rev-parse tests: don't ignore "git reflog" exit code
[alt-git.git] / merge-recursive.c
blob9ec1e6d043a2f85f61422138c97b379e1a5b70c9
1 /*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6 #include "cache.h"
7 #include "merge-recursive.h"
9 #include "advice.h"
10 #include "alloc.h"
11 #include "attr.h"
12 #include "blob.h"
13 #include "builtin.h"
14 #include "cache-tree.h"
15 #include "commit.h"
16 #include "commit-reach.h"
17 #include "config.h"
18 #include "diff.h"
19 #include "diffcore.h"
20 #include "dir.h"
21 #include "ll-merge.h"
22 #include "lockfile.h"
23 #include "object-store.h"
24 #include "repository.h"
25 #include "revision.h"
26 #include "string-list.h"
27 #include "submodule-config.h"
28 #include "submodule.h"
29 #include "tag.h"
30 #include "tree-walk.h"
31 #include "unpack-trees.h"
32 #include "xdiff-interface.h"
34 struct merge_options_internal {
35 int call_depth;
36 int needed_rename_limit;
37 struct hashmap current_file_dir_set;
38 struct string_list df_conflict_file_set;
39 struct unpack_trees_options unpack_opts;
40 struct index_state orig_index;
43 struct path_hashmap_entry {
44 struct hashmap_entry e;
45 char path[FLEX_ARRAY];
48 static int path_hashmap_cmp(const void *cmp_data,
49 const struct hashmap_entry *eptr,
50 const struct hashmap_entry *entry_or_key,
51 const void *keydata)
53 const struct path_hashmap_entry *a, *b;
54 const char *key = keydata;
56 a = container_of(eptr, const struct path_hashmap_entry, e);
57 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
59 return fspathcmp(a->path, key ? key : b->path);
63 * For dir_rename_entry, directory names are stored as a full path from the
64 * toplevel of the repository and do not include a trailing '/'. Also:
66 * dir: original name of directory being renamed
67 * non_unique_new_dir: if true, could not determine new_dir
68 * new_dir: final name of directory being renamed
69 * possible_new_dirs: temporary used to help determine new_dir; see comments
70 * in get_directory_renames() for details
72 struct dir_rename_entry {
73 struct hashmap_entry ent;
74 char *dir;
75 unsigned non_unique_new_dir:1;
76 struct strbuf new_dir;
77 struct string_list possible_new_dirs;
80 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
81 char *dir)
83 struct dir_rename_entry key;
85 if (dir == NULL)
86 return NULL;
87 hashmap_entry_init(&key.ent, strhash(dir));
88 key.dir = dir;
89 return hashmap_get_entry(hashmap, &key, ent, NULL);
92 static int dir_rename_cmp(const void *unused_cmp_data,
93 const struct hashmap_entry *eptr,
94 const struct hashmap_entry *entry_or_key,
95 const void *unused_keydata)
97 const struct dir_rename_entry *e1, *e2;
99 e1 = container_of(eptr, const struct dir_rename_entry, ent);
100 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
102 return strcmp(e1->dir, e2->dir);
105 static void dir_rename_init(struct hashmap *map)
107 hashmap_init(map, dir_rename_cmp, NULL, 0);
110 static void dir_rename_entry_init(struct dir_rename_entry *entry,
111 char *directory)
113 hashmap_entry_init(&entry->ent, strhash(directory));
114 entry->dir = directory;
115 entry->non_unique_new_dir = 0;
116 strbuf_init(&entry->new_dir, 0);
117 string_list_init_nodup(&entry->possible_new_dirs);
120 struct collision_entry {
121 struct hashmap_entry ent;
122 char *target_file;
123 struct string_list source_files;
124 unsigned reported_already:1;
127 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
128 char *target_file)
130 struct collision_entry key;
132 hashmap_entry_init(&key.ent, strhash(target_file));
133 key.target_file = target_file;
134 return hashmap_get_entry(hashmap, &key, ent, NULL);
137 static int collision_cmp(const void *unused_cmp_data,
138 const struct hashmap_entry *eptr,
139 const struct hashmap_entry *entry_or_key,
140 const void *unused_keydata)
142 const struct collision_entry *e1, *e2;
144 e1 = container_of(eptr, const struct collision_entry, ent);
145 e2 = container_of(entry_or_key, const struct collision_entry, ent);
147 return strcmp(e1->target_file, e2->target_file);
150 static void collision_init(struct hashmap *map)
152 hashmap_init(map, collision_cmp, NULL, 0);
155 static void flush_output(struct merge_options *opt)
157 if (opt->buffer_output < 2 && opt->obuf.len) {
158 fputs(opt->obuf.buf, stdout);
159 strbuf_reset(&opt->obuf);
163 __attribute__((format (printf, 2, 3)))
164 static int err(struct merge_options *opt, const char *err, ...)
166 va_list params;
168 if (opt->buffer_output < 2)
169 flush_output(opt);
170 else {
171 strbuf_complete(&opt->obuf, '\n');
172 strbuf_addstr(&opt->obuf, "error: ");
174 va_start(params, err);
175 strbuf_vaddf(&opt->obuf, err, params);
176 va_end(params);
177 if (opt->buffer_output > 1)
178 strbuf_addch(&opt->obuf, '\n');
179 else {
180 error("%s", opt->obuf.buf);
181 strbuf_reset(&opt->obuf);
184 return -1;
187 static struct tree *shift_tree_object(struct repository *repo,
188 struct tree *one, struct tree *two,
189 const char *subtree_shift)
191 struct object_id shifted;
193 if (!*subtree_shift) {
194 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
195 } else {
196 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
197 subtree_shift);
199 if (oideq(&two->object.oid, &shifted))
200 return two;
201 return lookup_tree(repo, &shifted);
204 static inline void set_commit_tree(struct commit *c, struct tree *t)
206 c->maybe_tree = t;
209 static struct commit *make_virtual_commit(struct repository *repo,
210 struct tree *tree,
211 const char *comment)
213 struct commit *commit = alloc_commit_node(repo);
215 set_merge_remote_desc(commit, comment, (struct object *)commit);
216 set_commit_tree(commit, tree);
217 commit->object.parsed = 1;
218 return commit;
221 enum rename_type {
222 RENAME_NORMAL = 0,
223 RENAME_VIA_DIR,
224 RENAME_ADD,
225 RENAME_DELETE,
226 RENAME_ONE_FILE_TO_ONE,
227 RENAME_ONE_FILE_TO_TWO,
228 RENAME_TWO_FILES_TO_ONE
232 * Since we want to write the index eventually, we cannot reuse the index
233 * for these (temporary) data.
235 struct stage_data {
236 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
237 struct rename_conflict_info *rename_conflict_info;
238 unsigned processed:1;
241 struct rename {
242 unsigned processed:1;
243 struct diff_filepair *pair;
244 const char *branch; /* branch that the rename occurred on */
246 * If directory rename detection affected this rename, what was its
247 * original type ('A' or 'R') and it's original destination before
248 * the directory rename (otherwise, '\0' and NULL for these two vars).
250 char dir_rename_original_type;
251 char *dir_rename_original_dest;
253 * Purpose of src_entry and dst_entry:
255 * If 'before' is renamed to 'after' then src_entry will contain
256 * the versions of 'before' from the merge_base, HEAD, and MERGE in
257 * stages 1, 2, and 3; dst_entry will contain the respective
258 * versions of 'after' in corresponding locations. Thus, we have a
259 * total of six modes and oids, though some will be null. (Stage 0
260 * is ignored; we're interested in handling conflicts.)
262 * Since we don't turn on break-rewrites by default, neither
263 * src_entry nor dst_entry can have all three of their stages have
264 * non-null oids, meaning at most four of the six will be non-null.
265 * Also, since this is a rename, both src_entry and dst_entry will
266 * have at least one non-null oid, meaning at least two will be
267 * non-null. Of the six oids, a typical rename will have three be
268 * non-null. Only two implies a rename/delete, and four implies a
269 * rename/add.
271 struct stage_data *src_entry;
272 struct stage_data *dst_entry;
275 struct rename_conflict_info {
276 enum rename_type rename_type;
277 struct rename *ren1;
278 struct rename *ren2;
281 static inline void setup_rename_conflict_info(enum rename_type rename_type,
282 struct merge_options *opt,
283 struct rename *ren1,
284 struct rename *ren2)
286 struct rename_conflict_info *ci;
289 * When we have two renames involved, it's easiest to get the
290 * correct things into stage 2 and 3, and to make sure that the
291 * content merge puts HEAD before the other branch if we just
292 * ensure that branch1 == opt->branch1. So, simply flip arguments
293 * around if we don't have that.
295 if (ren2 && ren1->branch != opt->branch1) {
296 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
297 return;
300 CALLOC_ARRAY(ci, 1);
301 ci->rename_type = rename_type;
302 ci->ren1 = ren1;
303 ci->ren2 = ren2;
305 ci->ren1->dst_entry->processed = 0;
306 ci->ren1->dst_entry->rename_conflict_info = ci;
307 if (ren2) {
308 ci->ren2->dst_entry->rename_conflict_info = ci;
312 static int show(struct merge_options *opt, int v)
314 return (!opt->priv->call_depth && opt->verbosity >= v) ||
315 opt->verbosity >= 5;
318 __attribute__((format (printf, 3, 4)))
319 static void output(struct merge_options *opt, int v, const char *fmt, ...)
321 va_list ap;
323 if (!show(opt, v))
324 return;
326 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
328 va_start(ap, fmt);
329 strbuf_vaddf(&opt->obuf, fmt, ap);
330 va_end(ap);
332 strbuf_addch(&opt->obuf, '\n');
333 if (!opt->buffer_output)
334 flush_output(opt);
337 static void repo_output_commit_title(struct merge_options *opt,
338 struct repository *repo,
339 struct commit *commit)
341 struct merge_remote_desc *desc;
343 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
344 desc = merge_remote_util(commit);
345 if (desc)
346 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
347 else {
348 strbuf_repo_add_unique_abbrev(&opt->obuf, repo,
349 &commit->object.oid,
350 DEFAULT_ABBREV);
351 strbuf_addch(&opt->obuf, ' ');
352 if (repo_parse_commit(repo, commit) != 0)
353 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
354 else {
355 const char *title;
356 const char *msg = repo_get_commit_buffer(repo, commit, NULL);
357 int len = find_commit_subject(msg, &title);
358 if (len)
359 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
360 repo_unuse_commit_buffer(repo, commit, msg);
363 flush_output(opt);
366 static void output_commit_title(struct merge_options *opt, struct commit *commit)
368 repo_output_commit_title(opt, the_repository, commit);
371 static int add_cacheinfo(struct merge_options *opt,
372 const struct diff_filespec *blob,
373 const char *path, int stage, int refresh, int options)
375 struct index_state *istate = opt->repo->index;
376 struct cache_entry *ce;
377 int ret;
379 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
380 if (!ce)
381 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
383 ret = add_index_entry(istate, ce, options);
384 if (refresh) {
385 struct cache_entry *nce;
387 nce = refresh_cache_entry(istate, ce,
388 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
389 if (!nce)
390 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
391 if (nce != ce)
392 ret = add_index_entry(istate, nce, options);
394 return ret;
397 static inline int merge_detect_rename(struct merge_options *opt)
399 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
402 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
404 parse_tree(tree);
405 init_tree_desc(desc, tree->buffer, tree->size);
408 static int unpack_trees_start(struct merge_options *opt,
409 struct tree *common,
410 struct tree *head,
411 struct tree *merge)
413 int rc;
414 struct tree_desc t[3];
415 struct index_state tmp_index = { NULL };
417 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
418 if (opt->priv->call_depth)
419 opt->priv->unpack_opts.index_only = 1;
420 else {
421 opt->priv->unpack_opts.update = 1;
422 /* FIXME: should only do this if !overwrite_ignore */
423 opt->priv->unpack_opts.preserve_ignored = 0;
425 opt->priv->unpack_opts.merge = 1;
426 opt->priv->unpack_opts.head_idx = 2;
427 opt->priv->unpack_opts.fn = threeway_merge;
428 opt->priv->unpack_opts.src_index = opt->repo->index;
429 opt->priv->unpack_opts.dst_index = &tmp_index;
430 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
431 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
433 init_tree_desc_from_tree(t+0, common);
434 init_tree_desc_from_tree(t+1, head);
435 init_tree_desc_from_tree(t+2, merge);
437 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
438 cache_tree_free(&opt->repo->index->cache_tree);
441 * Update opt->repo->index to match the new results, AFTER saving a
442 * copy in opt->priv->orig_index. Update src_index to point to the
443 * saved copy. (verify_uptodate() checks src_index, and the original
444 * index is the one that had the necessary modification timestamps.)
446 opt->priv->orig_index = *opt->repo->index;
447 *opt->repo->index = tmp_index;
448 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
450 return rc;
453 static void unpack_trees_finish(struct merge_options *opt)
455 discard_index(&opt->priv->orig_index);
456 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
459 static int save_files_dirs(const struct object_id *oid,
460 struct strbuf *base, const char *path,
461 unsigned int mode, void *context)
463 struct path_hashmap_entry *entry;
464 int baselen = base->len;
465 struct merge_options *opt = context;
467 strbuf_addstr(base, path);
469 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
470 hashmap_entry_init(&entry->e, fspathhash(entry->path));
471 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
473 strbuf_setlen(base, baselen);
474 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
477 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
479 struct pathspec match_all;
480 memset(&match_all, 0, sizeof(match_all));
481 read_tree(opt->repo, tree,
482 &match_all, save_files_dirs, opt);
485 static int get_tree_entry_if_blob(struct repository *r,
486 const struct object_id *tree,
487 const char *path,
488 struct diff_filespec *dfs)
490 int ret;
492 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
493 if (S_ISDIR(dfs->mode)) {
494 oidcpy(&dfs->oid, null_oid());
495 dfs->mode = 0;
497 return ret;
501 * Returns an index_entry instance which doesn't have to correspond to
502 * a real cache entry in Git's index.
504 static struct stage_data *insert_stage_data(struct repository *r,
505 const char *path,
506 struct tree *o, struct tree *a, struct tree *b,
507 struct string_list *entries)
509 struct string_list_item *item;
510 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
511 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
512 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
513 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
514 item = string_list_insert(entries, path);
515 item->util = e;
516 return e;
520 * Create a dictionary mapping file names to stage_data objects. The
521 * dictionary contains one entry for every path with a non-zero stage entry.
523 static struct string_list *get_unmerged(struct index_state *istate)
525 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
526 int i;
528 unmerged->strdup_strings = 1;
530 /* TODO: audit for interaction with sparse-index. */
531 ensure_full_index(istate);
532 for (i = 0; i < istate->cache_nr; i++) {
533 struct string_list_item *item;
534 struct stage_data *e;
535 const struct cache_entry *ce = istate->cache[i];
536 if (!ce_stage(ce))
537 continue;
539 item = string_list_lookup(unmerged, ce->name);
540 if (!item) {
541 item = string_list_insert(unmerged, ce->name);
542 item->util = xcalloc(1, sizeof(struct stage_data));
544 e = item->util;
545 e->stages[ce_stage(ce)].mode = ce->ce_mode;
546 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
549 return unmerged;
552 static int string_list_df_name_compare(const char *one, const char *two)
554 int onelen = strlen(one);
555 int twolen = strlen(two);
557 * Here we only care that entries for D/F conflicts are
558 * adjacent, in particular with the file of the D/F conflict
559 * appearing before files below the corresponding directory.
