pathspec: ensure full index
[git/debian.git] / merge-recursive.c
blob91d8597728c1a1efbdcfaf1014cf79a405e9b0b3
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, 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(opt->repo, tree,
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 /* TODO: audit for interaction with sparse-index. */
526 ensure_full_index(istate);
527 for (i = 0; i < istate->cache_nr; i++) {
528 struct string_list_item *item;
529 struct stage_data *e;
530 const struct cache_entry *ce = istate->cache[i];
531 if (!ce_stage(ce))
532 continue;
534 item = string_list_lookup(unmerged, ce->name);
535 if (!item) {
536 item = string_list_insert(unmerged, ce->name);
537 item->util = xcalloc(1, sizeof(struct stage_data));
539 e = item->util;
540 e->stages[ce_stage(ce)].mode = ce->ce_mode;
541 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
544 return unmerged;
547 static int string_list_df_name_compare(const char *one, const char *two)
549 int onelen = strlen(one);
550 int twolen = strlen(two);
552 * Here we only care that entries for D/F conflicts are
553 * adjacent, in particular with the file of the D/F conflict
554 * appearing before files below the corresponding directory.
555 * The order of the rest of the list is irrelevant for us.
557 * To achieve this, we sort with df_name_compare and provide
558 * the mode S_IFDIR so that D/F conflicts will sort correctly.
559 * We use the mode S_IFDIR for everything else for simplicity,
560 * since in other cases any changes in their order due to
561 * sorting cause no problems for us.
563 int cmp = df_name_compare(one, onelen, S_IFDIR,
564 two, twolen, S_IFDIR);
566 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
567 * that 'foo' comes before 'foo/bar'.
569 if (cmp)
570 return cmp;
571 return onelen - twolen;
574 static void record_df_conflict_files(struct merge_options *opt,
575 struct string_list *entries)
577 /* If there is a D/F conflict and the file for such a conflict
578 * currently exists in the working tree, we want to allow it to be
579 * removed to make room for the corresponding directory if needed.
580 * The files underneath the directories of such D/F conflicts will
581 * be processed before the corresponding file involved in the D/F
582 * conflict. If the D/F directory ends up being removed by the
583 * merge, then we won't have to touch the D/F file. If the D/F
584 * directory needs to be written to the working copy, then the D/F
585 * file will simply be removed (in make_room_for_path()) to make
586 * room for the necessary paths. Note that if both the directory
587 * and the file need to be present, then the D/F file will be
588 * reinstated with a new unique name at the time it is processed.
590 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
591 const char *last_file = NULL;
592 int last_len = 0;
593 int i;
596 * If we're merging merge-bases, we don't want to bother with
597 * any working directory changes.
599 if (opt->priv->call_depth)
600 return;
602 /* Ensure D/F conflicts are adjacent in the entries list. */
603 for (i = 0; i < entries->nr; i++) {
604 struct string_list_item *next = &entries->items[i];
605 string_list_append(&df_sorted_entries, next->string)->util =
606 next->util;
608 df_sorted_entries.cmp = string_list_df_name_compare;
609 string_list_sort(&df_sorted_entries);
611 string_list_clear(&opt->priv->df_conflict_file_set, 1);
612 for (i = 0; i < df_sorted_entries.nr; i++) {
613 const char *path = df_sorted_entries.items[i].string;
614 int len = strlen(path);
615 struct stage_data *e = df_sorted_entries.items[i].util;
618 * Check if last_file & path correspond to a D/F conflict;
619 * i.e. whether path is last_file+'/'+<something>.
620 * If so, record that it's okay to remove last_file to make
621 * room for path and friends if needed.
623 if (last_file &&
624 len > last_len &&
625 memcmp(path, last_file, last_len) == 0 &&
626 path[last_len] == '/') {
627 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
631 * Determine whether path could exist as a file in the
632 * working directory as a possible D/F conflict. This
633 * will only occur when it exists in stage 2 as a
634 * file.
636 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
637 last_file = path;
638 last_len = len;
639 } else {
640 last_file = NULL;
643 string_list_clear(&df_sorted_entries, 0);
646 static int update_stages(struct merge_options *opt, const char *path,
647 const struct diff_filespec *o,
648 const struct diff_filespec *a,
649 const struct diff_filespec *b)
653 * NOTE: It is usually a bad idea to call update_stages on a path
654 * before calling update_file on that same path, since it can
655 * sometimes lead to spurious "refusing to lose untracked file..."
656 * messages from update_file (via make_room_for path via
657 * would_lose_untracked). Instead, reverse the order of the calls
658 * (executing update_file first and then update_stages).
660 int clear = 1;
661 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
662 if (clear)
663 if (remove_file_from_index(opt->repo->index, path))
664 return -1;
665 if (o)
666 if (add_cacheinfo(opt, o, path, 1, 0, options))
667 return -1;
668 if (a)
669 if (add_cacheinfo(opt, a, path, 2, 0, options))
670 return -1;
671 if (b)
672 if (add_cacheinfo(opt, b, path, 3, 0, options))
673 return -1;
674 return 0;
677 static void update_entry(struct stage_data *entry,
678 struct diff_filespec *o,
679 struct diff_filespec *a,
680 struct diff_filespec *b)
682 entry->processed = 0;
683 entry->stages[1].mode = o->mode;
684 entry->stages[2].mode = a->mode;
685 entry->stages[3].mode = b->mode;
686 oidcpy(&entry->stages[1].oid, &o->oid);
687 oidcpy(&entry->stages[2].oid, &a->oid);
688 oidcpy(&entry->stages[3].oid, &b->oid);
691 static int remove_file(struct merge_options *opt, int clean,
692 const char *path, int no_wd)
694 int update_cache = opt->priv->call_depth || clean;
695 int update_working_directory = !opt->priv->call_depth && !no_wd;
697 if (update_cache) {
698 if (remove_file_from_index(opt->repo->index, path))
699 return -1;
701 if (update_working_directory) {
702 if (ignore_case) {
703 struct cache_entry *ce;
704 ce = index_file_exists(opt->repo->index, path, strlen(path),
705 ignore_case);
706 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
707 return 0;
709 if (remove_path(path))
710 return -1;
712 return 0;
715 /* add a string to a strbuf, but converting "/" to "_" */
716 static void add_flattened_path(struct strbuf *out, const char *s)
718 size_t i = out->len;
719 strbuf_addstr(out, s);
720 for (; i < out->len; i++)
721 if (out->buf[i] == '/')
722 out->buf[i] = '_';
725 static char *unique_path(struct merge_options *opt,
726 const char *path,
727 const char *branch)
729 struct path_hashmap_entry *entry;
730 struct strbuf newpath = STRBUF_INIT;
731 int suffix = 0;
732 size_t base_len;
734 strbuf_addf(&newpath, "%s~", path);
735 add_flattened_path(&newpath, branch);
737 base_len = newpath.len;
738 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
739 path_hash(newpath.buf), newpath.buf) ||
740 (!opt->priv->call_depth && file_exists(newpath.buf))) {
741 strbuf_setlen(&newpath, base_len);
742 strbuf_addf(&newpath, "_%d", suffix++);
745 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
746 hashmap_entry_init(&entry->e, path_hash(entry->path));
747 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
748 return strbuf_detach(&newpath, NULL);
752 * Check whether a directory in the index is in the way of an incoming
753 * file. Return 1 if so. If check_working_copy is non-zero, also
754 * check the working directory. If empty_ok is non-zero, also return
755 * 0 in the case where the working-tree dir exists but is empty.
757 static int dir_in_way(struct index_state *istate, const char *path,
758 int check_working_copy, int empty_ok)
760 int pos;
761 struct strbuf dirpath = STRBUF_INIT;
762 struct stat st;
764 strbuf_addstr(&dirpath, path);
765 strbuf_addch(&dirpath, '/');
767 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
769 if (pos < 0)
770 pos = -1 - pos;
771 if (pos < istate->cache_nr &&
772 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
773 strbuf_release(&dirpath);
774 return 1;
777 strbuf_release(&dirpath);
778 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
779 !(empty_ok && is_empty_dir(path)) &&
780 !has_symlink_leading_path(path, strlen(path));
784 * Returns whether path was tracked in the index before the merge started,
785 * and its oid and mode match the specified values
787 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
788 const struct diff_filespec *blob)
790 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
791 struct cache_entry *ce;
793 if (0 > pos)
794 /* we were not tracking this path before the merge */
795 return 0;
797 /* See if the file we were tracking before matches */
798 ce = opt->priv->orig_index.cache[pos];
799 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
803 * Returns whether path was tracked in the index before the merge started
805 static int was_tracked(struct merge_options *opt, const char *path)
807 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
809 if (0 <= pos)
810 /* we were tracking this path before the merge */
811 return 1;
813 return 0;
816 static int would_lose_untracked(struct merge_options *opt, const char *path)
818 struct index_state *istate = opt->repo->index;
821 * This may look like it can be simplified to:
822 * return !was_tracked(opt, path) && file_exists(path)
823 * but it can't. This function needs to know whether path was in
824 * the working tree due to EITHER having been tracked in the index
825 * before the merge OR having been put into the working copy and
826 * index by unpack_trees(). Due to that either-or requirement, we
827 * check the current index instead of the original one.
829 * Note that we do not need to worry about merge-recursive itself
830 * updating the index after unpack_trees() and before calling this
831 * function, because we strictly require all code paths in
832 * merge-recursive to update the working tree first and the index
833 * second. Doing otherwise would break
834 * update_file()/would_lose_untracked(); see every comment in this
835 * file which mentions "update_stages".
837 int pos = index_name_pos(istate, path, strlen(path));
839 if (pos < 0)
840 pos = -1 - pos;
841 while (pos < istate->cache_nr &&
842 !strcmp(path, istate->cache[pos]->name)) {
844 * If stage #0, it is definitely tracked.
845 * If it has stage #2 then it was tracked
846 * before this merge started. All other
847 * cases the path was not tracked.
849 switch (ce_stage(istate->cache[pos])) {
850 case 0:
851 case 2:
852 return 0;
854 pos++;
856 return file_exists(path);
859 static int was_dirty(struct merge_options *opt, const char *path)
861 struct cache_entry *ce;
862 int dirty = 1;
864 if (opt->priv->call_depth || !was_tracked(opt, path))
865 return !dirty;
867 ce = index_file_exists(opt->priv->unpack_opts.src_index,
868 path, strlen(path), ignore_case);
869 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
870 return dirty;
873 static int make_room_for_path(struct merge_options *opt, const char *path)
875 int status, i;
876 const char *msg = _("failed to create path '%s'%s");
878 /* Unlink any D/F conflict files that are in the way */
879 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
880 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
881 size_t pathlen = strlen(path);
882 size_t df_pathlen = strlen(df_path);
883 if (df_pathlen < pathlen &&
884 path[df_pathlen] == '/' &&
885 strncmp(path, df_path, df_pathlen) == 0) {
886 output(opt, 3,
887 _("Removing %s to make room for subdirectory\n"),
888 df_path);
889 unlink(df_path);
890 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
891 i, 0);
892 break;
896 /* Make sure leading directories are created */
897 status = safe_create_leading_directories_const(path);
898 if (status) {
899 if (status == SCLD_EXISTS)
900 /* something else exists */
901 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
902 return err(opt, msg, path, "");
906 * Do not unlink a file in the work tree if we are not
907 * tracking it.
909 if (would_lose_untracked(opt, path))
910 return err(opt, _("refusing to lose untracked file at '%s'"),
911 path);
913 /* Successful unlink is good.. */
914 if (!unlink(path))
915 return 0;
916 /* .. and so is no existing file */
917 if (errno == ENOENT)
918 return 0;
919 /* .. but not some other error (who really cares what?) */
920 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
923 static int update_file_flags(struct merge_options *opt,
924 const struct diff_filespec *contents,
925 const char *path,
926 int update_cache,
927 int update_wd)
929 int ret = 0;
931 if (opt->priv->call_depth)
932 update_wd = 0;
934 if (update_wd) {
935 enum object_type type;
936 void *buf;
937 unsigned long size;
939 if (S_ISGITLINK(contents->mode)) {
941 * We may later decide to recursively descend into
942 * the submodule directory and update its index
943 * and/or work tree, but we do not do that now.
