Fix various issues found in comments
[git/debian.git] / merge-recursive.c
blob99a197597db5c7ab83532661027774253c3abea7
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.h"
28 #include "tag.h"
29 #include "tree-walk.h"
30 #include "unpack-trees.h"
31 #include "xdiff-interface.h"
33 struct merge_options_internal {
34 int call_depth;
35 int needed_rename_limit;
36 struct hashmap current_file_dir_set;
37 struct string_list df_conflict_file_set;
38 struct unpack_trees_options unpack_opts;
39 struct index_state orig_index;
42 struct path_hashmap_entry {
43 struct hashmap_entry e;
44 char path[FLEX_ARRAY];
47 static int path_hashmap_cmp(const void *cmp_data,
48 const struct hashmap_entry *eptr,
49 const struct hashmap_entry *entry_or_key,
50 const void *keydata)
52 const struct path_hashmap_entry *a, *b;
53 const char *key = keydata;
55 a = container_of(eptr, const struct path_hashmap_entry, e);
56 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
58 if (ignore_case)
59 return strcasecmp(a->path, key ? key : b->path);
60 else
61 return strcmp(a->path, key ? key : b->path);
64 static unsigned int path_hash(const char *path)
66 return ignore_case ? strihash(path) : strhash(path);
70 * For dir_rename_entry, directory names are stored as a full path from the
71 * toplevel of the repository and do not include a trailing '/'. Also:
73 * dir: original name of directory being renamed
74 * non_unique_new_dir: if true, could not determine new_dir
75 * new_dir: final name of directory being renamed
76 * possible_new_dirs: temporary used to help determine new_dir; see comments
77 * in get_directory_renames() for details
79 struct dir_rename_entry {
80 struct hashmap_entry ent;
81 char *dir;
82 unsigned non_unique_new_dir:1;
83 struct strbuf new_dir;
84 struct string_list possible_new_dirs;
87 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
88 char *dir)
90 struct dir_rename_entry key;
92 if (dir == NULL)
93 return NULL;
94 hashmap_entry_init(&key.ent, strhash(dir));
95 key.dir = dir;
96 return hashmap_get_entry(hashmap, &key, ent, NULL);
99 static int dir_rename_cmp(const void *unused_cmp_data,
100 const struct hashmap_entry *eptr,
101 const struct hashmap_entry *entry_or_key,
102 const void *unused_keydata)
104 const struct dir_rename_entry *e1, *e2;
106 e1 = container_of(eptr, const struct dir_rename_entry, ent);
107 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
109 return strcmp(e1->dir, e2->dir);
112 static void dir_rename_init(struct hashmap *map)
114 hashmap_init(map, dir_rename_cmp, NULL, 0);
117 static void dir_rename_entry_init(struct dir_rename_entry *entry,
118 char *directory)
120 hashmap_entry_init(&entry->ent, strhash(directory));
121 entry->dir = directory;
122 entry->non_unique_new_dir = 0;
123 strbuf_init(&entry->new_dir, 0);
124 string_list_init(&entry->possible_new_dirs, 0);
127 struct collision_entry {
128 struct hashmap_entry ent;
129 char *target_file;
130 struct string_list source_files;
131 unsigned reported_already:1;
134 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
135 char *target_file)
137 struct collision_entry key;
139 hashmap_entry_init(&key.ent, strhash(target_file));
140 key.target_file = target_file;
141 return hashmap_get_entry(hashmap, &key, ent, NULL);
144 static int collision_cmp(const void *unused_cmp_data,
145 const struct hashmap_entry *eptr,
146 const struct hashmap_entry *entry_or_key,
147 const void *unused_keydata)
149 const struct collision_entry *e1, *e2;
151 e1 = container_of(eptr, const struct collision_entry, ent);
152 e2 = container_of(entry_or_key, const struct collision_entry, ent);
154 return strcmp(e1->target_file, e2->target_file);
157 static void collision_init(struct hashmap *map)
159 hashmap_init(map, collision_cmp, NULL, 0);
162 static void flush_output(struct merge_options *opt)
164 if (opt->buffer_output < 2 && opt->obuf.len) {
165 fputs(opt->obuf.buf, stdout);
166 strbuf_reset(&opt->obuf);
170 static int err(struct merge_options *opt, const char *err, ...)
172 va_list params;
174 if (opt->buffer_output < 2)
175 flush_output(opt);
176 else {
177 strbuf_complete(&opt->obuf, '\n');
178 strbuf_addstr(&opt->obuf, "error: ");
180 va_start(params, err);
181 strbuf_vaddf(&opt->obuf, err, params);
182 va_end(params);
183 if (opt->buffer_output > 1)
184 strbuf_addch(&opt->obuf, '\n');
185 else {
186 error("%s", opt->obuf.buf);
187 strbuf_reset(&opt->obuf);
190 return -1;
193 static struct tree *shift_tree_object(struct repository *repo,
194 struct tree *one, struct tree *two,
195 const char *subtree_shift)
197 struct object_id shifted;
199 if (!*subtree_shift) {
200 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
201 } else {
202 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
203 subtree_shift);
205 if (oideq(&two->object.oid, &shifted))
206 return two;
207 return lookup_tree(repo, &shifted);
210 static inline void set_commit_tree(struct commit *c, struct tree *t)
212 c->maybe_tree = t;
215 static struct commit *make_virtual_commit(struct repository *repo,
216 struct tree *tree,
217 const char *comment)
219 struct commit *commit = alloc_commit_node(repo);
221 set_merge_remote_desc(commit, comment, (struct object *)commit);
222 set_commit_tree(commit, tree);
223 commit->object.parsed = 1;
224 return commit;
227 enum rename_type {
228 RENAME_NORMAL = 0,
229 RENAME_VIA_DIR,
230 RENAME_ADD,
231 RENAME_DELETE,
232 RENAME_ONE_FILE_TO_ONE,
233 RENAME_ONE_FILE_TO_TWO,
234 RENAME_TWO_FILES_TO_ONE
238 * Since we want to write the index eventually, we cannot reuse the index
239 * for these (temporary) data.
241 struct stage_data {
242 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
243 struct rename_conflict_info *rename_conflict_info;
244 unsigned processed:1;
247 struct rename {
248 unsigned processed:1;
249 struct diff_filepair *pair;
250 const char *branch; /* branch that the rename occurred on */
252 * If directory rename detection affected this rename, what was its
253 * original type ('A' or 'R') and it's original destination before
254 * the directory rename (otherwise, '\0' and NULL for these two vars).
256 char dir_rename_original_type;
257 char *dir_rename_original_dest;
259 * Purpose of src_entry and dst_entry:
261 * If 'before' is renamed to 'after' then src_entry will contain
262 * the versions of 'before' from the merge_base, HEAD, and MERGE in
263 * stages 1, 2, and 3; dst_entry will contain the respective
264 * versions of 'after' in corresponding locations. Thus, we have a
265 * total of six modes and oids, though some will be null. (Stage 0
266 * is ignored; we're interested in handling conflicts.)
268 * Since we don't turn on break-rewrites by default, neither
269 * src_entry nor dst_entry can have all three of their stages have
270 * non-null oids, meaning at most four of the six will be non-null.
271 * Also, since this is a rename, both src_entry and dst_entry will
272 * have at least one non-null oid, meaning at least two will be
273 * non-null. Of the six oids, a typical rename will have three be
274 * non-null. Only two implies a rename/delete, and four implies a
275 * rename/add.
277 struct stage_data *src_entry;
278 struct stage_data *dst_entry;
281 struct rename_conflict_info {
282 enum rename_type rename_type;
283 struct rename *ren1;
284 struct rename *ren2;
287 static inline void setup_rename_conflict_info(enum rename_type rename_type,
288 struct merge_options *opt,
289 struct rename *ren1,
290 struct rename *ren2)
292 struct rename_conflict_info *ci;
295 * When we have two renames involved, it's easiest to get the
296 * correct things into stage 2 and 3, and to make sure that the
297 * content merge puts HEAD before the other branch if we just
298 * ensure that branch1 == opt->branch1. So, simply flip arguments
299 * around if we don't have that.
301 if (ren2 && ren1->branch != opt->branch1) {
302 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
303 return;
306 ci = xcalloc(1, sizeof(struct rename_conflict_info));
307 ci->rename_type = rename_type;
308 ci->ren1 = ren1;
309 ci->ren2 = ren2;
311 ci->ren1->dst_entry->processed = 0;
312 ci->ren1->dst_entry->rename_conflict_info = ci;
313 if (ren2) {
314 ci->ren2->dst_entry->rename_conflict_info = ci;
318 static int show(struct merge_options *opt, int v)
320 return (!opt->priv->call_depth && opt->verbosity >= v) ||
321 opt->verbosity >= 5;
324 __attribute__((format (printf, 3, 4)))
325 static void output(struct merge_options *opt, int v, const char *fmt, ...)
327 va_list ap;
329 if (!show(opt, v))
330 return;
332 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
334 va_start(ap, fmt);
335 strbuf_vaddf(&opt->obuf, fmt, ap);
336 va_end(ap);
338 strbuf_addch(&opt->obuf, '\n');
339 if (!opt->buffer_output)
340 flush_output(opt);
343 static void output_commit_title(struct merge_options *opt, struct commit *commit)
345 struct merge_remote_desc *desc;
347 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
348 desc = merge_remote_util(commit);
349 if (desc)
350 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
351 else {
352 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
353 DEFAULT_ABBREV);
354 strbuf_addch(&opt->obuf, ' ');
355 if (parse_commit(commit) != 0)
356 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
357 else {
358 const char *title;
359 const char *msg = get_commit_buffer(commit, NULL);
360 int len = find_commit_subject(msg, &title);
361 if (len)
362 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
363 unuse_commit_buffer(commit, msg);
366 flush_output(opt);
369 static int add_cacheinfo(struct merge_options *opt,
370 const struct diff_filespec *blob,
371 const char *path, int stage, int refresh, int options)
373 struct index_state *istate = opt->repo->index;
374 struct cache_entry *ce;
375 int ret;
377 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
378 if (!ce)
379 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
381 ret = add_index_entry(istate, ce, options);
382 if (refresh) {
383 struct cache_entry *nce;
385 nce = refresh_cache_entry(istate, ce,
386 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
387 if (!nce)
388 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
389 if (nce != ce)
390 ret = add_index_entry(istate, nce, options);
392 return ret;
395 static inline int merge_detect_rename(struct merge_options *opt)
397 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
400 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
402 parse_tree(tree);
403 init_tree_desc(desc, tree->buffer, tree->size);
406 static int unpack_trees_start(struct merge_options *opt,
407 struct tree *common,
408 struct tree *head,
409 struct tree *merge)
411 int rc;
412 struct tree_desc t[3];
413 struct index_state tmp_index = { NULL };
415 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
416 if (opt->priv->call_depth)
417 opt->priv->unpack_opts.index_only = 1;
418 else
419 opt->priv->unpack_opts.update = 1;
420 opt->priv->unpack_opts.merge = 1;
421 opt->priv->unpack_opts.head_idx = 2;
422 opt->priv->unpack_opts.fn = threeway_merge;
423 opt->priv->unpack_opts.src_index = opt->repo->index;
424 opt->priv->unpack_opts.dst_index = &tmp_index;
425 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
426 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
428 init_tree_desc_from_tree(t+0, common);
429 init_tree_desc_from_tree(t+1, head);
430 init_tree_desc_from_tree(t+2, merge);
432 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
433 cache_tree_free(&opt->repo->index->cache_tree);
436 * Update opt->repo->index to match the new results, AFTER saving a
437 * copy in opt->priv->orig_index. Update src_index to point to the
438 * saved copy. (verify_uptodate() checks src_index, and the original
439 * index is the one that had the necessary modification timestamps.)
441 opt->priv->orig_index = *opt->repo->index;
442 *opt->repo->index = tmp_index;
443 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
445 return rc;
448 static void unpack_trees_finish(struct merge_options *opt)
450 discard_index(&opt->priv->orig_index);
451 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
454 static int save_files_dirs(const struct object_id *oid,
455 struct strbuf *base, const char *path,
456 unsigned int mode, int stage, void *context)
458 struct path_hashmap_entry *entry;
459 int baselen = base->len;
460 struct merge_options *opt = context;
462 strbuf_addstr(base, path);
464 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
465 hashmap_entry_init(&entry->e, path_hash(entry->path));
466 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
468 strbuf_setlen(base, baselen);
469 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
472 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
474 struct pathspec match_all;
475 memset(&match_all, 0, sizeof(match_all));
476 read_tree_recursive(opt->repo, tree, "", 0, 0,
477 &match_all, save_files_dirs, opt);
480 static int get_tree_entry_if_blob(struct repository *r,
481 const struct object_id *tree,
482 const char *path,
483 struct diff_filespec *dfs)
485 int ret;
487 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
488 if (S_ISDIR(dfs->mode)) {
489 oidcpy(&dfs->oid, &null_oid);
490 dfs->mode = 0;
492 return ret;
496 * Returns an index_entry instance which doesn't have to correspond to
497 * a real cache entry in Git's index.
499 static struct stage_data *insert_stage_data(struct repository *r,
500 const char *path,
501 struct tree *o, struct tree *a, struct tree *b,
502 struct string_list *entries)
504 struct string_list_item *item;
505 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
506 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
507 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
508 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
509 item = string_list_insert(entries, path);
510 item->util = e;
511 return e;
515 * Create a dictionary mapping file names to stage_data objects. The
516 * dictionary contains one entry for every path with a non-zero stage entry.
518 static struct string_list *get_unmerged(struct index_state *istate)
520 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
521 int i;
523 unmerged->strdup_strings = 1;
525 for (i = 0; i < istate->cache_nr; i++) {
526 struct string_list_item *item;
527 struct stage_data *e;
528 const struct cache_entry *ce = istate->cache[i];
529 if (!ce_stage(ce))
530 continue;
532 item = string_list_lookup(unmerged, ce->name);
533 if (!item) {
534 item = string_list_insert(unmerged, ce->name);
535 item->util = xcalloc(1, sizeof(struct stage_data));
537 e = item->util;
538 e->stages[ce_stage(ce)].mode = ce->ce_mode;
539 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
542 return unmerged;
545 static int string_list_df_name_compare(const char *one, const char *two)
547 int onelen = strlen(one);
548 int twolen = strlen(two);
550 * Here we only care that entries for D/F conflicts are
551 * adjacent, in particular with the file of the D/F conflict
552 * appearing before files below the corresponding directory.
553 * The order of the rest of the list is irrelevant for us.
555 * To achieve this, we sort with df_name_compare and provide
556 * the mode S_IFDIR so that D/F conflicts will sort correctly.
557 * We use the mode S_IFDIR for everything else for simplicity,
558 * since in other cases any changes in their order due to
559 * sorting cause no problems for us.
561 int cmp = df_name_compare(one, onelen, S_IFDIR,
562 two, twolen, S_IFDIR);
564 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
565 * that 'foo' comes before 'foo/bar'.
567 if (cmp)
568 return cmp;
569 return onelen - twolen;
572 static void record_df_conflict_files(struct merge_options *opt,
573 struct string_list *entries)
575 /* If there is a D/F conflict and the file for such a conflict
576 * currently exists in the working tree, we want to allow it to be
577 * removed to make room for the corresponding directory if needed.