560 * The order of the rest of the list is irrelevant for us.
562 * To achieve this, we sort with df_name_compare and provide
563 * the mode S_IFDIR so that D/F conflicts will sort correctly.
564 * We use the mode S_IFDIR for everything else for simplicity,
565 * since in other cases any changes in their order due to
566 * sorting cause no problems for us.
568 int cmp = df_name_compare(one, onelen, S_IFDIR,
569 two, twolen, S_IFDIR);
571 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
572 * that 'foo' comes before 'foo/bar'.
574 if (cmp)
575 return cmp;
576 return onelen - twolen;
579 static void record_df_conflict_files(struct merge_options *opt,
580 struct string_list *entries)
582 /* If there is a D/F conflict and the file for such a conflict
583 * currently exists in the working tree, we want to allow it to be
584 * removed to make room for the corresponding directory if needed.
585 * The files underneath the directories of such D/F conflicts will
586 * be processed before the corresponding file involved in the D/F
587 * conflict. If the D/F directory ends up being removed by the
588 * merge, then we won't have to touch the D/F file. If the D/F
589 * directory needs to be written to the working copy, then the D/F
590 * file will simply be removed (in make_room_for_path()) to make
591 * room for the necessary paths. Note that if both the directory
592 * and the file need to be present, then the D/F file will be
593 * reinstated with a new unique name at the time it is processed.
595 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
596 const char *last_file = NULL;
597 int last_len = 0;
598 int i;
601 * If we're merging merge-bases, we don't want to bother with
602 * any working directory changes.
604 if (opt->priv->call_depth)
605 return;
607 /* Ensure D/F conflicts are adjacent in the entries list. */
608 for (i = 0; i < entries->nr; i++) {
609 struct string_list_item *next = &entries->items[i];
610 string_list_append(&df_sorted_entries, next->string)->util =
611 next->util;
613 df_sorted_entries.cmp = string_list_df_name_compare;
614 string_list_sort(&df_sorted_entries);
616 string_list_clear(&opt->priv->df_conflict_file_set, 1);
617 for (i = 0; i < df_sorted_entries.nr; i++) {
618 const char *path = df_sorted_entries.items[i].string;
619 int len = strlen(path);
620 struct stage_data *e = df_sorted_entries.items[i].util;
623 * Check if last_file & path correspond to a D/F conflict;
624 * i.e. whether path is last_file+'/'+<something>.
625 * If so, record that it's okay to remove last_file to make
626 * room for path and friends if needed.
628 if (last_file &&
629 len > last_len &&
630 memcmp(path, last_file, last_len) == 0 &&
631 path[last_len] == '/') {
632 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
636 * Determine whether path could exist as a file in the
637 * working directory as a possible D/F conflict. This
638 * will only occur when it exists in stage 2 as a
639 * file.
641 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
642 last_file = path;
643 last_len = len;
644 } else {
645 last_file = NULL;
648 string_list_clear(&df_sorted_entries, 0);
651 static int update_stages(struct merge_options *opt, const char *path,
652 const struct diff_filespec *o,
653 const struct diff_filespec *a,
654 const struct diff_filespec *b)
658 * NOTE: It is usually a bad idea to call update_stages on a path
659 * before calling update_file on that same path, since it can
660 * sometimes lead to spurious "refusing to lose untracked file..."
661 * messages from update_file (via make_room_for path via
662 * would_lose_untracked). Instead, reverse the order of the calls
663 * (executing update_file first and then update_stages).
665 int clear = 1;
666 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
667 if (clear)
668 if (remove_file_from_index(opt->repo->index, path))
669 return -1;
670 if (o)
671 if (add_cacheinfo(opt, o, path, 1, 0, options))
672 return -1;
673 if (a)
674 if (add_cacheinfo(opt, a, path, 2, 0, options))
675 return -1;
676 if (b)
677 if (add_cacheinfo(opt, b, path, 3, 0, options))
678 return -1;
679 return 0;
682 static void update_entry(struct stage_data *entry,
683 struct diff_filespec *o,
684 struct diff_filespec *a,
685 struct diff_filespec *b)
687 entry->processed = 0;
688 entry->stages[1].mode = o->mode;
689 entry->stages[2].mode = a->mode;
690 entry->stages[3].mode = b->mode;
691 oidcpy(&entry->stages[1].oid, &o->oid);
692 oidcpy(&entry->stages[2].oid, &a->oid);
693 oidcpy(&entry->stages[3].oid, &b->oid);
696 static int remove_file(struct merge_options *opt, int clean,
697 const char *path, int no_wd)
699 int update_cache = opt->priv->call_depth || clean;
700 int update_working_directory = !opt->priv->call_depth && !no_wd;
702 if (update_cache) {
703 if (remove_file_from_index(opt->repo->index, path))
704 return -1;
706 if (update_working_directory) {
707 if (ignore_case) {
708 struct cache_entry *ce;
709 ce = index_file_exists(opt->repo->index, path, strlen(path),
710 ignore_case);
711 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
712 return 0;
714 if (remove_path(path))
715 return -1;
717 return 0;
720 /* add a string to a strbuf, but converting "/" to "_" */
721 static void add_flattened_path(struct strbuf *out, const char *s)
723 size_t i = out->len;
724 strbuf_addstr(out, s);
725 for (; i < out->len; i++)
726 if (out->buf[i] == '/')
727 out->buf[i] = '_';
730 static char *unique_path(struct merge_options *opt,
731 const char *path,
732 const char *branch)
734 struct path_hashmap_entry *entry;
735 struct strbuf newpath = STRBUF_INIT;
736 int suffix = 0;
737 size_t base_len;
739 strbuf_addf(&newpath, "%s~", path);
740 add_flattened_path(&newpath, branch);
742 base_len = newpath.len;
743 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
744 fspathhash(newpath.buf), newpath.buf) ||
745 (!opt->priv->call_depth && file_exists(newpath.buf))) {
746 strbuf_setlen(&newpath, base_len);
747 strbuf_addf(&newpath, "_%d", suffix++);
750 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
751 hashmap_entry_init(&entry->e, fspathhash(entry->path));
752 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
753 return strbuf_detach(&newpath, NULL);
757 * Check whether a directory in the index is in the way of an incoming
758 * file. Return 1 if so. If check_working_copy is non-zero, also
759 * check the working directory. If empty_ok is non-zero, also return
760 * 0 in the case where the working-tree dir exists but is empty.
762 static int dir_in_way(struct index_state *istate, const char *path,
763 int check_working_copy, int empty_ok)
765 int pos;
766 struct strbuf dirpath = STRBUF_INIT;
767 struct stat st;
769 strbuf_addstr(&dirpath, path);
770 strbuf_addch(&dirpath, '/');
772 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
774 if (pos < 0)
775 pos = -1 - pos;
776 if (pos < istate->cache_nr &&
777 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
778 strbuf_release(&dirpath);
779 return 1;
782 strbuf_release(&dirpath);
783 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
784 !(empty_ok && is_empty_dir(path)) &&
785 !has_symlink_leading_path(path, strlen(path));
789 * Returns whether path was tracked in the index before the merge started,
790 * and its oid and mode match the specified values
792 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
793 const struct diff_filespec *blob)
795 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
796 struct cache_entry *ce;
798 if (0 > pos)
799 /* we were not tracking this path before the merge */
800 return 0;
802 /* See if the file we were tracking before matches */
803 ce = opt->priv->orig_index.cache[pos];
804 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
808 * Returns whether path was tracked in the index before the merge started
810 static int was_tracked(struct merge_options *opt, const char *path)
812 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
814 if (0 <= pos)
815 /* we were tracking this path before the merge */
816 return 1;
818 return 0;
821 static int would_lose_untracked(struct merge_options *opt, const char *path)
823 struct index_state *istate = opt->repo->index;
826 * This may look like it can be simplified to:
827 * return !was_tracked(opt, path) && file_exists(path)
828 * but it can't. This function needs to know whether path was in
829 * the working tree due to EITHER having been tracked in the index
830 * before the merge OR having been put into the working copy and
831 * index by unpack_trees(). Due to that either-or requirement, we
832 * check the current index instead of the original one.
834 * Note that we do not need to worry about merge-recursive itself
835 * updating the index after unpack_trees() and before calling this
836 * function, because we strictly require all code paths in
837 * merge-recursive to update the working tree first and the index
838 * second. Doing otherwise would break
839 * update_file()/would_lose_untracked(); see every comment in this
840 * file which mentions "update_stages".
842 int pos = index_name_pos(istate, path, strlen(path));
844 if (pos < 0)
845 pos = -1 - pos;
846 while (pos < istate->cache_nr &&
847 !strcmp(path, istate->cache[pos]->name)) {
849 * If stage #0, it is definitely tracked.
850 * If it has stage #2 then it was tracked
851 * before this merge started. All other
852 * cases the path was not tracked.
854 switch (ce_stage(istate->cache[pos])) {
855 case 0:
856 case 2:
857 return 0;
859 pos++;
861 return file_exists(path);
864 static int was_dirty(struct merge_options *opt, const char *path)
866 struct cache_entry *ce;
867 int dirty = 1;
869 if (opt->priv->call_depth || !was_tracked(opt, path))
870 return !dirty;
872 ce = index_file_exists(opt->priv->unpack_opts.src_index,
873 path, strlen(path), ignore_case);
874 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
875 return dirty;
878 static int make_room_for_path(struct merge_options *opt, const char *path)
880 int status, i;
881 const char *msg = _("failed to create path '%s'%s");
883 /* Unlink any D/F conflict files that are in the way */
884 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
885 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
886 size_t pathlen = strlen(path);
887 size_t df_pathlen = strlen(df_path);
888 if (df_pathlen < pathlen &&
889 path[df_pathlen] == '/' &&
890 strncmp(path, df_path, df_pathlen) == 0) {
891 output(opt, 3,
892 _("Removing %s to make room for subdirectory\n"),
893 df_path);
894 unlink(df_path);
895 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
896 i, 0);
897 break;
901 /* Make sure leading directories are created */
902 status = safe_create_leading_directories_const(path);
903 if (status) {
904 if (status == SCLD_EXISTS)
905 /* something else exists */
906 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
907 return err(opt, msg, path, "");
911 * Do not unlink a file in the work tree if we are not
912 * tracking it.
914 if (would_lose_untracked(opt, path))
915 return err(opt, _("refusing to lose untracked file at '%s'"),
916 path);
918 /* Successful unlink is good.. */
919 if (!unlink(path))
920 return 0;
921 /* .. and so is no existing file */
922 if (errno == ENOENT)
923 return 0;
924 /* .. but not some other error (who really cares what?) */
925 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
928 static int update_file_flags(struct merge_options *opt,
929 const struct diff_filespec *contents,
930 const char *path,
931 int update_cache,
932 int update_wd)
934 int ret = 0;
936 if (opt->priv->call_depth)
937 update_wd = 0;
939 if (update_wd) {
940 enum object_type type;
941 void *buf;
942 unsigned long size;
944 if (S_ISGITLINK(contents->mode)) {
946 * We may later decide to recursively descend into
947 * the submodule directory and update its index
948 * and/or work tree, but we do not do that now.
950 update_wd = 0;
951 goto update_index;
954 buf = read_object_file(&contents->oid, &type, &size);
955 if (!buf) {
956 ret = err(opt, _("cannot read object %s '%s'"),
957 oid_to_hex(&contents->oid), path);
958 goto free_buf;
960 if (type != OBJ_BLOB) {
961 ret = err(opt, _("blob expected for %s '%s'"),
962 oid_to_hex(&contents->oid), path);
963 goto free_buf;
965 if (S_ISREG(contents->mode)) {
966 struct strbuf strbuf = STRBUF_INIT;
967 if (convert_to_working_tree(opt->repo->index,
968 path, buf, size, &strbuf, NULL)) {
969 free(buf);
970 size = strbuf.len;
971 buf = strbuf_detach(&strbuf, NULL);
975 if (make_room_for_path(opt, path) < 0) {
976 update_wd = 0;
977 goto free_buf;
979 if (S_ISREG(contents->mode) ||
980 (!has_symlinks && S_ISLNK(contents->mode))) {
981 int fd;
982 int mode = (contents->mode & 0100 ? 0777 : 0666);
984 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
985 if (fd < 0) {
986 ret = err(opt, _("failed to open '%s': %s"),
987 path, strerror(errno));
988 goto free_buf;
990 write_in_full(fd, buf, size);
991 close(fd);
992 } else if (S_ISLNK(contents->mode)) {
993 char *lnk = xmemdupz(buf, size);
994 safe_create_leading_directories_const(path);
995 unlink(path);
996 if (symlink(lnk, path))
997 ret = err(opt, _("failed to symlink '%s': %s"),
998 path, strerror(errno));
999 free(lnk);
1000 } else
1001 ret = err(opt,
1002 _("do not know what to do with %06o %s '%s'"),
1003 contents->mode, oid_to_hex(&contents->oid), path);
1004 free_buf:
1005 free(buf);
1007 update_index:
1008 if (!ret && update_cache) {
1009 int refresh = (!opt->priv->call_depth &&
1010 contents->mode != S_IFGITLINK);
1011 if (add_cacheinfo(opt, contents, path, 0, refresh,
1012 ADD_CACHE_OK_TO_ADD))
1013 return -1;
1015 return ret;
1018 static int update_file(struct merge_options *opt,
1019 int clean,
1020 const struct diff_filespec *contents,
1021 const char *path)
1023 return update_file_flags(opt, contents, path,
1024 opt->priv->call_depth || clean, !opt->priv->call_depth);
1027 /* Low level file merging, update and removal */
1029 struct merge_file_info {
1030 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1031 unsigned clean:1,
1032 merge:1;
1035 static int merge_3way(struct merge_options *opt,
1036 mmbuffer_t *result_buf,
1037 const struct diff_filespec *o,
1038 const struct diff_filespec *a,
1039 const struct diff_filespec *b,
1040 const char *branch1,
1041 const char *branch2,
1042 const int extra_marker_size)
1044 mmfile_t orig, src1, src2;
1045 struct ll_merge_options ll_opts = {0};
1046 char *base, *name1, *name2;
1047 enum ll_merge_result merge_status;
1049 ll_opts.renormalize = opt->renormalize;
1050 ll_opts.extra_marker_size = extra_marker_size;
1051 ll_opts.xdl_opts = opt->xdl_opts;
1053 if (opt->priv->call_depth) {
1054 ll_opts.virtual_ancestor = 1;
1055 ll_opts.variant = 0;
1056 } else {
1057 switch (opt->recursive_variant) {
1058 case MERGE_VARIANT_OURS:
1059 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1060 break;
1061 case MERGE_VARIANT_THEIRS:
1062 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1063 break;
1064 default:
1065 ll_opts.variant = 0;
1066 break;
1070 assert(a->path && b->path && o->path && opt->ancestor);
1071 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1072 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1073 name1 = mkpathdup("%s:%s", branch1, a->path);
1074 name2 = mkpathdup("%s:%s", branch2, b->path);
1075 } else {
1076 base = mkpathdup("%s", opt->ancestor);
1077 name1 = mkpathdup("%s", branch1);
1078 name2 = mkpathdup("%s", branch2);
1081 read_mmblob(&orig, &o->oid);
1082 read_mmblob(&src1, &a->oid);
1083 read_mmblob(&src2, &b->oid);
1086 * FIXME: Using a->path for normalization rules in ll_merge could be
1087 * wrong if we renamed from a->path to b->path. We should use the
1088 * target path for where the file will be written.
