */
2701 const char *ren2_src
= ren2
->pair
->one
->path
;
2702 const char *ren2_dst
= ren2
->pair
->two
->path
;
2703 enum rename_type rename_type
;
2704 if (strcmp(ren1_src
, ren2_src
) != 0)
2705 BUG("ren1_src != ren2_src");
2706 ren2
->dst_entry
->processed
= 1;
2707 ren2
->processed
= 1;
2708 if (strcmp(ren1_dst
, ren2_dst
) != 0) {
2709 rename_type
= RENAME_ONE_FILE_TO_TWO
;
2712 rename_type
= RENAME_ONE_FILE_TO_ONE
;
2713 /* BUG: We should only remove ren1_src in
2714 * the base stage (think of rename +
2715 * add-source cases).
2717 remove_file(o
, 1, ren1_src
, 1);
2718 update_entry(ren1
->dst_entry
,
2723 setup_rename_conflict_info(rename_type
,
2733 } else if ((lookup
= string_list_lookup(renames2Dst
, ren1_dst
))) {
2734 /* Two different files renamed to the same thing */
2736 ren2
= lookup
->util
;
2737 ren2_dst
= ren2
->pair
->two
->path
;
2738 if (strcmp(ren1_dst
, ren2_dst
) != 0)
2739 BUG("ren1_dst != ren2_dst");
2742 ren2
->processed
= 1;
2744 * BUG: We should only mark src_entry as processed
2745 * if we are not dealing with a rename + add-source
2748 ren2
->src_entry
->processed
= 1;
2750 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE
,
2762 /* Renamed in 1, maybe changed in 2 */
2763 /* we only use sha1 and mode of these */
2764 struct diff_filespec src_other
, dst_other
;
2768 * unpack_trees loads entries from common-commit
2769 * into stage 1, from head-commit into stage 2, and
2770 * from merge-commit into stage 3. We keep track
2771 * of which side corresponds to the rename.
2773 int renamed_stage
= a_renames
== renames1
? 2 : 3;
2774 int other_stage
= a_renames
== renames1
? 3 : 2;
2776 /* BUG: We should only remove ren1_src in the base
2777 * stage and in other_stage (think of rename +
2780 remove_file(o
, 1, ren1_src
,
2781 renamed_stage
== 2 || !was_tracked(o
, ren1_src
));
2783 oidcpy(&src_other
.oid
,
2784 &ren1
->src_entry
->stages
[other_stage
].oid
);
2785 src_other
.mode
= ren1
->src_entry
->stages
[other_stage
].mode
;
2786 oidcpy(&dst_other
.oid
,
2787 &ren1
->dst_entry
->stages
[other_stage
].oid
);
2788 dst_other
.mode
= ren1
->dst_entry
->stages
[other_stage
].mode
;
2791 if (oid_eq(&src_other
.oid
, &null_oid
) &&
2792 ren1
->add_turned_into_rename
) {
2793 setup_rename_conflict_info(RENAME_VIA_DIR
,
2803 } else if (oid_eq(&src_other
.oid
, &null_oid
)) {
2804 setup_rename_conflict_info(RENAME_DELETE
,
2814 } else if ((dst_other
.mode
== ren1
->pair
->two
->mode
) &&
2815 oid_eq(&dst_other
.oid
, &ren1
->pair
->two
->oid
)) {
2817 * Added file on the other side identical to
2818 * the file being renamed: clean merge.
2819 * Also, there is no need to overwrite the
2820 * file already in the working copy, so call
2821 * update_file_flags() instead of
2824 if (update_file_flags(o
,
2825 &ren1
->pair
->two
->oid
,
2826 ren1
->pair
->two
->mode
,
2828 1, /* update_cache */
2831 } else if (!oid_eq(&dst_other
.oid
, &null_oid
)) {
2833 * Probably not a clean merge, but it's
2834 * premature to set clean_merge to 0 here,
2835 * because if the rename merges cleanly and
2836 * the merge exactly matches the newly added
2837 * file, then the merge will be clean.