578 * The files underneath the directories of such D/F conflicts will
579 * be processed before the corresponding file involved in the D/F
580 * conflict. If the D/F directory ends up being removed by the
581 * merge, then we won't have to touch the D/F file. If the D/F
582 * directory needs to be written to the working copy, then the D/F
583 * file will simply be removed (in make_room_for_path()) to make
584 * room for the necessary paths. Note that if both the directory
585 * and the file need to be present, then the D/F file will be
586 * reinstated with a new unique name at the time it is processed.
588 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
589 const char *last_file = NULL;
590 int last_len = 0;
591 int i;
594 * If we're merging merge-bases, we don't want to bother with
595 * any working directory changes.
597 if (opt->priv->call_depth)
598 return;
600 /* Ensure D/F conflicts are adjacent in the entries list. */
601 for (i = 0; i < entries->nr; i++) {
602 struct string_list_item *next = &entries->items[i];
603 string_list_append(&df_sorted_entries, next->string)->util =
604 next->util;
606 df_sorted_entries.cmp = string_list_df_name_compare;
607 string_list_sort(&df_sorted_entries);
609 string_list_clear(&opt->priv->df_conflict_file_set, 1);
610 for (i = 0; i < df_sorted_entries.nr; i++) {
611 const char *path = df_sorted_entries.items[i].string;
612 int len = strlen(path);
613 struct stage_data *e = df_sorted_entries.items[i].util;
616 * Check if last_file & path correspond to a D/F conflict;
617 * i.e. whether path is last_file+'/'+<something>.
618 * If so, record that it's okay to remove last_file to make
619 * room for path and friends if needed.
621 if (last_file &&
622 len > last_len &&
623 memcmp(path, last_file, last_len) == 0 &&
624 path[last_len] == '/') {
625 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
629 * Determine whether path could exist as a file in the
630 * working directory as a possible D/F conflict. This
631 * will only occur when it exists in stage 2 as a
632 * file.
634 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
635 last_file = path;
636 last_len = len;
637 } else {
638 last_file = NULL;
641 string_list_clear(&df_sorted_entries, 0);
644 static int update_stages(struct merge_options *opt, const char *path,
645 const struct diff_filespec *o,
646 const struct diff_filespec *a,
647 const struct diff_filespec *b)
651 * NOTE: It is usually a bad idea to call update_stages on a path
652 * before calling update_file on that same path, since it can
653 * sometimes lead to spurious "refusing to lose untracked file..."
654 * messages from update_file (via make_room_for path via
655 * would_lose_untracked). Instead, reverse the order of the calls
656 * (executing update_file first and then update_stages).
658 int clear = 1;
659 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
660 if (clear)
661 if (remove_file_from_index(opt->repo->index, path))
662 return -1;
663 if (o)
664 if (add_cacheinfo(opt, o, path, 1, 0, options))
665 return -1;
666 if (a)
667 if (add_cacheinfo(opt, a, path, 2, 0, options))
668 return -1;
669 if (b)
670 if (add_cacheinfo(opt, b, path, 3, 0, options))
671 return -1;
672 return 0;
675 static void update_entry(struct stage_data *entry,
676 struct diff_filespec *o,
677 struct diff_filespec *a,
678 struct diff_filespec *b)
680 entry->processed = 0;
681 entry->stages[1].mode = o->mode;
682 entry->stages[2].mode = a->mode;
683 entry->stages[3].mode = b->mode;
684 oidcpy(&entry->stages[1].oid, &o->oid);
685 oidcpy(&entry->stages[2].oid, &a->oid);
686 oidcpy(&entry->stages[3].oid, &b->oid);
689 static int remove_file(struct merge_options *opt, int clean,
690 const char *path, int no_wd)
692 int update_cache = opt->priv->call_depth || clean;
693 int update_working_directory = !opt->priv->call_depth && !no_wd;
695 if (update_cache) {
696 if (remove_file_from_index(opt->repo->index, path))
697 return -1;
699 if (update_working_directory) {
700 if (ignore_case) {
701 struct cache_entry *ce;
702 ce = index_file_exists(opt->repo->index, path, strlen(path),
703 ignore_case);
704 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
705 return 0;
707 if (remove_path(path))
708 return -1;
710 return 0;
713 /* add a string to a strbuf, but converting "/" to "_" */
714 static void add_flattened_path(struct strbuf *out, const char *s)
716 size_t i = out->len;
717 strbuf_addstr(out, s);
718 for (; i < out->len; i++)
719 if (out->buf[i] == '/')
720 out->buf[i] = '_';
723 static char *unique_path(struct merge_options *opt,
724 const char *path,
725 const char *branch)
727 struct path_hashmap_entry *entry;
728 struct strbuf newpath = STRBUF_INIT;
729 int suffix = 0;
730 size_t base_len;
732 strbuf_addf(&newpath, "%s~", path);
733 add_flattened_path(&newpath, branch);
735 base_len = newpath.len;
736 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
737 path_hash(newpath.buf), newpath.buf) ||
738 (!opt->priv->call_depth && file_exists(newpath.buf))) {
739 strbuf_setlen(&newpath, base_len);
740 strbuf_addf(&newpath, "_%d", suffix++);
743 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
744 hashmap_entry_init(&entry->e, path_hash(entry->path));
745 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
746 return strbuf_detach(&newpath, NULL);
750 * Check whether a directory in the index is in the way of an incoming
751 * file. Return 1 if so. If check_working_copy is non-zero, also
752 * check the working directory. If empty_ok is non-zero, also return
753 * 0 in the case where the working-tree dir exists but is empty.
755 static int dir_in_way(struct index_state *istate, const char *path,
756 int check_working_copy, int empty_ok)
758 int pos;
759 struct strbuf dirpath = STRBUF_INIT;
760 struct stat st;
762 strbuf_addstr(&dirpath, path);
763 strbuf_addch(&dirpath, '/');
765 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
767 if (pos < 0)
768 pos = -1 - pos;
769 if (pos < istate->cache_nr &&
770 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
771 strbuf_release(&dirpath);
772 return 1;
775 strbuf_release(&dirpath);
776 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
777 !(empty_ok && is_empty_dir(path)) &&
778 !has_symlink_leading_path(path, strlen(path));
782 * Returns whether path was tracked in the index before the merge started,
783 * and its oid and mode match the specified values
785 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
786 const struct diff_filespec *blob)
788 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
789 struct cache_entry *ce;
791 if (0 > pos)
792 /* we were not tracking this path before the merge */
793 return 0;
795 /* See if the file we were tracking before matches */
796 ce = opt->priv->orig_index.cache[pos];
797 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
801 * Returns whether path was tracked in the index before the merge started
803 static int was_tracked(struct merge_options *opt, const char *path)
805 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
807 if (0 <= pos)
808 /* we were tracking this path before the merge */
809 return 1;
811 return 0;
814 static int would_lose_untracked(struct merge_options *opt, const char *path)
816 struct index_state *istate = opt->repo->index;
819 * This may look like it can be simplified to:
820 * return !was_tracked(opt, path) && file_exists(path)
821 * but it can't. This function needs to know whether path was in
822 * the working tree due to EITHER having been tracked in the index
823 * before the merge OR having been put into the working copy and
824 * index by unpack_trees(). Due to that either-or requirement, we
825 * check the current index instead of the original one.
827 * Note that we do not need to worry about merge-recursive itself
828 * updating the index after unpack_trees() and before calling this
829 * function, because we strictly require all code paths in
830 * merge-recursive to update the working tree first and the index
831 * second. Doing otherwise would break
832 * update_file()/would_lose_untracked(); see every comment in this
833 * file which mentions "update_stages".
835 int pos = index_name_pos(istate, path, strlen(path));
837 if (pos < 0)
838 pos = -1 - pos;
839 while (pos < istate->cache_nr &&
840 !strcmp(path, istate->cache[pos]->name)) {
842 * If stage #0, it is definitely tracked.
843 * If it has stage #2 then it was tracked
844 * before this merge started. All other
845 * cases the path was not tracked.
847 switch (ce_stage(istate->cache[pos])) {
848 case 0:
849 case 2:
850 return 0;
852 pos++;
854 return file_exists(path);
857 static int was_dirty(struct merge_options *opt, const char *path)
859 struct cache_entry *ce;
860 int dirty = 1;
862 if (opt->priv->call_depth || !was_tracked(opt, path))
863 return !dirty;
865 ce = index_file_exists(opt->priv->unpack_opts.src_index,
866 path, strlen(path), ignore_case);
867 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
868 return dirty;
871 static int make_room_for_path(struct merge_options *opt, const char *path)
873 int status, i;
874 const char *msg = _("failed to create path '%s'%s");
876 /* Unlink any D/F conflict files that are in the way */
877 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
878 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
879 size_t pathlen = strlen(path);
880 size_t df_pathlen = strlen(df_path);
881 if (df_pathlen < pathlen &&
882 path[df_pathlen] == '/' &&
883 strncmp(path, df_path, df_pathlen) == 0) {
884 output(opt, 3,
885 _("Removing %s to make room for subdirectory\n"),
886 df_path);
887 unlink(df_path);
888 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
889 i, 0);
890 break;
894 /* Make sure leading directories are created */
895 status = safe_create_leading_directories_const(path);
896 if (status) {
897 if (status == SCLD_EXISTS)
898 /* something else exists */
899 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
900 return err(opt, msg, path, "");
904 * Do not unlink a file in the work tree if we are not
905 * tracking it.
907 if (would_lose_untracked(opt, path))
908 return err(opt, _("refusing to lose untracked file at '%s'"),
909 path);
911 /* Successful unlink is good.. */
912 if (!unlink(path))
913 return 0;
914 /* .. and so is no existing file */
915 if (errno == ENOENT)
916 return 0;
917 /* .. but not some other error (who really cares what?) */
918 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
921 static int update_file_flags(struct merge_options *opt,
922 const struct diff_filespec *contents,
923 const char *path,
924 int update_cache,
925 int update_wd)
927 int ret = 0;
929 if (opt->priv->call_depth)
930 update_wd = 0;
932 if (update_wd) {
933 enum object_type type;
934 void *buf;
935 unsigned long size;
937 if (S_ISGITLINK(contents->mode)) {
939 * We may later decide to recursively descend into
940 * the submodule directory and update its index
941 * and/or work tree, but we do not do that now.
943 update_wd = 0;
944 goto update_index;
947 buf = read_object_file(&contents->oid, &type, &size);
948 if (!buf) {
949 ret = err(opt, _("cannot read object %s '%s'"),
950 oid_to_hex(&contents->oid), path);
951 goto free_buf;
953 if (type != OBJ_BLOB) {
954 ret = err(opt, _("blob expected for %s '%s'"),
955 oid_to_hex(&contents->oid), path);
956 goto free_buf;
958 if (S_ISREG(contents->mode)) {
959 struct strbuf strbuf = STRBUF_INIT;
960 if (convert_to_working_tree(opt->repo->index,
961 path, buf, size, &strbuf, NULL)) {
962 free(buf);
963 size = strbuf.len;
964 buf = strbuf_detach(&strbuf, NULL);
968 if (make_room_for_path(opt, path) < 0) {
969 update_wd = 0;
970 goto free_buf;
972 if (S_ISREG(contents->mode) ||
973 (!has_symlinks && S_ISLNK(contents->mode))) {
974 int fd;
975 int mode = (contents->mode & 0100 ? 0777 : 0666);
977 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
978 if (fd < 0) {
979 ret = err(opt, _("failed to open '%s': %s"),
980 path, strerror(errno));
981 goto free_buf;
983 write_in_full(fd, buf, size);
984 close(fd);
985 } else if (S_ISLNK(contents->mode)) {
986 char *lnk = xmemdupz(buf, size);
987 safe_create_leading_directories_const(path);
988 unlink(path);
989 if (symlink(lnk, path))
990 ret = err(opt, _("failed to symlink '%s': %s"),
991 path, strerror(errno));
992 free(lnk);
993 } else
994 ret = err(opt,
995 _("do not know what to do with %06o %s '%s'"),
996 contents->mode, oid_to_hex(&contents->oid), path);
997 free_buf:
998 free(buf);
1000 update_index:
1001 if (!ret && update_cache) {
1002 int refresh = (!opt->priv->call_depth &&
1003 contents->mode != S_IFGITLINK);
1004 if (add_cacheinfo(opt, contents, path, 0, refresh,
1005 ADD_CACHE_OK_TO_ADD))
1006 return -1;
1008 return ret;
1011 static int update_file(struct merge_options *opt,
1012 int clean,
1013 const struct diff_filespec *contents,
1014 const char *path)
1016 return update_file_flags(opt, contents, path,
1017 opt->priv->call_depth || clean, !opt->priv->call_depth);
1020 /* Low level file merging, update and removal */
1022 struct merge_file_info {
1023 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1024 unsigned clean:1,
1025 merge:1;
1028 static int merge_3way(struct merge_options *opt,
1029 mmbuffer_t *result_buf,
1030 const struct diff_filespec *o,
1031 const struct diff_filespec *a,
1032 const struct diff_filespec *b,
1033 const char *branch1,
1034 const char *branch2,
1035 const int extra_marker_size)
1037 mmfile_t orig, src1, src2;
1038 struct ll_merge_options ll_opts = {0};
1039 char *base, *name1, *name2;
1040 int merge_status;
1042 ll_opts.renormalize = opt->renormalize;
1043 ll_opts.extra_marker_size = extra_marker_size;
1044 ll_opts.xdl_opts = opt->xdl_opts;
1046 if (opt->priv->call_depth) {
1047 ll_opts.virtual_ancestor = 1;
1048 ll_opts.variant = 0;
1049 } else {
1050 switch (opt->recursive_variant) {
1051 case MERGE_VARIANT_OURS:
1052 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1053 break;
1054 case MERGE_VARIANT_THEIRS:
1055 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1056 break;
1057 default:
1058 ll_opts.variant = 0;
1059 break;
1063 assert(a->path && b->path && o->path && opt->ancestor);
1064 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1065 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1066 name1 = mkpathdup("%s:%s", branch1, a->path);
1067 name2 = mkpathdup("%s:%s", branch2, b->path);
1068 } else {
1069 base = mkpathdup("%s", opt->ancestor);
1070 name1 = mkpathdup("%s", branch1);
1071 name2 = mkpathdup("%s", branch2);
1074 read_mmblob(&orig, &o->oid);
1075 read_mmblob(&src1, &a->oid);
1076 read_mmblob(&src2, &b->oid);
1079 * FIXME: Using a->path for normalization rules in ll_merge could be
1080 * wrong if we renamed from a->path to b->path. We should use the
1081 * target path for where the file will be written.