1090 merge_status = ll_merge(result_buf, a->path, &orig, base,
1091 &src1, name1, &src2, name2,
1092 opt->repo->index, &ll_opts);
1093 if (merge_status == LL_MERGE_BINARY_CONFLICT)
1094 warning("Cannot merge binary files: %s (%s vs. %s)",
1095 a->path, name1, name2);
1097 free(base);
1098 free(name1);
1099 free(name2);
1100 free(orig.ptr);
1101 free(src1.ptr);
1102 free(src2.ptr);
1103 return merge_status;
1106 static int find_first_merges(struct repository *repo,
1107 struct object_array *result, const char *path,
1108 struct commit *a, struct commit *b)
1110 int i, j;
1111 struct object_array merges = OBJECT_ARRAY_INIT;
1112 struct commit *commit;
1113 int contains_another;
1115 char merged_revision[GIT_MAX_HEXSZ + 2];
1116 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1117 "--all", merged_revision, NULL };
1118 struct rev_info revs;
1119 struct setup_revision_opt rev_opts;
1121 memset(result, 0, sizeof(struct object_array));
1122 memset(&rev_opts, 0, sizeof(rev_opts));
1124 /* get all revisions that merge commit a */
1125 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1126 oid_to_hex(&a->object.oid));
1127 repo_init_revisions(repo, &revs, NULL);
1128 /* FIXME: can't handle linked worktrees in submodules yet */
1129 revs.single_worktree = path != NULL;
1130 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1132 /* save all revisions from the above list that contain b */
1133 if (prepare_revision_walk(&revs))
1134 die("revision walk setup failed");
1135 while ((commit = get_revision(&revs)) != NULL) {
1136 struct object *o = &(commit->object);
1137 if (repo_in_merge_bases(repo, b, commit))
1138 add_object_array(o, NULL, &merges);
1140 reset_revision_walk();
1142 /* Now we've got all merges that contain a and b. Prune all
1143 * merges that contain another found merge and save them in
1144 * result.
1146 for (i = 0; i < merges.nr; i++) {
1147 struct commit *m1 = (struct commit *) merges.objects[i].item;
1149 contains_another = 0;
1150 for (j = 0; j < merges.nr; j++) {
1151 struct commit *m2 = (struct commit *) merges.objects[j].item;
1152 if (i != j && repo_in_merge_bases(repo, m2, m1)) {
1153 contains_another = 1;
1154 break;
1158 if (!contains_another)
1159 add_object_array(merges.objects[i].item, NULL, result);
1162 object_array_clear(&merges);
1163 return result->nr;
1166 static void print_commit(struct repository *repo, struct commit *commit)
1168 struct strbuf sb = STRBUF_INIT;
1169 struct pretty_print_context ctx = {0};
1170 ctx.date_mode.type = DATE_NORMAL;
1171 /* FIXME: Merge this with output_commit_title() */
1172 assert(!merge_remote_util(commit));
1173 repo_format_commit_message(repo, commit, " %h: %m %s", &sb, &ctx);
1174 fprintf(stderr, "%s\n", sb.buf);
1175 strbuf_release(&sb);
1178 static int is_valid(const struct diff_filespec *dfs)
1180 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1183 static int merge_submodule(struct merge_options *opt,
1184 struct object_id *result, const char *path,
1185 const struct object_id *base, const struct object_id *a,
1186 const struct object_id *b)
1188 struct repository subrepo;
1189 int ret = 0;
1190 struct commit *commit_base, *commit_a, *commit_b;
1191 int parent_count;
1192 struct object_array merges;
1194 int i;
1195 int search = !opt->priv->call_depth;
1197 /* store a in result in case we fail */
1198 /* FIXME: This is the WRONG resolution for the recursive case when
1199 * we need to be careful to avoid accidentally matching either side.
1200 * Should probably use o instead there, much like we do for merging
1201 * binaries.
1203 oidcpy(result, a);
1205 /* we can not handle deletion conflicts */
1206 if (is_null_oid(base))
1207 return 0;
1208 if (is_null_oid(a))
1209 return 0;
1210 if (is_null_oid(b))
1211 return 0;
1213 if (repo_submodule_init(&subrepo, opt->repo, path, null_oid())) {
1214 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1215 return 0;
1218 if (!(commit_base = lookup_commit_reference(&subrepo, base)) ||
1219 !(commit_a = lookup_commit_reference(&subrepo, a)) ||
1220 !(commit_b = lookup_commit_reference(&subrepo, b))) {
1221 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1222 goto cleanup;
1225 /* check whether both changes are forward */
1226 if (!repo_in_merge_bases(&subrepo, commit_base, commit_a) ||
1227 !repo_in_merge_bases(&subrepo, commit_base, commit_b)) {
1228 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1229 goto cleanup;
1232 /* Case #1: a is contained in b or vice versa */
1233 if (repo_in_merge_bases(&subrepo, commit_a, commit_b)) {
1234 oidcpy(result, b);
1235 if (show(opt, 3)) {
1236 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1237 repo_output_commit_title(opt, &subrepo, commit_b);
1238 } else if (show(opt, 2))
1239 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1240 else
1241 ; /* no output */
1243 ret = 1;
1244 goto cleanup;
1246 if (repo_in_merge_bases(&subrepo, commit_b, commit_a)) {
1247 oidcpy(result, a);
1248 if (show(opt, 3)) {
1249 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1250 repo_output_commit_title(opt, &subrepo, commit_a);
1251 } else if (show(opt, 2))
1252 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1253 else
1254 ; /* no output */
1256 ret = 1;
1257 goto cleanup;
1261 * Case #2: There are one or more merges that contain a and b in
1262 * the submodule. If there is only one, then present it as a
1263 * suggestion to the user, but leave it marked unmerged so the
1264 * user needs to confirm the resolution.
1267 /* Skip the search if makes no sense to the calling context. */
1268 if (!search)
1269 goto cleanup;
1271 /* find commit which merges them */
1272 parent_count = find_first_merges(&subrepo, &merges, path,
1273 commit_a, commit_b);
1274 switch (parent_count) {
1275 case 0:
1276 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1277 break;
1279 case 1:
1280 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1281 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1282 print_commit(&subrepo, (struct commit *) merges.objects[0].item);
1283 output(opt, 2, _(
1284 "If this is correct simply add it to the index "
1285 "for example\n"
1286 "by using:\n\n"
1287 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1288 "which will accept this suggestion.\n"),
1289 oid_to_hex(&merges.objects[0].item->oid), path);
1290 break;
1292 default:
1293 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1294 for (i = 0; i < merges.nr; i++)
1295 print_commit(&subrepo, (struct commit *) merges.objects[i].item);
1298 object_array_clear(&merges);
1299 cleanup:
1300 repo_clear(&subrepo);
1301 return ret;
1304 static int merge_mode_and_contents(struct merge_options *opt,
1305 const struct diff_filespec *o,
1306 const struct diff_filespec *a,
1307 const struct diff_filespec *b,
1308 const char *filename,
1309 const char *branch1,
1310 const char *branch2,
1311 const int extra_marker_size,
1312 struct merge_file_info *result)
1314 if (opt->branch1 != branch1) {
1316 * It's weird getting a reverse merge with HEAD on the bottom
1317 * side of the conflict markers and the other branch on the
1318 * top. Fix that.
1320 return merge_mode_and_contents(opt, o, b, a,
1321 filename,
1322 branch2, branch1,
1323 extra_marker_size, result);
1326 result->merge = 0;
1327 result->clean = 1;
1329 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1330 result->clean = 0;
1332 * FIXME: This is a bad resolution for recursive case; for
1333 * the recursive case we want something that is unlikely to
1334 * accidentally match either side. Also, while it makes
1335 * sense to prefer regular files over symlinks, it doesn't
1336 * make sense to prefer regular files over submodules.
1338 if (S_ISREG(a->mode)) {
1339 result->blob.mode = a->mode;
1340 oidcpy(&result->blob.oid, &a->oid);
1341 } else {
1342 result->blob.mode = b->mode;
1343 oidcpy(&result->blob.oid, &b->oid);
1345 } else {
1346 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1347 result->merge = 1;
1350 * Merge modes
1352 if (a->mode == b->mode || a->mode == o->mode)
1353 result->blob.mode = b->mode;
1354 else {
1355 result->blob.mode = a->mode;
1356 if (b->mode != o->mode) {
1357 result->clean = 0;
1358 result->merge = 1;
1362 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1363 oidcpy(&result->blob.oid, &b->oid);
1364 else if (oideq(&b->oid, &o->oid))
1365 oidcpy(&result->blob.oid, &a->oid);
1366 else if (S_ISREG(a->mode)) {
1367 mmbuffer_t result_buf;
1368 int ret = 0, merge_status;
1370 merge_status = merge_3way(opt, &result_buf, o, a, b,
1371 branch1, branch2,
1372 extra_marker_size);
1374 if ((merge_status < 0) || !result_buf.ptr)
1375 ret = err(opt, _("Failed to execute internal merge"));
1377 if (!ret &&
1378 write_object_file(result_buf.ptr, result_buf.size,
1379 blob_type, &result->blob.oid))
1380 ret = err(opt, _("Unable to add %s to database"),
1381 a->path);
1383 free(result_buf.ptr);
1384 if (ret)
1385 return ret;
1386 /* FIXME: bug, what if modes didn't match? */
1387 result->clean = (merge_status == 0);
1388 } else if (S_ISGITLINK(a->mode)) {
1389 result->clean = merge_submodule(opt, &result->blob.oid,
1390 o->path,
1391 &o->oid,
1392 &a->oid,
1393 &b->oid);
1394 } else if (S_ISLNK(a->mode)) {
1395 switch (opt->recursive_variant) {
1396 case MERGE_VARIANT_NORMAL:
1397 oidcpy(&result->blob.oid, &a->oid);
1398 if (!oideq(&a->oid, &b->oid))
1399 result->clean = 0;
1400 break;
1401 case MERGE_VARIANT_OURS:
1402 oidcpy(&result->blob.oid, &a->oid);
1403 break;
1404 case MERGE_VARIANT_THEIRS:
1405 oidcpy(&result->blob.oid, &b->oid);
1406 break;
1408 } else
1409 BUG("unsupported object type in the tree");
1412 if (result->merge)
1413 output(opt, 2, _("Auto-merging %s"), filename);
1415 return 0;
1418 static int handle_rename_via_dir(struct merge_options *opt,
1419 struct rename_conflict_info *ci)
1422 * Handle file adds that need to be renamed due to directory rename
1423 * detection. This differs from handle_rename_normal, because
1424 * there is no content merge to do; just move the file into the
1425 * desired final location.
1427 const struct rename *ren = ci->ren1;
1428 const struct diff_filespec *dest = ren->pair->two;
1429 char *file_path = dest->path;
1430 int mark_conflicted = (opt->detect_directory_renames ==
1431 MERGE_DIRECTORY_RENAMES_CONFLICT);
1432 assert(ren->dir_rename_original_dest);
1434 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1435 mark_conflicted = 1;
1436 file_path = unique_path(opt, dest->path, ren->branch);
1437 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1438 "writing to %s instead."),
1439 dest->path, file_path);
1442 if (mark_conflicted) {
1444 * Write the file in worktree at file_path. In the index,
1445 * only record the file at dest->path in the appropriate
1446 * higher stage.
1448 if (update_file(opt, 0, dest, file_path))
1449 return -1;
1450 if (file_path != dest->path)
1451 free(file_path);
1452 if (update_stages(opt, dest->path, NULL,
1453 ren->branch == opt->branch1 ? dest : NULL,
1454 ren->branch == opt->branch1 ? NULL : dest))
1455 return -1;
1456 return 0; /* not clean, but conflicted */
1457 } else {
1458 /* Update dest->path both in index and in worktree */
1459 if (update_file(opt, 1, dest, dest->path))
1460 return -1;
1461 return 1; /* clean */
1465 static int handle_change_delete(struct merge_options *opt,
1466 const char *path, const char *old_path,
1467 const struct diff_filespec *o,
1468 const struct diff_filespec *changed,
1469 const char *change_branch,
1470 const char *delete_branch,
1471 const char *change, const char *change_past)
1473 char *alt_path = NULL;
1474 const char *update_path = path;
1475 int ret = 0;
1477 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1478 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1479 update_path = alt_path = unique_path(opt, path, change_branch);
1482 if (opt->priv->call_depth) {
1484 * We cannot arbitrarily accept either a_sha or b_sha as
1485 * correct; since there is no true "middle point" between
1486 * them, simply reuse the base version for virtual merge base.
1488 ret = remove_file_from_index(opt->repo->index, path);
1489 if (!ret)
1490 ret = update_file(opt, 0, o, update_path);
1491 } else {
1493 * Despite the four nearly duplicate messages and argument
1494 * lists below and the ugliness of the nested if-statements,
1495 * having complete messages makes the job easier for
1496 * translators.
1498 * The slight variance among the cases is due to the fact
1499 * that:
1500 * 1) directory/file conflicts (in effect if
1501 * !alt_path) could cause us to need to write the
1502 * file to a different path.
1503 * 2) renames (in effect if !old_path) could mean that
1504 * there are two names for the path that the user
1505 * may know the file by.
1507 if (!alt_path) {
1508 if (!old_path) {
1509 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1510 "and %s in %s. Version %s of %s left in tree."),
1511 change, path, delete_branch, change_past,
1512 change_branch, change_branch, path);
1513 } else {
1514 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1515 "and %s to %s in %s. Version %s of %s left in tree."),
1516 change, old_path, delete_branch, change_past, path,
1517 change_branch, change_branch, path);
1519 } else {
1520 if (!old_path) {
1521 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1522 "and %s in %s. Version %s of %s left in tree at %s."),
1523 change, path, delete_branch, change_past,
1524 change_branch, change_branch, path, alt_path);
1525 } else {
1526 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1527 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1528 change, old_path, delete_branch, change_past, path,
1529 change_branch, change_branch, path, alt_path);
1533 * No need to call update_file() on path when change_branch ==
1534 * opt->branch1 && !alt_path, since that would needlessly touch
1535 * path. We could call update_file_flags() with update_cache=0
1536 * and update_wd=0, but that's a no-op.
1538 if (change_branch != opt->branch1 || alt_path)
1539 ret = update_file(opt, 0, changed, update_path);
1541 free(alt_path);
1543 return ret;
1546 static int handle_rename_delete(struct merge_options *opt,
1547 struct rename_conflict_info *ci)
1549 const struct rename *ren = ci->ren1;
1550 const struct diff_filespec *orig = ren->pair->one;
1551 const struct diff_filespec *dest = ren->pair->two;
1552 const char *rename_branch = ren->branch;
1553 const char *delete_branch = (opt->branch1 == ren->branch ?