2839 setup_rename_conflict_info(RENAME_ADD
,
2852 if (clean_merge
< 0)
2853 goto cleanup_and_return
;
2855 struct diff_filespec
*one
, *a
, *b
;
2856 src_other
.path
= (char *)ren1_src
;
2858 one
= ren1
->pair
->one
;
2859 if (a_renames
== renames1
) {
2860 a
= ren1
->pair
->two
;
2863 b
= ren1
->pair
->two
;
2866 update_entry(ren1
->dst_entry
, one
, a
, b
);
2867 setup_rename_conflict_info(RENAME_NORMAL
,
2881 string_list_clear(&a_by_dst
, 0);
2882 string_list_clear(&b_by_dst
, 0);
2887 struct rename_info
{
2888 struct string_list
*head_renames
;
2889 struct string_list
*merge_renames
;
2892 static void initial_cleanup_rename(struct diff_queue_struct
*pairs
,
2893 struct hashmap
*dir_renames
)
2895 struct hashmap_iter iter
;
2896 struct dir_rename_entry
*e
;
2898 hashmap_iter_init(dir_renames
, &iter
);
2899 while ((e
= hashmap_iter_next(&iter
))) {
2901 strbuf_release(&e
->new_dir
);
2902 /* possible_new_dirs already cleared in get_directory_renames */
2904 hashmap_free(dir_renames
, 1);
2911 static int detect_and_process_renames(struct merge_options
*o
,
2912 struct tree
*common
,
2915 struct string_list
*entries
,
2916 struct rename_info
*ri
)
2918 struct diff_queue_struct
*head_pairs
, *merge_pairs
;
2919 struct hashmap
*dir_re_head
, *dir_re_merge
;
2922 ri
->head_renames
= NULL
;
2923 ri
->merge_renames
= NULL
;
2925 if (!merge_detect_rename(o
))
2928 head_pairs
= get_diffpairs(o
, common
, head
);
2929 merge_pairs
= get_diffpairs(o
, common
, merge
);
2931 if (o
->detect_directory_renames
) {
2932 dir_re_head
= get_directory_renames(head_pairs
, head
);
2933 dir_re_merge
= get_directory_renames(merge_pairs
, merge
);
2935 handle_directory_level_conflicts(o
,
2937 dir_re_merge
, merge
);
2939 dir_re_head
= xmalloc(sizeof(*dir_re_head
));
2940 dir_re_merge
= xmalloc(sizeof(*dir_re_merge
));
2941 dir_rename_init(dir_re_head
);
2942 dir_rename_init(dir_re_merge
);
2945 ri
->head_renames
= get_renames(o
, head_pairs
,
2946 dir_re_merge
, dir_re_head
, head
,
2947 common
, head
, merge
, entries
,
2951 ri
->merge_renames
= get_renames(o
, merge_pairs
,
2952 dir_re_head
, dir_re_merge
, merge
,
2953 common
, head
, merge
, entries
,
2957 clean
&= process_renames(o
, ri
->head_renames
, ri
->merge_renames
);
2961 * Some cleanup is deferred until cleanup_renames() because the
2962 * data structures are still needed and referenced in
2963 * process_entry(). But there are a few things we can free now.