1083 merge_status = ll_merge(result_buf, a->path, &orig, base,
1084 &src1, name1, &src2, name2,
1085 opt->repo->index, &ll_opts);
1087 free(base);
1088 free(name1);
1089 free(name2);
1090 free(orig.ptr);
1091 free(src1.ptr);
1092 free(src2.ptr);
1093 return merge_status;
1096 static int find_first_merges(struct repository *repo,
1097 struct object_array *result, const char *path,
1098 struct commit *a, struct commit *b)
1100 int i, j;
1101 struct object_array merges = OBJECT_ARRAY_INIT;
1102 struct commit *commit;
1103 int contains_another;
1105 char merged_revision[GIT_MAX_HEXSZ + 2];
1106 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1107 "--all", merged_revision, NULL };
1108 struct rev_info revs;
1109 struct setup_revision_opt rev_opts;
1111 memset(result, 0, sizeof(struct object_array));
1112 memset(&rev_opts, 0, sizeof(rev_opts));
1114 /* get all revisions that merge commit a */
1115 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1116 oid_to_hex(&a->object.oid));
1117 repo_init_revisions(repo, &revs, NULL);
1118 rev_opts.submodule = path;
1119 /* FIXME: can't handle linked worktrees in submodules yet */
1120 revs.single_worktree = path != NULL;
1121 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1123 /* save all revisions from the above list that contain b */
1124 if (prepare_revision_walk(&revs))
1125 die("revision walk setup failed");
1126 while ((commit = get_revision(&revs)) != NULL) {
1127 struct object *o = &(commit->object);
1128 if (in_merge_bases(b, commit))
1129 add_object_array(o, NULL, &merges);
1131 reset_revision_walk();
1133 /* Now we've got all merges that contain a and b. Prune all
1134 * merges that contain another found merge and save them in
1135 * result.
1137 for (i = 0; i < merges.nr; i++) {
1138 struct commit *m1 = (struct commit *) merges.objects[i].item;
1140 contains_another = 0;
1141 for (j = 0; j < merges.nr; j++) {
1142 struct commit *m2 = (struct commit *) merges.objects[j].item;
1143 if (i != j && in_merge_bases(m2, m1)) {
1144 contains_another = 1;
1145 break;
1149 if (!contains_another)
1150 add_object_array(merges.objects[i].item, NULL, result);
1153 object_array_clear(&merges);
1154 return result->nr;
1157 static void print_commit(struct commit *commit)
1159 struct strbuf sb = STRBUF_INIT;
1160 struct pretty_print_context ctx = {0};
1161 ctx.date_mode.type = DATE_NORMAL;
1162 /* FIXME: Merge this with output_commit_title() */
1163 assert(!merge_remote_util(commit));
1164 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1165 fprintf(stderr, "%s\n", sb.buf);
1166 strbuf_release(&sb);
1169 static int is_valid(const struct diff_filespec *dfs)
1171 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1174 static int merge_submodule(struct merge_options *opt,
1175 struct object_id *result, const char *path,
1176 const struct object_id *base, const struct object_id *a,
1177 const struct object_id *b)
1179 struct commit *commit_base, *commit_a, *commit_b;
1180 int parent_count;
1181 struct object_array merges;
1183 int i;
1184 int search = !opt->priv->call_depth;
1186 /* store a in result in case we fail */
1187 /* FIXME: This is the WRONG resolution for the recursive case when
1188 * we need to be careful to avoid accidentally matching either side.
1189 * Should probably use o instead there, much like we do for merging
1190 * binaries.
1192 oidcpy(result, a);
1194 /* we can not handle deletion conflicts */
1195 if (is_null_oid(base))
1196 return 0;
1197 if (is_null_oid(a))
1198 return 0;
1199 if (is_null_oid(b))
1200 return 0;
1202 if (add_submodule_odb(path)) {
1203 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1204 return 0;
1207 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1208 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1209 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1210 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1211 return 0;
1214 /* check whether both changes are forward */
1215 if (!in_merge_bases(commit_base, commit_a) ||
1216 !in_merge_bases(commit_base, commit_b)) {
1217 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1218 return 0;
1221 /* Case #1: a is contained in b or vice versa */
1222 if (in_merge_bases(commit_a, commit_b)) {
1223 oidcpy(result, b);
1224 if (show(opt, 3)) {
1225 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1226 output_commit_title(opt, commit_b);
1227 } else if (show(opt, 2))
1228 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1229 else
1230 ; /* no output */
1232 return 1;
1234 if (in_merge_bases(commit_b, commit_a)) {
1235 oidcpy(result, a);
1236 if (show(opt, 3)) {
1237 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1238 output_commit_title(opt, commit_a);
1239 } else if (show(opt, 2))
1240 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1241 else
1242 ; /* no output */
1244 return 1;
1248 * Case #2: There are one or more merges that contain a and b in
1249 * the submodule. If there is only one, then present it as a
1250 * suggestion to the user, but leave it marked unmerged so the
1251 * user needs to confirm the resolution.
1254 /* Skip the search if makes no sense to the calling context. */
1255 if (!search)
1256 return 0;
1258 /* find commit which merges them */
1259 parent_count = find_first_merges(opt->repo, &merges, path,
1260 commit_a, commit_b);
1261 switch (parent_count) {
1262 case 0:
1263 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1264 break;
1266 case 1:
1267 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1268 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1269 print_commit((struct commit *) merges.objects[0].item);
1270 output(opt, 2, _(
1271 "If this is correct simply add it to the index "
1272 "for example\n"
1273 "by using:\n\n"
1274 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1275 "which will accept this suggestion.\n"),
1276 oid_to_hex(&merges.objects[0].item->oid), path);
1277 break;
1279 default:
1280 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1281 for (i = 0; i < merges.nr; i++)
1282 print_commit((struct commit *) merges.objects[i].item);
1285 object_array_clear(&merges);
1286 return 0;
1289 static int merge_mode_and_contents(struct merge_options *opt,
1290 const struct diff_filespec *o,
1291 const struct diff_filespec *a,
1292 const struct diff_filespec *b,
1293 const char *filename,
1294 const char *branch1,
1295 const char *branch2,
1296 const int extra_marker_size,
1297 struct merge_file_info *result)
1299 if (opt->branch1 != branch1) {
1301 * It's weird getting a reverse merge with HEAD on the bottom
1302 * side of the conflict markers and the other branch on the
1303 * top. Fix that.
1305 return merge_mode_and_contents(opt, o, b, a,
1306 filename,
1307 branch2, branch1,
1308 extra_marker_size, result);
1311 result->merge = 0;
1312 result->clean = 1;
1314 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1315 result->clean = 0;
1317 * FIXME: This is a bad resolution for recursive case; for
1318 * the recursive case we want something that is unlikely to
1319 * accidentally match either side. Also, while it makes
1320 * sense to prefer regular files over symlinks, it doesn't
1321 * make sense to prefer regular files over submodules.
1323 if (S_ISREG(a->mode)) {
1324 result->blob.mode = a->mode;
1325 oidcpy(&result->blob.oid, &a->oid);
1326 } else {
1327 result->blob.mode = b->mode;
1328 oidcpy(&result->blob.oid, &b->oid);
1330 } else {
1331 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1332 result->merge = 1;
1335 * Merge modes
1337 if (a->mode == b->mode || a->mode == o->mode)
1338 result->blob.mode = b->mode;
1339 else {
1340 result->blob.mode = a->mode;
1341 if (b->mode != o->mode) {
1342 result->clean = 0;
1343 result->merge = 1;
1347 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1348 oidcpy(&result->blob.oid, &b->oid);
1349 else if (oideq(&b->oid, &o->oid))
1350 oidcpy(&result->blob.oid, &a->oid);
1351 else if (S_ISREG(a->mode)) {
1352 mmbuffer_t result_buf;
1353 int ret = 0, merge_status;
1355 merge_status = merge_3way(opt, &result_buf, o, a, b,
1356 branch1, branch2,
1357 extra_marker_size);
1359 if ((merge_status < 0) || !result_buf.ptr)
1360 ret = err(opt, _("Failed to execute internal merge"));
1362 if (!ret &&
1363 write_object_file(result_buf.ptr, result_buf.size,
1364 blob_type, &result->blob.oid))
1365 ret = err(opt, _("Unable to add %s to database"),
1366 a->path);
1368 free(result_buf.ptr);
1369 if (ret)
1370 return ret;
1371 /* FIXME: bug, what if modes didn't match? */
1372 result->clean = (merge_status == 0);
1373 } else if (S_ISGITLINK(a->mode)) {
1374 result->clean = merge_submodule(opt, &result->blob.oid,
1375 o->path,
1376 &o->oid,
1377 &a->oid,
1378 &b->oid);
1379 } else if (S_ISLNK(a->mode)) {
1380 switch (opt->recursive_variant) {
1381 case MERGE_VARIANT_NORMAL:
1382 oidcpy(&result->blob.oid, &a->oid);
1383 if (!oideq(&a->oid, &b->oid))
1384 result->clean = 0;
1385 break;
1386 case MERGE_VARIANT_OURS:
1387 oidcpy(&result->blob.oid, &a->oid);
1388 break;
1389 case MERGE_VARIANT_THEIRS:
1390 oidcpy(&result->blob.oid, &b->oid);
1391 break;
1393 } else
1394 BUG("unsupported object type in the tree");
1397 if (result->merge)
1398 output(opt, 2, _("Auto-merging %s"), filename);
1400 return 0;
1403 static int handle_rename_via_dir(struct merge_options *opt,
1404 struct rename_conflict_info *ci)
1407 * Handle file adds that need to be renamed due to directory rename
1408 * detection. This differs from handle_rename_normal, because
1409 * there is no content merge to do; just move the file into the
1410 * desired final location.
1412 const struct rename *ren = ci->ren1;
1413 const struct diff_filespec *dest = ren->pair->two;
1414 char *file_path = dest->path;
1415 int mark_conflicted = (opt->detect_directory_renames ==
1416 MERGE_DIRECTORY_RENAMES_CONFLICT);
1417 assert(ren->dir_rename_original_dest);
1419 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1420 mark_conflicted = 1;
1421 file_path = unique_path(opt, dest->path, ren->branch);
1422 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1423 "writing to %s instead."),
1424 dest->path, file_path);
1427 if (mark_conflicted) {
1429 * Write the file in worktree at file_path. In the index,
1430 * only record the file at dest->path in the appropriate
1431 * higher stage.
1433 if (update_file(opt, 0, dest, file_path))
1434 return -1;
1435 if (file_path != dest->path)
1436 free(file_path);
1437 if (update_stages(opt, dest->path, NULL,
1438 ren->branch == opt->branch1 ? dest : NULL,
1439 ren->branch == opt->branch1 ? NULL : dest))
1440 return -1;
1441 return 0; /* not clean, but conflicted */
1442 } else {
1443 /* Update dest->path both in index and in worktree */
1444 if (update_file(opt, 1, dest, dest->path))
1445 return -1;
1446 return 1; /* clean */
1450 static int handle_change_delete(struct merge_options *opt,
1451 const char *path, const char *old_path,
1452 const struct diff_filespec *o,
1453 const struct diff_filespec *changed,
1454 const char *change_branch,
1455 const char *delete_branch,
1456 const char *change, const char *change_past)
1458 char *alt_path = NULL;
1459 const char *update_path = path;
1460 int ret = 0;
1462 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1463 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1464 update_path = alt_path = unique_path(opt, path, change_branch);
1467 if (opt->priv->call_depth) {
1469 * We cannot arbitrarily accept either a_sha or b_sha as
1470 * correct; since there is no true "middle point" between
1471 * them, simply reuse the base version for virtual merge base.
1473 ret = remove_file_from_index(opt->repo->index, path);
1474 if (!ret)
1475 ret = update_file(opt, 0, o, update_path);
1476 } else {
1478 * Despite the four nearly duplicate messages and argument
1479 * lists below and the ugliness of the nested if-statements,
1480 * having complete messages makes the job easier for
1481 * translators.
1483 * The slight variance among the cases is due to the fact
1484 * that:
1485 * 1) directory/file conflicts (in effect if
1486 * !alt_path) could cause us to need to write the
1487 * file to a different path.
1488 * 2) renames (in effect if !old_path) could mean that
1489 * there are two names for the path that the user
1490 * may know the file by.
1492 if (!alt_path) {
1493 if (!old_path) {
1494 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1495 "and %s in %s. Version %s of %s left in tree."),
1496 change, path, delete_branch, change_past,
1497 change_branch, change_branch, path);
1498 } else {
1499 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1500 "and %s to %s in %s. Version %s of %s left in tree."),
1501 change, old_path, delete_branch, change_past, path,
1502 change_branch, change_branch, path);
1504 } else {
1505 if (!old_path) {
1506 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1507 "and %s in %s. Version %s of %s left in tree at %s."),
1508 change, path, delete_branch, change_past,
1509 change_branch, change_branch, path, alt_path);
1510 } else {
1511 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1512 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1513 change, old_path, delete_branch, change_past, path,
1514 change_branch, change_branch, path, alt_path);
1518 * No need to call update_file() on path when change_branch ==
1519 * opt->branch1 && !alt_path, since that would needlessly touch
1520 * path. We could call update_file_flags() with update_cache=0
1521 * and update_wd=0, but that's a no-op.
1523 if (change_branch != opt->branch1 || alt_path)
1524 ret = update_file(opt, 0, changed, update_path);
1526 free(alt_path);
1528 return ret;
1531 static int handle_rename_delete(struct merge_options *opt,
1532 struct rename_conflict_info *ci)
1534 const struct rename *ren = ci->ren1;
1535 const struct diff_filespec *orig = ren->pair->one;
1536 const struct diff_filespec *dest = ren->pair->two;
1537 const char *rename_branch = ren->branch;
1538 const char *delete_branch = (opt->branch1 == ren->branch ?
1539 opt->branch2 : opt->branch1);
1541 if (handle_change_delete(opt,
1542 opt->priv->call_depth ? orig->path : dest->path,
1543 opt->priv->call_depth ? NULL : orig->path,
1544 orig, dest,
1545 rename_branch, delete_branch,
1546 _("rename"), _("renamed")))
1547 return -1;
1549 if (opt->priv->call_depth)
1550 return remove_file_from_index(opt->repo->index, dest->path);
1551 else
1552 return update_stages(opt, dest->path, NULL,
1553 rename_branch == opt->branch1 ? dest : NULL,
1554 rename_branch == opt->branch1 ? NULL : dest);
1557 static int handle_file_collision(struct merge_options *opt,
1558 const char *collide_path,
1559 const char *prev_path1,
1560 const char *prev_path2,
1561 const char *branch1, const char *branch2,
1562 struct diff_filespec *a,
1563 struct diff_filespec *b)
1565 struct merge_file_info mfi;
1566 struct diff_filespec null;
1567 char *alt_path = NULL;
1568 const char *update_path = collide_path;
1571 * It's easiest to get the correct things into stage 2 and 3, and
1572 * to make sure that the content merge puts HEAD before the other
1573 * branch if we just ensure that branch1 == opt->branch1. So, simply
1574 * flip arguments around if we don't have that.
1576 if (branch1 != opt->branch1) {
1577 return handle_file_collision(opt, collide_path,
1578 prev_path2, prev_path1,
1579 branch2, branch1,
1580 b, a);
1583 /* Remove rename sources if rename/add or rename/rename(2to1) */
1584 if (prev_path1)
1585 remove_file(opt, 1, prev_path1,
1586 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1587 if (prev_path2)
1588 remove_file(opt, 1, prev_path2,
1589 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1592 * Remove the collision path, if it wouldn't cause dirty contents
1593 * or an untracked file to get lost. We'll either overwrite with
1594 * merged contents, or just write out to differently named files.