945 update_wd = 0;
946 goto update_index;
949 buf = read_object_file(&contents->oid, &type, &size);
950 if (!buf) {
951 ret = err(opt, _("cannot read object %s '%s'"),
952 oid_to_hex(&contents->oid), path);
953 goto free_buf;
955 if (type != OBJ_BLOB) {
956 ret = err(opt, _("blob expected for %s '%s'"),
957 oid_to_hex(&contents->oid), path);
958 goto free_buf;
960 if (S_ISREG(contents->mode)) {
961 struct strbuf strbuf = STRBUF_INIT;
962 if (convert_to_working_tree(opt->repo->index,
963 path, buf, size, &strbuf, NULL)) {
964 free(buf);
965 size = strbuf.len;
966 buf = strbuf_detach(&strbuf, NULL);
970 if (make_room_for_path(opt, path) < 0) {
971 update_wd = 0;
972 goto free_buf;
974 if (S_ISREG(contents->mode) ||
975 (!has_symlinks && S_ISLNK(contents->mode))) {
976 int fd;
977 int mode = (contents->mode & 0100 ? 0777 : 0666);
979 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
980 if (fd < 0) {
981 ret = err(opt, _("failed to open '%s': %s"),
982 path, strerror(errno));
983 goto free_buf;
985 write_in_full(fd, buf, size);
986 close(fd);
987 } else if (S_ISLNK(contents->mode)) {
988 char *lnk = xmemdupz(buf, size);
989 safe_create_leading_directories_const(path);
990 unlink(path);
991 if (symlink(lnk, path))
992 ret = err(opt, _("failed to symlink '%s': %s"),
993 path, strerror(errno));
994 free(lnk);
995 } else
996 ret = err(opt,
997 _("do not know what to do with %06o %s '%s'"),
998 contents->mode, oid_to_hex(&contents->oid), path);
999 free_buf:
1000 free(buf);
1002 update_index:
1003 if (!ret && update_cache) {
1004 int refresh = (!opt->priv->call_depth &&
1005 contents->mode != S_IFGITLINK);
1006 if (add_cacheinfo(opt, contents, path, 0, refresh,
1007 ADD_CACHE_OK_TO_ADD))
1008 return -1;
1010 return ret;
1013 static int update_file(struct merge_options *opt,
1014 int clean,
1015 const struct diff_filespec *contents,
1016 const char *path)
1018 return update_file_flags(opt, contents, path,
1019 opt->priv->call_depth || clean, !opt->priv->call_depth);
1022 /* Low level file merging, update and removal */
1024 struct merge_file_info {
1025 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1026 unsigned clean:1,
1027 merge:1;
1030 static int merge_3way(struct merge_options *opt,
1031 mmbuffer_t *result_buf,
1032 const struct diff_filespec *o,
1033 const struct diff_filespec *a,
1034 const struct diff_filespec *b,
1035 const char *branch1,
1036 const char *branch2,
1037 const int extra_marker_size)
1039 mmfile_t orig, src1, src2;
1040 struct ll_merge_options ll_opts = {0};
1041 char *base, *name1, *name2;
1042 int merge_status;
1044 ll_opts.renormalize = opt->renormalize;
1045 ll_opts.extra_marker_size = extra_marker_size;
1046 ll_opts.xdl_opts = opt->xdl_opts;
1048 if (opt->priv->call_depth) {
1049 ll_opts.virtual_ancestor = 1;
1050 ll_opts.variant = 0;
1051 } else {
1052 switch (opt->recursive_variant) {
1053 case MERGE_VARIANT_OURS:
1054 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1055 break;
1056 case MERGE_VARIANT_THEIRS:
1057 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1058 break;
1059 default:
1060 ll_opts.variant = 0;
1061 break;
1065 assert(a->path && b->path && o->path && opt->ancestor);
1066 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1067 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1068 name1 = mkpathdup("%s:%s", branch1, a->path);
1069 name2 = mkpathdup("%s:%s", branch2, b->path);
1070 } else {
1071 base = mkpathdup("%s", opt->ancestor);
1072 name1 = mkpathdup("%s", branch1);
1073 name2 = mkpathdup("%s", branch2);
1076 read_mmblob(&orig, &o->oid);
1077 read_mmblob(&src1, &a->oid);
1078 read_mmblob(&src2, &b->oid);
1080 merge_status = ll_merge(result_buf, a->path, &orig, base,
1081 &src1, name1, &src2, name2,
1082 opt->repo->index, &ll_opts);
1084 free(base);
1085 free(name1);
1086 free(name2);
1087 free(orig.ptr);
1088 free(src1.ptr);
1089 free(src2.ptr);
1090 return merge_status;
1093 static int find_first_merges(struct repository *repo,
1094 struct object_array *result, const char *path,
1095 struct commit *a, struct commit *b)
1097 int i, j;
1098 struct object_array merges = OBJECT_ARRAY_INIT;
1099 struct commit *commit;
1100 int contains_another;
1102 char merged_revision[GIT_MAX_HEXSZ + 2];
1103 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1104 "--all", merged_revision, NULL };
1105 struct rev_info revs;
1106 struct setup_revision_opt rev_opts;
1108 memset(result, 0, sizeof(struct object_array));
1109 memset(&rev_opts, 0, sizeof(rev_opts));
1111 /* get all revisions that merge commit a */
1112 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1113 oid_to_hex(&a->object.oid));
1114 repo_init_revisions(repo, &revs, NULL);
1115 rev_opts.submodule = path;
1116 /* FIXME: can't handle linked worktrees in submodules yet */
1117 revs.single_worktree = path != NULL;
1118 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1120 /* save all revisions from the above list that contain b */
1121 if (prepare_revision_walk(&revs))
1122 die("revision walk setup failed");
1123 while ((commit = get_revision(&revs)) != NULL) {
1124 struct object *o = &(commit->object);
1125 if (in_merge_bases(b, commit))
1126 add_object_array(o, NULL, &merges);
1128 reset_revision_walk();
1130 /* Now we've got all merges that contain a and b. Prune all
1131 * merges that contain another found merge and save them in
1132 * result.
1134 for (i = 0; i < merges.nr; i++) {
1135 struct commit *m1 = (struct commit *) merges.objects[i].item;
1137 contains_another = 0;
1138 for (j = 0; j < merges.nr; j++) {
1139 struct commit *m2 = (struct commit *) merges.objects[j].item;
1140 if (i != j && in_merge_bases(m2, m1)) {
1141 contains_another = 1;
1142 break;
1146 if (!contains_another)
1147 add_object_array(merges.objects[i].item, NULL, result);
1150 object_array_clear(&merges);
1151 return result->nr;
1154 static void print_commit(struct commit *commit)
1156 struct strbuf sb = STRBUF_INIT;
1157 struct pretty_print_context ctx = {0};
1158 ctx.date_mode.type = DATE_NORMAL;
1159 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1160 fprintf(stderr, "%s\n", sb.buf);
1161 strbuf_release(&sb);
1164 static int is_valid(const struct diff_filespec *dfs)
1166 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1169 static int merge_submodule(struct merge_options *opt,
1170 struct object_id *result, const char *path,
1171 const struct object_id *base, const struct object_id *a,
1172 const struct object_id *b)
1174 struct commit *commit_base, *commit_a, *commit_b;
1175 int parent_count;
1176 struct object_array merges;
1178 int i;
1179 int search = !opt->priv->call_depth;
1181 /* store a in result in case we fail */
1182 oidcpy(result, a);
1184 /* we can not handle deletion conflicts */
1185 if (is_null_oid(base))
1186 return 0;
1187 if (is_null_oid(a))
1188 return 0;
1189 if (is_null_oid(b))
1190 return 0;
1192 if (add_submodule_odb(path)) {
1193 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1194 return 0;
1197 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1198 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1199 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1200 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1201 return 0;
1204 /* check whether both changes are forward */
1205 if (!in_merge_bases(commit_base, commit_a) ||
1206 !in_merge_bases(commit_base, commit_b)) {
1207 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1208 return 0;
1211 /* Case #1: a is contained in b or vice versa */
1212 if (in_merge_bases(commit_a, commit_b)) {
1213 oidcpy(result, b);
1214 if (show(opt, 3)) {
1215 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1216 output_commit_title(opt, commit_b);
1217 } else if (show(opt, 2))
1218 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1219 else
1220 ; /* no output */
1222 return 1;
1224 if (in_merge_bases(commit_b, commit_a)) {
1225 oidcpy(result, a);
1226 if (show(opt, 3)) {
1227 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1228 output_commit_title(opt, commit_a);
1229 } else if (show(opt, 2))
1230 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1231 else
1232 ; /* no output */
1234 return 1;
1238 * Case #2: There are one or more merges that contain a and b in
1239 * the submodule. If there is only one, then present it as a
1240 * suggestion to the user, but leave it marked unmerged so the
1241 * user needs to confirm the resolution.
1244 /* Skip the search if makes no sense to the calling context. */
1245 if (!search)
1246 return 0;
1248 /* find commit which merges them */
1249 parent_count = find_first_merges(opt->repo, &merges, path,
1250 commit_a, commit_b);
1251 switch (parent_count) {
1252 case 0:
1253 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1254 break;
1256 case 1:
1257 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1258 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1259 print_commit((struct commit *) merges.objects[0].item);
1260 output(opt, 2, _(
1261 "If this is correct simply add it to the index "
1262 "for example\n"
1263 "by using:\n\n"
1264 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1265 "which will accept this suggestion.\n"),
1266 oid_to_hex(&merges.objects[0].item->oid), path);
1267 break;
1269 default:
1270 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1271 for (i = 0; i < merges.nr; i++)
1272 print_commit((struct commit *) merges.objects[i].item);
1275 object_array_clear(&merges);
1276 return 0;
1279 static int merge_mode_and_contents(struct merge_options *opt,
1280 const struct diff_filespec *o,
1281 const struct diff_filespec *a,
1282 const struct diff_filespec *b,
1283 const char *filename,
1284 const char *branch1,
1285 const char *branch2,
1286 const int extra_marker_size,
1287 struct merge_file_info *result)
1289 if (opt->branch1 != branch1) {
1291 * It's weird getting a reverse merge with HEAD on the bottom
1292 * side of the conflict markers and the other branch on the
1293 * top. Fix that.
1295 return merge_mode_and_contents(opt, o, b, a,
1296 filename,
1297 branch2, branch1,
1298 extra_marker_size, result);
1301 result->merge = 0;
1302 result->clean = 1;
1304 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1305 result->clean = 0;
1306 if (S_ISREG(a->mode)) {
1307 result->blob.mode = a->mode;
1308 oidcpy(&result->blob.oid, &a->oid);
1309 } else {
1310 result->blob.mode = b->mode;
1311 oidcpy(&result->blob.oid, &b->oid);
1313 } else {
1314 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1315 result->merge = 1;
1318 * Merge modes
1320 if (a->mode == b->mode || a->mode == o->mode)
1321 result->blob.mode = b->mode;
1322 else {
1323 result->blob.mode = a->mode;
1324 if (b->mode != o->mode) {
1325 result->clean = 0;
1326 result->merge = 1;
1330 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1331 oidcpy(&result->blob.oid, &b->oid);
1332 else if (oideq(&b->oid, &o->oid))
1333 oidcpy(&result->blob.oid, &a->oid);
1334 else if (S_ISREG(a->mode)) {
1335 mmbuffer_t result_buf;
1336 int ret = 0, merge_status;
1338 merge_status = merge_3way(opt, &result_buf, o, a, b,
1339 branch1, branch2,
1340 extra_marker_size);
1342 if ((merge_status < 0) || !result_buf.ptr)
1343 ret = err(opt, _("Failed to execute internal merge"));
1345 if (!ret &&
1346 write_object_file(result_buf.ptr, result_buf.size,
1347 blob_type, &result->blob.oid))
1348 ret = err(opt, _("Unable to add %s to database"),
1349 a->path);
1351 free(result_buf.ptr);
1352 if (ret)
1353 return ret;
1354 result->clean = (merge_status == 0);
1355 } else if (S_ISGITLINK(a->mode)) {
1356 result->clean = merge_submodule(opt, &result->blob.oid,
1357 o->path,
1358 &o->oid,
1359 &a->oid,
1360 &b->oid);
1361 } else if (S_ISLNK(a->mode)) {
1362 switch (opt->recursive_variant) {
1363 case MERGE_VARIANT_NORMAL:
1364 oidcpy(&result->blob.oid, &a->oid);
1365 if (!oideq(&a->oid, &b->oid))
1366 result->clean = 0;
1367 break;
1368 case MERGE_VARIANT_OURS:
1369 oidcpy(&result->blob.oid, &a->oid);
1370 break;
1371 case MERGE_VARIANT_THEIRS:
1372 oidcpy(&result->blob.oid, &b->oid);
1373 break;
1375 } else
1376 BUG("unsupported object type in the tree");
1379 if (result->merge)
1380 output(opt, 2, _("Auto-merging %s"), filename);
1382 return 0;
1385 static int handle_rename_via_dir(struct merge_options *opt,
1386 struct rename_conflict_info *ci)
1389 * Handle file adds that need to be renamed due to directory rename
1390 * detection. This differs from handle_rename_normal, because
1391 * there is no content merge to do; just move the file into the
1392 * desired final location.