2965 initial_cleanup_rename(head_pairs
, dir_re_head
);
2966 initial_cleanup_rename(merge_pairs
, dir_re_merge
);
2971 static void final_cleanup_rename(struct string_list
*rename
)
2973 const struct rename
*re
;
2979 for (i
= 0; i
< rename
->nr
; i
++) {
2980 re
= rename
->items
[i
].util
;
2981 diff_free_filepair(re
->pair
);
2983 string_list_clear(rename
, 1);
2987 static void final_cleanup_renames(struct rename_info
*re_info
)
2989 final_cleanup_rename(re_info
->head_renames
);
2990 final_cleanup_rename(re_info
->merge_renames
);
2993 static struct object_id
*stage_oid(const struct object_id
*oid
, unsigned mode
)
2995 return (is_null_oid(oid
) || mode
== 0) ? NULL
: (struct object_id
*)oid
;
2998 static int read_oid_strbuf(struct merge_options
*o
,
2999 const struct object_id
*oid
,
3003 enum object_type type
;
3005 buf
= read_object_file(oid
, &type
, &size
);
3007 return err(o
, _("cannot read object %s"), oid_to_hex(oid
));
3008 if (type
!= OBJ_BLOB
) {
3010 return err(o
, _("object %s is not a blob"), oid_to_hex(oid
));
3012 strbuf_attach(dst
, buf
, size
, size
+ 1);
3016 static int blob_unchanged(struct merge_options
*opt
,
3017 const struct object_id
*o_oid
,
3019 const struct object_id
*a_oid
,
3021 int renormalize
, const char *path
)
3023 struct strbuf o
= STRBUF_INIT
;
3024 struct strbuf a
= STRBUF_INIT
;
3025 int ret
= 0; /* assume changed for safety */
3027 if (a_mode
!= o_mode
)
3029 if (oid_eq(o_oid
, a_oid
))
3034 assert(o_oid
&& a_oid
);
3035 if (read_oid_strbuf(opt
, o_oid
, &o
) || read_oid_strbuf(opt
, a_oid
, &a
))
3038 * Note: binary | is used so that both renormalizations are
3039 * performed. Comparison can be skipped if both files are
3040 * unchanged since their sha1s have already been compared.
3042 if (renormalize_buffer(&the_index
, path
, o
.buf
, o
.len
, &o
) |
3043 renormalize_buffer(&the_index
, path
, a
.buf
, a
.len
, &a
))
3044 ret
= (o
.len
== a
.len
&& !memcmp(o
.buf
, a
.buf
, o
.len
));
3052 static int handle_modify_delete(struct merge_options
*o
,
3054 struct object_id
*o_oid
, int o_mode
,
3055 struct object_id
*a_oid
, int a_mode
,
3056 struct object_id
*b_oid
, int b_mode
)
3058 const char *modify_branch
, *delete_branch
;
3059 struct object_id
*changed_oid
;
3063 modify_branch
= o
->branch1
;
3064 delete_branch
= o
->branch2
;
3065 changed_oid
= a_oid
;
3066 changed_mode
= a_mode
;
3068 modify_branch
= o
->branch2
;
3069 delete_branch
= o
->branch1
;
3070 changed_oid
= b_oid
;
3071 changed_mode
= b_mode
;
3074 return handle_change_delete(o
,
3077 changed_oid
, changed_mode
,
3078 modify_branch
, delete_branch
,
3079 _("modify"), _("modified"));
3082 static int handle_content_merge(struct merge_options
*o
,
3085 struct object_id
*o_oid
, int o_mode
,
3086 struct object_id
*a_oid
, int a_mode
,
3087 struct object_id
*b_oid
, int b_mode
,
3088 struct rename_conflict_info
*rename_conflict_info
)
3090 const char *reason
= _("content");
3091 const char *path1
= NULL
, *path2
= NULL
;
3092 struct merge_file_info mfi
;
3093 struct diff_filespec one
, a
, b
;
3094 unsigned df_conflict_remains
= 0;
3097 reason
= _("add/add");
3098 o_oid
= (struct object_id
*)&null_oid
;
3100 one
.path
= a
.path
= b
.path
= (char *)path
;
3101 oidcpy(&one
.oid
, o_oid
);
3103 oidcpy(&a
.oid
, a_oid
);
3105 oidcpy(&b
.oid
, b_oid
);
3108 if (rename_conflict_info
) {
3109 struct diff_filepair
*pair1
= rename_conflict_info
->pair1
;
3111 path1
= (o
->branch1
== rename_conflict_info
->branch1
) ?
3112 pair1
->two
->path
: pair1
->one
->path
;
3113 /* If rename_conflict_info->pair2 != NULL, we are in
3114 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
3117 path2
= (rename_conflict_info
->pair2
||
3118 o
->branch2
== rename_conflict_info
->branch1
) ?