1554 opt->branch2 : opt->branch1);
1556 if (handle_change_delete(opt,
1557 opt->priv->call_depth ? orig->path : dest->path,
1558 opt->priv->call_depth ? NULL : orig->path,
1559 orig, dest,
1560 rename_branch, delete_branch,
1561 _("rename"), _("renamed")))
1562 return -1;
1564 if (opt->priv->call_depth)
1565 return remove_file_from_index(opt->repo->index, dest->path);
1566 else
1567 return update_stages(opt, dest->path, NULL,
1568 rename_branch == opt->branch1 ? dest : NULL,
1569 rename_branch == opt->branch1 ? NULL : dest);
1572 static int handle_file_collision(struct merge_options *opt,
1573 const char *collide_path,
1574 const char *prev_path1,
1575 const char *prev_path2,
1576 const char *branch1, const char *branch2,
1577 struct diff_filespec *a,
1578 struct diff_filespec *b)
1580 struct merge_file_info mfi;
1581 struct diff_filespec null;
1582 char *alt_path = NULL;
1583 const char *update_path = collide_path;
1586 * It's easiest to get the correct things into stage 2 and 3, and
1587 * to make sure that the content merge puts HEAD before the other
1588 * branch if we just ensure that branch1 == opt->branch1. So, simply
1589 * flip arguments around if we don't have that.
1591 if (branch1 != opt->branch1) {
1592 return handle_file_collision(opt, collide_path,
1593 prev_path2, prev_path1,
1594 branch2, branch1,
1595 b, a);
1598 /* Remove rename sources if rename/add or rename/rename(2to1) */
1599 if (prev_path1)
1600 remove_file(opt, 1, prev_path1,
1601 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1602 if (prev_path2)
1603 remove_file(opt, 1, prev_path2,
1604 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1607 * Remove the collision path, if it wouldn't cause dirty contents
1608 * or an untracked file to get lost. We'll either overwrite with
1609 * merged contents, or just write out to differently named files.
1611 if (was_dirty(opt, collide_path)) {
1612 output(opt, 1, _("Refusing to lose dirty file at %s"),
1613 collide_path);
1614 update_path = alt_path = unique_path(opt, collide_path, "merged");
1615 } else if (would_lose_untracked(opt, collide_path)) {
1617 * Only way we get here is if both renames were from
1618 * a directory rename AND user had an untracked file
1619 * at the location where both files end up after the
1620 * two directory renames. See testcase 10d of t6043.
1622 output(opt, 1, _("Refusing to lose untracked file at "
1623 "%s, even though it's in the way."),
1624 collide_path);
1625 update_path = alt_path = unique_path(opt, collide_path, "merged");
1626 } else {
1628 * FIXME: It's possible that the two files are identical
1629 * and that the current working copy happens to match, in
1630 * which case we are unnecessarily touching the working
1631 * tree file. It's not a likely enough scenario that I
1632 * want to code up the checks for it and a better fix is
1633 * available if we restructure how unpack_trees() and
1634 * merge-recursive interoperate anyway, so punting for
1635 * now...
1637 remove_file(opt, 0, collide_path, 0);
1640 /* Store things in diff_filespecs for functions that need it */
1641 null.path = (char *)collide_path;
1642 oidcpy(&null.oid, null_oid());
1643 null.mode = 0;
1645 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1646 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1647 return -1;
1648 mfi.clean &= !alt_path;
1649 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1650 return -1;
1651 if (!mfi.clean && !opt->priv->call_depth &&
1652 update_stages(opt, collide_path, NULL, a, b))
1653 return -1;
1654 free(alt_path);
1656 * FIXME: If both a & b both started with conflicts (only possible
1657 * if they came from a rename/rename(2to1)), but had IDENTICAL
1658 * contents including those conflicts, then in the next line we claim
1659 * it was clean. If someone cares about this case, we should have the
1660 * caller notify us if we started with conflicts.
1662 return mfi.clean;
1665 static int handle_rename_add(struct merge_options *opt,
1666 struct rename_conflict_info *ci)
1668 /* a was renamed to c, and a separate c was added. */
1669 struct diff_filespec *a = ci->ren1->pair->one;
1670 struct diff_filespec *c = ci->ren1->pair->two;
1671 char *path = c->path;
1672 char *prev_path_desc;
1673 struct merge_file_info mfi;
1675 const char *rename_branch = ci->ren1->branch;
1676 const char *add_branch = (opt->branch1 == rename_branch ?
1677 opt->branch2 : opt->branch1);
1678 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1680 output(opt, 1, _("CONFLICT (rename/add): "
1681 "Rename %s->%s in %s. Added %s in %s"),
1682 a->path, c->path, rename_branch,
1683 c->path, add_branch);
1685 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1686 ci->ren1->src_entry->stages[other_stage].path = a->path;
1687 if (merge_mode_and_contents(opt, a, c,
1688 &ci->ren1->src_entry->stages[other_stage],
1689 prev_path_desc,
1690 opt->branch1, opt->branch2,
1691 1 + opt->priv->call_depth * 2, &mfi))
1692 return -1;
1693 free(prev_path_desc);
1695 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1696 return handle_file_collision(opt,
1697 c->path, a->path, NULL,
1698 rename_branch, add_branch,
1699 &mfi.blob,
1700 &ci->ren1->dst_entry->stages[other_stage]);
1703 static char *find_path_for_conflict(struct merge_options *opt,
1704 const char *path,
1705 const char *branch1,
1706 const char *branch2)
1708 char *new_path = NULL;
1709 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1710 new_path = unique_path(opt, path, branch1);
1711 output(opt, 1, _("%s is a directory in %s adding "
1712 "as %s instead"),
1713 path, branch2, new_path);
1714 } else if (would_lose_untracked(opt, path)) {
1715 new_path = unique_path(opt, path, branch1);
1716 output(opt, 1, _("Refusing to lose untracked file"
1717 " at %s; adding as %s instead"),
1718 path, new_path);
1721 return new_path;
1725 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1727 static inline int flip_stage(int stage)
1729 return (2 + 3) - stage;
1732 static int handle_rename_rename_1to2(struct merge_options *opt,
1733 struct rename_conflict_info *ci)
1735 /* One file was renamed in both branches, but to different names. */
1736 struct merge_file_info mfi;
1737 struct diff_filespec *add;
1738 struct diff_filespec *o = ci->ren1->pair->one;
1739 struct diff_filespec *a = ci->ren1->pair->two;
1740 struct diff_filespec *b = ci->ren2->pair->two;
1741 char *path_desc;
1743 output(opt, 1, _("CONFLICT (rename/rename): "
1744 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1745 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1746 o->path, a->path, ci->ren1->branch,
1747 o->path, b->path, ci->ren2->branch,
1748 opt->priv->call_depth ? _(" (left unresolved)") : "");
1750 path_desc = xstrfmt("%s and %s, both renamed from %s",
1751 a->path, b->path, o->path);
1752 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1753 ci->ren1->branch, ci->ren2->branch,
1754 opt->priv->call_depth * 2, &mfi))
1755 return -1;
1756 free(path_desc);
1758 if (opt->priv->call_depth)
1759 remove_file_from_index(opt->repo->index, o->path);
1762 * For each destination path, we need to see if there is a
1763 * rename/add collision. If not, we can write the file out
1764 * to the specified location.
1766 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1767 if (is_valid(add)) {
1768 add->path = mfi.blob.path = a->path;
1769 if (handle_file_collision(opt, a->path,
1770 NULL, NULL,
1771 ci->ren1->branch,
1772 ci->ren2->branch,
1773 &mfi.blob, add) < 0)
1774 return -1;
1775 } else {
1776 char *new_path = find_path_for_conflict(opt, a->path,
1777 ci->ren1->branch,
1778 ci->ren2->branch);
1779 if (update_file(opt, 0, &mfi.blob,
1780 new_path ? new_path : a->path))
1781 return -1;
1782 free(new_path);
1783 if (!opt->priv->call_depth &&
1784 update_stages(opt, a->path, NULL, a, NULL))
1785 return -1;
1788 if (!mfi.clean && mfi.blob.mode == a->mode &&
1789 oideq(&mfi.blob.oid, &a->oid)) {
1791 * Getting here means we were attempting to merge a binary
1792 * blob. Since we can't merge binaries, the merge algorithm
1793 * just takes one side. But we don't want to copy the
1794 * contents of one side to both paths; we'd rather use the
1795 * original content at the given path for each path.
1797 oidcpy(&mfi.blob.oid, &b->oid);
1798 mfi.blob.mode = b->mode;
1800 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1801 if (is_valid(add)) {
1802 add->path = mfi.blob.path = b->path;
1803 if (handle_file_collision(opt, b->path,
1804 NULL, NULL,
1805 ci->ren1->branch,
1806 ci->ren2->branch,
1807 add, &mfi.blob) < 0)
1808 return -1;
1809 } else {
1810 char *new_path = find_path_for_conflict(opt, b->path,
1811 ci->ren2->branch,
1812 ci->ren1->branch);
1813 if (update_file(opt, 0, &mfi.blob,
1814 new_path ? new_path : b->path))
1815 return -1;
1816 free(new_path);
1817 if (!opt->priv->call_depth &&
1818 update_stages(opt, b->path, NULL, NULL, b))
1819 return -1;
1822 return 0;
1825 static int handle_rename_rename_2to1(struct merge_options *opt,
1826 struct rename_conflict_info *ci)
1828 /* Two files, a & b, were renamed to the same thing, c. */
1829 struct diff_filespec *a = ci->ren1->pair->one;
1830 struct diff_filespec *b = ci->ren2->pair->one;
1831 struct diff_filespec *c1 = ci->ren1->pair->two;
1832 struct diff_filespec *c2 = ci->ren2->pair->two;
1833 char *path = c1->path; /* == c2->path */
1834 char *path_side_1_desc;
1835 char *path_side_2_desc;
1836 struct merge_file_info mfi_c1;
1837 struct merge_file_info mfi_c2;
1838 int ostage1, ostage2;
1840 output(opt, 1, _("CONFLICT (rename/rename): "
1841 "Rename %s->%s in %s. "
1842 "Rename %s->%s in %s"),
1843 a->path, c1->path, ci->ren1->branch,
1844 b->path, c2->path, ci->ren2->branch);
1846 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1847 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1848 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1849 ostage2 = flip_stage(ostage1);
1850 ci->ren1->src_entry->stages[ostage1].path = a->path;
1851 ci->ren2->src_entry->stages[ostage2].path = b->path;
1852 if (merge_mode_and_contents(opt, a, c1,
1853 &ci->ren1->src_entry->stages[ostage1],
1854 path_side_1_desc,
1855 opt->branch1, opt->branch2,
1856 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1857 merge_mode_and_contents(opt, b,
1858 &ci->ren2->src_entry->stages[ostage2],
1859 c2, path_side_2_desc,
1860 opt->branch1, opt->branch2,
1861 1 + opt->priv->call_depth * 2, &mfi_c2))
1862 return -1;
1863 free(path_side_1_desc);
1864 free(path_side_2_desc);
1865 mfi_c1.blob.path = path;
1866 mfi_c2.blob.path = path;
1868 return handle_file_collision(opt, path, a->path, b->path,
1869 ci->ren1->branch, ci->ren2->branch,
1870 &mfi_c1.blob, &mfi_c2.blob);
1874 * Get the diff_filepairs changed between o_tree and tree.
1876 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1877 struct tree *o_tree,
1878 struct tree *tree)
1880 struct diff_queue_struct *ret;
1881 struct diff_options opts;
1883 repo_diff_setup(opt->repo, &opts);
1884 opts.flags.recursive = 1;
1885 opts.flags.rename_empty = 0;
1886 opts.detect_rename = merge_detect_rename(opt);
1888 * We do not have logic to handle the detection of copies. In
1889 * fact, it may not even make sense to add such logic: would we
1890 * really want a change to a base file to be propagated through
1891 * multiple other files by a merge?
1893 if (opts.detect_rename > DIFF_DETECT_RENAME)
1894 opts.detect_rename = DIFF_DETECT_RENAME;
1895 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 7000;
1896 opts.rename_score = opt->rename_score;
1897 opts.show_rename_progress = opt->show_rename_progress;
1898 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1899 diff_setup_done(&opts);
1900 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1901 diffcore_std(&opts);
1902 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1903 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1905 ret = xmalloc(sizeof(*ret));
1906 *ret = diff_queued_diff;
1908 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1909 diff_queued_diff.nr = 0;
1910 diff_queued_diff.queue = NULL;
1911 diff_flush(&opts);
1912 return ret;
1915 static int tree_has_path(struct repository *r, struct tree *tree,
1916 const char *path)
1918 struct object_id hashy;
1919 unsigned short mode_o;
1921 return !get_tree_entry(r,
1922 &tree->object.oid, path,
1923 &hashy, &mode_o);
1927 * Return a new string that replaces the beginning portion (which matches
1928 * entry->dir), with entry->new_dir. In perl-speak:
1929 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1930 * NOTE:
1931 * Caller must ensure that old_path starts with entry->dir + '/'.
1933 static char *apply_dir_rename(struct dir_rename_entry *entry,
1934 const char *old_path)
1936 struct strbuf new_path = STRBUF_INIT;
1937 int oldlen, newlen;
1939 if (entry->non_unique_new_dir)
1940 return NULL;
1942 oldlen = strlen(entry->dir);
1943 if (entry->new_dir.len == 0)
1945 * If someone renamed/merged a subdirectory into the root
1946 * directory (e.g. 'some/subdir' -> ''), then we want to
1947 * avoid returning
1948 * '' + '/filename'
1949 * as the rename; we need to make old_path + oldlen advance
1950 * past the '/' character.
1952 oldlen++;
1953 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1954 strbuf_grow(&new_path, newlen);
1955 strbuf_addbuf(&new_path, &entry->new_dir);
1956 strbuf_addstr(&new_path, &old_path[oldlen]);
1958 return strbuf_detach(&new_path, NULL);
1961 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1962 char **old_dir, char **new_dir)
1964 char *end_of_old, *end_of_new;
1966 /* Default return values: NULL, meaning no rename */
1967 *old_dir = NULL;
1968 *new_dir = NULL;
1971 * For
1972 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1973 * the "e/foo.c" part is the same, we just want to know that
1974 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1975 * so, for this example, this function returns "a/b/c/d" in
1976 * *old_dir and "a/b/some/thing/else" in *new_dir.
1980 * If the basename of the file changed, we don't care. We want
1981 * to know which portion of the directory, if any, changed.
1983 end_of_old = strrchr(old_path, '/');
1984 end_of_new = strrchr(new_path, '/');
1987 * If end_of_old is NULL, old_path wasn't in a directory, so there
1988 * could not be a directory rename (our rule elsewhere that a
1989 * directory which still exists is not considered to have been
1990 * renamed means the root directory can never be renamed -- because
1991 * the root directory always exists).
1993 if (end_of_old == NULL)
1994 return; /* Note: *old_dir and *new_dir are still NULL */
1997 * If new_path contains no directory (end_of_new is NULL), then we
1998 * have a rename of old_path's directory to the root directory.
2000 if (end_of_new == NULL) {
2001 *old_dir = xstrndup(old_path, end_of_old - old_path);
2002 *new_dir = xstrdup("");
2003 return;
2006 /* Find the first non-matching character traversing backwards */
2007 while (*--end_of_new == *--end_of_old &&
2008 end_of_old != old_path &&
2009 end_of_new != new_path)
2010 ; /* Do nothing; all in the while loop */
2013 * If both got back to the beginning of their strings, then the
2014 * directory didn't change at all, only the basename did.
2016 if (end_of_old == old_path && end_of_new == new_path &&
2017 *end_of_old == *end_of_new)
2018 return; /* Note: *old_dir and *new_dir are still NULL */
2021 * If end_of_new got back to the beginning of its string, and
2022 * end_of_old got back to the beginning of some subdirectory, then
2023 * we have a rename/merge of a subdirectory into the root, which
2024 * needs slightly special handling.