1394 const struct rename *ren = ci->ren1;
1395 const struct diff_filespec *dest = ren->pair->two;
1396 char *file_path = dest->path;
1397 int mark_conflicted = (opt->detect_directory_renames ==
1398 MERGE_DIRECTORY_RENAMES_CONFLICT);
1399 assert(ren->dir_rename_original_dest);
1401 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1402 mark_conflicted = 1;
1403 file_path = unique_path(opt, dest->path, ren->branch);
1404 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1405 "writing to %s instead."),
1406 dest->path, file_path);
1409 if (mark_conflicted) {
1411 * Write the file in worktree at file_path. In the index,
1412 * only record the file at dest->path in the appropriate
1413 * higher stage.
1415 if (update_file(opt, 0, dest, file_path))
1416 return -1;
1417 if (file_path != dest->path)
1418 free(file_path);
1419 if (update_stages(opt, dest->path, NULL,
1420 ren->branch == opt->branch1 ? dest : NULL,
1421 ren->branch == opt->branch1 ? NULL : dest))
1422 return -1;
1423 return 0; /* not clean, but conflicted */
1424 } else {
1425 /* Update dest->path both in index and in worktree */
1426 if (update_file(opt, 1, dest, dest->path))
1427 return -1;
1428 return 1; /* clean */
1432 static int handle_change_delete(struct merge_options *opt,
1433 const char *path, const char *old_path,
1434 const struct diff_filespec *o,
1435 const struct diff_filespec *changed,
1436 const char *change_branch,
1437 const char *delete_branch,
1438 const char *change, const char *change_past)
1440 char *alt_path = NULL;
1441 const char *update_path = path;
1442 int ret = 0;
1444 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1445 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1446 update_path = alt_path = unique_path(opt, path, change_branch);
1449 if (opt->priv->call_depth) {
1451 * We cannot arbitrarily accept either a_sha or b_sha as
1452 * correct; since there is no true "middle point" between
1453 * them, simply reuse the base version for virtual merge base.
1455 ret = remove_file_from_index(opt->repo->index, path);
1456 if (!ret)
1457 ret = update_file(opt, 0, o, update_path);
1458 } else {
1460 * Despite the four nearly duplicate messages and argument
1461 * lists below and the ugliness of the nested if-statements,
1462 * having complete messages makes the job easier for
1463 * translators.
1465 * The slight variance among the cases is due to the fact
1466 * that:
1467 * 1) directory/file conflicts (in effect if
1468 * !alt_path) could cause us to need to write the
1469 * file to a different path.
1470 * 2) renames (in effect if !old_path) could mean that
1471 * there are two names for the path that the user
1472 * may know the file by.
1474 if (!alt_path) {
1475 if (!old_path) {
1476 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1477 "and %s in %s. Version %s of %s left in tree."),
1478 change, path, delete_branch, change_past,
1479 change_branch, change_branch, path);
1480 } else {
1481 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1482 "and %s to %s in %s. Version %s of %s left in tree."),
1483 change, old_path, delete_branch, change_past, path,
1484 change_branch, change_branch, path);
1486 } else {
1487 if (!old_path) {
1488 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1489 "and %s in %s. Version %s of %s left in tree at %s."),
1490 change, path, delete_branch, change_past,
1491 change_branch, change_branch, path, alt_path);
1492 } else {
1493 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1494 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1495 change, old_path, delete_branch, change_past, path,
1496 change_branch, change_branch, path, alt_path);
1500 * No need to call update_file() on path when change_branch ==
1501 * opt->branch1 && !alt_path, since that would needlessly touch
1502 * path. We could call update_file_flags() with update_cache=0
1503 * and update_wd=0, but that's a no-op.
1505 if (change_branch != opt->branch1 || alt_path)
1506 ret = update_file(opt, 0, changed, update_path);
1508 free(alt_path);
1510 return ret;
1513 static int handle_rename_delete(struct merge_options *opt,
1514 struct rename_conflict_info *ci)
1516 const struct rename *ren = ci->ren1;
1517 const struct diff_filespec *orig = ren->pair->one;
1518 const struct diff_filespec *dest = ren->pair->two;
1519 const char *rename_branch = ren->branch;
1520 const char *delete_branch = (opt->branch1 == ren->branch ?
1521 opt->branch2 : opt->branch1);
1523 if (handle_change_delete(opt,
1524 opt->priv->call_depth ? orig->path : dest->path,
1525 opt->priv->call_depth ? NULL : orig->path,
1526 orig, dest,
1527 rename_branch, delete_branch,
1528 _("rename"), _("renamed")))
1529 return -1;
1531 if (opt->priv->call_depth)
1532 return remove_file_from_index(opt->repo->index, dest->path);
1533 else
1534 return update_stages(opt, dest->path, NULL,
1535 rename_branch == opt->branch1 ? dest : NULL,
1536 rename_branch == opt->branch1 ? NULL : dest);
1539 static int handle_file_collision(struct merge_options *opt,
1540 const char *collide_path,
1541 const char *prev_path1,
1542 const char *prev_path2,
1543 const char *branch1, const char *branch2,
1544 struct diff_filespec *a,
1545 struct diff_filespec *b)
1547 struct merge_file_info mfi;
1548 struct diff_filespec null;
1549 char *alt_path = NULL;
1550 const char *update_path = collide_path;
1553 * It's easiest to get the correct things into stage 2 and 3, and
1554 * to make sure that the content merge puts HEAD before the other
1555 * branch if we just ensure that branch1 == opt->branch1. So, simply
1556 * flip arguments around if we don't have that.
1558 if (branch1 != opt->branch1) {
1559 return handle_file_collision(opt, collide_path,
1560 prev_path2, prev_path1,
1561 branch2, branch1,
1562 b, a);
1565 /* Remove rename sources if rename/add or rename/rename(2to1) */
1566 if (prev_path1)
1567 remove_file(opt, 1, prev_path1,
1568 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1569 if (prev_path2)
1570 remove_file(opt, 1, prev_path2,
1571 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1574 * Remove the collision path, if it wouldn't cause dirty contents
1575 * or an untracked file to get lost. We'll either overwrite with
1576 * merged contents, or just write out to differently named files.
1578 if (was_dirty(opt, collide_path)) {
1579 output(opt, 1, _("Refusing to lose dirty file at %s"),
1580 collide_path);
1581 update_path = alt_path = unique_path(opt, collide_path, "merged");
1582 } else if (would_lose_untracked(opt, collide_path)) {
1584 * Only way we get here is if both renames were from
1585 * a directory rename AND user had an untracked file
1586 * at the location where both files end up after the
1587 * two directory renames. See testcase 10d of t6043.
1589 output(opt, 1, _("Refusing to lose untracked file at "
1590 "%s, even though it's in the way."),
1591 collide_path);
1592 update_path = alt_path = unique_path(opt, collide_path, "merged");
1593 } else {
1595 * FIXME: It's possible that the two files are identical
1596 * and that the current working copy happens to match, in
1597 * which case we are unnecessarily touching the working
1598 * tree file. It's not a likely enough scenario that I
1599 * want to code up the checks for it and a better fix is
1600 * available if we restructure how unpack_trees() and
1601 * merge-recursive interoperate anyway, so punting for
1602 * now...
1604 remove_file(opt, 0, collide_path, 0);
1607 /* Store things in diff_filespecs for functions that need it */
1608 null.path = (char *)collide_path;
1609 oidcpy(&null.oid, &null_oid);
1610 null.mode = 0;
1612 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1613 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1614 return -1;
1615 mfi.clean &= !alt_path;
1616 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1617 return -1;
1618 if (!mfi.clean && !opt->priv->call_depth &&
1619 update_stages(opt, collide_path, NULL, a, b))
1620 return -1;
1621 free(alt_path);
1623 * FIXME: If both a & b both started with conflicts (only possible
1624 * if they came from a rename/rename(2to1)), but had IDENTICAL
1625 * contents including those conflicts, then in the next line we claim
1626 * it was clean. If someone cares about this case, we should have the
1627 * caller notify us if we started with conflicts.
1629 return mfi.clean;
1632 static int handle_rename_add(struct merge_options *opt,
1633 struct rename_conflict_info *ci)
1635 /* a was renamed to c, and a separate c was added. */
1636 struct diff_filespec *a = ci->ren1->pair->one;
1637 struct diff_filespec *c = ci->ren1->pair->two;
1638 char *path = c->path;
1639 char *prev_path_desc;
1640 struct merge_file_info mfi;
1642 const char *rename_branch = ci->ren1->branch;
1643 const char *add_branch = (opt->branch1 == rename_branch ?
1644 opt->branch2 : opt->branch1);
1645 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1647 output(opt, 1, _("CONFLICT (rename/add): "
1648 "Rename %s->%s in %s. Added %s in %s"),
1649 a->path, c->path, rename_branch,
1650 c->path, add_branch);
1652 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1653 ci->ren1->src_entry->stages[other_stage].path = a->path;
1654 if (merge_mode_and_contents(opt, a, c,
1655 &ci->ren1->src_entry->stages[other_stage],
1656 prev_path_desc,
1657 opt->branch1, opt->branch2,
1658 1 + opt->priv->call_depth * 2, &mfi))
1659 return -1;
1660 free(prev_path_desc);
1662 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1663 return handle_file_collision(opt,
1664 c->path, a->path, NULL,
1665 rename_branch, add_branch,
1666 &mfi.blob,
1667 &ci->ren1->dst_entry->stages[other_stage]);
1670 static char *find_path_for_conflict(struct merge_options *opt,
1671 const char *path,
1672 const char *branch1,
1673 const char *branch2)
1675 char *new_path = NULL;
1676 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1677 new_path = unique_path(opt, path, branch1);
1678 output(opt, 1, _("%s is a directory in %s adding "
1679 "as %s instead"),
1680 path, branch2, new_path);
1681 } else if (would_lose_untracked(opt, path)) {
1682 new_path = unique_path(opt, path, branch1);
1683 output(opt, 1, _("Refusing to lose untracked file"
1684 " at %s; adding as %s instead"),
1685 path, new_path);
1688 return new_path;
1692 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1694 static inline int flip_stage(int stage)
1696 return (2 + 3) - stage;
1699 static int handle_rename_rename_1to2(struct merge_options *opt,
1700 struct rename_conflict_info *ci)
1702 /* One file was renamed in both branches, but to different names. */
1703 struct merge_file_info mfi;
1704 struct diff_filespec *add;
1705 struct diff_filespec *o = ci->ren1->pair->one;
1706 struct diff_filespec *a = ci->ren1->pair->two;
1707 struct diff_filespec *b = ci->ren2->pair->two;
1708 char *path_desc;
1710 output(opt, 1, _("CONFLICT (rename/rename): "
1711 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1712 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1713 o->path, a->path, ci->ren1->branch,
1714 o->path, b->path, ci->ren2->branch,
1715 opt->priv->call_depth ? _(" (left unresolved)") : "");
1717 path_desc = xstrfmt("%s and %s, both renamed from %s",
1718 a->path, b->path, o->path);
1719 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1720 ci->ren1->branch, ci->ren2->branch,
1721 opt->priv->call_depth * 2, &mfi))
1722 return -1;
1723 free(path_desc);
1725 if (opt->priv->call_depth)
1726 remove_file_from_index(opt->repo->index, o->path);
1729 * For each destination path, we need to see if there is a
1730 * rename/add collision. If not, we can write the file out
1731 * to the specified location.