3119 pair1
->two
->path
: pair1
->one
->path
;
3120 one
.path
= pair1
->one
->path
;
3121 a
.path
= (char *)path1
;
3122 b
.path
= (char *)path2
;
3124 if (dir_in_way(path
, !o
->call_depth
,
3125 S_ISGITLINK(pair1
->two
->mode
)))
3126 df_conflict_remains
= 1;
3128 if (merge_mode_and_contents(o
, &one
, &a
, &b
, path
,
3129 o
->branch1
, o
->branch2
,
3130 o
->call_depth
* 2, &mfi
))
3134 * We can skip updating the working tree file iff:
3135 * a) The merge is clean
3136 * b) The merge matches what was in HEAD (content, mode, pathname)
3137 * c) The target path is usable (i.e. not involved in D/F conflict)
3140 was_tracked_and_matches(o
, path
, &mfi
.oid
, mfi
.mode
) &&
3141 !df_conflict_remains
) {
3143 struct cache_entry
*ce
;
3145 output(o
, 3, _("Skipped %s (merged same as existing)"), path
);
3146 if (add_cacheinfo(o
, mfi
.mode
, &mfi
.oid
, path
,
3147 0, (!o
->call_depth
&& !is_dirty
), 0))
3150 * However, add_cacheinfo() will delete the old cache entry
3151 * and add a new one. We need to copy over any skip_worktree
3152 * flag to avoid making the file appear as if it were
3153 * deleted by the user.
3155 pos
= index_name_pos(&o
->orig_index
, path
, strlen(path
));
3156 ce
= o
->orig_index
.cache
[pos
];
3157 if (ce_skip_worktree(ce
)) {
3158 pos
= index_name_pos(&the_index
, path
, strlen(path
));
3159 ce
= the_index
.cache
[pos
];
3160 ce
->ce_flags
|= CE_SKIP_WORKTREE
;
3166 if (S_ISGITLINK(mfi
.mode
))
3167 reason
= _("submodule");
3168 output(o
, 1, _("CONFLICT (%s): Merge conflict in %s"),
3170 if (rename_conflict_info
&& !df_conflict_remains
)
3171 if (update_stages(o
, path
, &one
, &a
, &b
))
3175 if (df_conflict_remains
|| is_dirty
) {
3177 if (o
->call_depth
) {
3178 remove_file_from_cache(path
);
3181 if (update_stages(o
, path
, &one
, &a
, &b
))
3184 int file_from_stage2
= was_tracked(o
, path
);
3185 struct diff_filespec merged
;
3186 oidcpy(&merged
.oid
, &mfi
.oid
);
3187 merged
.mode
= mfi
.mode
;
3189 if (update_stages(o
, path
, NULL
,
3190 file_from_stage2
? &merged
: NULL
,
3191 file_from_stage2
? NULL
: &merged
))
3196 new_path
= unique_path(o
, path
, rename_conflict_info
->branch1
);
3198 output(o
, 1, _("Refusing to lose dirty file at %s"),
3201 output(o
, 1, _("Adding as %s instead"), new_path
);
3202 if (update_file(o
, 0, &mfi
.oid
, mfi
.mode
, new_path
)) {
3208 } else if (update_file(o
, mfi
.clean
, &mfi
.oid
, mfi
.mode
, path
))
3210 return !is_dirty
&& mfi
.clean
;
3213 static int handle_rename_normal(struct merge_options
*o
,
3215 struct object_id
*o_oid
, unsigned int o_mode
,
3216 struct object_id
*a_oid
, unsigned int a_mode
,
3217 struct object_id
*b_oid
, unsigned int b_mode
,
3218 struct rename_conflict_info
*ci
)
3220 /* Merge the content and write it out */
3221 return handle_content_merge(o
, path
, was_dirty(o
, path
),
3222 o_oid
, o_mode
, a_oid
, a_mode
, b_oid
, b_mode
,
3226 /* Per entry merge function */
3227 static int process_entry(struct merge_options
*o
,
3228 const char *path
, struct stage_data
*entry
)
3230 int clean_merge
= 1;
3231 int normalize
= o
->renormalize
;
3232 unsigned o_mode
= entry
->stages
[1].mode
;
3233 unsigned a_mode
= entry
->stages
[2].mode
;
3234 unsigned b_mode
= entry
->stages
[3].mode
;
3235 struct object_id
*o_oid
= stage_oid(&entry
->stages
[1].oid
, o_mode
);
3236 struct object_id
*a_oid
= stage_oid(&entry
->stages
[2].oid
, a_mode
);
3237 struct object_id
*b_oid
= stage_oid(&entry
->stages
[3].oid
, b_mode
);
3239 entry
->processed
= 1;
3240 if (entry
->rename_conflict_info
) {
3241 struct rename_conflict_info
*conflict_info
= entry
->rename_conflict_info
;
3242 switch (conflict_info
->rename_type
) {
3244 case RENAME_ONE_FILE_TO_ONE
:
3245 clean_merge
= handle_rename_normal(o
,
3252 case RENAME_VIA_DIR
:
3254 if (handle_rename_via_dir(o
,
3255 conflict_info
->pair1
,
3256 conflict_info
->branch1
,
3257 conflict_info
->branch2
))
3262 * Probably unclean merge, but if the renamed file
3263 * merges cleanly and the result can then be
3264 * two-way merged cleanly with the added file, I
3265 * guess it's a clean merge?