1596 if (was_dirty(opt, collide_path)) {
1597 output(opt, 1, _("Refusing to lose dirty file at %s"),
1598 collide_path);
1599 update_path = alt_path = unique_path(opt, collide_path, "merged");
1600 } else if (would_lose_untracked(opt, collide_path)) {
1602 * Only way we get here is if both renames were from
1603 * a directory rename AND user had an untracked file
1604 * at the location where both files end up after the
1605 * two directory renames. See testcase 10d of t6043.
1607 output(opt, 1, _("Refusing to lose untracked file at "
1608 "%s, even though it's in the way."),
1609 collide_path);
1610 update_path = alt_path = unique_path(opt, collide_path, "merged");
1611 } else {
1613 * FIXME: It's possible that the two files are identical
1614 * and that the current working copy happens to match, in
1615 * which case we are unnecessarily touching the working
1616 * tree file. It's not a likely enough scenario that I
1617 * want to code up the checks for it and a better fix is
1618 * available if we restructure how unpack_trees() and
1619 * merge-recursive interoperate anyway, so punting for
1620 * now...
1622 remove_file(opt, 0, collide_path, 0);
1625 /* Store things in diff_filespecs for functions that need it */
1626 null.path = (char *)collide_path;
1627 oidcpy(&null.oid, &null_oid);
1628 null.mode = 0;
1630 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1631 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1632 return -1;
1633 mfi.clean &= !alt_path;
1634 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1635 return -1;
1636 if (!mfi.clean && !opt->priv->call_depth &&
1637 update_stages(opt, collide_path, NULL, a, b))
1638 return -1;
1639 free(alt_path);
1641 * FIXME: If both a & b both started with conflicts (only possible
1642 * if they came from a rename/rename(2to1)), but had IDENTICAL
1643 * contents including those conflicts, then in the next line we claim
1644 * it was clean. If someone cares about this case, we should have the
1645 * caller notify us if we started with conflicts.
1647 return mfi.clean;
1650 static int handle_rename_add(struct merge_options *opt,
1651 struct rename_conflict_info *ci)
1653 /* a was renamed to c, and a separate c was added. */
1654 struct diff_filespec *a = ci->ren1->pair->one;
1655 struct diff_filespec *c = ci->ren1->pair->two;
1656 char *path = c->path;
1657 char *prev_path_desc;
1658 struct merge_file_info mfi;
1660 const char *rename_branch = ci->ren1->branch;
1661 const char *add_branch = (opt->branch1 == rename_branch ?
1662 opt->branch2 : opt->branch1);
1663 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1665 output(opt, 1, _("CONFLICT (rename/add): "
1666 "Rename %s->%s in %s. Added %s in %s"),
1667 a->path, c->path, rename_branch,
1668 c->path, add_branch);
1670 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1671 ci->ren1->src_entry->stages[other_stage].path = a->path;
1672 if (merge_mode_and_contents(opt, a, c,
1673 &ci->ren1->src_entry->stages[other_stage],
1674 prev_path_desc,
1675 opt->branch1, opt->branch2,
1676 1 + opt->priv->call_depth * 2, &mfi))
1677 return -1;
1678 free(prev_path_desc);
1680 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1681 return handle_file_collision(opt,
1682 c->path, a->path, NULL,
1683 rename_branch, add_branch,
1684 &mfi.blob,
1685 &ci->ren1->dst_entry->stages[other_stage]);
1688 static char *find_path_for_conflict(struct merge_options *opt,
1689 const char *path,
1690 const char *branch1,
1691 const char *branch2)
1693 char *new_path = NULL;
1694 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1695 new_path = unique_path(opt, path, branch1);
1696 output(opt, 1, _("%s is a directory in %s adding "
1697 "as %s instead"),
1698 path, branch2, new_path);
1699 } else if (would_lose_untracked(opt, path)) {
1700 new_path = unique_path(opt, path, branch1);
1701 output(opt, 1, _("Refusing to lose untracked file"
1702 " at %s; adding as %s instead"),
1703 path, new_path);
1706 return new_path;
1710 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1712 static inline int flip_stage(int stage)
1714 return (2 + 3) - stage;
1717 static int handle_rename_rename_1to2(struct merge_options *opt,
1718 struct rename_conflict_info *ci)
1720 /* One file was renamed in both branches, but to different names. */
1721 struct merge_file_info mfi;
1722 struct diff_filespec *add;
1723 struct diff_filespec *o = ci->ren1->pair->one;
1724 struct diff_filespec *a = ci->ren1->pair->two;
1725 struct diff_filespec *b = ci->ren2->pair->two;
1726 char *path_desc;
1728 output(opt, 1, _("CONFLICT (rename/rename): "
1729 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1730 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1731 o->path, a->path, ci->ren1->branch,
1732 o->path, b->path, ci->ren2->branch,
1733 opt->priv->call_depth ? _(" (left unresolved)") : "");
1735 path_desc = xstrfmt("%s and %s, both renamed from %s",
1736 a->path, b->path, o->path);
1737 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1738 ci->ren1->branch, ci->ren2->branch,
1739 opt->priv->call_depth * 2, &mfi))
1740 return -1;
1741 free(path_desc);
1743 if (opt->priv->call_depth)
1744 remove_file_from_index(opt->repo->index, o->path);
1747 * For each destination path, we need to see if there is a
1748 * rename/add collision. If not, we can write the file out
1749 * to the specified location.
1751 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1752 if (is_valid(add)) {
1753 add->path = mfi.blob.path = a->path;
1754 if (handle_file_collision(opt, a->path,
1755 NULL, NULL,
1756 ci->ren1->branch,
1757 ci->ren2->branch,
1758 &mfi.blob, add) < 0)
1759 return -1;
1760 } else {
1761 char *new_path = find_path_for_conflict(opt, a->path,
1762 ci->ren1->branch,
1763 ci->ren2->branch);
1764 if (update_file(opt, 0, &mfi.blob,
1765 new_path ? new_path : a->path))
1766 return -1;
1767 free(new_path);
1768 if (!opt->priv->call_depth &&
1769 update_stages(opt, a->path, NULL, a, NULL))
1770 return -1;
1773 if (!mfi.clean && mfi.blob.mode == a->mode &&
1774 oideq(&mfi.blob.oid, &a->oid)) {
1776 * Getting here means we were attempting to merge a binary
1777 * blob. Since we can't merge binaries, the merge algorithm
1778 * just takes one side. But we don't want to copy the
1779 * contents of one side to both paths; we'd rather use the
1780 * original content at the given path for each path.
1782 oidcpy(&mfi.blob.oid, &b->oid);
1783 mfi.blob.mode = b->mode;
1785 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1786 if (is_valid(add)) {
1787 add->path = mfi.blob.path = b->path;
1788 if (handle_file_collision(opt, b->path,
1789 NULL, NULL,
1790 ci->ren1->branch,
1791 ci->ren2->branch,
1792 add, &mfi.blob) < 0)
1793 return -1;
1794 } else {
1795 char *new_path = find_path_for_conflict(opt, b->path,
1796 ci->ren2->branch,
1797 ci->ren1->branch);
1798 if (update_file(opt, 0, &mfi.blob,
1799 new_path ? new_path : b->path))
1800 return -1;
1801 free(new_path);
1802 if (!opt->priv->call_depth &&
1803 update_stages(opt, b->path, NULL, NULL, b))
1804 return -1;
1807 return 0;
1810 static int handle_rename_rename_2to1(struct merge_options *opt,
1811 struct rename_conflict_info *ci)
1813 /* Two files, a & b, were renamed to the same thing, c. */
1814 struct diff_filespec *a = ci->ren1->pair->one;
1815 struct diff_filespec *b = ci->ren2->pair->one;
1816 struct diff_filespec *c1 = ci->ren1->pair->two;
1817 struct diff_filespec *c2 = ci->ren2->pair->two;
1818 char *path = c1->path; /* == c2->path */
1819 char *path_side_1_desc;
1820 char *path_side_2_desc;
1821 struct merge_file_info mfi_c1;
1822 struct merge_file_info mfi_c2;
1823 int ostage1, ostage2;
1825 output(opt, 1, _("CONFLICT (rename/rename): "
1826 "Rename %s->%s in %s. "
1827 "Rename %s->%s in %s"),
1828 a->path, c1->path, ci->ren1->branch,
1829 b->path, c2->path, ci->ren2->branch);
1831 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1832 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1833 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1834 ostage2 = flip_stage(ostage1);
1835 ci->ren1->src_entry->stages[ostage1].path = a->path;
1836 ci->ren2->src_entry->stages[ostage2].path = b->path;
1837 if (merge_mode_and_contents(opt, a, c1,
1838 &ci->ren1->src_entry->stages[ostage1],
1839 path_side_1_desc,
1840 opt->branch1, opt->branch2,
1841 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1842 merge_mode_and_contents(opt, b,
1843 &ci->ren2->src_entry->stages[ostage2],
1844 c2, path_side_2_desc,
1845 opt->branch1, opt->branch2,
1846 1 + opt->priv->call_depth * 2, &mfi_c2))
1847 return -1;
1848 free(path_side_1_desc);
1849 free(path_side_2_desc);
1850 mfi_c1.blob.path = path;
1851 mfi_c2.blob.path = path;
1853 return handle_file_collision(opt, path, a->path, b->path,
1854 ci->ren1->branch, ci->ren2->branch,
1855 &mfi_c1.blob, &mfi_c2.blob);
1859 * Get the diff_filepairs changed between o_tree and tree.
1861 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1862 struct tree *o_tree,
1863 struct tree *tree)
1865 struct diff_queue_struct *ret;
1866 struct diff_options opts;
1868 repo_diff_setup(opt->repo, &opts);
1869 opts.flags.recursive = 1;
1870 opts.flags.rename_empty = 0;
1871 opts.detect_rename = merge_detect_rename(opt);
1873 * We do not have logic to handle the detection of copies. In
1874 * fact, it may not even make sense to add such logic: would we
1875 * really want a change to a base file to be propagated through
1876 * multiple other files by a merge?
1878 if (opts.detect_rename > DIFF_DETECT_RENAME)
1879 opts.detect_rename = DIFF_DETECT_RENAME;
1880 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 1000;
1881 opts.rename_score = opt->rename_score;
1882 opts.show_rename_progress = opt->show_rename_progress;
1883 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1884 diff_setup_done(&opts);
1885 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1886 diffcore_std(&opts);
1887 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1888 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1890 ret = xmalloc(sizeof(*ret));
1891 *ret = diff_queued_diff;
1893 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1894 diff_queued_diff.nr = 0;
1895 diff_queued_diff.queue = NULL;
1896 diff_flush(&opts);
1897 return ret;
1900 static int tree_has_path(struct repository *r, struct tree *tree,
1901 const char *path)
1903 struct object_id hashy;
1904 unsigned short mode_o;
1906 return !get_tree_entry(r,
1907 &tree->object.oid, path,
1908 &hashy, &mode_o);
1912 * Return a new string that replaces the beginning portion (which matches
1913 * entry->dir), with entry->new_dir. In perl-speak:
1914 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1915 * NOTE:
1916 * Caller must ensure that old_path starts with entry->dir + '/'.
1918 static char *apply_dir_rename(struct dir_rename_entry *entry,
1919 const char *old_path)
1921 struct strbuf new_path = STRBUF_INIT;
1922 int oldlen, newlen;
1924 if (entry->non_unique_new_dir)
1925 return NULL;
1927 oldlen = strlen(entry->dir);
1928 if (entry->new_dir.len == 0)
1930 * If someone renamed/merged a subdirectory into the root
1931 * directory (e.g. 'some/subdir' -> ''), then we want to
1932 * avoid returning
1933 * '' + '/filename'
1934 * as the rename; we need to make old_path + oldlen advance
1935 * past the '/' character.
1937 oldlen++;
1938 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1939 strbuf_grow(&new_path, newlen);
1940 strbuf_addbuf(&new_path, &entry->new_dir);
1941 strbuf_addstr(&new_path, &old_path[oldlen]);
1943 return strbuf_detach(&new_path, NULL);
1946 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1947 char **old_dir, char **new_dir)
1949 char *end_of_old, *end_of_new;
1951 /* Default return values: NULL, meaning no rename */
1952 *old_dir = NULL;
1953 *new_dir = NULL;
1956 * For
1957 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1958 * the "e/foo.c" part is the same, we just want to know that
1959 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1960 * so, for this example, this function returns "a/b/c/d" in
1961 * *old_dir and "a/b/some/thing/else" in *new_dir.
1965 * If the basename of the file changed, we don't care. We want
1966 * to know which portion of the directory, if any, changed.
1968 end_of_old = strrchr(old_path, '/');
1969 end_of_new = strrchr(new_path, '/');
1972 * If end_of_old is NULL, old_path wasn't in a directory, so there
1973 * could not be a directory rename (our rule elsewhere that a
1974 * directory which still exists is not considered to have been
1975 * renamed means the root directory can never be renamed -- because
1976 * the root directory always exists).
1978 if (end_of_old == NULL)
1979 return; /* Note: *old_dir and *new_dir are still NULL */
1982 * If new_path contains no directory (end_of_new is NULL), then we
1983 * have a rename of old_path's directory to the root directory.
1985 if (end_of_new == NULL) {
1986 *old_dir = xstrndup(old_path, end_of_old - old_path);
1987 *new_dir = xstrdup("");
1988 return;
1991 /* Find the first non-matching character traversing backwards */
1992 while (*--end_of_new == *--end_of_old &&
1993 end_of_old != old_path &&
1994 end_of_new != new_path)
1995 ; /* Do nothing; all in the while loop */
1998 * If both got back to the beginning of their strings, then the
1999 * directory didn't change at all, only the basename did.
2001 if (end_of_old == old_path && end_of_new == new_path &&
2002 *end_of_old == *end_of_new)
2003 return; /* Note: *old_dir and *new_dir are still NULL */
2006 * If end_of_new got back to the beginning of its string, and
2007 * end_of_old got back to the beginning of some subdirectory, then
2008 * we have a rename/merge of a subdirectory into the root, which
2009 * needs slightly special handling.
2011 * Note: There is no need to consider the opposite case, with a
2012 * rename/merge of the root directory into some subdirectory
2013 * because as noted above the root directory always exists so it
2014 * cannot be considered to be renamed.
2016 if (end_of_new == new_path &&
2017 end_of_old != old_path && end_of_old[-1] == '/') {
2018 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2019 *new_dir = xstrdup("");
2020 return;
2024 * We've found the first non-matching character in the directory
2025 * paths. That means the current characters we were looking at
2026 * were part of the first non-matching subdir name going back from
2027 * the end of the strings. Get the whole name by advancing both
2028 * end_of_old and end_of_new to the NEXT '/' character. That will
2029 * represent the entire directory rename.