1733 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1734 if (is_valid(add)) {
1735 add->path = mfi.blob.path = a->path;
1736 if (handle_file_collision(opt, a->path,
1737 NULL, NULL,
1738 ci->ren1->branch,
1739 ci->ren2->branch,
1740 &mfi.blob, add) < 0)
1741 return -1;
1742 } else {
1743 char *new_path = find_path_for_conflict(opt, a->path,
1744 ci->ren1->branch,
1745 ci->ren2->branch);
1746 if (update_file(opt, 0, &mfi.blob,
1747 new_path ? new_path : a->path))
1748 return -1;
1749 free(new_path);
1750 if (!opt->priv->call_depth &&
1751 update_stages(opt, a->path, NULL, a, NULL))
1752 return -1;
1755 if (!mfi.clean && mfi.blob.mode == a->mode &&
1756 oideq(&mfi.blob.oid, &a->oid)) {
1758 * Getting here means we were attempting to merge a binary
1759 * blob. Since we can't merge binaries, the merge algorithm
1760 * just takes one side. But we don't want to copy the
1761 * contents of one side to both paths; we'd rather use the
1762 * original content at the given path for each path.
1764 oidcpy(&mfi.blob.oid, &b->oid);
1765 mfi.blob.mode = b->mode;
1767 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1768 if (is_valid(add)) {
1769 add->path = mfi.blob.path = b->path;
1770 if (handle_file_collision(opt, b->path,
1771 NULL, NULL,
1772 ci->ren1->branch,
1773 ci->ren2->branch,
1774 add, &mfi.blob) < 0)
1775 return -1;
1776 } else {
1777 char *new_path = find_path_for_conflict(opt, b->path,
1778 ci->ren2->branch,
1779 ci->ren1->branch);
1780 if (update_file(opt, 0, &mfi.blob,
1781 new_path ? new_path : b->path))
1782 return -1;
1783 free(new_path);
1784 if (!opt->priv->call_depth &&
1785 update_stages(opt, b->path, NULL, NULL, b))
1786 return -1;
1789 return 0;
1792 static int handle_rename_rename_2to1(struct merge_options *opt,
1793 struct rename_conflict_info *ci)
1795 /* Two files, a & b, were renamed to the same thing, c. */
1796 struct diff_filespec *a = ci->ren1->pair->one;
1797 struct diff_filespec *b = ci->ren2->pair->one;
1798 struct diff_filespec *c1 = ci->ren1->pair->two;
1799 struct diff_filespec *c2 = ci->ren2->pair->two;
1800 char *path = c1->path; /* == c2->path */
1801 char *path_side_1_desc;
1802 char *path_side_2_desc;
1803 struct merge_file_info mfi_c1;
1804 struct merge_file_info mfi_c2;
1805 int ostage1, ostage2;
1807 output(opt, 1, _("CONFLICT (rename/rename): "
1808 "Rename %s->%s in %s. "
1809 "Rename %s->%s in %s"),
1810 a->path, c1->path, ci->ren1->branch,
1811 b->path, c2->path, ci->ren2->branch);
1813 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1814 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1815 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1816 ostage2 = flip_stage(ostage1);
1817 ci->ren1->src_entry->stages[ostage1].path = a->path;
1818 ci->ren2->src_entry->stages[ostage2].path = b->path;
1819 if (merge_mode_and_contents(opt, a, c1,
1820 &ci->ren1->src_entry->stages[ostage1],
1821 path_side_1_desc,
1822 opt->branch1, opt->branch2,
1823 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1824 merge_mode_and_contents(opt, b,
1825 &ci->ren2->src_entry->stages[ostage2],
1826 c2, path_side_2_desc,
1827 opt->branch1, opt->branch2,
1828 1 + opt->priv->call_depth * 2, &mfi_c2))
1829 return -1;
1830 free(path_side_1_desc);
1831 free(path_side_2_desc);
1832 mfi_c1.blob.path = path;
1833 mfi_c2.blob.path = path;
1835 return handle_file_collision(opt, path, a->path, b->path,
1836 ci->ren1->branch, ci->ren2->branch,
1837 &mfi_c1.blob, &mfi_c2.blob);
1841 * Get the diff_filepairs changed between o_tree and tree.
1843 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1844 struct tree *o_tree,
1845 struct tree *tree)
1847 struct diff_queue_struct *ret;
1848 struct diff_options opts;
1850 repo_diff_setup(opt->repo, &opts);
1851 opts.flags.recursive = 1;
1852 opts.flags.rename_empty = 0;
1853 opts.detect_rename = merge_detect_rename(opt);
1855 * We do not have logic to handle the detection of copies. In
1856 * fact, it may not even make sense to add such logic: would we
1857 * really want a change to a base file to be propagated through
1858 * multiple other files by a merge?
1860 if (opts.detect_rename > DIFF_DETECT_RENAME)
1861 opts.detect_rename = DIFF_DETECT_RENAME;
1862 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 1000;
1863 opts.rename_score = opt->rename_score;
1864 opts.show_rename_progress = opt->show_rename_progress;
1865 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1866 diff_setup_done(&opts);
1867 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1868 diffcore_std(&opts);
1869 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1870 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1872 ret = xmalloc(sizeof(*ret));
1873 *ret = diff_queued_diff;
1875 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1876 diff_queued_diff.nr = 0;
1877 diff_queued_diff.queue = NULL;
1878 diff_flush(&opts);
1879 return ret;
1882 static int tree_has_path(struct repository *r, struct tree *tree,
1883 const char *path)
1885 struct object_id hashy;
1886 unsigned short mode_o;
1888 return !get_tree_entry(r,
1889 &tree->object.oid, path,
1890 &hashy, &mode_o);
1894 * Return a new string that replaces the beginning portion (which matches
1895 * entry->dir), with entry->new_dir. In perl-speak:
1896 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1897 * NOTE:
1898 * Caller must ensure that old_path starts with entry->dir + '/'.
1900 static char *apply_dir_rename(struct dir_rename_entry *entry,
1901 const char *old_path)
1903 struct strbuf new_path = STRBUF_INIT;
1904 int oldlen, newlen;
1906 if (entry->non_unique_new_dir)
1907 return NULL;
1909 oldlen = strlen(entry->dir);
1910 if (entry->new_dir.len == 0)
1912 * If someone renamed/merged a subdirectory into the root
1913 * directory (e.g. 'some/subdir' -> ''), then we want to
1914 * avoid returning
1915 * '' + '/filename'
1916 * as the rename; we need to make old_path + oldlen advance
1917 * past the '/' character.
1919 oldlen++;
1920 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1921 strbuf_grow(&new_path, newlen);
1922 strbuf_addbuf(&new_path, &entry->new_dir);
1923 strbuf_addstr(&new_path, &old_path[oldlen]);
1925 return strbuf_detach(&new_path, NULL);
1928 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1929 char **old_dir, char **new_dir)
1931 char *end_of_old, *end_of_new;
1933 /* Default return values: NULL, meaning no rename */
1934 *old_dir = NULL;
1935 *new_dir = NULL;
1938 * For
1939 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1940 * the "e/foo.c" part is the same, we just want to know that
1941 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1942 * so, for this example, this function returns "a/b/c/d" in
1943 * *old_dir and "a/b/some/thing/else" in *new_dir.
1947 * If the basename of the file changed, we don't care. We want
1948 * to know which portion of the directory, if any, changed.
1950 end_of_old = strrchr(old_path, '/');
1951 end_of_new = strrchr(new_path, '/');
1954 * If end_of_old is NULL, old_path wasn't in a directory, so there
1955 * could not be a directory rename (our rule elsewhere that a
1956 * directory which still exists is not considered to have been
1957 * renamed means the root directory can never be renamed -- because
1958 * the root directory always exists).
1960 if (end_of_old == NULL)
1961 return; /* Note: *old_dir and *new_dir are still NULL */
1964 * If new_path contains no directory (end_of_new is NULL), then we
1965 * have a rename of old_path's directory to the root directory.
1967 if (end_of_new == NULL) {
1968 *old_dir = xstrndup(old_path, end_of_old - old_path);
1969 *new_dir = xstrdup("");
1970 return;
1973 /* Find the first non-matching character traversing backwards */
1974 while (*--end_of_new == *--end_of_old &&
1975 end_of_old != old_path &&
1976 end_of_new != new_path)
1977 ; /* Do nothing; all in the while loop */
1980 * If both got back to the beginning of their strings, then the
1981 * directory didn't change at all, only the basename did.
1983 if (end_of_old == old_path && end_of_new == new_path &&
1984 *end_of_old == *end_of_new)
1985 return; /* Note: *old_dir and *new_dir are still NULL */
1988 * If end_of_new got back to the beginning of its string, and
1989 * end_of_old got back to the beginning of some subdirectory, then
1990 * we have a rename/merge of a subdirectory into the root, which
1991 * needs slightly special handling.
1993 * Note: There is no need to consider the opposite case, with a
1994 * rename/merge of the root directory into some subdirectory
1995 * because as noted above the root directory always exists so it
1996 * cannot be considered to be renamed.
1998 if (end_of_new == new_path &&
1999 end_of_old != old_path && end_of_old[-1] == '/') {
2000 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2001 *new_dir = xstrdup("");
2002 return;
2006 * We've found the first non-matching character in the directory
2007 * paths. That means the current characters we were looking at
2008 * were part of the first non-matching subdir name going back from
2009 * the end of the strings. Get the whole name by advancing both
2010 * end_of_old and end_of_new to the NEXT '/' character. That will
2011 * represent the entire directory rename.
2013 * The reason for the increment is cases like
2014 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2015 * After dropping the basename and going back to the first
2016 * non-matching character, we're now comparing:
2017 * a/b/s and a/b/
2018 * and we want to be comparing:
2019 * a/b/star/ and a/b/tar/
2020 * but without the pre-increment, the one on the right would stay
2021 * a/b/.
2023 end_of_old = strchr(++end_of_old, '/');
2024 end_of_new = strchr(++end_of_new, '/');
2026 /* Copy the old and new directories into *old_dir and *new_dir. */
2027 *old_dir = xstrndup(old_path, end_of_old - old_path);
2028 *new_dir = xstrndup(new_path, end_of_new - new_path);
2031 static void remove_hashmap_entries(struct hashmap *dir_renames,
2032 struct string_list *items_to_remove)
2034 int i;
2035 struct dir_rename_entry *entry;
2037 for (i = 0; i < items_to_remove->nr; i++) {
2038 entry = items_to_remove->items[i].util;
2039 hashmap_remove(dir_renames, &entry->ent, NULL);
2041 string_list_clear(items_to_remove, 0);
2045 * See if there is a directory rename for path, and if there are any file
2046 * level conflicts for the renamed location. If there is a rename and
2047 * there are no conflicts, return the new name. Otherwise, return NULL.
2049 static char *handle_path_level_conflicts(struct merge_options *opt,
2050 const char *path,
2051 struct dir_rename_entry *entry,
2052 struct hashmap *collisions,
2053 struct tree *tree)
2055 char *new_path = NULL;
2056 struct collision_entry *collision_ent;
2057 int clean = 1;
2058 struct strbuf collision_paths = STRBUF_INIT;
2061 * entry has the mapping of old directory name to new directory name
2062 * that we want to apply to path.