3267 clean_merge
= handle_rename_add(o
, conflict_info
);
3271 if (handle_rename_delete(o
,
3272 conflict_info
->pair1
,
3273 conflict_info
->branch1
,
3274 conflict_info
->branch2
))
3277 case RENAME_ONE_FILE_TO_TWO
:
3279 if (handle_rename_rename_1to2(o
, conflict_info
))
3282 case RENAME_TWO_FILES_TO_ONE
:
3284 * Probably unclean merge, but if the two renamed
3285 * files merge cleanly and the two resulting files
3286 * can then be two-way merged cleanly, I guess it's
3289 clean_merge
= handle_rename_rename_2to1(o
,
3293 entry
->processed
= 0;
3296 } else if (o_oid
&& (!a_oid
|| !b_oid
)) {
3297 /* Case A: Deleted in one */
3298 if ((!a_oid
&& !b_oid
) ||
3299 (!b_oid
&& blob_unchanged(o
, o_oid
, o_mode
, a_oid
, a_mode
, normalize
, path
)) ||
3300 (!a_oid
&& blob_unchanged(o
, o_oid
, o_mode
, b_oid
, b_mode
, normalize
, path
))) {
3301 /* Deleted in both or deleted in one and
3302 * unchanged in the other */
3304 output(o
, 2, _("Removing %s"), path
);
3305 /* do not touch working file if it did not exist */
3306 remove_file(o
, 1, path
, !a_oid
);
3308 /* Modify/delete; deleted side may have put a directory in the way */
3310 if (handle_modify_delete(o
, path
, o_oid
, o_mode
,
3311 a_oid
, a_mode
, b_oid
, b_mode
))
3314 } else if ((!o_oid
&& a_oid
&& !b_oid
) ||
3315 (!o_oid
&& !a_oid
&& b_oid
)) {
3316 /* Case B: Added in one. */
3317 /* [nothing|directory] -> ([nothing|directory], file) */
3319 const char *add_branch
;
3320 const char *other_branch
;
3322 const struct object_id
*oid
;
3326 add_branch
= o
->branch1
;
3327 other_branch
= o
->branch2
;
3330 conf
= _("file/directory");
3332 add_branch
= o
->branch2
;
3333 other_branch
= o
->branch1
;
3336 conf
= _("directory/file");
3338 if (dir_in_way(path
,
3339 !o
->call_depth
&& !S_ISGITLINK(a_mode
),
3341 char *new_path
= unique_path(o
, path
, add_branch
);
3343 output(o
, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3345 conf
, path
, other_branch
, path
, new_path
);
3346 if (update_file(o
, 0, oid
, mode
, new_path
))
3348 else if (o
->call_depth
)
3349 remove_file_from_cache(path
);
3352 output(o
, 2, _("Adding %s"), path
);
3353 /* do not overwrite file if already present */
3354 if (update_file_flags(o
, oid
, mode
, path
, 1, !a_oid
))
3357 } else if (a_oid
&& b_oid
) {
3359 /* Case C: Added in both (check for same permissions) */
3361 _("CONFLICT (add/add): Merge conflict in %s"),
3363 clean_merge
= handle_file_collision(o
,
3370 /* case D: Modified in both, but differently. */
3371 int is_dirty
= 0; /* unpack_trees would have bailed if dirty */
3372 clean_merge
= handle_content_merge(o
, path
,
3379 } else if (!o_oid
&& !a_oid
&& !b_oid
) {
3381 * this entry was deleted altogether. a_mode == 0 means
3382 * we had that path and want to actively remove it.