2031 * The reason for the increment is cases like
2032 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2033 * After dropping the basename and going back to the first
2034 * non-matching character, we're now comparing:
2035 * a/b/s and a/b/
2036 * and we want to be comparing:
2037 * a/b/star/ and a/b/tar/
2038 * but without the pre-increment, the one on the right would stay
2039 * a/b/.
2041 end_of_old = strchr(++end_of_old, '/');
2042 end_of_new = strchr(++end_of_new, '/');
2044 /* Copy the old and new directories into *old_dir and *new_dir. */
2045 *old_dir = xstrndup(old_path, end_of_old - old_path);
2046 *new_dir = xstrndup(new_path, end_of_new - new_path);
2049 static void remove_hashmap_entries(struct hashmap *dir_renames,
2050 struct string_list *items_to_remove)
2052 int i;
2053 struct dir_rename_entry *entry;
2055 for (i = 0; i < items_to_remove->nr; i++) {
2056 entry = items_to_remove->items[i].util;
2057 hashmap_remove(dir_renames, &entry->ent, NULL);
2059 string_list_clear(items_to_remove, 0);
2063 * See if there is a directory rename for path, and if there are any file
2064 * level conflicts for the renamed location. If there is a rename and
2065 * there are no conflicts, return the new name. Otherwise, return NULL.
2067 static char *handle_path_level_conflicts(struct merge_options *opt,
2068 const char *path,
2069 struct dir_rename_entry *entry,
2070 struct hashmap *collisions,
2071 struct tree *tree)
2073 char *new_path = NULL;
2074 struct collision_entry *collision_ent;
2075 int clean = 1;
2076 struct strbuf collision_paths = STRBUF_INIT;
2079 * entry has the mapping of old directory name to new directory name
2080 * that we want to apply to path.
2082 new_path = apply_dir_rename(entry, path);
2084 if (!new_path) {
2085 /* This should only happen when entry->non_unique_new_dir set */
2086 if (!entry->non_unique_new_dir)
2087 BUG("entry->non_unqiue_dir not set and !new_path");
2088 output(opt, 1, _("CONFLICT (directory rename split): "
2089 "Unclear where to place %s because directory "
2090 "%s was renamed to multiple other directories, "
2091 "with no destination getting a majority of the "
2092 "files."),
2093 path, entry->dir);
2094 clean = 0;
2095 return NULL;
2099 * The caller needs to have ensured that it has pre-populated
2100 * collisions with all paths that map to new_path. Do a quick check
2101 * to ensure that's the case.
2103 collision_ent = collision_find_entry(collisions, new_path);
2104 if (collision_ent == NULL)
2105 BUG("collision_ent is NULL");
2108 * Check for one-sided add/add/.../add conflicts, i.e.
2109 * where implicit renames from the other side doing
2110 * directory rename(s) can affect this side of history
2111 * to put multiple paths into the same location. Warn
2112 * and bail on directory renames for such paths.
2114 if (collision_ent->reported_already) {
2115 clean = 0;
2116 } else if (tree_has_path(opt->repo, tree, new_path)) {
2117 collision_ent->reported_already = 1;
2118 strbuf_add_separated_string_list(&collision_paths, ", ",
2119 &collision_ent->source_files);
2120 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2121 "file/dir at %s in the way of implicit "
2122 "directory rename(s) putting the following "
2123 "path(s) there: %s."),
2124 new_path, collision_paths.buf);
2125 clean = 0;
2126 } else if (collision_ent->source_files.nr > 1) {
2127 collision_ent->reported_already = 1;
2128 strbuf_add_separated_string_list(&collision_paths, ", ",
2129 &collision_ent->source_files);
2130 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2131 "more than one path to %s; implicit directory "
2132 "renames tried to put these paths there: %s"),
2133 new_path, collision_paths.buf);
2134 clean = 0;
2137 /* Free memory we no longer need */
2138 strbuf_release(&collision_paths);
2139 if (!clean && new_path) {
2140 free(new_path);
2141 return NULL;
2144 return new_path;
2148 * There are a couple things we want to do at the directory level:
2149 * 1. Check for both sides renaming to the same thing, in order to avoid
2150 * implicit renaming of files that should be left in place. (See
2151 * testcase 6b in t6043 for details.)
2152 * 2. Prune directory renames if there are still files left in the
2153 * the original directory. These represent a partial directory rename,
2154 * i.e. a rename where only some of the files within the directory
2155 * were renamed elsewhere. (Technically, this could be done earlier
2156 * in get_directory_renames(), except that would prevent us from
2157 * doing the previous check and thus failing testcase 6b.)
2158 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2159 * In the future, we could potentially record this info as well and
2160 * omit reporting rename/rename(1to2) conflicts for each path within
2161 * the affected directories, thus cleaning up the merge output.
2162 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2163 * directory level, because merging directories is fine. If it
2164 * causes conflicts for files within those merged directories, then
2165 * that should be detected at the individual path level.
2167 static void handle_directory_level_conflicts(struct merge_options *opt,
2168 struct hashmap *dir_re_head,
2169 struct tree *head,
2170 struct hashmap *dir_re_merge,
2171 struct tree *merge)
2173 struct hashmap_iter iter;
2174 struct dir_rename_entry *head_ent;
2175 struct dir_rename_entry *merge_ent;
2177 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2178 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2180 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2181 ent /* member name */) {
2182 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2183 if (merge_ent &&
2184 !head_ent->non_unique_new_dir &&
2185 !merge_ent->non_unique_new_dir &&
2186 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2187 /* 1. Renamed identically; remove it from both sides */
2188 string_list_append(&remove_from_head,
2189 head_ent->dir)->util = head_ent;
2190 strbuf_release(&head_ent->new_dir);
2191 string_list_append(&remove_from_merge,
2192 merge_ent->dir)->util = merge_ent;
2193 strbuf_release(&merge_ent->new_dir);
2194 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2195 /* 2. This wasn't a directory rename after all */
2196 string_list_append(&remove_from_head,
2197 head_ent->dir)->util = head_ent;
2198 strbuf_release(&head_ent->new_dir);
2202 remove_hashmap_entries(dir_re_head, &remove_from_head);
2203 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2205 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2206 ent /* member name */) {
2207 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2208 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2209 /* 2. This wasn't a directory rename after all */
2210 string_list_append(&remove_from_merge,
2211 merge_ent->dir)->util = merge_ent;
2212 } else if (head_ent &&
2213 !head_ent->non_unique_new_dir &&
2214 !merge_ent->non_unique_new_dir) {
2215 /* 3. rename/rename(1to2) */
2217 * We can assume it's not rename/rename(1to1) because
2218 * that was case (1), already checked above. So we
2219 * know that head_ent->new_dir and merge_ent->new_dir
2220 * are different strings.
2222 output(opt, 1, _("CONFLICT (rename/rename): "
2223 "Rename directory %s->%s in %s. "
2224 "Rename directory %s->%s in %s"),
2225 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2226 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2227 string_list_append(&remove_from_head,
2228 head_ent->dir)->util = head_ent;
2229 strbuf_release(&head_ent->new_dir);
2230 string_list_append(&remove_from_merge,
2231 merge_ent->dir)->util = merge_ent;
2232 strbuf_release(&merge_ent->new_dir);
2236 remove_hashmap_entries(dir_re_head, &remove_from_head);
2237 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2240 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2242 struct hashmap *dir_renames;
2243 struct hashmap_iter iter;
2244 struct dir_rename_entry *entry;
2245 int i;
2248 * Typically, we think of a directory rename as all files from a
2249 * certain directory being moved to a target directory. However,
2250 * what if someone first moved two files from the original
2251 * directory in one commit, and then renamed the directory
2252 * somewhere else in a later commit? At merge time, we just know
2253 * that files from the original directory went to two different
2254 * places, and that the bulk of them ended up in the same place.
2255 * We want each directory rename to represent where the bulk of the
2256 * files from that directory end up; this function exists to find
2257 * where the bulk of the files went.
2259 * The first loop below simply iterates through the list of file
2260 * renames, finding out how often each directory rename pair
2261 * possibility occurs.
2263 dir_renames = xmalloc(sizeof(*dir_renames));
2264 dir_rename_init(dir_renames);
2265 for (i = 0; i < pairs->nr; ++i) {
2266 struct string_list_item *item;
2267 int *count;
2268 struct diff_filepair *pair = pairs->queue[i];
2269 char *old_dir, *new_dir;
2271 /* File not part of directory rename if it wasn't renamed */
2272 if (pair->status != 'R')
2273 continue;
2275 get_renamed_dir_portion(pair->one->path, pair->two->path,
2276 &old_dir, &new_dir);
2277 if (!old_dir)
2278 /* Directory didn't change at all; ignore this one. */
2279 continue;
2281 entry = dir_rename_find_entry(dir_renames, old_dir);
2282 if (!entry) {
2283 entry = xmalloc(sizeof(*entry));
2284 dir_rename_entry_init(entry, old_dir);
2285 hashmap_put(dir_renames, &entry->ent);
2286 } else {
2287 free(old_dir);
2289 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2290 if (!item) {
2291 item = string_list_insert(&entry->possible_new_dirs,
2292 new_dir);
2293 item->util = xcalloc(1, sizeof(int));
2294 } else {
2295 free(new_dir);
2297 count = item->util;
2298 *count += 1;
2302 * For each directory with files moved out of it, we find out which
2303 * target directory received the most files so we can declare it to
2304 * be the "winning" target location for the directory rename. This
2305 * winner gets recorded in new_dir. If there is no winner
2306 * (multiple target directories received the same number of files),
2307 * we set non_unique_new_dir. Once we've determined the winner (or
2308 * that there is no winner), we no longer need possible_new_dirs.
2310 hashmap_for_each_entry(dir_renames, &iter, entry,
2311 ent /* member name */) {
2312 int max = 0;
2313 int bad_max = 0;
2314 char *best = NULL;
2316 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2317 int *count = entry->possible_new_dirs.items[i].util;
2319 if (*count == max)
2320 bad_max = max;
2321 else if (*count > max) {
2322 max = *count;
2323 best = entry->possible_new_dirs.items[i].string;
2326 if (bad_max == max)
2327 entry->non_unique_new_dir = 1;
2328 else {
2329 assert(entry->new_dir.len == 0);
2330 strbuf_addstr(&entry->new_dir, best);
2333 * The relevant directory sub-portion of the original full
2334 * filepaths were xstrndup'ed before inserting into
2335 * possible_new_dirs, and instead of manually iterating the
2336 * list and free'ing each, just lie and tell
2337 * possible_new_dirs that it did the strdup'ing so that it
2338 * will free them for us.
2340 entry->possible_new_dirs.strdup_strings = 1;
2341 string_list_clear(&entry->possible_new_dirs, 1);
2344 return dir_renames;
2347 static struct dir_rename_entry *check_dir_renamed(const char *path,
2348 struct hashmap *dir_renames)
2350 char *temp = xstrdup(path);
2351 char *end;
2352 struct dir_rename_entry *entry = NULL;
2354 while ((end = strrchr(temp, '/'))) {
2355 *end = '\0';
2356 entry = dir_rename_find_entry(dir_renames, temp);
2357 if (entry)
2358 break;
2360 free(temp);
2361 return entry;
2364 static void compute_collisions(struct hashmap *collisions,
2365 struct hashmap *dir_renames,
2366 struct diff_queue_struct *pairs)
2368 int i;
2371 * Multiple files can be mapped to the same path due to directory
2372 * renames done by the other side of history. Since that other
2373 * side of history could have merged multiple directories into one,
2374 * if our side of history added the same file basename to each of
2375 * those directories, then all N of them would get implicitly
2376 * renamed by the directory rename detection into the same path,
2377 * and we'd get an add/add/.../add conflict, and all those adds
2378 * from *this* side of history. This is not representable in the
2379 * index, and users aren't going to easily be able to make sense of
2380 * it. So we need to provide a good warning about what's
2381 * happening, and fall back to no-directory-rename detection
2382 * behavior for those paths.
2384 * See testcases 9e and all of section 5 from t6043 for examples.
2386 collision_init(collisions);
2388 for (i = 0; i < pairs->nr; ++i) {
2389 struct dir_rename_entry *dir_rename_ent;
2390 struct collision_entry *collision_ent;
2391 char *new_path;
2392 struct diff_filepair *pair = pairs->queue[i];
2394 if (pair->status != 'A' && pair->status != 'R')
2395 continue;
2396 dir_rename_ent = check_dir_renamed(pair->two->path,
2397 dir_renames);
2398 if (!dir_rename_ent)
2399 continue;
2401 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2402 if (!new_path)
2404 * dir_rename_ent->non_unique_new_path is true, which
2405 * means there is no directory rename for us to use,
2406 * which means it won't cause us any additional
2407 * collisions.
2409 continue;
2410 collision_ent = collision_find_entry(collisions, new_path);
2411 if (!collision_ent) {
2412 collision_ent = xcalloc(1,
2413 sizeof(struct collision_entry));
2414 hashmap_entry_init(&collision_ent->ent,
2415 strhash(new_path));
2416 hashmap_put(collisions, &collision_ent->ent);
2417 collision_ent->target_file = new_path;
2418 } else {
2419 free(new_path);
2421 string_list_insert(&collision_ent->source_files,
2422 pair->two->path);
2426 static char *check_for_directory_rename(struct merge_options *opt,
2427 const char *path,
2428 struct tree *tree,
2429 struct hashmap *dir_renames,
2430 struct hashmap *dir_rename_exclusions,
2431 struct hashmap *collisions,
2432 int *clean_merge)
2434 char *new_path = NULL;
2435 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2436 struct dir_rename_entry *oentry = NULL;
2438 if (!entry)
2439 return new_path;
2442 * This next part is a little weird. We do not want to do an
2443 * implicit rename into a directory we renamed on our side, because
2444 * that will result in a spurious rename/rename(1to2) conflict. An
2445 * example:
2446 * Base commit: dumbdir/afile, otherdir/bfile
2447 * Side 1: smrtdir/afile, otherdir/bfile
2448 * Side 2: dumbdir/afile, dumbdir/bfile
2449 * Here, while working on Side 1, we could notice that otherdir was
2450 * renamed/merged to dumbdir, and change the diff_filepair for
2451 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2452 * 2 will notice the rename from dumbdir to smrtdir, and do the
2453 * transitive rename to move it from dumbdir/bfile to
2454 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2455 * smrtdir, a rename/rename(1to2) conflict. We really just want
2456 * the file to end up in smrtdir. And the way to achieve that is
2457 * to not let Side1 do the rename to dumbdir, since we know that is
2458 * the source of one of our directory renames.
2460 * That's why oentry and dir_rename_exclusions is here.
2462 * As it turns out, this also prevents N-way transient rename
2463 * confusion; See testcases 9c and 9d of t6043.