2064 new_path = apply_dir_rename(entry, path);
2066 if (!new_path) {
2067 /* This should only happen when entry->non_unique_new_dir set */
2068 if (!entry->non_unique_new_dir)
2069 BUG("entry->non_unqiue_dir not set and !new_path");
2070 output(opt, 1, _("CONFLICT (directory rename split): "
2071 "Unclear where to place %s because directory "
2072 "%s was renamed to multiple other directories, "
2073 "with no destination getting a majority of the "
2074 "files."),
2075 path, entry->dir);
2076 clean = 0;
2077 return NULL;
2081 * The caller needs to have ensured that it has pre-populated
2082 * collisions with all paths that map to new_path. Do a quick check
2083 * to ensure that's the case.
2085 collision_ent = collision_find_entry(collisions, new_path);
2086 if (collision_ent == NULL)
2087 BUG("collision_ent is NULL");
2090 * Check for one-sided add/add/.../add conflicts, i.e.
2091 * where implicit renames from the other side doing
2092 * directory rename(s) can affect this side of history
2093 * to put multiple paths into the same location. Warn
2094 * and bail on directory renames for such paths.
2096 if (collision_ent->reported_already) {
2097 clean = 0;
2098 } else if (tree_has_path(opt->repo, tree, new_path)) {
2099 collision_ent->reported_already = 1;
2100 strbuf_add_separated_string_list(&collision_paths, ", ",
2101 &collision_ent->source_files);
2102 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2103 "file/dir at %s in the way of implicit "
2104 "directory rename(s) putting the following "
2105 "path(s) there: %s."),
2106 new_path, collision_paths.buf);
2107 clean = 0;
2108 } else if (collision_ent->source_files.nr > 1) {
2109 collision_ent->reported_already = 1;
2110 strbuf_add_separated_string_list(&collision_paths, ", ",
2111 &collision_ent->source_files);
2112 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2113 "more than one path to %s; implicit directory "
2114 "renames tried to put these paths there: %s"),
2115 new_path, collision_paths.buf);
2116 clean = 0;
2119 /* Free memory we no longer need */
2120 strbuf_release(&collision_paths);
2121 if (!clean && new_path) {
2122 free(new_path);
2123 return NULL;
2126 return new_path;
2130 * There are a couple things we want to do at the directory level:
2131 * 1. Check for both sides renaming to the same thing, in order to avoid
2132 * implicit renaming of files that should be left in place. (See
2133 * testcase 6b in t6043 for details.)
2134 * 2. Prune directory renames if there are still files left in the
2135 * the original directory. These represent a partial directory rename,
2136 * i.e. a rename where only some of the files within the directory
2137 * were renamed elsewhere. (Technically, this could be done earlier
2138 * in get_directory_renames(), except that would prevent us from
2139 * doing the previous check and thus failing testcase 6b.)
2140 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2141 * In the future, we could potentially record this info as well and
2142 * omit reporting rename/rename(1to2) conflicts for each path within
2143 * the affected directories, thus cleaning up the merge output.
2144 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2145 * directory level, because merging directories is fine. If it
2146 * causes conflicts for files within those merged directories, then
2147 * that should be detected at the individual path level.
2149 static void handle_directory_level_conflicts(struct merge_options *opt,
2150 struct hashmap *dir_re_head,
2151 struct tree *head,
2152 struct hashmap *dir_re_merge,
2153 struct tree *merge)
2155 struct hashmap_iter iter;
2156 struct dir_rename_entry *head_ent;
2157 struct dir_rename_entry *merge_ent;
2159 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2160 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2162 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2163 ent /* member name */) {
2164 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2165 if (merge_ent &&
2166 !head_ent->non_unique_new_dir &&
2167 !merge_ent->non_unique_new_dir &&
2168 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2169 /* 1. Renamed identically; remove it from both sides */
2170 string_list_append(&remove_from_head,
2171 head_ent->dir)->util = head_ent;
2172 strbuf_release(&head_ent->new_dir);
2173 string_list_append(&remove_from_merge,
2174 merge_ent->dir)->util = merge_ent;
2175 strbuf_release(&merge_ent->new_dir);
2176 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2177 /* 2. This wasn't a directory rename after all */
2178 string_list_append(&remove_from_head,
2179 head_ent->dir)->util = head_ent;
2180 strbuf_release(&head_ent->new_dir);
2184 remove_hashmap_entries(dir_re_head, &remove_from_head);
2185 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2187 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2188 ent /* member name */) {
2189 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2190 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2191 /* 2. This wasn't a directory rename after all */
2192 string_list_append(&remove_from_merge,
2193 merge_ent->dir)->util = merge_ent;
2194 } else if (head_ent &&
2195 !head_ent->non_unique_new_dir &&
2196 !merge_ent->non_unique_new_dir) {
2197 /* 3. rename/rename(1to2) */
2199 * We can assume it's not rename/rename(1to1) because
2200 * that was case (1), already checked above. So we
2201 * know that head_ent->new_dir and merge_ent->new_dir
2202 * are different strings.
2204 output(opt, 1, _("CONFLICT (rename/rename): "
2205 "Rename directory %s->%s in %s. "
2206 "Rename directory %s->%s in %s"),
2207 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2208 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2209 string_list_append(&remove_from_head,
2210 head_ent->dir)->util = head_ent;
2211 strbuf_release(&head_ent->new_dir);
2212 string_list_append(&remove_from_merge,
2213 merge_ent->dir)->util = merge_ent;
2214 strbuf_release(&merge_ent->new_dir);
2218 remove_hashmap_entries(dir_re_head, &remove_from_head);
2219 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2222 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2224 struct hashmap *dir_renames;
2225 struct hashmap_iter iter;
2226 struct dir_rename_entry *entry;
2227 int i;
2230 * Typically, we think of a directory rename as all files from a
2231 * certain directory being moved to a target directory. However,
2232 * what if someone first moved two files from the original
2233 * directory in one commit, and then renamed the directory
2234 * somewhere else in a later commit? At merge time, we just know
2235 * that files from the original directory went to two different
2236 * places, and that the bulk of them ended up in the same place.
2237 * We want each directory rename to represent where the bulk of the
2238 * files from that directory end up; this function exists to find
2239 * where the bulk of the files went.
2241 * The first loop below simply iterates through the list of file
2242 * renames, finding out how often each directory rename pair
2243 * possibility occurs.
2245 dir_renames = xmalloc(sizeof(*dir_renames));
2246 dir_rename_init(dir_renames);
2247 for (i = 0; i < pairs->nr; ++i) {
2248 struct string_list_item *item;
2249 int *count;
2250 struct diff_filepair *pair = pairs->queue[i];
2251 char *old_dir, *new_dir;
2253 /* File not part of directory rename if it wasn't renamed */
2254 if (pair->status != 'R')
2255 continue;
2257 get_renamed_dir_portion(pair->one->path, pair->two->path,
2258 &old_dir, &new_dir);
2259 if (!old_dir)
2260 /* Directory didn't change at all; ignore this one. */
2261 continue;
2263 entry = dir_rename_find_entry(dir_renames, old_dir);
2264 if (!entry) {
2265 entry = xmalloc(sizeof(*entry));
2266 dir_rename_entry_init(entry, old_dir);
2267 hashmap_put(dir_renames, &entry->ent);
2268 } else {
2269 free(old_dir);
2271 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2272 if (!item) {
2273 item = string_list_insert(&entry->possible_new_dirs,
2274 new_dir);
2275 item->util = xcalloc(1, sizeof(int));
2276 } else {
2277 free(new_dir);
2279 count = item->util;
2280 *count += 1;
2284 * For each directory with files moved out of it, we find out which
2285 * target directory received the most files so we can declare it to
2286 * be the "winning" target location for the directory rename. This
2287 * winner gets recorded in new_dir. If there is no winner
2288 * (multiple target directories received the same number of files),
2289 * we set non_unique_new_dir. Once we've determined the winner (or
2290 * that there is no winner), we no longer need possible_new_dirs.
2292 hashmap_for_each_entry(dir_renames, &iter, entry,
2293 ent /* member name */) {
2294 int max = 0;
2295 int bad_max = 0;
2296 char *best = NULL;
2298 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2299 int *count = entry->possible_new_dirs.items[i].util;
2301 if (*count == max)
2302 bad_max = max;
2303 else if (*count > max) {
2304 max = *count;
2305 best = entry->possible_new_dirs.items[i].string;
2308 if (bad_max == max)
2309 entry->non_unique_new_dir = 1;
2310 else {
2311 assert(entry->new_dir.len == 0);
2312 strbuf_addstr(&entry->new_dir, best);
2315 * The relevant directory sub-portion of the original full
2316 * filepaths were xstrndup'ed before inserting into
2317 * possible_new_dirs, and instead of manually iterating the
2318 * list and free'ing each, just lie and tell
2319 * possible_new_dirs that it did the strdup'ing so that it
2320 * will free them for us.
2322 entry->possible_new_dirs.strdup_strings = 1;
2323 string_list_clear(&entry->possible_new_dirs, 1);
2326 return dir_renames;
2329 static struct dir_rename_entry *check_dir_renamed(const char *path,
2330 struct hashmap *dir_renames)
2332 char *temp = xstrdup(path);
2333 char *end;
2334 struct dir_rename_entry *entry = NULL;
2336 while ((end = strrchr(temp, '/'))) {
2337 *end = '\0';
2338 entry = dir_rename_find_entry(dir_renames, temp);
2339 if (entry)
2340 break;
2342 free(temp);
2343 return entry;
2346 static void compute_collisions(struct hashmap *collisions,
2347 struct hashmap *dir_renames,
2348 struct diff_queue_struct *pairs)
2350 int i;
2353 * Multiple files can be mapped to the same path due to directory
2354 * renames done by the other side of history. Since that other
2355 * side of history could have merged multiple directories into one,
2356 * if our side of history added the same file basename to each of
2357 * those directories, then all N of them would get implicitly
2358 * renamed by the directory rename detection into the same path,
2359 * and we'd get an add/add/.../add conflict, and all those adds
2360 * from *this* side of history. This is not representable in the
2361 * index, and users aren't going to easily be able to make sense of
2362 * it. So we need to provide a good warning about what's
2363 * happening, and fall back to no-directory-rename detection
2364 * behavior for those paths.
2366 * See testcases 9e and all of section 5 from t6043 for examples.
2368 collision_init(collisions);
2370 for (i = 0; i < pairs->nr; ++i) {
2371 struct dir_rename_entry *dir_rename_ent;
2372 struct collision_entry *collision_ent;
2373 char *new_path;
2374 struct diff_filepair *pair = pairs->queue[i];
2376 if (pair->status != 'A' && pair->status != 'R')
2377 continue;
2378 dir_rename_ent = check_dir_renamed(pair->two->path,
2379 dir_renames);
2380 if (!dir_rename_ent)
2381 continue;
2383 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2384 if (!new_path)
2386 * dir_rename_ent->non_unique_new_path is true, which
2387 * means there is no directory rename for us to use,
2388 * which means it won't cause us any additional
2389 * collisions.
2391 continue;
2392 collision_ent = collision_find_entry(collisions, new_path);
2393 if (!collision_ent) {
2394 collision_ent = xcalloc(1,
2395 sizeof(struct collision_entry));
2396 hashmap_entry_init(&collision_ent->ent,
2397 strhash(new_path));
2398 hashmap_put(collisions, &collision_ent->ent);
2399 collision_ent->target_file = new_path;
2400 } else {
2401 free(new_path);
2403 string_list_insert(&collision_ent->source_files,
2404 pair->two->path);
2408 static char *check_for_directory_rename(struct merge_options *opt,
2409 const char *path,
2410 struct tree *tree,
2411 struct hashmap *dir_renames,
2412 struct hashmap *dir_rename_exclusions,
2413 struct hashmap *collisions,
2414 int *clean_merge)
2416 char *new_path = NULL;
2417 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2418 struct dir_rename_entry *oentry = NULL;
2420 if (!entry)
2421 return new_path;
2424 * This next part is a little weird. We do not want to do an
2425 * implicit rename into a directory we renamed on our side, because
2426 * that will result in a spurious rename/rename(1to2) conflict. An
2427 * example:
2428 * Base commit: dumbdir/afile, otherdir/bfile
2429 * Side 1: smrtdir/afile, otherdir/bfile
2430 * Side 2: dumbdir/afile, dumbdir/bfile
2431 * Here, while working on Side 1, we could notice that otherdir was
2432 * renamed/merged to dumbdir, and change the diff_filepair for
2433 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2434 * 2 will notice the rename from dumbdir to smrtdir, and do the
2435 * transitive rename to move it from dumbdir/bfile to
2436 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2437 * smrtdir, a rename/rename(1to2) conflict. We really just want
2438 * the file to end up in smrtdir. And the way to achieve that is
2439 * to not let Side1 do the rename to dumbdir, since we know that is
2440 * the source of one of our directory renames.