3384 remove_file(o
, 1, path
, !a_mode
);
3386 BUG("fatal merge failure, shouldn't happen.");
3391 int merge_trees(struct merge_options
*o
,
3394 struct tree
*common
,
3395 struct tree
**result
)
3398 struct strbuf sb
= STRBUF_INIT
;
3400 if (!o
->call_depth
&& index_has_changes(&the_index
, head
, &sb
)) {
3401 err(o
, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3406 if (o
->subtree_shift
) {
3407 merge
= shift_tree_object(head
, merge
, o
->subtree_shift
);
3408 common
= shift_tree_object(head
, common
, o
->subtree_shift
);
3411 if (oid_eq(&common
->object
.oid
, &merge
->object
.oid
)) {
3412 output(o
, 0, _("Already up to date!"));
3417 code
= unpack_trees_start(o
, common
, head
, merge
);
3420 if (show(o
, 4) || o
->call_depth
)
3421 err(o
, _("merging of trees %s and %s failed"),
3422 oid_to_hex(&head
->object
.oid
),
3423 oid_to_hex(&merge
->object
.oid
));
3424 unpack_trees_finish(o
);
3428 if (unmerged_cache()) {
3429 struct string_list
*entries
;
3430 struct rename_info re_info
;
3433 * Only need the hashmap while processing entries, so
3434 * initialize it here and free it when we are done running
3435 * through the entries. Keeping it in the merge_options as
3436 * opposed to decaring a local hashmap is for convenience
3437 * so that we don't have to pass it to around.
3439 hashmap_init(&o
->current_file_dir_set
, path_hashmap_cmp
, NULL
, 512);
3440 get_files_dirs(o
, head
);
3441 get_files_dirs(o
, merge
);
3443 entries
= get_unmerged();
3444 clean
= detect_and_process_renames(o
, common
, head
, merge
,
3446 record_df_conflict_files(o
, entries
);
3449 for (i
= entries
->nr
-1; 0 <= i
; i
--) {
3450 const char *path
= entries
->items
[i
].string
;
3451 struct stage_data
*e
= entries
->items
[i
].util
;
3452 if (!e
->processed
) {
3453 int ret
= process_entry(o
, path
, e
);
3462 for (i
= 0; i
< entries
->nr
; i
++) {
3463 struct stage_data
*e
= entries
->items
[i
].util
;
3465 BUG("unprocessed path??? %s",
3466 entries
->items
[i
].string
);
3470 final_cleanup_renames(&re_info
);
3472 string_list_clear(entries
, 1);
3475 hashmap_free(&o
->current_file_dir_set
, 1);
3478 unpack_trees_finish(o
);
3485 unpack_trees_finish(o
);
3487 if (o
->call_depth
&& !(*result
= write_tree_from_memory(o
)))
3493 static struct commit_list
*reverse_commit_list(struct commit_list
*list
)
3495 struct commit_list
*next
= NULL
, *current
, *backup
;
3496 for (current
= list
; current
; current
= backup
) {
3497 backup
= current
->next
;
3498 current
->next
= next
;
3505 * Merge the commits h1 and h2, return the resulting virtual
3506 * commit object and a flag indicating the cleanness of the merge.