2026 * Note: There is no need to consider the opposite case, with a
2027 * rename/merge of the root directory into some subdirectory
2028 * because as noted above the root directory always exists so it
2029 * cannot be considered to be renamed.
2031 if (end_of_new == new_path &&
2032 end_of_old != old_path && end_of_old[-1] == '/') {
2033 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2034 *new_dir = xstrdup("");
2035 return;
2039 * We've found the first non-matching character in the directory
2040 * paths. That means the current characters we were looking at
2041 * were part of the first non-matching subdir name going back from
2042 * the end of the strings. Get the whole name by advancing both
2043 * end_of_old and end_of_new to the NEXT '/' character. That will
2044 * represent the entire directory rename.
2046 * The reason for the increment is cases like
2047 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2048 * After dropping the basename and going back to the first
2049 * non-matching character, we're now comparing:
2050 * a/b/s and a/b/
2051 * and we want to be comparing:
2052 * a/b/star/ and a/b/tar/
2053 * but without the pre-increment, the one on the right would stay
2054 * a/b/.
2056 end_of_old = strchr(++end_of_old, '/');
2057 end_of_new = strchr(++end_of_new, '/');
2059 /* Copy the old and new directories into *old_dir and *new_dir. */
2060 *old_dir = xstrndup(old_path, end_of_old - old_path);
2061 *new_dir = xstrndup(new_path, end_of_new - new_path);
2064 static void remove_hashmap_entries(struct hashmap *dir_renames,
2065 struct string_list *items_to_remove)
2067 int i;
2068 struct dir_rename_entry *entry;
2070 for (i = 0; i < items_to_remove->nr; i++) {
2071 entry = items_to_remove->items[i].util;
2072 hashmap_remove(dir_renames, &entry->ent, NULL);
2074 string_list_clear(items_to_remove, 0);
2078 * See if there is a directory rename for path, and if there are any file
2079 * level conflicts for the renamed location. If there is a rename and
2080 * there are no conflicts, return the new name. Otherwise, return NULL.
2082 static char *handle_path_level_conflicts(struct merge_options *opt,
2083 const char *path,
2084 struct dir_rename_entry *entry,
2085 struct hashmap *collisions,
2086 struct tree *tree)
2088 char *new_path = NULL;
2089 struct collision_entry *collision_ent;
2090 int clean = 1;
2091 struct strbuf collision_paths = STRBUF_INIT;
2094 * entry has the mapping of old directory name to new directory name
2095 * that we want to apply to path.
2097 new_path = apply_dir_rename(entry, path);
2099 if (!new_path) {
2100 /* This should only happen when entry->non_unique_new_dir set */
2101 if (!entry->non_unique_new_dir)
2102 BUG("entry->non_unqiue_dir not set and !new_path");
2103 output(opt, 1, _("CONFLICT (directory rename split): "
2104 "Unclear where to place %s because directory "
2105 "%s was renamed to multiple other directories, "
2106 "with no destination getting a majority of the "
2107 "files."),
2108 path, entry->dir);
2109 clean = 0;
2110 return NULL;
2114 * The caller needs to have ensured that it has pre-populated
2115 * collisions with all paths that map to new_path. Do a quick check
2116 * to ensure that's the case.
2118 collision_ent = collision_find_entry(collisions, new_path);
2119 if (collision_ent == NULL)
2120 BUG("collision_ent is NULL");
2123 * Check for one-sided add/add/.../add conflicts, i.e.
2124 * where implicit renames from the other side doing
2125 * directory rename(s) can affect this side of history
2126 * to put multiple paths into the same location. Warn
2127 * and bail on directory renames for such paths.
2129 if (collision_ent->reported_already) {
2130 clean = 0;
2131 } else if (tree_has_path(opt->repo, tree, new_path)) {
2132 collision_ent->reported_already = 1;
2133 strbuf_add_separated_string_list(&collision_paths, ", ",
2134 &collision_ent->source_files);
2135 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2136 "file/dir at %s in the way of implicit "
2137 "directory rename(s) putting the following "
2138 "path(s) there: %s."),
2139 new_path, collision_paths.buf);
2140 clean = 0;
2141 } else if (collision_ent->source_files.nr > 1) {
2142 collision_ent->reported_already = 1;
2143 strbuf_add_separated_string_list(&collision_paths, ", ",
2144 &collision_ent->source_files);
2145 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2146 "more than one path to %s; implicit directory "
2147 "renames tried to put these paths there: %s"),
2148 new_path, collision_paths.buf);
2149 clean = 0;
2152 /* Free memory we no longer need */
2153 strbuf_release(&collision_paths);
2154 if (!clean && new_path) {
2155 free(new_path);
2156 return NULL;
2159 return new_path;
2163 * There are a couple things we want to do at the directory level:
2164 * 1. Check for both sides renaming to the same thing, in order to avoid
2165 * implicit renaming of files that should be left in place. (See
2166 * testcase 6b in t6043 for details.)
2167 * 2. Prune directory renames if there are still files left in the
2168 * original directory. These represent a partial directory rename,
2169 * i.e. a rename where only some of the files within the directory
2170 * were renamed elsewhere. (Technically, this could be done earlier
2171 * in get_directory_renames(), except that would prevent us from
2172 * doing the previous check and thus failing testcase 6b.)
2173 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2174 * In the future, we could potentially record this info as well and
2175 * omit reporting rename/rename(1to2) conflicts for each path within
2176 * the affected directories, thus cleaning up the merge output.
2177 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2178 * directory level, because merging directories is fine. If it
2179 * causes conflicts for files within those merged directories, then
2180 * that should be detected at the individual path level.
2182 static void handle_directory_level_conflicts(struct merge_options *opt,
2183 struct hashmap *dir_re_head,
2184 struct tree *head,
2185 struct hashmap *dir_re_merge,
2186 struct tree *merge)
2188 struct hashmap_iter iter;
2189 struct dir_rename_entry *head_ent;
2190 struct dir_rename_entry *merge_ent;
2192 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2193 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2195 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2196 ent /* member name */) {
2197 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2198 if (merge_ent &&
2199 !head_ent->non_unique_new_dir &&
2200 !merge_ent->non_unique_new_dir &&
2201 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2202 /* 1. Renamed identically; remove it from both sides */
2203 string_list_append(&remove_from_head,
2204 head_ent->dir)->util = head_ent;
2205 strbuf_release(&head_ent->new_dir);
2206 string_list_append(&remove_from_merge,
2207 merge_ent->dir)->util = merge_ent;
2208 strbuf_release(&merge_ent->new_dir);
2209 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2210 /* 2. This wasn't a directory rename after all */
2211 string_list_append(&remove_from_head,
2212 head_ent->dir)->util = head_ent;
2213 strbuf_release(&head_ent->new_dir);
2217 remove_hashmap_entries(dir_re_head, &remove_from_head);
2218 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2220 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2221 ent /* member name */) {
2222 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2223 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2224 /* 2. This wasn't a directory rename after all */
2225 string_list_append(&remove_from_merge,
2226 merge_ent->dir)->util = merge_ent;
2227 } else if (head_ent &&
2228 !head_ent->non_unique_new_dir &&
2229 !merge_ent->non_unique_new_dir) {
2230 /* 3. rename/rename(1to2) */
2232 * We can assume it's not rename/rename(1to1) because
2233 * that was case (1), already checked above. So we
2234 * know that head_ent->new_dir and merge_ent->new_dir
2235 * are different strings.
2237 output(opt, 1, _("CONFLICT (rename/rename): "
2238 "Rename directory %s->%s in %s. "
2239 "Rename directory %s->%s in %s"),
2240 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2241 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2242 string_list_append(&remove_from_head,
2243 head_ent->dir)->util = head_ent;
2244 strbuf_release(&head_ent->new_dir);
2245 string_list_append(&remove_from_merge,
2246 merge_ent->dir)->util = merge_ent;
2247 strbuf_release(&merge_ent->new_dir);
2251 remove_hashmap_entries(dir_re_head, &remove_from_head);
2252 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2255 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2257 struct hashmap *dir_renames;
2258 struct hashmap_iter iter;
2259 struct dir_rename_entry *entry;
2260 int i;
2263 * Typically, we think of a directory rename as all files from a
2264 * certain directory being moved to a target directory. However,
2265 * what if someone first moved two files from the original
2266 * directory in one commit, and then renamed the directory
2267 * somewhere else in a later commit? At merge time, we just know
2268 * that files from the original directory went to two different
2269 * places, and that the bulk of them ended up in the same place.
2270 * We want each directory rename to represent where the bulk of the
2271 * files from that directory end up; this function exists to find
2272 * where the bulk of the files went.
2274 * The first loop below simply iterates through the list of file
2275 * renames, finding out how often each directory rename pair
2276 * possibility occurs.
2278 dir_renames = xmalloc(sizeof(*dir_renames));
2279 dir_rename_init(dir_renames);
2280 for (i = 0; i < pairs->nr; ++i) {
2281 struct string_list_item *item;
2282 int *count;
2283 struct diff_filepair *pair = pairs->queue[i];
2284 char *old_dir, *new_dir;
2286 /* File not part of directory rename if it wasn't renamed */
2287 if (pair->status != 'R')
2288 continue;
2290 get_renamed_dir_portion(pair->one->path, pair->two->path,
2291 &old_dir, &new_dir);
2292 if (!old_dir)
2293 /* Directory didn't change at all; ignore this one. */
2294 continue;
2296 entry = dir_rename_find_entry(dir_renames, old_dir);
2297 if (!entry) {
2298 entry = xmalloc(sizeof(*entry));
2299 dir_rename_entry_init(entry, old_dir);
2300 hashmap_put(dir_renames, &entry->ent);
2301 } else {
2302 free(old_dir);
2304 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2305 if (!item) {
2306 item = string_list_insert(&entry->possible_new_dirs,
2307 new_dir);
2308 item->util = xcalloc(1, sizeof(int));
2309 } else {
2310 free(new_dir);
2312 count = item->util;
2313 *count += 1;
2317 * For each directory with files moved out of it, we find out which
2318 * target directory received the most files so we can declare it to
2319 * be the "winning" target location for the directory rename. This
2320 * winner gets recorded in new_dir. If there is no winner
2321 * (multiple target directories received the same number of files),
2322 * we set non_unique_new_dir. Once we've determined the winner (or
2323 * that there is no winner), we no longer need possible_new_dirs.
2325 hashmap_for_each_entry(dir_renames, &iter, entry,
2326 ent /* member name */) {
2327 int max = 0;
2328 int bad_max = 0;
2329 char *best = NULL;
2331 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2332 int *count = entry->possible_new_dirs.items[i].util;
2334 if (*count == max)
2335 bad_max = max;
2336 else if (*count > max) {
2337 max = *count;
2338 best = entry->possible_new_dirs.items[i].string;
2341 if (bad_max == max)
2342 entry->non_unique_new_dir = 1;
2343 else {
2344 assert(entry->new_dir.len == 0);
2345 strbuf_addstr(&entry->new_dir, best);
2348 * The relevant directory sub-portion of the original full
2349 * filepaths were xstrndup'ed before inserting into
2350 * possible_new_dirs, and instead of manually iterating the
2351 * list and free'ing each, just lie and tell
2352 * possible_new_dirs that it did the strdup'ing so that it
2353 * will free them for us.
2355 entry->possible_new_dirs.strdup_strings = 1;
2356 string_list_clear(&entry->possible_new_dirs, 1);
2359 return dir_renames;
2362 static struct dir_rename_entry *check_dir_renamed(const char *path,
2363 struct hashmap *dir_renames)
2365 char *temp = xstrdup(path);
2366 char *end;
2367 struct dir_rename_entry *entry = NULL;
2369 while ((end = strrchr(temp, '/'))) {
2370 *end = '\0';
2371 entry = dir_rename_find_entry(dir_renames, temp);
2372 if (entry)
2373 break;
2375 free(temp);
2376 return entry;
2379 static void compute_collisions(struct hashmap *collisions,
2380 struct hashmap *dir_renames,
2381 struct diff_queue_struct *pairs)
2383 int i;
2386 * Multiple files can be mapped to the same path due to directory
2387 * renames done by the other side of history. Since that other
2388 * side of history could have merged multiple directories into one,
2389 * if our side of history added the same file basename to each of
2390 * those directories, then all N of them would get implicitly
2391 * renamed by the directory rename detection into the same path,
2392 * and we'd get an add/add/.../add conflict, and all those adds
2393 * from *this* side of history. This is not representable in the
2394 * index, and users aren't going to easily be able to make sense of
2395 * it. So we need to provide a good warning about what's
2396 * happening, and fall back to no-directory-rename detection
2397 * behavior for those paths.
2399 * See testcases 9e and all of section 5 from t6043 for examples.
2401 collision_init(collisions);
2403 for (i = 0; i < pairs->nr; ++i) {
2404 struct dir_rename_entry *dir_rename_ent;
2405 struct collision_entry *collision_ent;
2406 char *new_path;
2407 struct diff_filepair *pair = pairs->queue[i];
2409 if (pair->status != 'A' && pair->status != 'R')
2410 continue;
2411 dir_rename_ent = check_dir_renamed(pair->two->path,
2412 dir_renames);
2413 if (!dir_rename_ent)
2414 continue;
2416 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2417 if (!new_path)
2419 * dir_rename_ent->non_unique_new_path is true, which
2420 * means there is no directory rename for us to use,
2421 * which means it won't cause us any additional
2422 * collisions.
2424 continue;
2425 collision_ent = collision_find_entry(collisions, new_path);
2426 if (!collision_ent) {
2427 CALLOC_ARRAY(collision_ent, 1);
2428 hashmap_entry_init(&collision_ent->ent,
2429 strhash(new_path));
2430 hashmap_put(collisions, &collision_ent->ent);
2431 collision_ent->target_file = new_path;
2432 } else {
2433 free(new_path);
2435 string_list_insert(&collision_ent->source_files,
2436 pair->two->path);
2440 static char *check_for_directory_rename(struct merge_options *opt,
2441 const char *path,
2442 struct tree *tree,
2443 struct hashmap *dir_renames,
2444 struct hashmap *dir_rename_exclusions,
2445 struct hashmap *collisions,
2446 int *clean_merge)
2448 char *new_path = NULL;
2449 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2450 struct dir_rename_entry *oentry = NULL;
2452 if (!entry)
2453 return new_path;
2456 * This next part is a little weird. We do not want to do an
2457 * implicit rename into a directory we renamed on our side, because
2458 * that will result in a spurious rename/rename(1to2) conflict. An
2459 * example:
2460 * Base commit: dumbdir/afile, otherdir/bfile
2461 * Side 1: smrtdir/afile, otherdir/bfile
2462 * Side 2: dumbdir/afile, dumbdir/bfile
2463 * Here, while working on Side 1, we could notice that otherdir was
2464 * renamed/merged to dumbdir, and change the diff_filepair for
2465 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2466 * 2 will notice the rename from dumbdir to smrtdir, and do the
2467 * transitive rename to move it from dumbdir/bfile to
2468 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2469 * smrtdir, a rename/rename(1to2) conflict. We really just want
2470 * the file to end up in smrtdir. And the way to achieve that is
2471 * to not let Side1 do the rename to dumbdir, since we know that is
2472 * the source of one of our directory renames.
2474 * That's why oentry and dir_rename_exclusions is here.
2476 * As it turns out, this also prevents N-way transient rename
2477 * confusion; See testcases 9c and 9d of t6043.