2442 * That's why oentry and dir_rename_exclusions is here.
2444 * As it turns out, this also prevents N-way transient rename
2445 * confusion; See testcases 9c and 9d of t6043.
2447 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2448 if (oentry) {
2449 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2450 "to %s, because %s itself was renamed."),
2451 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2452 } else {
2453 new_path = handle_path_level_conflicts(opt, path, entry,
2454 collisions, tree);
2455 *clean_merge &= (new_path != NULL);
2458 return new_path;
2461 static void apply_directory_rename_modifications(struct merge_options *opt,
2462 struct diff_filepair *pair,
2463 char *new_path,
2464 struct rename *re,
2465 struct tree *tree,
2466 struct tree *o_tree,
2467 struct tree *a_tree,
2468 struct tree *b_tree,
2469 struct string_list *entries)
2471 struct string_list_item *item;
2472 int stage = (tree == a_tree ? 2 : 3);
2473 int update_wd;
2476 * In all cases where we can do directory rename detection,
2477 * unpack_trees() will have read pair->two->path into the
2478 * index and the working copy. We need to remove it so that
2479 * we can instead place it at new_path. It is guaranteed to
2480 * not be untracked (unpack_trees() would have errored out
2481 * saying the file would have been overwritten), but it might
2482 * be dirty, though.
2484 update_wd = !was_dirty(opt, pair->two->path);
2485 if (!update_wd)
2486 output(opt, 1, _("Refusing to lose dirty file at %s"),
2487 pair->two->path);
2488 remove_file(opt, 1, pair->two->path, !update_wd);
2490 /* Find or create a new re->dst_entry */
2491 item = string_list_lookup(entries, new_path);
2492 if (item) {
2494 * Since we're renaming on this side of history, and it's
2495 * due to a directory rename on the other side of history
2496 * (which we only allow when the directory in question no
2497 * longer exists on the other side of history), the
2498 * original entry for re->dst_entry is no longer
2499 * necessary...
2501 re->dst_entry->processed = 1;
2504 * ...because we'll be using this new one.
2506 re->dst_entry = item->util;
2507 } else {
2509 * re->dst_entry is for the before-dir-rename path, and we
2510 * need it to hold information for the after-dir-rename
2511 * path. Before creating a new entry, we need to mark the
2512 * old one as unnecessary (...unless it is shared by
2513 * src_entry, i.e. this didn't use to be a rename, in which
2514 * case we can just allow the normal processing to happen
2515 * for it).
2517 if (pair->status == 'R')
2518 re->dst_entry->processed = 1;
2520 re->dst_entry = insert_stage_data(opt->repo, new_path,
2521 o_tree, a_tree, b_tree,
2522 entries);
2523 item = string_list_insert(entries, new_path);
2524 item->util = re->dst_entry;
2528 * Update the stage_data with the information about the path we are
2529 * moving into place. That slot will be empty and available for us
2530 * to write to because of the collision checks in
2531 * handle_path_level_conflicts(). In other words,
2532 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2533 * open for us to write to.
2535 * It may be tempting to actually update the index at this point as
2536 * well, using update_stages_for_stage_data(), but as per the big
2537 * "NOTE" in update_stages(), doing so will modify the current
2538 * in-memory index which will break calls to would_lose_untracked()
2539 * that we need to make. Instead, we need to just make sure that
2540 * the various handle_rename_*() functions update the index
2541 * explicitly rather than relying on unpack_trees() to have done it.
2543 get_tree_entry(opt->repo,
2544 &tree->object.oid,
2545 pair->two->path,
2546 &re->dst_entry->stages[stage].oid,
2547 &re->dst_entry->stages[stage].mode);
2550 * Record the original change status (or 'type' of change). If it
2551 * was originally an add ('A'), this lets us differentiate later
2552 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2553 * otherwise look the same). If it was originally a rename ('R'),
2554 * this lets us remember and report accurately about the transitive
2555 * renaming that occurred via the directory rename detection. Also,
2556 * record the original destination name.
2558 re->dir_rename_original_type = pair->status;
2559 re->dir_rename_original_dest = pair->two->path;
2562 * We don't actually look at pair->status again, but it seems
2563 * pedagogically correct to adjust it.
2565 pair->status = 'R';
2568 * Finally, record the new location.
2570 pair->two->path = new_path;
2574 * Get information of all renames which occurred in 'pairs', making use of
2575 * any implicit directory renames inferred from the other side of history.
2576 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2577 * to be able to associate the correct cache entries with the rename
2578 * information; tree is always equal to either a_tree or b_tree.
2580 static struct string_list *get_renames(struct merge_options *opt,
2581 const char *branch,
2582 struct diff_queue_struct *pairs,
2583 struct hashmap *dir_renames,
2584 struct hashmap *dir_rename_exclusions,
2585 struct tree *tree,
2586 struct tree *o_tree,
2587 struct tree *a_tree,
2588 struct tree *b_tree,
2589 struct string_list *entries,
2590 int *clean_merge)
2592 int i;
2593 struct hashmap collisions;
2594 struct hashmap_iter iter;
2595 struct collision_entry *e;
2596 struct string_list *renames;
2598 compute_collisions(&collisions, dir_renames, pairs);
2599 renames = xcalloc(1, sizeof(struct string_list));
2601 for (i = 0; i < pairs->nr; ++i) {
2602 struct string_list_item *item;
2603 struct rename *re;
2604 struct diff_filepair *pair = pairs->queue[i];
2605 char *new_path; /* non-NULL only with directory renames */
2607 if (pair->status != 'A' && pair->status != 'R') {
2608 diff_free_filepair(pair);
2609 continue;
2611 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2612 dir_renames,
2613 dir_rename_exclusions,
2614 &collisions,
2615 clean_merge);
2616 if (pair->status != 'R' && !new_path) {
2617 diff_free_filepair(pair);
2618 continue;
2621 re = xmalloc(sizeof(*re));
2622 re->processed = 0;
2623 re->pair = pair;
2624 re->branch = branch;
2625 re->dir_rename_original_type = '\0';
2626 re->dir_rename_original_dest = NULL;
2627 item = string_list_lookup(entries, re->pair->one->path);
2628 if (!item)
2629 re->src_entry = insert_stage_data(opt->repo,
2630 re->pair->one->path,
2631 o_tree, a_tree, b_tree, entries);
2632 else
2633 re->src_entry = item->util;
2635 item = string_list_lookup(entries, re->pair->two->path);
2636 if (!item)
2637 re->dst_entry = insert_stage_data(opt->repo,
2638 re->pair->two->path,
2639 o_tree, a_tree, b_tree, entries);
2640 else
2641 re->dst_entry = item->util;
2642 item = string_list_insert(renames, pair->one->path);
2643 item->util = re;
2644 if (new_path)
2645 apply_directory_rename_modifications(opt, pair, new_path,
2646 re, tree, o_tree,
2647 a_tree, b_tree,
2648 entries);
2651 hashmap_for_each_entry(&collisions, &iter, e,
2652 ent /* member name */) {
2653 free(e->target_file);
2654 string_list_clear(&e->source_files, 0);
2656 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2657 return renames;
2660 static int process_renames(struct merge_options *opt,
2661 struct string_list *a_renames,
2662 struct string_list *b_renames)
2664 int clean_merge = 1, i, j;
2665 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2666 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2667 const struct rename *sre;
2669 for (i = 0; i < a_renames->nr; i++) {
2670 sre = a_renames->items[i].util;
2671 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2672 = (void *)sre;
2674 for (i = 0; i < b_renames->nr; i++) {
2675 sre = b_renames->items[i].util;
2676 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2677 = (void *)sre;
2680 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2681 struct string_list *renames1, *renames2Dst;
2682 struct rename *ren1 = NULL, *ren2 = NULL;
2683 const char *ren1_src, *ren1_dst;
2684 struct string_list_item *lookup;
2686 if (i >= a_renames->nr) {
2687 ren2 = b_renames->items[j++].util;
2688 } else if (j >= b_renames->nr) {
2689 ren1 = a_renames->items[i++].util;
2690 } else {
2691 int compare = strcmp(a_renames->items[i].string,
2692 b_renames->items[j].string);
2693 if (compare <= 0)
2694 ren1 = a_renames->items[i++].util;
2695 if (compare >= 0)
2696 ren2 = b_renames->items[j++].util;
2699 /* TODO: refactor, so that 1/2 are not needed */
2700 if (ren1) {
2701 renames1 = a_renames;
2702 renames2Dst = &b_by_dst;
2703 } else {
2704 renames1 = b_renames;
2705 renames2Dst = &a_by_dst;
2706 SWAP(ren2, ren1);
2709 if (ren1->processed)
2710 continue;
2711 ren1->processed = 1;
2712 ren1->dst_entry->processed = 1;
2713 /* BUG: We should only mark src_entry as processed if we
2714 * are not dealing with a rename + add-source case.
2716 ren1->src_entry->processed = 1;
2718 ren1_src = ren1->pair->one->path;
2719 ren1_dst = ren1->pair->two->path;
2721 if (ren2) {
2722 /* One file renamed on both sides */
2723 const char *ren2_src = ren2->pair->one->path;
2724 const char *ren2_dst = ren2->pair->two->path;
2725 enum rename_type rename_type;
2726 if (strcmp(ren1_src, ren2_src) != 0)
2727 BUG("ren1_src != ren2_src");
2728 ren2->dst_entry->processed = 1;
2729 ren2->processed = 1;
2730 if (strcmp(ren1_dst, ren2_dst) != 0) {
2731 rename_type = RENAME_ONE_FILE_TO_TWO;
2732 clean_merge = 0;
2733 } else {
2734 rename_type = RENAME_ONE_FILE_TO_ONE;
2735 /* BUG: We should only remove ren1_src in
2736 * the base stage (think of rename +
2737 * add-source cases).
2739 remove_file(opt, 1, ren1_src, 1);
2740 update_entry(ren1->dst_entry,
2741 ren1->pair->one,
2742 ren1->pair->two,
2743 ren2->pair->two);
2745 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2746 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2747 /* Two different files renamed to the same thing */
2748 char *ren2_dst;
2749 ren2 = lookup->util;
2750 ren2_dst = ren2->pair->two->path;
2751 if (strcmp(ren1_dst, ren2_dst) != 0)
2752 BUG("ren1_dst != ren2_dst");
2754 clean_merge = 0;
2755 ren2->processed = 1;
2757 * BUG: We should only mark src_entry as processed
2758 * if we are not dealing with a rename + add-source
2759 * case.
2761 ren2->src_entry->processed = 1;
2763 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2764 opt, ren1, ren2);
2765 } else {
2766 /* Renamed in 1, maybe changed in 2 */
2767 /* we only use sha1 and mode of these */
2768 struct diff_filespec src_other, dst_other;
2769 int try_merge;
2772 * unpack_trees loads entries from common-commit
2773 * into stage 1, from head-commit into stage 2, and
2774 * from merge-commit into stage 3. We keep track
2775 * of which side corresponds to the rename.
2777 int renamed_stage = a_renames == renames1 ? 2 : 3;
2778 int other_stage = a_renames == renames1 ? 3 : 2;
2780 /* BUG: We should only remove ren1_src in the base
2781 * stage and in other_stage (think of rename +
2782 * add-source case).