3508 int merge_recursive(struct merge_options
*o
,
3511 struct commit_list
*ca
,
3512 struct commit
**result
)
3514 struct commit_list
*iter
;
3515 struct commit
*merged_common_ancestors
;
3516 struct tree
*mrtree
;
3520 output(o
, 4, _("Merging:"));
3521 output_commit_title(o
, h1
);
3522 output_commit_title(o
, h2
);
3526 ca
= get_merge_bases(h1
, h2
);
3527 ca
= reverse_commit_list(ca
);
3531 unsigned cnt
= commit_list_count(ca
);
3533 output(o
, 5, Q_("found %u common ancestor:",
3534 "found %u common ancestors:", cnt
), cnt
);
3535 for (iter
= ca
; iter
; iter
= iter
->next
)
3536 output_commit_title(o
, iter
->item
);
3539 merged_common_ancestors
= pop_commit(&ca
);
3540 if (merged_common_ancestors
== NULL
) {
3541 /* if there is no common ancestor, use an empty tree */
3544 tree
= lookup_tree(the_repository
, the_repository
->hash_algo
->empty_tree
);
3545 merged_common_ancestors
= make_virtual_commit(tree
, "ancestor");
3548 for (iter
= ca
; iter
; iter
= iter
->next
) {
3549 const char *saved_b1
, *saved_b2
;
3552 * When the merge fails, the result contains files
3553 * with conflict markers. The cleanness flag is
3554 * ignored (unless indicating an error), it was never
3555 * actually used, as result of merge_trees has always
3556 * overwritten it: the committed "conflicts" were
3560 saved_b1
= o
->branch1
;
3561 saved_b2
= o
->branch2
;
3562 o
->branch1
= "Temporary merge branch 1";
3563 o
->branch2
= "Temporary merge branch 2";
3564 if (merge_recursive(o
, merged_common_ancestors
, iter
->item
,
3565 NULL
, &merged_common_ancestors
) < 0)
3567 o
->branch1
= saved_b1
;
3568 o
->branch2
= saved_b2
;
3571 if (!merged_common_ancestors
)
3572 return err(o
, _("merge returned no commit"));
3579 o
->ancestor
= "merged common ancestors";
3580 clean
= merge_trees(o
, get_commit_tree(h1
), get_commit_tree(h2
),
3581 get_commit_tree(merged_common_ancestors
),
3588 if (o
->call_depth
) {
3589 *result
= make_virtual_commit(mrtree
, "merged tree");
3590 commit_list_insert(h1
, &(*result
)->parents
);
3591 commit_list_insert(h2
, &(*result
)->parents
->next
);
3594 if (!o
->call_depth
&& o
->buffer_output
< 2)
3595 strbuf_release(&o
->obuf
);
3597 diff_warn_rename_limit("merge.renamelimit",
3598 o
->needed_rename_limit
, 0);
3602 static struct commit
*get_ref(const struct object_id
*oid
, const char *name
)
3604 struct object
*object
;
3606 object
= deref_tag(the_repository
, parse_object(the_repository
, oid
),
3611 if (object
->type
== OBJ_TREE
)
3612 return make_virtual_commit((struct tree
*)object
, name
);
3613 if (object
->type
!= OBJ_COMMIT
)
3615 if (parse_commit((struct commit
*)object
))
3617 return (struct commit
*)object
;
3620 int merge_recursive_generic(struct merge_options
*o
,
3621 const struct object_id
*head
,
3622 const struct object_id
*merge
,
3624 const struct object_id
**base_list
,
3625 struct commit
**result
)
3628 struct lock_file lock
= LOCK_INIT
;
3629 struct commit
*head_commit
= get_ref(head
, o
->branch1
);
3630 struct commit
*next_commit
= get_ref(merge
, o
->branch2
);
3631 struct commit_list
*ca
= NULL
;
3635 for (i
= 0; i
< num_base_list
; ++i
) {
3636 struct commit
*base
;
3637 if (!(base
= get_ref(base_list
[i
], oid_to_hex(base_list
[i
]))))
3638 return err(o
, _("Could not parse object '%s'"),
3639 oid_to_hex(base_list