2479 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2480 if (oentry) {
2481 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2482 "to %s, because %s itself was renamed."),
2483 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2484 } else {
2485 new_path = handle_path_level_conflicts(opt, path, entry,
2486 collisions, tree);
2487 *clean_merge &= (new_path != NULL);
2490 return new_path;
2493 static void apply_directory_rename_modifications(struct merge_options *opt,
2494 struct diff_filepair *pair,
2495 char *new_path,
2496 struct rename *re,
2497 struct tree *tree,
2498 struct tree *o_tree,
2499 struct tree *a_tree,
2500 struct tree *b_tree,
2501 struct string_list *entries)
2503 struct string_list_item *item;
2504 int stage = (tree == a_tree ? 2 : 3);
2505 int update_wd;
2508 * In all cases where we can do directory rename detection,
2509 * unpack_trees() will have read pair->two->path into the
2510 * index and the working copy. We need to remove it so that
2511 * we can instead place it at new_path. It is guaranteed to
2512 * not be untracked (unpack_trees() would have errored out
2513 * saying the file would have been overwritten), but it might
2514 * be dirty, though.
2516 update_wd = !was_dirty(opt, pair->two->path);
2517 if (!update_wd)
2518 output(opt, 1, _("Refusing to lose dirty file at %s"),
2519 pair->two->path);
2520 remove_file(opt, 1, pair->two->path, !update_wd);
2522 /* Find or create a new re->dst_entry */
2523 item = string_list_lookup(entries, new_path);
2524 if (item) {
2526 * Since we're renaming on this side of history, and it's
2527 * due to a directory rename on the other side of history
2528 * (which we only allow when the directory in question no
2529 * longer exists on the other side of history), the
2530 * original entry for re->dst_entry is no longer
2531 * necessary...
2533 re->dst_entry->processed = 1;
2536 * ...because we'll be using this new one.
2538 re->dst_entry = item->util;
2539 } else {
2541 * re->dst_entry is for the before-dir-rename path, and we
2542 * need it to hold information for the after-dir-rename
2543 * path. Before creating a new entry, we need to mark the
2544 * old one as unnecessary (...unless it is shared by
2545 * src_entry, i.e. this didn't use to be a rename, in which
2546 * case we can just allow the normal processing to happen
2547 * for it).
2549 if (pair->status == 'R')
2550 re->dst_entry->processed = 1;
2552 re->dst_entry = insert_stage_data(opt->repo, new_path,
2553 o_tree, a_tree, b_tree,
2554 entries);
2555 item = string_list_insert(entries, new_path);
2556 item->util = re->dst_entry;
2560 * Update the stage_data with the information about the path we are
2561 * moving into place. That slot will be empty and available for us
2562 * to write to because of the collision checks in
2563 * handle_path_level_conflicts(). In other words,
2564 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2565 * open for us to write to.
2567 * It may be tempting to actually update the index at this point as
2568 * well, using update_stages_for_stage_data(), but as per the big
2569 * "NOTE" in update_stages(), doing so will modify the current
2570 * in-memory index which will break calls to would_lose_untracked()
2571 * that we need to make. Instead, we need to just make sure that
2572 * the various handle_rename_*() functions update the index
2573 * explicitly rather than relying on unpack_trees() to have done it.
2575 get_tree_entry(opt->repo,
2576 &tree->object.oid,
2577 pair->two->path,
2578 &re->dst_entry->stages[stage].oid,
2579 &re->dst_entry->stages[stage].mode);
2582 * Record the original change status (or 'type' of change). If it
2583 * was originally an add ('A'), this lets us differentiate later
2584 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2585 * otherwise look the same). If it was originally a rename ('R'),
2586 * this lets us remember and report accurately about the transitive
2587 * renaming that occurred via the directory rename detection. Also,
2588 * record the original destination name.
2590 re->dir_rename_original_type = pair->status;
2591 re->dir_rename_original_dest = pair->two->path;
2594 * We don't actually look at pair->status again, but it seems
2595 * pedagogically correct to adjust it.
2597 pair->status = 'R';
2600 * Finally, record the new location.
2602 pair->two->path = new_path;
2606 * Get information of all renames which occurred in 'pairs', making use of
2607 * any implicit directory renames inferred from the other side of history.
2608 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2609 * to be able to associate the correct cache entries with the rename
2610 * information; tree is always equal to either a_tree or b_tree.
2612 static struct string_list *get_renames(struct merge_options *opt,
2613 const char *branch,
2614 struct diff_queue_struct *pairs,
2615 struct hashmap *dir_renames,
2616 struct hashmap *dir_rename_exclusions,
2617 struct tree *tree,
2618 struct tree *o_tree,
2619 struct tree *a_tree,
2620 struct tree *b_tree,
2621 struct string_list *entries,
2622 int *clean_merge)
2624 int i;
2625 struct hashmap collisions;
2626 struct hashmap_iter iter;
2627 struct collision_entry *e;
2628 struct string_list *renames;
2630 compute_collisions(&collisions, dir_renames, pairs);
2631 CALLOC_ARRAY(renames, 1);
2633 for (i = 0; i < pairs->nr; ++i) {
2634 struct string_list_item *item;
2635 struct rename *re;
2636 struct diff_filepair *pair = pairs->queue[i];
2637 char *new_path; /* non-NULL only with directory renames */
2639 if (pair->status != 'A' && pair->status != 'R') {
2640 diff_free_filepair(pair);
2641 continue;
2643 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2644 dir_renames,
2645 dir_rename_exclusions,
2646 &collisions,
2647 clean_merge);
2648 if (pair->status != 'R' && !new_path) {
2649 diff_free_filepair(pair);
2650 continue;
2653 re = xmalloc(sizeof(*re));
2654 re->processed = 0;
2655 re->pair = pair;
2656 re->branch = branch;
2657 re->dir_rename_original_type = '\0';
2658 re->dir_rename_original_dest = NULL;
2659 item = string_list_lookup(entries, re->pair->one->path);
2660 if (!item)
2661 re->src_entry = insert_stage_data(opt->repo,
2662 re->pair->one->path,
2663 o_tree, a_tree, b_tree, entries);
2664 else
2665 re->src_entry = item->util;
2667 item = string_list_lookup(entries, re->pair->two->path);
2668 if (!item)
2669 re->dst_entry = insert_stage_data(opt->repo,
2670 re->pair->two->path,
2671 o_tree, a_tree, b_tree, entries);
2672 else
2673 re->dst_entry = item->util;
2674 item = string_list_insert(renames, pair->one->path);
2675 item->util = re;
2676 if (new_path)
2677 apply_directory_rename_modifications(opt, pair, new_path,
2678 re, tree, o_tree,
2679 a_tree, b_tree,
2680 entries);
2683 hashmap_for_each_entry(&collisions, &iter, e,
2684 ent /* member name */) {
2685 free(e->target_file);
2686 string_list_clear(&e->source_files, 0);
2688 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2689 return renames;
2692 static int process_renames(struct merge_options *opt,
2693 struct string_list *a_renames,
2694 struct string_list *b_renames)
2696 int clean_merge = 1, i, j;
2697 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2698 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2699 const struct rename *sre;
2702 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2703 * string_list one-by-one, but O(n log n) to build it unsorted and
2704 * then sort it. Note that as we build the list, we do not need to
2705 * check if the existing destination path is already in the list,
2706 * because the structure of diffcore_rename guarantees we won't
2707 * have duplicates.
2709 for (i = 0; i < a_renames->nr; i++) {
2710 sre = a_renames->items[i].util;
2711 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2712 = (void *)sre;
2714 for (i = 0; i < b_renames->nr; i++) {
2715 sre = b_renames->items[i].util;
2716 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2717 = (void *)sre;
2720 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2721 struct string_list *renames1, *renames2Dst;
2722 struct rename *ren1 = NULL, *ren2 = NULL;
2723 const char *ren1_src, *ren1_dst;
2724 struct string_list_item *lookup;
2726 if (i >= a_renames->nr) {
2727 ren2 = b_renames->items[j++].util;
2728 } else if (j >= b_renames->nr) {
2729 ren1 = a_renames->items[i++].util;
2730 } else {
2731 int compare = strcmp(a_renames->items[i].string,
2732 b_renames->items[j].string);
2733 if (compare <= 0)
2734 ren1 = a_renames->items[i++].util;
2735 if (compare >= 0)
2736 ren2 = b_renames->items[j++].util;
2739 /* TODO: refactor, so that 1/2 are not needed */
2740 if (ren1) {
2741 renames1 = a_renames;
2742 renames2Dst = &b_by_dst;
2743 } else {
2744 renames1 = b_renames;
2745 renames2Dst = &a_by_dst;
2746 SWAP(ren2, ren1);
2749 if (ren1->processed)
2750 continue;
2751 ren1->processed = 1;
2752 ren1->dst_entry->processed = 1;
2753 /* BUG: We should only mark src_entry as processed if we
2754 * are not dealing with a rename + add-source case.
2756 ren1->src_entry->processed = 1;
2758 ren1_src = ren1->pair->one->path;
2759 ren1_dst = ren1->pair->two->path;
2761 if (ren2) {
2762 /* One file renamed on both sides */
2763 const char *ren2_src = ren2->pair->one->path;
2764 const char *ren2_dst = ren2->pair->two->path;
2765 enum rename_type rename_type;
2766 if (strcmp(ren1_src, ren2_src) != 0)
2767 BUG("ren1_src != ren2_src");
2768 ren2->dst_entry->processed = 1;
2769 ren2->processed = 1;
2770 if (strcmp(ren1_dst, ren2_dst) != 0) {
2771 rename_type = RENAME_ONE_FILE_TO_TWO;
2772 clean_merge = 0;
2773 } else {
2774 rename_type = RENAME_ONE_FILE_TO_ONE;
2775 /* BUG: We should only remove ren1_src in
2776 * the base stage (think of rename +
2777 * add-source cases).
2779 remove_file(opt, 1, ren1_src, 1);
2780 update_entry(ren1->dst_entry,
2781 ren1->pair->one,
2782 ren1->pair->two,
2783 ren2->pair->two);
2785 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2786 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2787 /* Two different files renamed to the same thing */
2788 char *ren2_dst;
2789 ren2 = lookup->util;
2790 ren2_dst = ren2->pair->two->path;
2791 if (strcmp(ren1_dst, ren2_dst) != 0)
2792 BUG("ren1_dst != ren2_dst");
2794 clean_merge = 0;
2795 ren2->processed = 1;
2797 * BUG: We should only mark src_entry as processed
2798 * if we are not dealing with a rename + add-source
2799 * case.
2801 ren2->src_entry->processed = 1;
2803 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2804 opt, ren1, ren2);
2805 } else {
2806 /* Renamed in 1, maybe changed in 2 */
2807 /* we only use sha1 and mode of these */
2808 struct diff_filespec src_other, dst_other;
2809 int try_merge;
2812 * unpack_trees loads entries from common-commit
2813 * into stage 1, from head-commit into stage 2, and
2814 * from merge-commit into stage 3. We keep track
2815 * of which side corresponds to the rename.
2817 int renamed_stage = a_renames == renames1 ? 2 : 3;
2818 int other_stage = a_renames == renames1 ? 3 : 2;
2821 * Directory renames have a funny corner case...
2823 int renamed_to_self = !strcmp(ren1_src, ren1_dst);
2825 /* BUG: We should only remove ren1_src in the base
2826 * stage and in other_stage (think of rename +
2827 * add-source case).
2829 if (!renamed_to_self)
2830 remove_file(opt, 1, ren1_src,
2831 renamed_stage == 2 ||
2832 !was_tracked(opt, ren1_src));
2834 oidcpy(&src_other.oid,
2835 &ren1->src_entry->stages[other_stage].oid);
2836 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2837 oidcpy(&dst_other.oid,
2838 &ren1->dst_entry->stages[other_stage].oid);
2839 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2840 try_merge = 0;
2842 if (oideq(&src_other.oid, null_oid()) &&
2843 ren1->dir_rename_original_type == 'A') {
2844 setup_rename_conflict_info(RENAME_VIA_DIR,
2845 opt, ren1, NULL);
2846 } else if (renamed_to_self) {
2847 setup_rename_conflict_info(RENAME_NORMAL,
2848 opt, ren1, NULL);
2849 } else if (oideq(&src_other.oid, null_oid())) {
2850 setup_rename_conflict_info(RENAME_DELETE,
2851 opt, ren1, NULL);
2852 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2853 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2855 * Added file on the other side identical to
2856 * the file being renamed: clean merge.
2857 * Also, there is no need to overwrite the
2858 * file already in the working copy, so call
2859 * update_file_flags() instead of
2860 * update_file().
2862 if (update_file_flags(opt,
2863 ren1->pair->two,
2864 ren1_dst,
2865 1, /* update_cache */
2866 0 /* update_wd */))
2867 clean_merge = -1;
2868 } else if (!oideq(&dst_other.oid, null_oid())) {
2870 * Probably not a clean merge, but it's
2871 * premature to set clean_merge to 0 here,
2872 * because if the rename merges cleanly and
2873 * the merge exactly matches the newly added
2874 * file, then the merge will be clean.
2876 setup_rename_conflict_info(RENAME_ADD,
2877 opt, ren1, NULL);
2878 } else
2879 try_merge = 1;
2881 if (clean_merge < 0)
2882 goto cleanup_and_return;
2883 if (try_merge) {
2884 struct diff_filespec *o, *a, *b;
2885 src_other.path = (char *)ren1_src;
2887 o = ren1->pair->one;
2888 if (a_renames == renames1) {
2889 a = ren1->pair->two;
2890 b = &src_other;
2891 } else {
2892 b = ren1->pair->two;
2893 a = &src_other;
2895 update_entry(ren1->dst_entry, o, a, b);
2896 setup_rename_conflict_info(RENAME_NORMAL,
2897 opt, ren1, NULL);
2901 cleanup_and_return:
2902 string_list_clear(&a_by_dst, 0);
2903 string_list_clear(&b_by_dst, 0);
2905 return clean_merge;
2908 struct rename_info {
2909 struct string_list *head_renames;
2910 struct string_list *merge_renames;
2913 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2914 struct hashmap *dir_renames)
2916 struct hashmap_iter iter;
2917 struct dir_rename_entry *e;
2919 hashmap_for_each_entry(dir_renames, &iter, e,
2920 ent /* member name */) {
2921 free(e->dir);
2922 strbuf_release(&e->new_dir);
2923 /* possible_new_dirs already cleared in get_directory_renames */
2925 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2926 free(dir_renames);
2928 free(pairs->queue);
2929 free(pairs);
2932 static int detect_and_process_renames(struct merge_options *opt,
2933 struct tree *common,
2934 struct tree *head,
2935 struct tree *merge,
2936 struct string_list *entries,
2937 struct rename_info *ri)
2939 struct diff_queue_struct *head_pairs, *merge_pairs;
2940 struct hashmap *dir_re_head, *dir_re_merge;
2941 int clean = 1;
2943 ri->head_renames = NULL;
2944 ri->merge_renames = NULL;
2946 if (!merge_detect_rename(opt))
2947 return 1;
2949 head_pairs = get_diffpairs(opt, common, head);
2950 merge_pairs = get_diffpairs(opt, common, merge);
2952 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2953 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2954 !opt->priv->call_depth)) {
2955 dir_re_head = get_directory_renames(head_pairs);
2956 dir_re_merge = get_directory_renames(merge_pairs);
2958 handle_directory_level_conflicts(opt,
2959 dir_re_head, head,
2960 dir_re_merge, merge);
2961 } else {
2962 dir_re_head = xmalloc(sizeof(*dir_re_head));
2963 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2964 dir_rename_init(dir_re_head);
2965 dir_rename_init(dir_re_merge);
2968 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2969 dir_re_merge, dir_re_head, head,
2970 common, head, merge, entries,
2971 &clean);
2972 if (clean < 0)
2973 goto cleanup;
2974 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2975 dir_re_head, dir_re_merge, merge,
2976 common, head, merge, entries,
2977 &clean);
2978 if (clean < 0)
2979 goto cleanup;
2980 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2982 cleanup:
2984 * Some cleanup is deferred until cleanup_renames() because the
2985 * data structures are still needed and referenced in
2986 * process_entry(). But there are a few things we can free now.