2465 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2466 if (oentry) {
2467 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2468 "to %s, because %s itself was renamed."),
2469 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2470 } else {
2471 new_path = handle_path_level_conflicts(opt, path, entry,
2472 collisions, tree);
2473 *clean_merge &= (new_path != NULL);
2476 return new_path;
2479 static void apply_directory_rename_modifications(struct merge_options *opt,
2480 struct diff_filepair *pair,
2481 char *new_path,
2482 struct rename *re,
2483 struct tree *tree,
2484 struct tree *o_tree,
2485 struct tree *a_tree,
2486 struct tree *b_tree,
2487 struct string_list *entries)
2489 struct string_list_item *item;
2490 int stage = (tree == a_tree ? 2 : 3);
2491 int update_wd;
2494 * In all cases where we can do directory rename detection,
2495 * unpack_trees() will have read pair->two->path into the
2496 * index and the working copy. We need to remove it so that
2497 * we can instead place it at new_path. It is guaranteed to
2498 * not be untracked (unpack_trees() would have errored out
2499 * saying the file would have been overwritten), but it might
2500 * be dirty, though.
2502 update_wd = !was_dirty(opt, pair->two->path);
2503 if (!update_wd)
2504 output(opt, 1, _("Refusing to lose dirty file at %s"),
2505 pair->two->path);
2506 remove_file(opt, 1, pair->two->path, !update_wd);
2508 /* Find or create a new re->dst_entry */
2509 item = string_list_lookup(entries, new_path);
2510 if (item) {
2512 * Since we're renaming on this side of history, and it's
2513 * due to a directory rename on the other side of history
2514 * (which we only allow when the directory in question no
2515 * longer exists on the other side of history), the
2516 * original entry for re->dst_entry is no longer
2517 * necessary...
2519 re->dst_entry->processed = 1;
2522 * ...because we'll be using this new one.
2524 re->dst_entry = item->util;
2525 } else {
2527 * re->dst_entry is for the before-dir-rename path, and we
2528 * need it to hold information for the after-dir-rename
2529 * path. Before creating a new entry, we need to mark the
2530 * old one as unnecessary (...unless it is shared by
2531 * src_entry, i.e. this didn't use to be a rename, in which
2532 * case we can just allow the normal processing to happen
2533 * for it).
2535 if (pair->status == 'R')
2536 re->dst_entry->processed = 1;
2538 re->dst_entry = insert_stage_data(opt->repo, new_path,
2539 o_tree, a_tree, b_tree,
2540 entries);
2541 item = string_list_insert(entries, new_path);
2542 item->util = re->dst_entry;
2546 * Update the stage_data with the information about the path we are
2547 * moving into place. That slot will be empty and available for us
2548 * to write to because of the collision checks in
2549 * handle_path_level_conflicts(). In other words,
2550 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2551 * open for us to write to.
2553 * It may be tempting to actually update the index at this point as
2554 * well, using update_stages_for_stage_data(), but as per the big
2555 * "NOTE" in update_stages(), doing so will modify the current
2556 * in-memory index which will break calls to would_lose_untracked()
2557 * that we need to make. Instead, we need to just make sure that
2558 * the various handle_rename_*() functions update the index
2559 * explicitly rather than relying on unpack_trees() to have done it.
2561 get_tree_entry(opt->repo,
2562 &tree->object.oid,
2563 pair->two->path,
2564 &re->dst_entry->stages[stage].oid,
2565 &re->dst_entry->stages[stage].mode);
2568 * Record the original change status (or 'type' of change). If it
2569 * was originally an add ('A'), this lets us differentiate later
2570 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2571 * otherwise look the same). If it was originally a rename ('R'),
2572 * this lets us remember and report accurately about the transitive
2573 * renaming that occurred via the directory rename detection. Also,
2574 * record the original destination name.
2576 re->dir_rename_original_type = pair->status;
2577 re->dir_rename_original_dest = pair->two->path;
2580 * We don't actually look at pair->status again, but it seems
2581 * pedagogically correct to adjust it.
2583 pair->status = 'R';
2586 * Finally, record the new location.
2588 pair->two->path = new_path;
2592 * Get information of all renames which occurred in 'pairs', making use of
2593 * any implicit directory renames inferred from the other side of history.
2594 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2595 * to be able to associate the correct cache entries with the rename
2596 * information; tree is always equal to either a_tree or b_tree.
2598 static struct string_list *get_renames(struct merge_options *opt,
2599 const char *branch,
2600 struct diff_queue_struct *pairs,
2601 struct hashmap *dir_renames,
2602 struct hashmap *dir_rename_exclusions,
2603 struct tree *tree,
2604 struct tree *o_tree,
2605 struct tree *a_tree,
2606 struct tree *b_tree,
2607 struct string_list *entries,
2608 int *clean_merge)
2610 int i;
2611 struct hashmap collisions;
2612 struct hashmap_iter iter;
2613 struct collision_entry *e;
2614 struct string_list *renames;
2616 compute_collisions(&collisions, dir_renames, pairs);
2617 renames = xcalloc(1, sizeof(struct string_list));
2619 for (i = 0; i < pairs->nr; ++i) {
2620 struct string_list_item *item;
2621 struct rename *re;
2622 struct diff_filepair *pair = pairs->queue[i];
2623 char *new_path; /* non-NULL only with directory renames */
2625 if (pair->status != 'A' && pair->status != 'R') {
2626 diff_free_filepair(pair);
2627 continue;
2629 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2630 dir_renames,
2631 dir_rename_exclusions,
2632 &collisions,
2633 clean_merge);
2634 if (pair->status != 'R' && !new_path) {
2635 diff_free_filepair(pair);
2636 continue;
2639 re = xmalloc(sizeof(*re));
2640 re->processed = 0;
2641 re->pair = pair;
2642 re->branch = branch;
2643 re->dir_rename_original_type = '\0';
2644 re->dir_rename_original_dest = NULL;
2645 item = string_list_lookup(entries, re->pair->one->path);
2646 if (!item)
2647 re->src_entry = insert_stage_data(opt->repo,
2648 re->pair->one->path,
2649 o_tree, a_tree, b_tree, entries);
2650 else
2651 re->src_entry = item->util;
2653 item = string_list_lookup(entries, re->pair->two->path);
2654 if (!item)
2655 re->dst_entry = insert_stage_data(opt->repo,
2656 re->pair->two->path,
2657 o_tree, a_tree, b_tree, entries);
2658 else
2659 re->dst_entry = item->util;
2660 item = string_list_insert(renames, pair->one->path);
2661 item->util = re;
2662 if (new_path)
2663 apply_directory_rename_modifications(opt, pair, new_path,
2664 re, tree, o_tree,
2665 a_tree, b_tree,
2666 entries);
2669 hashmap_for_each_entry(&collisions, &iter, e,
2670 ent /* member name */) {
2671 free(e->target_file);
2672 string_list_clear(&e->source_files, 0);
2674 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2675 return renames;
2678 static int process_renames(struct merge_options *opt,
2679 struct string_list *a_renames,
2680 struct string_list *b_renames)
2682 int clean_merge = 1, i, j;
2683 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2684 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2685 const struct rename *sre;
2688 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2689 * string_list one-by-one, but O(n log n) to build it unsorted and
2690 * then sort it. Note that as we build the list, we do not need to
2691 * check if the existing destination path is already in the list,
2692 * because the structure of diffcore_rename guarantees we won't
2693 * have duplicates.
2695 for (i = 0; i < a_renames->nr; i++) {
2696 sre = a_renames->items[i].util;
2697 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2698 = (void *)sre;
2700 for (i = 0; i < b_renames->nr; i++) {
2701 sre = b_renames->items[i].util;
2702 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2703 = (void *)sre;
2706 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2707 struct string_list *renames1, *renames2Dst;
2708 struct rename *ren1 = NULL, *ren2 = NULL;
2709 const char *ren1_src, *ren1_dst;
2710 struct string_list_item *lookup;
2712 if (i >= a_renames->nr) {
2713 ren2 = b_renames->items[j++].util;
2714 } else if (j >= b_renames->nr) {
2715 ren1 = a_renames->items[i++].util;
2716 } else {
2717 int compare = strcmp(a_renames->items[i].string,
2718 b_renames->items[j].string);
2719 if (compare <= 0)
2720 ren1 = a_renames->items[i++].util;
2721 if (compare >= 0)
2722 ren2 = b_renames->items[j++].util;
2725 /* TODO: refactor, so that 1/2 are not needed */
2726 if (ren1) {
2727 renames1 = a_renames;
2728 renames2Dst = &b_by_dst;
2729 } else {
2730 renames1 = b_renames;
2731 renames2Dst = &a_by_dst;
2732 SWAP(ren2, ren1);
2735 if (ren1->processed)
2736 continue;
2737 ren1->processed = 1;
2738 ren1->dst_entry->processed = 1;
2739 /* BUG: We should only mark src_entry as processed if we
2740 * are not dealing with a rename + add-source case.
2742 ren1->src_entry->processed = 1;
2744 ren1_src = ren1->pair->one->path;
2745 ren1_dst = ren1->pair->two->path;
2747 if (ren2) {
2748 /* One file renamed on both sides */
2749 const char *ren2_src = ren2->pair->one->path;
2750 const char *ren2_dst = ren2->pair->two->path;
2751 enum rename_type rename_type;
2752 if (strcmp(ren1_src, ren2_src) != 0)
2753 BUG("ren1_src != ren2_src");
2754 ren2->dst_entry->processed = 1;
2755 ren2->processed = 1;
2756 if (strcmp(ren1_dst, ren2_dst) != 0) {
2757 rename_type = RENAME_ONE_FILE_TO_TWO;
2758 clean_merge = 0;
2759 } else {
2760 rename_type = RENAME_ONE_FILE_TO_ONE;
2761 /* BUG: We should only remove ren1_src in
2762 * the base stage (think of rename +
2763 * add-source cases).
2765 remove_file(opt, 1, ren1_src, 1);
2766 update_entry(ren1->dst_entry,
2767 ren1->pair->one,
2768 ren1->pair->two,
2769 ren2->pair->two);
2771 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2772 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2773 /* Two different files renamed to the same thing */
2774 char *ren2_dst;
2775 ren2 = lookup->util;
2776 ren2_dst = ren2->pair->two->path;
2777 if (strcmp(ren1_dst, ren2_dst) != 0)
2778 BUG("ren1_dst != ren2_dst");
2780 clean_merge = 0;
2781 ren2->processed = 1;
2783 * BUG: We should only mark src_entry as processed
2784 * if we are not dealing with a rename + add-source
2785 * case.
2787 ren2->src_entry->processed = 1;
2789 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2790 opt, ren1, ren2);
2791 } else {
2792 /* Renamed in 1, maybe changed in 2 */
2793 /* we only use sha1 and mode of these */
2794 struct diff_filespec src_other, dst_other;
2795 int try_merge;
2798 * unpack_trees loads entries from common-commit
2799 * into stage 1, from head-commit into stage 2, and
2800 * from merge-commit into stage 3. We keep track
2801 * of which side corresponds to the rename.
2803 int renamed_stage = a_renames == renames1 ? 2 : 3;
2804 int other_stage = a_renames == renames1 ? 3 : 2;
2806 /* BUG: We should only remove ren1_src in the base
2807 * stage and in other_stage (think of rename +
2808 * add-source case).
2810 remove_file(opt, 1, ren1_src,
2811 renamed_stage == 2 || !was_tracked(opt, ren1_src));
2813 oidcpy(&src_other.oid,
2814 &ren1->src_entry->stages[other_stage].oid);
2815 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2816 oidcpy(&dst_other.oid,
2817 &ren1->dst_entry->stages[other_stage].oid);
2818 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2819 try_merge = 0;
2821 if (oideq(&src_other.oid, &null_oid) &&
2822 ren1->dir_rename_original_type == 'A') {
2823 setup_rename_conflict_info(RENAME_VIA_DIR,
2824 opt, ren1, NULL);
2825 } else if (oideq(&src_other.oid, &null_oid)) {
2826 setup_rename_conflict_info(RENAME_DELETE,
2827 opt, ren1, NULL);
2828 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2829 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2831 * Added file on the other side identical to
2832 * the file being renamed: clean merge.
2833 * Also, there is no need to overwrite the
2834 * file already in the working copy, so call
2835 * update_file_flags() instead of
2836 * update_file().
2838 if (update_file_flags(opt,
2839 ren1->pair->two,
2840 ren1_dst,
2841 1, /* update_cache */
2842 0 /* update_wd */))
2843 clean_merge = -1;
2844 } else if (!oideq(&dst_other.oid, &null_oid)) {
2846 * Probably not a clean merge, but it's
2847 * premature to set clean_merge to 0 here,
2848 * because if the rename merges cleanly and
2849 * the merge exactly matches the newly added
2850 * file, then the merge will be clean.
2852 setup_rename_conflict_info(RENAME_ADD,
2853 opt, ren1, NULL);
2854 } else
2855 try_merge = 1;
2857 if (clean_merge < 0)
2858 goto cleanup_and_return;
2859 if (try_merge) {
2860 struct diff_filespec *o, *a, *b;
2861 src_other.path = (char *)ren1_src;
2863 o = ren1->pair->one;
2864 if (a_renames == renames1) {
2865 a = ren1->pair->two;
2866 b = &src_other;
2867 } else {
2868 b = ren1->pair->two;
2869 a = &src_other;
2871 update_entry(ren1->dst_entry, o, a, b);
2872 setup_rename_conflict_info(RENAME_NORMAL,
2873 opt, ren1, NULL);
2877 cleanup_and_return:
2878 string_list_clear(&a_by_dst, 0);
2879 string_list_clear(&b_by_dst, 0);
2881 return clean_merge;
2884 struct rename_info {
2885 struct string_list *head_renames;
2886 struct string_list *merge_renames;
2889 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2890 struct hashmap *dir_renames)
2892 struct hashmap_iter iter;
2893 struct dir_rename_entry *e;
2895 hashmap_for_each_entry(dir_renames, &iter, e,
2896 ent /* member name */) {
2897 free(e->dir);
2898 strbuf_release(&e->new_dir);
2899 /* possible_new_dirs already cleared in get_directory_renames */
2901 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2902 free(dir_renames);
2904 free(pairs->queue);
2905 free(pairs);
2908 static int detect_and_process_renames(struct merge_options *opt,
2909 struct tree *common,
2910 struct tree *head,
2911 struct tree *merge,
2912 struct string_list *entries,
2913 struct rename_info *ri)
2915 struct diff_queue_struct *head_pairs, *merge_pairs;
2916 struct hashmap *dir_re_head, *dir_re_merge;
2917 int clean = 1;
2919 ri->head_renames = NULL;
2920 ri->merge_renames = NULL;
2922 if (!merge_detect_rename(opt))
2923 return 1;
2925 head_pairs = get_diffpairs(opt, common, head);
2926 merge_pairs = get_diffpairs(opt, common, merge);
2928 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2929 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2930 !opt->priv->call_depth)) {
2931 dir_re_head = get_directory_renames(head_pairs);
2932 dir_re_merge = get_directory_renames(merge_pairs);
2934 handle_directory_level_conflicts(opt,
2935 dir_re_head, head,
2936 dir_re_merge, merge);
2937 } else {
2938 dir_re_head = xmalloc(sizeof(*dir_re_head));
2939 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2940 dir_rename_init(dir_re_head);
2941 dir_rename_init(dir_re_merge);
2944 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2945 dir_re_merge, dir_re_head, head,
2946 common, head, merge, entries,
2947 &clean);
2948 if (clean < 0)
2949 goto cleanup;
2950 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2951 dir_re_head, dir_re_merge, merge,
2952 common, head, merge, entries,
2953 &clean);
2954 if (clean < 0)
2955 goto cleanup;
2956 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2958 cleanup:
2960 * Some cleanup is deferred until cleanup_renames() because the
2961 * data structures are still needed and referenced in
2962 * process_entry(). But there are a few things we can free now.