2784 remove_file(opt, 1, ren1_src,
2785 renamed_stage == 2 || !was_tracked(opt, ren1_src));
2787 oidcpy(&src_other.oid,
2788 &ren1->src_entry->stages[other_stage].oid);
2789 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2790 oidcpy(&dst_other.oid,
2791 &ren1->dst_entry->stages[other_stage].oid);
2792 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2793 try_merge = 0;
2795 if (oideq(&src_other.oid, &null_oid) &&
2796 ren1->dir_rename_original_type == 'A') {
2797 setup_rename_conflict_info(RENAME_VIA_DIR,
2798 opt, ren1, NULL);
2799 } else if (oideq(&src_other.oid, &null_oid)) {
2800 setup_rename_conflict_info(RENAME_DELETE,
2801 opt, ren1, NULL);
2802 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2803 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2805 * Added file on the other side identical to
2806 * the file being renamed: clean merge.
2807 * Also, there is no need to overwrite the
2808 * file already in the working copy, so call
2809 * update_file_flags() instead of
2810 * update_file().
2812 if (update_file_flags(opt,
2813 ren1->pair->two,
2814 ren1_dst,
2815 1, /* update_cache */
2816 0 /* update_wd */))
2817 clean_merge = -1;
2818 } else if (!oideq(&dst_other.oid, &null_oid)) {
2820 * Probably not a clean merge, but it's
2821 * premature to set clean_merge to 0 here,
2822 * because if the rename merges cleanly and
2823 * the merge exactly matches the newly added
2824 * file, then the merge will be clean.
2826 setup_rename_conflict_info(RENAME_ADD,
2827 opt, ren1, NULL);
2828 } else
2829 try_merge = 1;
2831 if (clean_merge < 0)
2832 goto cleanup_and_return;
2833 if (try_merge) {
2834 struct diff_filespec *o, *a, *b;
2835 src_other.path = (char *)ren1_src;
2837 o = ren1->pair->one;
2838 if (a_renames == renames1) {
2839 a = ren1->pair->two;
2840 b = &src_other;
2841 } else {
2842 b = ren1->pair->two;
2843 a = &src_other;
2845 update_entry(ren1->dst_entry, o, a, b);
2846 setup_rename_conflict_info(RENAME_NORMAL,
2847 opt, ren1, NULL);
2851 cleanup_and_return:
2852 string_list_clear(&a_by_dst, 0);
2853 string_list_clear(&b_by_dst, 0);
2855 return clean_merge;
2858 struct rename_info {
2859 struct string_list *head_renames;
2860 struct string_list *merge_renames;
2863 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2864 struct hashmap *dir_renames)
2866 struct hashmap_iter iter;
2867 struct dir_rename_entry *e;
2869 hashmap_for_each_entry(dir_renames, &iter, e,
2870 ent /* member name */) {
2871 free(e->dir);
2872 strbuf_release(&e->new_dir);
2873 /* possible_new_dirs already cleared in get_directory_renames */
2875 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2876 free(dir_renames);
2878 free(pairs->queue);
2879 free(pairs);
2882 static int detect_and_process_renames(struct merge_options *opt,
2883 struct tree *common,
2884 struct tree *head,
2885 struct tree *merge,
2886 struct string_list *entries,
2887 struct rename_info *ri)
2889 struct diff_queue_struct *head_pairs, *merge_pairs;
2890 struct hashmap *dir_re_head, *dir_re_merge;
2891 int clean = 1;
2893 ri->head_renames = NULL;
2894 ri->merge_renames = NULL;
2896 if (!merge_detect_rename(opt))
2897 return 1;
2899 head_pairs = get_diffpairs(opt, common, head);
2900 merge_pairs = get_diffpairs(opt, common, merge);
2902 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2903 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2904 !opt->priv->call_depth)) {
2905 dir_re_head = get_directory_renames(head_pairs);
2906 dir_re_merge = get_directory_renames(merge_pairs);
2908 handle_directory_level_conflicts(opt,
2909 dir_re_head, head,
2910 dir_re_merge, merge);
2911 } else {
2912 dir_re_head = xmalloc(sizeof(*dir_re_head));
2913 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2914 dir_rename_init(dir_re_head);
2915 dir_rename_init(dir_re_merge);
2918 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2919 dir_re_merge, dir_re_head, head,
2920 common, head, merge, entries,
2921 &clean);
2922 if (clean < 0)
2923 goto cleanup;
2924 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2925 dir_re_head, dir_re_merge, merge,
2926 common, head, merge, entries,
2927 &clean);
2928 if (clean < 0)
2929 goto cleanup;
2930 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2932 cleanup:
2934 * Some cleanup is deferred until cleanup_renames() because the
2935 * data structures are still needed and referenced in
2936 * process_entry(). But there are a few things we can free now.
2938 initial_cleanup_rename(head_pairs, dir_re_head);
2939 initial_cleanup_rename(merge_pairs, dir_re_merge);
2941 return clean;
2944 static void final_cleanup_rename(struct string_list *rename)
2946 const struct rename *re;
2947 int i;
2949 if (rename == NULL)
2950 return;
2952 for (i = 0; i < rename->nr; i++) {
2953 re = rename->items[i].util;
2954 diff_free_filepair(re->pair);
2956 string_list_clear(rename, 1);
2957 free(rename);
2960 static void final_cleanup_renames(struct rename_info *re_info)
2962 final_cleanup_rename(re_info->head_renames);
2963 final_cleanup_rename(re_info->merge_renames);
2966 static int read_oid_strbuf(struct merge_options *opt,
2967 const struct object_id *oid,
2968 struct strbuf *dst)
2970 void *buf;
2971 enum object_type type;
2972 unsigned long size;
2973 buf = read_object_file(oid, &type, &size);
2974 if (!buf)
2975 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
2976 if (type != OBJ_BLOB) {
2977 free(buf);
2978 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
2980 strbuf_attach(dst, buf, size, size + 1);
2981 return 0;
2984 static int blob_unchanged(struct merge_options *opt,
2985 const struct diff_filespec *o,
2986 const struct diff_filespec *a,
2987 int renormalize, const char *path)
2989 struct strbuf obuf = STRBUF_INIT;
2990 struct strbuf abuf = STRBUF_INIT;
2991 int ret = 0; /* assume changed for safety */
2992 struct index_state *idx = opt->repo->index;
2994 if (a->mode != o->mode)
2995 return 0;
2996 if (oideq(&o->oid, &a->oid))
2997 return 1;
2998 if (!renormalize)
2999 return 0;
3001 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3002 read_oid_strbuf(opt, &a->oid, &abuf))
3003 goto error_return;
3005 * Note: binary | is used so that both renormalizations are
3006 * performed. Comparison can be skipped if both files are
3007 * unchanged since their sha1s have already been compared.
3009 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3010 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3011 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3013 error_return:
3014 strbuf_release(&obuf);
3015 strbuf_release(&abuf);
3016 return ret;
3019 static int handle_modify_delete(struct merge_options *opt,
3020 const char *path,
3021 const struct diff_filespec *o,
3022 const struct diff_filespec *a,
3023 const struct diff_filespec *b)
3025 const char *modify_branch, *delete_branch;
3026 const struct diff_filespec *changed;
3028 if (is_valid(a)) {
3029 modify_branch = opt->branch1;
3030 delete_branch = opt->branch2;
3031 changed = a;
3032 } else {
3033 modify_branch = opt->branch2;
3034 delete_branch = opt->branch1;
3035 changed = b;
3038 return handle_change_delete(opt,
3039 path, NULL,
3040 o, changed,
3041 modify_branch, delete_branch,
3042 _("modify"), _("modified"));
3045 static int handle_content_merge(struct merge_file_info *mfi,
3046 struct merge_options *opt,
3047 const char *path,
3048 int is_dirty,
3049 const struct diff_filespec *o,
3050 const struct diff_filespec *a,
3051 const struct diff_filespec *b,
3052 struct rename_conflict_info *ci)
3054 const char *reason = _("content");
3055 unsigned df_conflict_remains = 0;
3057 if (!is_valid(o))
3058 reason = _("add/add");
3060 assert(o->path && a->path && b->path);
3061 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3062 S_ISGITLINK(ci->ren1->pair->two->mode)))
3063 df_conflict_remains = 1;
3065 if (merge_mode_and_contents(opt, o, a, b, path,
3066 opt->branch1, opt->branch2,
3067 opt->priv->call_depth * 2, mfi))
3068 return -1;
3071 * We can skip updating the working tree file iff:
3072 * a) The merge is clean
3073 * b) The merge matches what was in HEAD (content, mode, pathname)
3074 * c) The target path is usable (i.e. not involved in D/F conflict)
3076 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3077 !df_conflict_remains) {
3078 int pos;
3079 struct cache_entry *ce;
3081 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3082 if (add_cacheinfo(opt, &mfi->blob, path,
3083 0, (!opt->priv->call_depth && !is_dirty), 0))
3084 return -1;
3086 * However, add_cacheinfo() will delete the old cache entry
3087 * and add a new one. We need to copy over any skip_worktree
3088 * flag to avoid making the file appear as if it were
3089 * deleted by the user.
3091 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3092 ce = opt->priv->orig_index.cache[pos];
3093 if (ce_skip_worktree(ce)) {
3094 pos = index_name_pos(opt->repo->index, path, strlen(path));
3095 ce = opt->repo->index->cache[pos];
3096 ce->ce_flags |= CE_SKIP_WORKTREE;
3098 return mfi->clean;
3101 if (!mfi->clean) {
3102 if (S_ISGITLINK(mfi->blob.mode))
3103 reason = _("submodule");
3104 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3105 reason, path);
3106 if (ci && !df_conflict_remains)
3107 if (update_stages(opt, path, o, a, b))
3108 return -1;
3111 if (df_conflict_remains || is_dirty) {
3112 char *new_path;
3113 if (opt->priv->call_depth) {
3114 remove_file_from_index(opt->repo->index, path);
3115 } else {
3116 if (!mfi->clean) {
3117 if (update_stages(opt, path, o, a, b))
3118 return -1;
3119 } else {
3120 int file_from_stage2 = was_tracked(opt, path);
3122 if (update_stages(opt, path, NULL,
3123 file_from_stage2 ? &mfi->blob : NULL,
3124 file_from_stage2 ? NULL : &mfi->blob))
3125 return -1;
3129 new_path = unique_path(opt, path, ci->ren1->branch);
3130 if (is_dirty) {
3131 output(opt, 1, _("Refusing to lose dirty file at %s"),
3132 path);
3134 output(opt, 1, _("Adding as %s instead"), new_path);
3135 if (update_file(opt, 0, &mfi->blob, new_path)) {
3136 free(new_path);
3137 return -1;
3139 free(new_path);
3140 mfi->clean = 0;
3141 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3142 return -1;
3143 return !is_dirty && mfi->clean;
3146 static int handle_rename_normal(struct merge_options *opt,
3147 const char *path,
3148 const struct diff_filespec *o,
3149 const struct diff_filespec *a,
3150 const struct diff_filespec *b,
3151 struct rename_conflict_info *ci)
3153 struct rename *ren = ci->ren1;
3154 struct merge_file_info mfi;
3155 int clean;
3156 int side = (ren->branch == opt->branch1 ? 2 : 3);
3158 /* Merge the content and write it out */
3159 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3160 o, a, b, ci);
3162 if (clean &&
3163 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3164 ren->dir_rename_original_dest) {
3165 if (update_stages(opt, path,
3166 NULL,
3167 side == 2 ? &mfi.blob : NULL,
3168 side == 2 ? NULL : &mfi.blob))
3169 return -1;
3170 clean = 0; /* not clean, but conflicted */
3172 return clean;
3175 static void dir_rename_warning(const char *msg,
3176 int is_add,
3177 int clean,
3178 struct merge_options *opt,
3179 struct rename *ren)
3181 const char *other_branch;
3182 other_branch = (ren->branch == opt->branch1 ?