2988 initial_cleanup_rename(head_pairs, dir_re_head);
2989 initial_cleanup_rename(merge_pairs, dir_re_merge);
2991 return clean;
2994 static void final_cleanup_rename(struct string_list *rename)
2996 const struct rename *re;
2997 int i;
2999 if (rename == NULL)
3000 return;
3002 for (i = 0; i < rename->nr; i++) {
3003 re = rename->items[i].util;
3004 diff_free_filepair(re->pair);
3006 string_list_clear(rename, 1);
3007 free(rename);
3010 static void final_cleanup_renames(struct rename_info *re_info)
3012 final_cleanup_rename(re_info->head_renames);
3013 final_cleanup_rename(re_info->merge_renames);
3016 static int read_oid_strbuf(struct merge_options *opt,
3017 const struct object_id *oid,
3018 struct strbuf *dst)
3020 void *buf;
3021 enum object_type type;
3022 unsigned long size;
3023 buf = read_object_file(oid, &type, &size);
3024 if (!buf)
3025 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3026 if (type != OBJ_BLOB) {
3027 free(buf);
3028 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3030 strbuf_attach(dst, buf, size, size + 1);
3031 return 0;
3034 static int blob_unchanged(struct merge_options *opt,
3035 const struct diff_filespec *o,
3036 const struct diff_filespec *a,
3037 int renormalize, const char *path)
3039 struct strbuf obuf = STRBUF_INIT;
3040 struct strbuf abuf = STRBUF_INIT;
3041 int ret = 0; /* assume changed for safety */
3042 struct index_state *idx = opt->repo->index;
3044 if (a->mode != o->mode)
3045 return 0;
3046 if (oideq(&o->oid, &a->oid))
3047 return 1;
3048 if (!renormalize)
3049 return 0;
3051 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3052 read_oid_strbuf(opt, &a->oid, &abuf))
3053 goto error_return;
3055 * Note: binary | is used so that both renormalizations are
3056 * performed. Comparison can be skipped if both files are
3057 * unchanged since their sha1s have already been compared.
3059 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3060 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3061 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3063 error_return:
3064 strbuf_release(&obuf);
3065 strbuf_release(&abuf);
3066 return ret;
3069 static int handle_modify_delete(struct merge_options *opt,
3070 const char *path,
3071 const struct diff_filespec *o,
3072 const struct diff_filespec *a,
3073 const struct diff_filespec *b)
3075 const char *modify_branch, *delete_branch;
3076 const struct diff_filespec *changed;
3078 if (is_valid(a)) {
3079 modify_branch = opt->branch1;
3080 delete_branch = opt->branch2;
3081 changed = a;
3082 } else {
3083 modify_branch = opt->branch2;
3084 delete_branch = opt->branch1;
3085 changed = b;
3088 return handle_change_delete(opt,
3089 path, NULL,
3090 o, changed,
3091 modify_branch, delete_branch,
3092 _("modify"), _("modified"));
3095 static int handle_content_merge(struct merge_file_info *mfi,
3096 struct merge_options *opt,
3097 const char *path,
3098 int is_dirty,
3099 const struct diff_filespec *o,
3100 const struct diff_filespec *a,
3101 const struct diff_filespec *b,
3102 struct rename_conflict_info *ci)
3104 const char *reason = _("content");
3105 unsigned df_conflict_remains = 0;
3107 if (!is_valid(o))
3108 reason = _("add/add");
3110 assert(o->path && a->path && b->path);
3111 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3112 S_ISGITLINK(ci->ren1->pair->two->mode)))
3113 df_conflict_remains = 1;
3115 if (merge_mode_and_contents(opt, o, a, b, path,
3116 opt->branch1, opt->branch2,
3117 opt->priv->call_depth * 2, mfi))
3118 return -1;
3121 * We can skip updating the working tree file iff:
3122 * a) The merge is clean
3123 * b) The merge matches what was in HEAD (content, mode, pathname)
3124 * c) The target path is usable (i.e. not involved in D/F conflict)
3126 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3127 !df_conflict_remains) {
3128 int pos;
3129 struct cache_entry *ce;
3131 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3132 if (add_cacheinfo(opt, &mfi->blob, path,
3133 0, (!opt->priv->call_depth && !is_dirty), 0))
3134 return -1;
3136 * However, add_cacheinfo() will delete the old cache entry
3137 * and add a new one. We need to copy over any skip_worktree
3138 * flag to avoid making the file appear as if it were
3139 * deleted by the user.
3141 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3142 ce = opt->priv->orig_index.cache[pos];
3143 if (ce_skip_worktree(ce)) {
3144 pos = index_name_pos(opt->repo->index, path, strlen(path));
3145 ce = opt->repo->index->cache[pos];
3146 ce->ce_flags |= CE_SKIP_WORKTREE;
3148 return mfi->clean;
3151 if (!mfi->clean) {
3152 if (S_ISGITLINK(mfi->blob.mode))
3153 reason = _("submodule");
3154 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3155 reason, path);
3156 if (ci && !df_conflict_remains)
3157 if (update_stages(opt, path, o, a, b))
3158 return -1;
3161 if (df_conflict_remains || is_dirty) {
3162 char *new_path;
3163 if (opt->priv->call_depth) {
3164 remove_file_from_index(opt->repo->index, path);
3165 } else {
3166 if (!mfi->clean) {
3167 if (update_stages(opt, path, o, a, b))
3168 return -1;
3169 } else {
3170 int file_from_stage2 = was_tracked(opt, path);
3172 if (update_stages(opt, path, NULL,
3173 file_from_stage2 ? &mfi->blob : NULL,
3174 file_from_stage2 ? NULL : &mfi->blob))
3175 return -1;
3179 new_path = unique_path(opt, path, ci->ren1->branch);
3180 if (is_dirty) {
3181 output(opt, 1, _("Refusing to lose dirty file at %s"),
3182 path);
3184 output(opt, 1, _("Adding as %s instead"), new_path);
3185 if (update_file(opt, 0, &mfi->blob, new_path)) {
3186 free(new_path);
3187 return -1;
3189 free(new_path);
3190 mfi->clean = 0;
3191 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3192 return -1;
3193 return !is_dirty && mfi->clean;
3196 static int handle_rename_normal(struct merge_options *opt,
3197 const char *path,
3198 const struct diff_filespec *o,
3199 const struct diff_filespec *a,
3200 const struct diff_filespec *b,
3201 struct rename_conflict_info *ci)
3203 struct rename *ren = ci->ren1;
3204 struct merge_file_info mfi;
3205 int clean;
3207 /* Merge the content and write it out */
3208 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3209 o, a, b, ci);
3211 if (clean &&
3212 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3213 ren->dir_rename_original_dest) {
3214 if (update_stages(opt, path,
3215 &mfi.blob, &mfi.blob, &mfi.blob))
3216 return -1;
3217 clean = 0; /* not clean, but conflicted */
3219 return clean;
3222 static void dir_rename_warning(const char *msg,
3223 int is_add,
3224 int clean,
3225 struct merge_options *opt,
3226 struct rename *ren)
3228 const char *other_branch;
3229 other_branch = (ren->branch == opt->branch1 ?
3230 opt->branch2 : opt->branch1);
3231 if (is_add) {
3232 output(opt, clean ? 2 : 1, msg,
3233 ren->pair->one->path, ren->branch,
3234 other_branch, ren->pair->two->path);
3235 return;
3237 output(opt, clean ? 2 : 1, msg,
3238 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3239 other_branch, ren->pair->two->path);
3241 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3242 struct rename *ren)
3244 const char *msg;
3245 int clean = 1, is_add;
3247 if (!ren)
3248 return clean;
3250 /* Return early if ren was not affected/created by a directory rename */
3251 if (!ren->dir_rename_original_dest)
3252 return clean;
3254 /* Sanity checks */
3255 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3256 assert(ren->dir_rename_original_type == 'A' ||
3257 ren->dir_rename_original_type == 'R');
3259 /* Check whether to treat directory renames as a conflict */
3260 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3262 is_add = (ren->dir_rename_original_type == 'A');
3263 if (ren->dir_rename_original_type == 'A' && clean) {
3264 msg = _("Path updated: %s added in %s inside a "
3265 "directory that was renamed in %s; moving it to %s.");
3266 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3267 msg = _("CONFLICT (file location): %s added in %s "
3268 "inside a directory that was renamed in %s, "
3269 "suggesting it should perhaps be moved to %s.");
3270 } else if (ren->dir_rename_original_type == 'R' && clean) {
3271 msg = _("Path updated: %s renamed to %s in %s, inside a "
3272 "directory that was renamed in %s; moving it to %s.");
3273 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3274 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3275 "inside a directory that was renamed in %s, "
3276 "suggesting it should perhaps be moved to %s.");
3277 } else {
3278 BUG("Impossible dir_rename_original_type/clean combination");
3280 dir_rename_warning(msg, is_add, clean, opt, ren);
3282 return clean;
3285 /* Per entry merge function */
3286 static int process_entry(struct merge_options *opt,
3287 const char *path, struct stage_data *entry)
3289 int clean_merge = 1;
3290 int normalize = opt->renormalize;
3292 struct diff_filespec *o = &entry->stages[1];
3293 struct diff_filespec *a = &entry->stages[2];
3294 struct diff_filespec *b = &entry->stages[3];
3295 int o_valid = is_valid(o);
3296 int a_valid = is_valid(a);
3297 int b_valid = is_valid(b);
3298 o->path = a->path = b->path = (char*)path;
3300 entry->processed = 1;
3301 if (entry->rename_conflict_info) {
3302 struct rename_conflict_info *ci = entry->rename_conflict_info;
3303 struct diff_filespec *temp;
3304 int path_clean;
3306 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3307 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3310 * For cases with a single rename, {o,a,b}->path have all been
3311 * set to the rename target path; we need to set two of these
3312 * back to the rename source.
3313 * For rename/rename conflicts, we'll manually fix paths below.
3315 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3316 o->path = temp->path = ci->ren1->pair->one->path;
3317 if (ci->ren2) {
3318 assert(opt->branch1 == ci->ren1->branch);
3321 switch (ci->rename_type) {
3322 case RENAME_NORMAL:
3323 case RENAME_ONE_FILE_TO_ONE:
3324 clean_merge = handle_rename_normal(opt, path, o, a, b,
3325 ci);
3326 break;
3327 case RENAME_VIA_DIR:
3328 clean_merge = handle_rename_via_dir(opt, ci);
3329 break;
3330 case RENAME_ADD:
3332 * Probably unclean merge, but if the renamed file
3333 * merges cleanly and the result can then be
3334 * two-way merged cleanly with the added file, I
3335 * guess it's a clean merge?
3337 clean_merge = handle_rename_add(opt, ci);
3338 break;
3339 case RENAME_DELETE:
3340 clean_merge = 0;
3341 if (handle_rename_delete(opt, ci))
3342 clean_merge = -1;
3343 break;
3344 case RENAME_ONE_FILE_TO_TWO:
3346 * Manually fix up paths; note:
3347 * ren[12]->pair->one->path are equal.
3349 o->path = ci->ren1->pair->one->path;
3350 a->path = ci->ren1->pair->two->path;
3351 b->path = ci->ren2->pair->two->path;
3353 clean_merge = 0;
3354 if (handle_rename_rename_1to2(opt, ci))
3355 clean_merge = -1;
3356 break;
3357 case RENAME_TWO_FILES_TO_ONE:
3359 * Manually fix up paths; note,
3360 * ren[12]->pair->two->path are actually equal.
3362 o->path = NULL;
3363 a->path = ci->ren1->pair->two->path;
3364 b->path = ci->ren2->pair->two->path;
3367 * Probably unclean merge, but if the two renamed
3368 * files merge cleanly and the two resulting files
3369 * can then be two-way merged cleanly, I guess it's
3370 * a clean merge?
3372 clean_merge = handle_rename_rename_2to1(opt, ci);
3373 break;
3374 default:
3375 entry->processed = 0;
3376 break;
3378 if (path_clean < clean_merge)
3379 clean_merge = path_clean;
3380 } else if (o_valid && (!a_valid || !b_valid)) {
3381 /* Case A: Deleted in one */
3382 if ((!a_valid && !b_valid) ||
3383 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3384 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3385 /* Deleted in both or deleted in one and
3386 * unchanged in the other */
3387 if (a_valid)
3388 output(opt, 2, _("Removing %s"), path);
3389 /* do not touch working file if it did not exist */
3390 remove_file(opt, 1, path, !a_valid);
3391 } else {
3392 /* Modify/delete; deleted side may have put a directory in the way */
3393 clean_merge = 0;
3394 if (handle_modify_delete(opt, path, o, a, b))
3395 clean_merge = -1;
3397 } else if ((!o_valid && a_valid && !b_valid) ||
3398 (!o_valid && !a_valid && b_valid)) {
3399 /* Case B: Added in one. */
3400 /* [nothing|directory] -> ([nothing|directory], file) */
3402 const char *add_branch;
3403 const char *other_branch;
3404 const char *conf;
3405 const struct diff_filespec *contents;
3407 if (a_valid) {
3408 add_branch = opt->branch1;
3409 other_branch = opt->branch2;
3410 contents = a;
3411 conf = _("file/directory");
3412 } else {
3413 add_branch = opt->branch2;
3414 other_branch = opt->branch1;
3415 contents = b;
3416 conf = _("directory/file");
3418 if (dir_in_way(opt->repo->index, path,
3419 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3420 0)) {
3421 char *new_path = unique_path(opt, path, add_branch);
3422 clean_merge = 0;
3423 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3424 "Adding %s as %s"),
3425 conf, path, other_branch, path, new_path);
3426 if (update_file(opt, 0, contents, new_path))
3427 clean_merge = -1;
3428 else if (opt->priv->call_depth)
3429 remove_file_from_index(opt->repo->index, path);
3430 free(new_path);
3431 } else {
3432 output(opt, 2, _("Adding %s"), path);
3433 /* do not overwrite file if already present */
3434 if (update_file_flags(opt, contents, path, 1, !a_valid))
3435 clean_merge = -1;
3437 } else if (a_valid && b_valid) {
3438 if (!o_valid) {
3439 /* Case C: Added in both (check for same permissions) */
3440 output(opt, 1,
3441 _("CONFLICT (add/add): Merge conflict in %s"),
3442 path);
3443 clean_merge = handle_file_collision(opt,
3444 path, NULL, NULL,
3445 opt->branch1,
3446 opt->branch2,
3447 a, b);
3448 } else {
3449 /* case D: Modified in both, but differently. */
3450 struct merge_file_info mfi;
3451 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3452 clean_merge = handle_content_merge(&mfi, opt, path,
3453 is_dirty,
3454 o, a, b, NULL);
3456 } else if (!o_valid && !a_valid && !b_valid) {
3458 * this entry was deleted altogether. a_mode == 0 means
3459 * we had that path and want to actively remove it.