[i
]));
3640 commit_list_insert(base
, &ca
);
3644 hold_locked_index(&lock
, LOCK_DIE_ON_ERROR
);
3645 clean
= merge_recursive(o
, head_commit
, next_commit
, ca
,
3648 rollback_lock_file(&lock
);
3652 if (write_locked_index(&the_index
, &lock
,
3653 COMMIT_LOCK
| SKIP_IF_UNCHANGED
))
3654 return err(o
, _("Unable to write index."));
3656 return clean
? 0 : 1;
3659 static void merge_recursive_config(struct merge_options
*o
)
3662 git_config_get_int("merge.verbosity", &o
->verbosity
);
3663 git_config_get_int("diff.renamelimit", &o
->diff_rename_limit
);
3664 git_config_get_int("merge.renamelimit", &o
->merge_rename_limit
);
3665 if (!git_config_get_string("diff.renames", &value
)) {
3666 o
->diff_detect_rename
= git_config_rename("diff.renames", value
);
3669 if (!git_config_get_string("merge.renames", &value
)) {
3670 o
->merge_detect_rename
= git_config_rename("merge.renames", value
);
3673 git_config(git_xmerge_config
, NULL
);
3676 void init_merge_options(struct merge_options
*o
)
3678 const char *merge_verbosity
;
3679 memset(o
, 0, sizeof(struct merge_options
));
3681 o
->buffer_output
= 1;
3682 o
->diff_rename_limit
= -1;
3683 o
->merge_rename_limit
= -1;
3685 o
->diff_detect_rename
= -1;
3686 o
->merge_detect_rename
= -1;
3687 o
->detect_directory_renames
= 1;
3688 merge_recursive_config(o
);
3689 merge_verbosity
= getenv("GIT_MERGE_VERBOSITY");
3690 if (merge_verbosity
)
3691 o
->verbosity
= strtol(merge_verbosity
, NULL
, 10);
3692 if (o
->verbosity
>= 5)
3693 o
->buffer_output
= 0;
3694 strbuf_init(&o
->obuf
, 0);
3695 string_list_init(&o
->df_conflict_file_set
, 1);
3698 int parse_merge_opt(struct merge_options
*o
, const char *s
)
3704 if (!strcmp(s
, "ours"))
3705 o
->recursive_variant
= MERGE_RECURSIVE_OURS
;
3706 else if (!strcmp(s
, "theirs"))
3707 o
->recursive_variant
= MERGE_RECURSIVE_THEIRS
;
3708 else if (!strcmp(s
, "subtree"))
3709 o
->subtree_shift
= "";
3710 else if (skip_prefix(s
, "subtree=", &arg
))
3711 o
->subtree_shift
= arg
;
3712 else if (!strcmp(s
, "patience"))
3713 o
->xdl_opts
= DIFF_WITH_ALG(o
, PATIENCE_DIFF
);
3714 else if (!strcmp(s
, "histogram"))
3715 o
->xdl_opts
= DIFF_WITH_ALG(o
, HISTOGRAM_DIFF
);
3716 else if (skip_prefix(s
, "diff-algorithm=", &arg
)) {
3717 long value
= parse_algorithm_value(arg
);
3720 /* clear out previous settings */
3721 DIFF_XDL_CLR(o
, NEED_MINIMAL
);
3722 o
->xdl_opts
&= ~XDF_DIFF_ALGORITHM_MASK
;
3723 o
->xdl_opts
|= value
;
3725 else if (!strcmp(s
, "ignore-space-change"))
3726 DIFF_XDL_SET(o
, IGNORE_WHITESPACE_CHANGE
);
3727 else if (!strcmp(s
, "ignore-all-space"))
3728 DIFF_XDL_SET(o
, IGNORE_WHITESPACE
);
3729 else if (!strcmp(s
, "ignore-space-at-eol"))
3730 DIFF_XDL_SET(o
, IGNORE_WHITESPACE_AT_EOL
);
3731 else if (!strcmp(s
, "ignore-cr-at-eol"))
3732 DIFF_XDL_SET(o
, IGNORE_CR_AT_EOL
);
3733 else if (!strcmp(s
, "renormalize"))
3735 else if (!strcmp(s
, "no-renormalize"))
3737 else if (!strcmp(s
, "no-renames"))
3738 o
->merge_detect_rename
= 0;
3739 else if (!strcmp(s
, "find-renames")) {
3740 o
->merge_detect_rename
= 1;
3741 o
->rename_score
= 0;
3743 else if (skip_prefix(s
, "find-renames=", &arg
) ||
3744 skip_prefix(s
, "rename-threshold=", &arg
)) {
3745 if ((o
->rename_score
= parse_rename_score(&arg
)) == -1 || *arg
!= 0)
3747 o
->merge_detect_rename
= 1;