2964 initial_cleanup_rename(head_pairs, dir_re_head);
2965 initial_cleanup_rename(merge_pairs, dir_re_merge);
2967 return clean;
2970 static void final_cleanup_rename(struct string_list *rename)
2972 const struct rename *re;
2973 int i;
2975 if (rename == NULL)
2976 return;
2978 for (i = 0; i < rename->nr; i++) {
2979 re = rename->items[i].util;
2980 diff_free_filepair(re->pair);
2982 string_list_clear(rename, 1);
2983 free(rename);
2986 static void final_cleanup_renames(struct rename_info *re_info)
2988 final_cleanup_rename(re_info->head_renames);
2989 final_cleanup_rename(re_info->merge_renames);
2992 static int read_oid_strbuf(struct merge_options *opt,
2993 const struct object_id *oid,
2994 struct strbuf *dst)
2996 void *buf;
2997 enum object_type type;
2998 unsigned long size;
2999 buf = read_object_file(oid, &type, &size);
3000 if (!buf)
3001 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3002 if (type != OBJ_BLOB) {
3003 free(buf);
3004 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3006 strbuf_attach(dst, buf, size, size + 1);
3007 return 0;
3010 static int blob_unchanged(struct merge_options *opt,
3011 const struct diff_filespec *o,
3012 const struct diff_filespec *a,
3013 int renormalize, const char *path)
3015 struct strbuf obuf = STRBUF_INIT;
3016 struct strbuf abuf = STRBUF_INIT;
3017 int ret = 0; /* assume changed for safety */
3018 const struct index_state *idx = opt->repo->index;
3020 if (a->mode != o->mode)
3021 return 0;
3022 if (oideq(&o->oid, &a->oid))
3023 return 1;
3024 if (!renormalize)
3025 return 0;
3027 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3028 read_oid_strbuf(opt, &a->oid, &abuf))
3029 goto error_return;
3031 * Note: binary | is used so that both renormalizations are
3032 * performed. Comparison can be skipped if both files are
3033 * unchanged since their sha1s have already been compared.
3035 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3036 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3037 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3039 error_return:
3040 strbuf_release(&obuf);
3041 strbuf_release(&abuf);
3042 return ret;
3045 static int handle_modify_delete(struct merge_options *opt,
3046 const char *path,
3047 const struct diff_filespec *o,
3048 const struct diff_filespec *a,
3049 const struct diff_filespec *b)
3051 const char *modify_branch, *delete_branch;
3052 const struct diff_filespec *changed;
3054 if (is_valid(a)) {
3055 modify_branch = opt->branch1;
3056 delete_branch = opt->branch2;
3057 changed = a;
3058 } else {
3059 modify_branch = opt->branch2;
3060 delete_branch = opt->branch1;
3061 changed = b;
3064 return handle_change_delete(opt,
3065 path, NULL,
3066 o, changed,
3067 modify_branch, delete_branch,
3068 _("modify"), _("modified"));
3071 static int handle_content_merge(struct merge_file_info *mfi,
3072 struct merge_options *opt,
3073 const char *path,
3074 int is_dirty,
3075 const struct diff_filespec *o,
3076 const struct diff_filespec *a,
3077 const struct diff_filespec *b,
3078 struct rename_conflict_info *ci)
3080 const char *reason = _("content");
3081 unsigned df_conflict_remains = 0;
3083 if (!is_valid(o))
3084 reason = _("add/add");
3086 assert(o->path && a->path && b->path);
3087 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3088 S_ISGITLINK(ci->ren1->pair->two->mode)))
3089 df_conflict_remains = 1;
3091 if (merge_mode_and_contents(opt, o, a, b, path,
3092 opt->branch1, opt->branch2,
3093 opt->priv->call_depth * 2, mfi))
3094 return -1;
3097 * We can skip updating the working tree file iff:
3098 * a) The merge is clean
3099 * b) The merge matches what was in HEAD (content, mode, pathname)
3100 * c) The target path is usable (i.e. not involved in D/F conflict)
3102 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3103 !df_conflict_remains) {
3104 int pos;
3105 struct cache_entry *ce;
3107 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3108 if (add_cacheinfo(opt, &mfi->blob, path,
3109 0, (!opt->priv->call_depth && !is_dirty), 0))
3110 return -1;
3112 * However, add_cacheinfo() will delete the old cache entry
3113 * and add a new one. We need to copy over any skip_worktree
3114 * flag to avoid making the file appear as if it were
3115 * deleted by the user.
3117 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3118 ce = opt->priv->orig_index.cache[pos];
3119 if (ce_skip_worktree(ce)) {
3120 pos = index_name_pos(opt->repo->index, path, strlen(path));
3121 ce = opt->repo->index->cache[pos];
3122 ce->ce_flags |= CE_SKIP_WORKTREE;
3124 return mfi->clean;
3127 if (!mfi->clean) {
3128 if (S_ISGITLINK(mfi->blob.mode))
3129 reason = _("submodule");
3130 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3131 reason, path);
3132 if (ci && !df_conflict_remains)
3133 if (update_stages(opt, path, o, a, b))
3134 return -1;
3137 if (df_conflict_remains || is_dirty) {
3138 char *new_path;
3139 if (opt->priv->call_depth) {
3140 remove_file_from_index(opt->repo->index, path);
3141 } else {
3142 if (!mfi->clean) {
3143 if (update_stages(opt, path, o, a, b))
3144 return -1;
3145 } else {
3146 int file_from_stage2 = was_tracked(opt, path);
3148 if (update_stages(opt, path, NULL,
3149 file_from_stage2 ? &mfi->blob : NULL,
3150 file_from_stage2 ? NULL : &mfi->blob))
3151 return -1;
3155 new_path = unique_path(opt, path, ci->ren1->branch);
3156 if (is_dirty) {
3157 output(opt, 1, _("Refusing to lose dirty file at %s"),
3158 path);
3160 output(opt, 1, _("Adding as %s instead"), new_path);
3161 if (update_file(opt, 0, &mfi->blob, new_path)) {
3162 free(new_path);
3163 return -1;
3165 free(new_path);
3166 mfi->clean = 0;
3167 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3168 return -1;
3169 return !is_dirty && mfi->clean;
3172 static int handle_rename_normal(struct merge_options *opt,
3173 const char *path,
3174 const struct diff_filespec *o,
3175 const struct diff_filespec *a,
3176 const struct diff_filespec *b,
3177 struct rename_conflict_info *ci)
3179 struct rename *ren = ci->ren1;
3180 struct merge_file_info mfi;
3181 int clean;
3182 int side = (ren->branch == opt->branch1 ? 2 : 3);
3184 /* Merge the content and write it out */
3185 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3186 o, a, b, ci);
3188 if (clean &&
3189 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3190 ren->dir_rename_original_dest) {
3191 if (update_stages(opt, path,
3192 NULL,
3193 side == 2 ? &mfi.blob : NULL,
3194 side == 2 ? NULL : &mfi.blob))
3195 return -1;
3196 clean = 0; /* not clean, but conflicted */
3198 return clean;
3201 static void dir_rename_warning(const char *msg,
3202 int is_add,
3203 int clean,
3204 struct merge_options *opt,
3205 struct rename *ren)
3207 const char *other_branch;
3208 other_branch = (ren->branch == opt->branch1 ?
3209 opt->branch2 : opt->branch1);
3210 if (is_add) {
3211 output(opt, clean ? 2 : 1, msg,
3212 ren->pair->one->path, ren->branch,
3213 other_branch, ren->pair->two->path);
3214 return;
3216 output(opt, clean ? 2 : 1, msg,
3217 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3218 other_branch, ren->pair->two->path);
3220 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3221 struct rename *ren)
3223 const char *msg;
3224 int clean = 1, is_add;
3226 if (!ren)
3227 return clean;
3229 /* Return early if ren was not affected/created by a directory rename */
3230 if (!ren->dir_rename_original_dest)
3231 return clean;
3233 /* Sanity checks */
3234 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3235 assert(ren->dir_rename_original_type == 'A' ||
3236 ren->dir_rename_original_type == 'R');
3238 /* Check whether to treat directory renames as a conflict */
3239 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3241 is_add = (ren->dir_rename_original_type == 'A');
3242 if (ren->dir_rename_original_type == 'A' && clean) {
3243 msg = _("Path updated: %s added in %s inside a "
3244 "directory that was renamed in %s; moving it to %s.");
3245 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3246 msg = _("CONFLICT (file location): %s added in %s "
3247 "inside a directory that was renamed in %s, "
3248 "suggesting it should perhaps be moved to %s.");
3249 } else if (ren->dir_rename_original_type == 'R' && clean) {
3250 msg = _("Path updated: %s renamed to %s in %s, inside a "
3251 "directory that was renamed in %s; moving it to %s.");
3252 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3253 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3254 "inside a directory that was renamed in %s, "
3255 "suggesting it should perhaps be moved to %s.");
3256 } else {
3257 BUG("Impossible dir_rename_original_type/clean combination");
3259 dir_rename_warning(msg, is_add, clean, opt, ren);
3261 return clean;
3264 /* Per entry merge function */
3265 static int process_entry(struct merge_options *opt,
3266 const char *path, struct stage_data *entry)
3268 int clean_merge = 1;
3269 int normalize = opt->renormalize;
3271 struct diff_filespec *o = &entry->stages[1];
3272 struct diff_filespec *a = &entry->stages[2];
3273 struct diff_filespec *b = &entry->stages[3];
3274 int o_valid = is_valid(o);
3275 int a_valid = is_valid(a);
3276 int b_valid = is_valid(b);
3277 o->path = a->path = b->path = (char*)path;
3279 entry->processed = 1;
3280 if (entry->rename_conflict_info) {
3281 struct rename_conflict_info *ci = entry->rename_conflict_info;
3282 struct diff_filespec *temp;
3283 int path_clean;
3285 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3286 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3289 * For cases with a single rename, {o,a,b}->path have all been
3290 * set to the rename target path; we need to set two of these
3291 * back to the rename source.
3292 * For rename/rename conflicts, we'll manually fix paths below.
3294 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3295 o->path = temp->path = ci->ren1->pair->one->path;
3296 if (ci->ren2) {
3297 assert(opt->branch1 == ci->ren1->branch);
3300 switch (ci->rename_type) {
3301 case RENAME_NORMAL:
3302 case RENAME_ONE_FILE_TO_ONE:
3303 clean_merge = handle_rename_normal(opt, path, o, a, b,
3304 ci);
3305 break;
3306 case RENAME_VIA_DIR:
3307 clean_merge = handle_rename_via_dir(opt, ci);
3308 break;
3309 case RENAME_ADD:
3311 * Probably unclean merge, but if the renamed file
3312 * merges cleanly and the result can then be
3313 * two-way merged cleanly with the added file, I
3314 * guess it's a clean merge?
3316 clean_merge = handle_rename_add(opt, ci);
3317 break;
3318 case RENAME_DELETE:
3319 clean_merge = 0;
3320 if (handle_rename_delete(opt, ci))
3321 clean_merge = -1;
3322 break;
3323 case RENAME_ONE_FILE_TO_TWO:
3325 * Manually fix up paths; note:
3326 * ren[12]->pair->one->path are equal.
3328 o->path = ci->ren1->pair->one->path;
3329 a->path = ci->ren1->pair->two->path;
3330 b->path = ci->ren2->pair->two->path;
3332 clean_merge = 0;
3333 if (handle_rename_rename_1to2(opt, ci))
3334 clean_merge = -1;
3335 break;
3336 case RENAME_TWO_FILES_TO_ONE:
3338 * Manually fix up paths; note,
3339 * ren[12]->pair->two->path are actually equal.
3341 o->path = NULL;
3342 a->path = ci->ren1->pair->two->path;
3343 b->path = ci->ren2->pair->two->path;
3346 * Probably unclean merge, but if the two renamed
3347 * files merge cleanly and the two resulting files
3348 * can then be two-way merged cleanly, I guess it's
3349 * a clean merge?
3351 clean_merge = handle_rename_rename_2to1(opt, ci);
3352 break;
3353 default:
3354 entry->processed = 0;
3355 break;
3357 if (path_clean < clean_merge)
3358 clean_merge = path_clean;
3359 } else if (o_valid && (!a_valid || !b_valid)) {
3360 /* Case A: Deleted in one */
3361 if ((!a_valid && !b_valid) ||
3362 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3363 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3364 /* Deleted in both or deleted in one and
3365 * unchanged in the other */
3366 if (a_valid)
3367 output(opt, 2, _("Removing %s"), path);
3368 /* do not touch working file if it did not exist */
3369 remove_file(opt, 1, path, !a_valid);
3370 } else {
3371 /* Modify/delete; deleted side may have put a directory in the way */
3372 clean_merge = 0;
3373 if (handle_modify_delete(opt, path, o, a, b))
3374 clean_merge = -1;
3376 } else if ((!o_valid && a_valid && !b_valid) ||
3377 (!o_valid && !a_valid && b_valid)) {
3378 /* Case B: Added in one. */
3379 /* [nothing|directory] -> ([nothing|directory], file) */
3381 const char *add_branch;
3382 const char *other_branch;
3383 const char *conf;
3384 const struct diff_filespec *contents;
3386 if (a_valid) {
3387 add_branch = opt->branch1;
3388 other_branch = opt->branch2;
3389 contents = a;
3390 conf = _("file/directory");
3391 } else {
3392 add_branch = opt->branch2;
3393 other_branch = opt->branch1;
3394 contents = b;
3395 conf = _("directory/file");
3397 if (dir_in_way(opt->repo->index, path,
3398 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3399 0)) {
3400 char *new_path = unique_path(opt, path, add_branch);
3401 clean_merge = 0;
3402 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3403 "Adding %s as %s"),
3404 conf, path, other_branch, path, new_path);
3405 if (update_file(opt, 0, contents, new_path))
3406 clean_merge = -1;
3407 else if (opt->priv->call_depth)
3408 remove_file_from_index(opt->repo->index, path);
3409 free(new_path);
3410 } else {
3411 output(opt, 2, _("Adding %s"), path);
3412 /* do not overwrite file if already present */
3413 if (update_file_flags(opt, contents, path, 1, !a_valid))
3414 clean_merge = -1;
3416 } else if (a_valid && b_valid) {
3417 if (!o_valid) {
3418 /* Case C: Added in both (check for same permissions) */
3419 output(opt, 1,
3420 _("CONFLICT (add/add): Merge conflict in %s"),
3421 path);
3422 clean_merge = handle_file_collision(opt,
3423 path, NULL, NULL,
3424 opt->branch1,
3425 opt->branch2,
3426 a, b);
3427 } else {
3428 /* case D: Modified in both, but differently. */
3429 struct merge_file_info mfi;
3430 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3431 clean_merge = handle_content_merge(&mfi, opt, path,
3432 is_dirty,
3433 o, a, b, NULL);
3435 } else if (!o_valid && !a_valid && !b_valid) {
3437 * this entry was deleted altogether. a_mode == 0 means
3438 * we had that path and want to actively remove it.