3461 remove_file(opt, 1, path, !a->mode);
3462 } else
3463 BUG("fatal merge failure, shouldn't happen.");
3465 return clean_merge;
3468 static int merge_trees_internal(struct merge_options *opt,
3469 struct tree *head,
3470 struct tree *merge,
3471 struct tree *merge_base,
3472 struct tree **result)
3474 struct index_state *istate = opt->repo->index;
3475 int code, clean;
3477 if (opt->subtree_shift) {
3478 merge = shift_tree_object(opt->repo, head, merge,
3479 opt->subtree_shift);
3480 merge_base = shift_tree_object(opt->repo, head, merge_base,
3481 opt->subtree_shift);
3484 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3485 output(opt, 0, _("Already up to date."));
3486 *result = head;
3487 return 1;
3490 code = unpack_trees_start(opt, merge_base, head, merge);
3492 if (code != 0) {
3493 if (show(opt, 4) || opt->priv->call_depth)
3494 err(opt, _("merging of trees %s and %s failed"),
3495 oid_to_hex(&head->object.oid),
3496 oid_to_hex(&merge->object.oid));
3497 unpack_trees_finish(opt);
3498 return -1;
3501 if (unmerged_index(istate)) {
3502 struct string_list *entries;
3503 struct rename_info re_info;
3504 int i;
3506 * Only need the hashmap while processing entries, so
3507 * initialize it here and free it when we are done running
3508 * through the entries. Keeping it in the merge_options as
3509 * opposed to decaring a local hashmap is for convenience
3510 * so that we don't have to pass it to around.
3512 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3513 NULL, 512);
3514 get_files_dirs(opt, head);
3515 get_files_dirs(opt, merge);
3517 entries = get_unmerged(opt->repo->index);
3518 clean = detect_and_process_renames(opt, merge_base, head, merge,
3519 entries, &re_info);
3520 record_df_conflict_files(opt, entries);
3521 if (clean < 0)
3522 goto cleanup;
3523 for (i = entries->nr-1; 0 <= i; i--) {
3524 const char *path = entries->items[i].string;
3525 struct stage_data *e = entries->items[i].util;
3526 if (!e->processed) {
3527 int ret = process_entry(opt, path, e);
3528 if (!ret)
3529 clean = 0;
3530 else if (ret < 0) {
3531 clean = ret;
3532 goto cleanup;
3536 for (i = 0; i < entries->nr; i++) {
3537 struct stage_data *e = entries->items[i].util;
3538 if (!e->processed)
3539 BUG("unprocessed path??? %s",
3540 entries->items[i].string);
3543 cleanup:
3544 final_cleanup_renames(&re_info);
3546 string_list_clear(entries, 1);
3547 free(entries);
3549 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3550 struct path_hashmap_entry, e);
3552 if (clean < 0) {
3553 unpack_trees_finish(opt);
3554 return clean;
3557 else
3558 clean = 1;
3560 unpack_trees_finish(opt);
3562 if (opt->priv->call_depth &&
3563 !(*result = write_in_core_index_as_tree(opt->repo)))
3564 return -1;
3566 return clean;
3570 * Merge the commits h1 and h2, returning a flag (int) indicating the
3571 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3572 * virtual commit and write its location to *result.
3574 static int merge_recursive_internal(struct merge_options *opt,
3575 struct commit *h1,
3576 struct commit *h2,
3577 struct commit_list *merge_bases,
3578 struct commit **result)
3580 struct commit_list *iter;
3581 struct commit *merged_merge_bases;
3582 struct tree *result_tree;
3583 int clean;
3584 const char *ancestor_name;
3585 struct strbuf merge_base_abbrev = STRBUF_INIT;
3587 if (show(opt, 4)) {
3588 output(opt, 4, _("Merging:"));
3589 output_commit_title(opt, h1);
3590 output_commit_title(opt, h2);
3593 if (!merge_bases) {
3594 merge_bases = get_merge_bases(h1, h2);
3595 merge_bases = reverse_commit_list(merge_bases);
3598 if (show(opt, 5)) {
3599 unsigned cnt = commit_list_count(merge_bases);
3601 output(opt, 5, Q_("found %u common ancestor:",
3602 "found %u common ancestors:", cnt), cnt);
3603 for (iter = merge_bases; iter; iter = iter->next)
3604 output_commit_title(opt, iter->item);
3607 merged_merge_bases = pop_commit(&merge_bases);
3608 if (merged_merge_bases == NULL) {
3609 /* if there is no common ancestor, use an empty tree */
3610 struct tree *tree;
3612 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3613 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3614 "ancestor");
3615 ancestor_name = "empty tree";
3616 } else if (opt->ancestor && !opt->priv->call_depth) {
3617 ancestor_name = opt->ancestor;
3618 } else if (merge_bases) {
3619 ancestor_name = "merged common ancestors";
3620 } else {
3621 strbuf_add_unique_abbrev(&merge_base_abbrev,
3622 &merged_merge_bases->object.oid,
3623 DEFAULT_ABBREV);
3624 ancestor_name = merge_base_abbrev.buf;
3627 for (iter = merge_bases; iter; iter = iter->next) {
3628 const char *saved_b1, *saved_b2;
3629 opt->priv->call_depth++;
3631 * When the merge fails, the result contains files
3632 * with conflict markers. The cleanness flag is
3633 * ignored (unless indicating an error), it was never
3634 * actually used, as result of merge_trees has always
3635 * overwritten it: the committed "conflicts" were
3636 * already resolved.
3638 discard_index(opt->repo->index);
3639 saved_b1 = opt->branch1;
3640 saved_b2 = opt->branch2;
3641 opt->branch1 = "Temporary merge branch 1";
3642 opt->branch2 = "Temporary merge branch 2";
3643 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3644 NULL, &merged_merge_bases) < 0)
3645 return -1;
3646 opt->branch1 = saved_b1;
3647 opt->branch2 = saved_b2;
3648 opt->priv->call_depth--;
3650 if (!merged_merge_bases)
3651 return err(opt, _("merge returned no commit"));
3655 * FIXME: Since merge_recursive_internal() is only ever called by
3656 * places that ensure the index is loaded first
3657 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3658 * case where the merge base was unique that means when we get here
3659 * we immediately discard the index and re-read it, which is a
3660 * complete waste of time. We should only be discarding and
3661 * re-reading if we were forced to recurse.
3663 discard_index(opt->repo->index);
3664 if (!opt->priv->call_depth)
3665 repo_read_index(opt->repo);
3667 opt->ancestor = ancestor_name;
3668 clean = merge_trees_internal(opt,
3669 repo_get_commit_tree(opt->repo, h1),
3670 repo_get_commit_tree(opt->repo, h2),
3671 repo_get_commit_tree(opt->repo,
3672 merged_merge_bases),
3673 &result_tree);
3674 strbuf_release(&merge_base_abbrev);
3675 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3676 if (clean < 0) {
3677 flush_output(opt);
3678 return clean;
3681 if (opt->priv->call_depth) {
3682 *result = make_virtual_commit(opt->repo, result_tree,
3683 "merged tree");
3684 commit_list_insert(h1, &(*result)->parents);
3685 commit_list_insert(h2, &(*result)->parents->next);
3687 return clean;
3690 static int merge_start(struct merge_options *opt, struct tree *head)
3692 struct strbuf sb = STRBUF_INIT;
3694 /* Sanity checks on opt */
3695 assert(opt->repo);
3697 assert(opt->branch1 && opt->branch2);
3699 assert(opt->detect_renames >= -1 &&
3700 opt->detect_renames <= DIFF_DETECT_COPY);
3701 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3702 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3703 assert(opt->rename_limit >= -1);
3704 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3705 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3707 assert(opt->xdl_opts >= 0);
3708 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3709 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3711 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3712 assert(opt->buffer_output <= 2);
3713 assert(opt->obuf.len == 0);
3715 assert(opt->priv == NULL);
3717 /* Not supported; option specific to merge-ort */
3718 assert(!opt->record_conflict_msgs_as_headers);
3719 assert(!opt->msg_header_prefix);
3721 /* Sanity check on repo state; index must match head */
3722 if (repo_index_has_changes(opt->repo, head, &sb)) {
3723 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3724 sb.buf);
3725 strbuf_release(&sb);
3726 return -1;
3729 CALLOC_ARRAY(opt->priv, 1);
3730 string_list_init_dup(&opt->priv->df_conflict_file_set);
3731 return 0;
3734 static void merge_finalize(struct merge_options *opt)
3736 flush_output(opt);
3737 if (!opt->priv->call_depth && opt->buffer_output < 2)
3738 strbuf_release(&opt->obuf);
3739 if (show(opt, 2))
3740 diff_warn_rename_limit("merge.renamelimit",
3741 opt->priv->needed_rename_limit, 0);
3742 FREE_AND_NULL(opt->priv);
3745 int merge_trees(struct merge_options *opt,
3746 struct tree *head,
3747 struct tree *merge,
3748 struct tree *merge_base)
3750 int clean;
3751 struct tree *ignored;
3753 assert(opt->ancestor != NULL);
3755 if (merge_start(opt, head))
3756 return -1;
3757 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3758 merge_finalize(opt);
3760 return clean;
3763 int merge_recursive(struct merge_options *opt,
3764 struct commit *h1,
3765 struct commit *h2,
3766 struct commit_list *merge_bases,
3767 struct commit **result)
3769 int clean;
3771 assert(opt->ancestor == NULL ||
3772 !strcmp(opt->ancestor, "constructed merge base"));
3774 prepare_repo_settings(opt->repo);
3775 opt->repo->settings.command_requires_full_index = 1;
3777 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3778 return -1;
3779 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3780 merge_finalize(opt);
3782 return clean;
3785 static struct commit *get_ref(struct repository *repo,
3786 const struct object_id *oid,
3787 const char *name)
3789 struct object *object;
3791 object = deref_tag(repo, parse_object(repo, oid),
3792 name, strlen(name));
3793 if (!object)
3794 return NULL;
3795 if (object->type == OBJ_TREE)
3796 return make_virtual_commit(repo, (struct tree*)object, name);
3797 if (object->type != OBJ_COMMIT)
3798 return NULL;
3799 if (parse_commit((struct commit *)object))
3800 return NULL;
3801 return (struct commit *)object;
3804 int merge_recursive_generic(struct merge_options *opt,
3805 const struct object_id *head,
3806 const struct object_id *merge,
3807 int num_merge_bases,
3808 const struct object_id **merge_bases,
3809 struct commit **result)
3811 int clean;
3812 struct lock_file lock = LOCK_INIT;
3813 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3814 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3815 struct commit_list *ca = NULL;
3817 if (merge_bases) {
3818 int i;
3819 for (i = 0; i < num_merge_bases; ++i) {
3820 struct commit *base;
3821 if (!(base = get_ref(opt->repo, merge_bases[i],
3822 oid_to_hex(merge_bases[i]))))
3823 return err(opt, _("Could not parse object '%s'"),
3824 oid_to_hex(merge_bases[i]));
3825 commit_list_insert(base, &ca);
3827 if (num_merge_bases == 1)
3828 opt->ancestor = "constructed merge base";
3831 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3832 clean = merge_recursive(opt, head_commit, next_commit, ca,
3833 result);
3834 if (clean < 0) {
3835 rollback_lock_file(&lock);
3836 return clean;
3839 if (write_locked_index(opt->repo->index, &lock,
3840 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3841 return err(opt, _("Unable to write index."));
3843 return clean ? 0 : 1;
3846 static void merge_recursive_config(struct merge_options *opt)
3848 char *value = NULL;
3849 int renormalize = 0;
3850 git_config_get_int("merge.verbosity", &opt->verbosity);
3851 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3852 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3853 git_config_get_bool("merge.renormalize", &renormalize);
3854 opt->renormalize = renormalize;
3855 if (!git_config_get_string("diff.renames", &value)) {
3856 opt->detect_renames = git_config_rename("diff.renames", value);
3857 free(value);
3859 if (!git_config_get_string("merge.renames", &value)) {
3860 opt->detect_renames = git_config_rename("merge.renames", value);
3861 free(value);
3863 if (!git_config_get_string("merge.directoryrenames", &value)) {
3864 int boolval = git_parse_maybe_bool(value);
3865 if (0 <= boolval) {
3866 opt->detect_directory_renames = boolval ?
3867 MERGE_DIRECTORY_RENAMES_TRUE :
3868 MERGE_DIRECTORY_RENAMES_NONE;
3869 } else if (!strcasecmp(value, "conflict")) {
3870 opt->detect_directory_renames =
3871 MERGE_DIRECTORY_RENAMES_CONFLICT;
3872 } /* avoid erroring on values from future versions of git */
3873 free(value);
3875 git_config(git_xmerge_config, NULL);
3878 void init_merge_options(struct merge_options *opt,
3879 struct repository *repo)
3881 const char *merge_verbosity;
3882 memset(opt, 0, sizeof(struct merge_options));
3884 opt->repo = repo;
3886 opt->detect_renames = -1;
3887 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3888 opt->rename_limit = -1;
3890 opt->verbosity = 2;
3891 opt->buffer_output = 1;
3892 strbuf_init(&opt->obuf, 0);
3894 opt->renormalize = 0;
3896 merge_recursive_config(opt);
3897 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3898 if (merge_verbosity)
3899 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3900 if (opt->verbosity >= 5)
3901 opt->buffer_output = 0;
3904 int parse_merge_opt(struct merge_options *opt, const char *s)
3906 const char *arg;
3908 if (!s || !*s)
3909 return -1;
3910 if (!strcmp(s, "ours"))
3911 opt->recursive_variant = MERGE_VARIANT_OURS;
3912 else if (!strcmp(s, "theirs"))
3913 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3914 else if (!strcmp(s, "subtree"))
3915 opt->subtree_shift = "";
3916 else if (skip_prefix(s, "subtree=", &arg))
3917 opt->subtree_shift = arg;
3918 else if (!strcmp(s, "patience"))
3919 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3920 else if (!strcmp(s, "histogram"))
3921 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3922 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3923 long value = parse_algorithm_value(arg);
3924 if (value < 0)
3925 return -1;
3926 /* clear out previous settings */
3927 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3928 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3929 opt->xdl_opts |= value;
3931 else if (!strcmp(s, "ignore-space-change"))
3932 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3933 else if (!strcmp(s, "ignore-all-space"))
3934 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3935 else if (!strcmp(s, "ignore-space-at-eol"))
3936 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3937 else if (!strcmp(s, "ignore-cr-at-eol"))
3938 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3939 else if (!strcmp(s, "renormalize"))
3940 opt->renormalize = 1;
3941 else if (!strcmp(s, "no-renormalize"))
3942 opt->renormalize = 0;
3943 else if (!strcmp(s, "no-renames"))
3944 opt->detect_renames = 0;
3945 else if (!strcmp(s, "find-renames")) {
3946 opt->detect_renames = 1;
3947 opt->rename_score = 0;
3949 else if (skip_prefix(s, "find-renames=", &arg) ||
3950 skip_prefix(s, "rename-threshold=", &arg)) {
3951 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3952 return -1;
3953 opt->detect_renames = 1;
3956 * Please update $__git_merge_strategy_options in
3957 * git-completion.bash when you add new options
3959 else
3960 return -1;
3961 return 0;