3440 remove_file(opt, 1, path, !a->mode);
3441 } else
3442 BUG("fatal merge failure, shouldn't happen.");
3444 return clean_merge;
3447 static int merge_trees_internal(struct merge_options *opt,
3448 struct tree *head,
3449 struct tree *merge,
3450 struct tree *merge_base,
3451 struct tree **result)
3453 struct index_state *istate = opt->repo->index;
3454 int code, clean;
3456 if (opt->subtree_shift) {
3457 merge = shift_tree_object(opt->repo, head, merge,
3458 opt->subtree_shift);
3459 merge_base = shift_tree_object(opt->repo, head, merge_base,
3460 opt->subtree_shift);
3463 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3464 output(opt, 0, _("Already up to date!"));
3465 *result = head;
3466 return 1;
3469 code = unpack_trees_start(opt, merge_base, head, merge);
3471 if (code != 0) {
3472 if (show(opt, 4) || opt->priv->call_depth)
3473 err(opt, _("merging of trees %s and %s failed"),
3474 oid_to_hex(&head->object.oid),
3475 oid_to_hex(&merge->object.oid));
3476 unpack_trees_finish(opt);
3477 return -1;
3480 if (unmerged_index(istate)) {
3481 struct string_list *entries;
3482 struct rename_info re_info;
3483 int i;
3485 * Only need the hashmap while processing entries, so
3486 * initialize it here and free it when we are done running
3487 * through the entries. Keeping it in the merge_options as
3488 * opposed to decaring a local hashmap is for convenience
3489 * so that we don't have to pass it to around.
3491 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3492 NULL, 512);
3493 get_files_dirs(opt, head);
3494 get_files_dirs(opt, merge);
3496 entries = get_unmerged(opt->repo->index);
3497 clean = detect_and_process_renames(opt, merge_base, head, merge,
3498 entries, &re_info);
3499 record_df_conflict_files(opt, entries);
3500 if (clean < 0)
3501 goto cleanup;
3502 for (i = entries->nr-1; 0 <= i; i--) {
3503 const char *path = entries->items[i].string;
3504 struct stage_data *e = entries->items[i].util;
3505 if (!e->processed) {
3506 int ret = process_entry(opt, path, e);
3507 if (!ret)
3508 clean = 0;
3509 else if (ret < 0) {
3510 clean = ret;
3511 goto cleanup;
3515 for (i = 0; i < entries->nr; i++) {
3516 struct stage_data *e = entries->items[i].util;
3517 if (!e->processed)
3518 BUG("unprocessed path??? %s",
3519 entries->items[i].string);
3522 cleanup:
3523 final_cleanup_renames(&re_info);
3525 string_list_clear(entries, 1);
3526 free(entries);
3528 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3529 struct path_hashmap_entry, e);
3531 if (clean < 0) {
3532 unpack_trees_finish(opt);
3533 return clean;
3536 else
3537 clean = 1;
3539 unpack_trees_finish(opt);
3541 if (opt->priv->call_depth &&
3542 !(*result = write_in_core_index_as_tree(opt->repo)))
3543 return -1;
3545 return clean;
3549 * Merge the commits h1 and h2, returning a flag (int) indicating the
3550 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3551 * virtual commit and write its location to *result.
3553 static int merge_recursive_internal(struct merge_options *opt,
3554 struct commit *h1,
3555 struct commit *h2,
3556 struct commit_list *merge_bases,
3557 struct commit **result)
3559 struct commit_list *iter;
3560 struct commit *merged_merge_bases;
3561 struct tree *result_tree;
3562 int clean;
3563 const char *ancestor_name;
3564 struct strbuf merge_base_abbrev = STRBUF_INIT;
3566 if (show(opt, 4)) {
3567 output(opt, 4, _("Merging:"));
3568 output_commit_title(opt, h1);
3569 output_commit_title(opt, h2);
3572 if (!merge_bases) {
3573 merge_bases = get_merge_bases(h1, h2);
3574 merge_bases = reverse_commit_list(merge_bases);
3577 if (show(opt, 5)) {
3578 unsigned cnt = commit_list_count(merge_bases);
3580 output(opt, 5, Q_("found %u common ancestor:",
3581 "found %u common ancestors:", cnt), cnt);
3582 for (iter = merge_bases; iter; iter = iter->next)
3583 output_commit_title(opt, iter->item);
3586 merged_merge_bases = pop_commit(&merge_bases);
3587 if (merged_merge_bases == NULL) {
3588 /* if there is no common ancestor, use an empty tree */
3589 struct tree *tree;
3591 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3592 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3593 "ancestor");
3594 ancestor_name = "empty tree";
3595 } else if (opt->ancestor && !opt->priv->call_depth) {
3596 ancestor_name = opt->ancestor;
3597 } else if (merge_bases) {
3598 ancestor_name = "merged common ancestors";
3599 } else {
3600 strbuf_add_unique_abbrev(&merge_base_abbrev,
3601 &merged_merge_bases->object.oid,
3602 DEFAULT_ABBREV);
3603 ancestor_name = merge_base_abbrev.buf;
3606 for (iter = merge_bases; iter; iter = iter->next) {
3607 const char *saved_b1, *saved_b2;
3608 opt->priv->call_depth++;
3610 * When the merge fails, the result contains files
3611 * with conflict markers. The cleanness flag is
3612 * ignored (unless indicating an error), it was never
3613 * actually used, as result of merge_trees has always
3614 * overwritten it: the committed "conflicts" were
3615 * already resolved.
3617 discard_index(opt->repo->index);
3618 saved_b1 = opt->branch1;
3619 saved_b2 = opt->branch2;
3620 opt->branch1 = "Temporary merge branch 1";
3621 opt->branch2 = "Temporary merge branch 2";
3622 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3623 NULL, &merged_merge_bases) < 0)
3624 return -1;
3625 opt->branch1 = saved_b1;
3626 opt->branch2 = saved_b2;
3627 opt->priv->call_depth--;
3629 if (!merged_merge_bases)
3630 return err(opt, _("merge returned no commit"));
3634 * FIXME: Since merge_recursive_internal() is only ever called by
3635 * places that ensure the index is loaded first
3636 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3637 * case where the merge base was unique that means when we get here
3638 * we immediately discard the index and re-read it, which is a
3639 * complete waste of time. We should only be discarding and
3640 * re-reading if we were forced to recurse.
3642 discard_index(opt->repo->index);
3643 if (!opt->priv->call_depth)
3644 repo_read_index(opt->repo);
3646 opt->ancestor = ancestor_name;
3647 clean = merge_trees_internal(opt,
3648 repo_get_commit_tree(opt->repo, h1),
3649 repo_get_commit_tree(opt->repo, h2),
3650 repo_get_commit_tree(opt->repo,
3651 merged_merge_bases),
3652 &result_tree);
3653 strbuf_release(&merge_base_abbrev);
3654 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3655 if (clean < 0) {
3656 flush_output(opt);
3657 return clean;
3660 if (opt->priv->call_depth) {
3661 *result = make_virtual_commit(opt->repo, result_tree,
3662 "merged tree");
3663 commit_list_insert(h1, &(*result)->parents);
3664 commit_list_insert(h2, &(*result)->parents->next);
3666 return clean;
3669 static int merge_start(struct merge_options *opt, struct tree *head)
3671 struct strbuf sb = STRBUF_INIT;
3673 /* Sanity checks on opt */
3674 assert(opt->repo);
3676 assert(opt->branch1 && opt->branch2);
3678 assert(opt->detect_renames >= -1 &&
3679 opt->detect_renames <= DIFF_DETECT_COPY);
3680 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3681 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3682 assert(opt->rename_limit >= -1);
3683 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3684 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3686 assert(opt->xdl_opts >= 0);
3687 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3688 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3690 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3691 assert(opt->buffer_output <= 2);
3692 assert(opt->obuf.len == 0);
3694 assert(opt->priv == NULL);
3696 /* Sanity check on repo state; index must match head */
3697 if (repo_index_has_changes(opt->repo, head, &sb)) {
3698 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3699 sb.buf);
3700 strbuf_release(&sb);
3701 return -1;
3704 opt->priv = xcalloc(1, sizeof(*opt->priv));
3705 string_list_init(&opt->priv->df_conflict_file_set, 1);
3706 return 0;
3709 static void merge_finalize(struct merge_options *opt)
3711 flush_output(opt);
3712 if (!opt->priv->call_depth && opt->buffer_output < 2)
3713 strbuf_release(&opt->obuf);
3714 if (show(opt, 2))
3715 diff_warn_rename_limit("merge.renamelimit",
3716 opt->priv->needed_rename_limit, 0);
3717 FREE_AND_NULL(opt->priv);
3720 int merge_trees(struct merge_options *opt,
3721 struct tree *head,
3722 struct tree *merge,
3723 struct tree *merge_base)
3725 int clean;
3726 struct tree *ignored;
3728 assert(opt->ancestor != NULL);
3730 if (merge_start(opt, head))
3731 return -1;
3732 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3733 merge_finalize(opt);
3735 return clean;
3738 int merge_recursive(struct merge_options *opt,
3739 struct commit *h1,
3740 struct commit *h2,
3741 struct commit_list *merge_bases,
3742 struct commit **result)
3744 int clean;
3746 assert(opt->ancestor == NULL ||
3747 !strcmp(opt->ancestor, "constructed merge base"));
3749 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3750 return -1;
3751 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3752 merge_finalize(opt);
3754 return clean;
3757 static struct commit *get_ref(struct repository *repo,
3758 const struct object_id *oid,
3759 const char *name)
3761 struct object *object;
3763 object = deref_tag(repo, parse_object(repo, oid),
3764 name, strlen(name));
3765 if (!object)
3766 return NULL;
3767 if (object->type == OBJ_TREE)
3768 return make_virtual_commit(repo, (struct tree*)object, name);
3769 if (object->type != OBJ_COMMIT)
3770 return NULL;
3771 if (parse_commit((struct commit *)object))
3772 return NULL;
3773 return (struct commit *)object;
3776 int merge_recursive_generic(struct merge_options *opt,
3777 const struct object_id *head,
3778 const struct object_id *merge,
3779 int num_merge_bases,
3780 const struct object_id **merge_bases,
3781 struct commit **result)
3783 int clean;
3784 struct lock_file lock = LOCK_INIT;
3785 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3786 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3787 struct commit_list *ca = NULL;
3789 if (merge_bases) {
3790 int i;
3791 for (i = 0; i < num_merge_bases; ++i) {
3792 struct commit *base;
3793 if (!(base = get_ref(opt->repo, merge_bases[i],
3794 oid_to_hex(merge_bases[i]))))
3795 return err(opt, _("Could not parse object '%s'"),
3796 oid_to_hex(merge_bases[i]));
3797 commit_list_insert(base, &ca);
3799 if (num_merge_bases == 1)
3800 opt->ancestor = "constructed merge base";
3803 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3804 clean = merge_recursive(opt, head_commit, next_commit, ca,
3805 result);
3806 if (clean < 0) {
3807 rollback_lock_file(&lock);
3808 return clean;
3811 if (write_locked_index(opt->repo->index, &lock,
3812 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3813 return err(opt, _("Unable to write index."));
3815 return clean ? 0 : 1;
3818 static void merge_recursive_config(struct merge_options *opt)
3820 char *value = NULL;
3821 int renormalize = 0;
3822 git_config_get_int("merge.verbosity", &opt->verbosity);
3823 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3824 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3825 git_config_get_bool("merge.renormalize", &renormalize);
3826 opt->renormalize = renormalize;
3827 if (!git_config_get_string("diff.renames", &value)) {
3828 opt->detect_renames = git_config_rename("diff.renames", value);
3829 free(value);
3831 if (!git_config_get_string("merge.renames", &value)) {
3832 opt->detect_renames = git_config_rename("merge.renames", value);
3833 free(value);
3835 if (!git_config_get_string("merge.directoryrenames", &value)) {
3836 int boolval = git_parse_maybe_bool(value);
3837 if (0 <= boolval) {
3838 opt->detect_directory_renames = boolval ?
3839 MERGE_DIRECTORY_RENAMES_TRUE :
3840 MERGE_DIRECTORY_RENAMES_NONE;
3841 } else if (!strcasecmp(value, "conflict")) {
3842 opt->detect_directory_renames =
3843 MERGE_DIRECTORY_RENAMES_CONFLICT;
3844 } /* avoid erroring on values from future versions of git */
3845 free(value);
3847 git_config(git_xmerge_config, NULL);
3850 void init_merge_options(struct merge_options *opt,
3851 struct repository *repo)
3853 const char *merge_verbosity;
3854 memset(opt, 0, sizeof(struct merge_options));
3856 opt->repo = repo;
3858 opt->detect_renames = -1;
3859 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3860 opt->rename_limit = -1;
3862 opt->verbosity = 2;
3863 opt->buffer_output = 1;
3864 strbuf_init(&opt->obuf, 0);
3866 opt->renormalize = 0;
3868 merge_recursive_config(opt);
3869 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3870 if (merge_verbosity)
3871 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3872 if (opt->verbosity >= 5)
3873 opt->buffer_output = 0;
3876 int parse_merge_opt(struct merge_options *opt, const char *s)
3878 const char *arg;
3880 if (!s || !*s)
3881 return -1;
3882 if (!strcmp(s, "ours"))
3883 opt->recursive_variant = MERGE_VARIANT_OURS;
3884 else if (!strcmp(s, "theirs"))
3885 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3886 else if (!strcmp(s, "subtree"))
3887 opt->subtree_shift = "";
3888 else if (skip_prefix(s, "subtree=", &arg))
3889 opt->subtree_shift = arg;
3890 else if (!strcmp(s, "patience"))
3891 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3892 else if (!strcmp(s, "histogram"))
3893 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3894 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3895 long value = parse_algorithm_value(arg);
3896 if (value < 0)
3897 return -1;
3898 /* clear out previous settings */
3899 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3900 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3901 opt->xdl_opts |= value;
3903 else if (!strcmp(s, "ignore-space-change"))
3904 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3905 else if (!strcmp(s, "ignore-all-space"))
3906 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3907 else if (!strcmp(s, "ignore-space-at-eol"))
3908 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3909 else if (!strcmp(s, "ignore-cr-at-eol"))
3910 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3911 else if (!strcmp(s, "renormalize"))
3912 opt->renormalize = 1;
3913 else if (!strcmp(s, "no-renormalize"))
3914 opt->renormalize = 0;
3915 else if (!strcmp(s, "no-renames"))
3916 opt->detect_renames = 0;
3917 else if (!strcmp(s, "find-renames")) {
3918 opt->detect_renames = 1;
3919 opt->rename_score = 0;
3921 else if (skip_prefix(s, "find-renames=", &arg) ||
3922 skip_prefix(s, "rename-threshold=", &arg)) {
3923 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3924 return -1;
3925 opt->detect_renames = 1;
3928 * Please update $__git_merge_strategy_options in
3929 * git-completion.bash when you add new options
3931 else
3932 return -1;
3933 return 0;