3183 opt->branch2 : opt->branch1);
3184 if (is_add) {
3185 output(opt, clean ? 2 : 1, msg,
3186 ren->pair->one->path, ren->branch,
3187 other_branch, ren->pair->two->path);
3188 return;
3190 output(opt, clean ? 2 : 1, msg,
3191 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3192 other_branch, ren->pair->two->path);
3194 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3195 struct rename *ren)
3197 const char *msg;
3198 int clean = 1, is_add;
3200 if (!ren)
3201 return clean;
3203 /* Return early if ren was not affected/created by a directory rename */
3204 if (!ren->dir_rename_original_dest)
3205 return clean;
3207 /* Sanity checks */
3208 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3209 assert(ren->dir_rename_original_type == 'A' ||
3210 ren->dir_rename_original_type == 'R');
3212 /* Check whether to treat directory renames as a conflict */
3213 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3215 is_add = (ren->dir_rename_original_type == 'A');
3216 if (ren->dir_rename_original_type == 'A' && clean) {
3217 msg = _("Path updated: %s added in %s inside a "
3218 "directory that was renamed in %s; moving it to %s.");
3219 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3220 msg = _("CONFLICT (file location): %s added in %s "
3221 "inside a directory that was renamed in %s, "
3222 "suggesting it should perhaps be moved to %s.");
3223 } else if (ren->dir_rename_original_type == 'R' && clean) {
3224 msg = _("Path updated: %s renamed to %s in %s, inside a "
3225 "directory that was renamed in %s; moving it to %s.");
3226 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3227 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3228 "inside a directory that was renamed in %s, "
3229 "suggesting it should perhaps be moved to %s.");
3230 } else {
3231 BUG("Impossible dir_rename_original_type/clean combination");
3233 dir_rename_warning(msg, is_add, clean, opt, ren);
3235 return clean;
3238 /* Per entry merge function */
3239 static int process_entry(struct merge_options *opt,
3240 const char *path, struct stage_data *entry)
3242 int clean_merge = 1;
3243 int normalize = opt->renormalize;
3245 struct diff_filespec *o = &entry->stages[1];
3246 struct diff_filespec *a = &entry->stages[2];
3247 struct diff_filespec *b = &entry->stages[3];
3248 int o_valid = is_valid(o);
3249 int a_valid = is_valid(a);
3250 int b_valid = is_valid(b);
3251 o->path = a->path = b->path = (char*)path;
3253 entry->processed = 1;
3254 if (entry->rename_conflict_info) {
3255 struct rename_conflict_info *ci = entry->rename_conflict_info;
3256 struct diff_filespec *temp;
3257 int path_clean;
3259 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3260 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3263 * For cases with a single rename, {o,a,b}->path have all been
3264 * set to the rename target path; we need to set two of these
3265 * back to the rename source.
3266 * For rename/rename conflicts, we'll manually fix paths below.
3268 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3269 o->path = temp->path = ci->ren1->pair->one->path;
3270 if (ci->ren2) {
3271 assert(opt->branch1 == ci->ren1->branch);
3274 switch (ci->rename_type) {
3275 case RENAME_NORMAL:
3276 case RENAME_ONE_FILE_TO_ONE:
3277 clean_merge = handle_rename_normal(opt, path, o, a, b,
3278 ci);
3279 break;
3280 case RENAME_VIA_DIR:
3281 clean_merge = handle_rename_via_dir(opt, ci);
3282 break;
3283 case RENAME_ADD:
3285 * Probably unclean merge, but if the renamed file
3286 * merges cleanly and the result can then be
3287 * two-way merged cleanly with the added file, I
3288 * guess it's a clean merge?
3290 clean_merge = handle_rename_add(opt, ci);
3291 break;
3292 case RENAME_DELETE:
3293 clean_merge = 0;
3294 if (handle_rename_delete(opt, ci))
3295 clean_merge = -1;
3296 break;
3297 case RENAME_ONE_FILE_TO_TWO:
3299 * Manually fix up paths; note:
3300 * ren[12]->pair->one->path are equal.
3302 o->path = ci->ren1->pair->one->path;
3303 a->path = ci->ren1->pair->two->path;
3304 b->path = ci->ren2->pair->two->path;
3306 clean_merge = 0;
3307 if (handle_rename_rename_1to2(opt, ci))
3308 clean_merge = -1;
3309 break;
3310 case RENAME_TWO_FILES_TO_ONE:
3312 * Manually fix up paths; note,
3313 * ren[12]->pair->two->path are actually equal.
3315 o->path = NULL;
3316 a->path = ci->ren1->pair->two->path;
3317 b->path = ci->ren2->pair->two->path;
3320 * Probably unclean merge, but if the two renamed
3321 * files merge cleanly and the two resulting files
3322 * can then be two-way merged cleanly, I guess it's
3323 * a clean merge?
3325 clean_merge = handle_rename_rename_2to1(opt, ci);
3326 break;
3327 default:
3328 entry->processed = 0;
3329 break;
3331 if (path_clean < clean_merge)
3332 clean_merge = path_clean;
3333 } else if (o_valid && (!a_valid || !b_valid)) {
3334 /* Case A: Deleted in one */
3335 if ((!a_valid && !b_valid) ||
3336 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3337 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3338 /* Deleted in both or deleted in one and
3339 * unchanged in the other */
3340 if (a_valid)
3341 output(opt, 2, _("Removing %s"), path);
3342 /* do not touch working file if it did not exist */
3343 remove_file(opt, 1, path, !a_valid);
3344 } else {
3345 /* Modify/delete; deleted side may have put a directory in the way */
3346 clean_merge = 0;
3347 if (handle_modify_delete(opt, path, o, a, b))
3348 clean_merge = -1;
3350 } else if ((!o_valid && a_valid && !b_valid) ||
3351 (!o_valid && !a_valid && b_valid)) {
3352 /* Case B: Added in one. */
3353 /* [nothing|directory] -> ([nothing|directory], file) */
3355 const char *add_branch;
3356 const char *other_branch;
3357 const char *conf;
3358 const struct diff_filespec *contents;
3360 if (a_valid) {
3361 add_branch = opt->branch1;
3362 other_branch = opt->branch2;
3363 contents = a;
3364 conf = _("file/directory");
3365 } else {
3366 add_branch = opt->branch2;
3367 other_branch = opt->branch1;
3368 contents = b;
3369 conf = _("directory/file");
3371 if (dir_in_way(opt->repo->index, path,
3372 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3373 0)) {
3374 char *new_path = unique_path(opt, path, add_branch);
3375 clean_merge = 0;
3376 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3377 "Adding %s as %s"),
3378 conf, path, other_branch, path, new_path);
3379 if (update_file(opt, 0, contents, new_path))
3380 clean_merge = -1;
3381 else if (opt->priv->call_depth)
3382 remove_file_from_index(opt->repo->index, path);
3383 free(new_path);
3384 } else {
3385 output(opt, 2, _("Adding %s"), path);
3386 /* do not overwrite file if already present */
3387 if (update_file_flags(opt, contents, path, 1, !a_valid))
3388 clean_merge = -1;
3390 } else if (a_valid && b_valid) {
3391 if (!o_valid) {
3392 /* Case C: Added in both (check for same permissions) */
3393 output(opt, 1,
3394 _("CONFLICT (add/add): Merge conflict in %s"),
3395 path);
3396 clean_merge = handle_file_collision(opt,
3397 path, NULL, NULL,
3398 opt->branch1,
3399 opt->branch2,
3400 a, b);
3401 } else {
3402 /* case D: Modified in both, but differently. */
3403 struct merge_file_info mfi;
3404 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3405 clean_merge = handle_content_merge(&mfi, opt, path,
3406 is_dirty,
3407 o, a, b, NULL);
3409 } else if (!o_valid && !a_valid && !b_valid) {
3411 * this entry was deleted altogether. a_mode == 0 means
3412 * we had that path and want to actively remove it.
3414 remove_file(opt, 1, path, !a->mode);
3415 } else
3416 BUG("fatal merge failure, shouldn't happen.");
3418 return clean_merge;
3421 static int merge_trees_internal(struct merge_options *opt,
3422 struct tree *head,
3423 struct tree *merge,
3424 struct tree *merge_base,
3425 struct tree **result)
3427 struct index_state *istate = opt->repo->index;
3428 int code, clean;
3430 if (opt->subtree_shift) {
3431 merge = shift_tree_object(opt->repo, head, merge,
3432 opt->subtree_shift);
3433 merge_base = shift_tree_object(opt->repo, head, merge_base,
3434 opt->subtree_shift);
3437 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3438 output(opt, 0, _("Already up to date!"));
3439 *result = head;
3440 return 1;
3443 code = unpack_trees_start(opt, merge_base, head, merge);
3445 if (code != 0) {
3446 if (show(opt, 4) || opt->priv->call_depth)
3447 err(opt, _("merging of trees %s and %s failed"),
3448 oid_to_hex(&head->object.oid),
3449 oid_to_hex(&merge->object.oid));
3450 unpack_trees_finish(opt);
3451 return -1;
3454 if (unmerged_index(istate)) {
3455 struct string_list *entries;
3456 struct rename_info re_info;
3457 int i;
3459 * Only need the hashmap while processing entries, so
3460 * initialize it here and free it when we are done running
3461 * through the entries. Keeping it in the merge_options as
3462 * opposed to decaring a local hashmap is for convenience
3463 * so that we don't have to pass it to around.
3465 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3466 NULL, 512);
3467 get_files_dirs(opt, head);
3468 get_files_dirs(opt, merge);
3470 entries = get_unmerged(opt->repo->index);
3471 clean = detect_and_process_renames(opt, merge_base, head, merge,
3472 entries, &re_info);
3473 record_df_conflict_files(opt, entries);
3474 if (clean < 0)
3475 goto cleanup;
3476 for (i = entries->nr-1; 0 <= i; i--) {
3477 const char *path = entries->items[i].string;
3478 struct stage_data *e = entries->items[i].util;
3479 if (!e->processed) {
3480 int ret = process_entry(opt, path, e);
3481 if (!ret)
3482 clean = 0;
3483 else if (ret < 0) {
3484 clean = ret;
3485 goto cleanup;
3489 for (i = 0; i < entries->nr; i++) {
3490 struct stage_data *e = entries->items[i].util;
3491 if (!e->processed)
3492 BUG("unprocessed path??? %s",
3493 entries->items[i].string);
3496 cleanup:
3497 final_cleanup_renames(&re_info);
3499 string_list_clear(entries, 1);
3500 free(entries);
3502 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3503 struct path_hashmap_entry, e);
3505 if (clean < 0) {
3506 unpack_trees_finish(opt);
3507 return clean;
3510 else
3511 clean = 1;
3513 unpack_trees_finish(opt);
3515 if (opt->priv->call_depth &&
3516 !(*result = write_in_core_index_as_tree(opt->repo)))
3517 return -1;
3519 return clean;
3523 * Merge the commits h1 and h2, returning a flag (int) indicating the
3524 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3525 * virtual commit and write its location to *result.
3527 static int merge_recursive_internal(struct merge_options *opt,
3528 struct commit *h1,
3529 struct commit *h2,
3530 struct commit_list *merge_bases,
3531 struct commit **result)
3533 struct commit_list *iter;
3534 struct commit *merged_merge_bases;
3535 struct tree *result_tree;
3536 int clean;
3537 const char *ancestor_name;
3538 struct strbuf merge_base_abbrev = STRBUF_INIT;
3540 if (show(opt, 4)) {
3541 output(opt, 4, _("Merging:"));
3542 output_commit_title(opt, h1);
3543 output_commit_title(opt, h2);
3546 if (!merge_bases) {
3547 merge_bases = get_merge_bases(h1, h2);
3548 merge_bases = reverse_commit_list(merge_bases);
3551 if (show(opt, 5)) {
3552 unsigned cnt = commit_list_count(merge_bases);
3554 output(opt, 5, Q_("found %u common ancestor:",
3555 "found %u common ancestors:", cnt), cnt);
3556 for (iter = merge_bases; iter; iter = iter->next)
3557 output_commit_title(opt, iter->item);
3560 merged_merge_bases = pop_commit(&merge_bases);
3561 if (merged_merge_bases == NULL) {
3562 /* if there is no common ancestor, use an empty tree */
3563 struct tree *tree;
3565 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3566 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3567 "ancestor");
3568 ancestor_name = "empty tree";
3569 } else if (opt->ancestor && !opt->priv->call_depth) {
3570 ancestor_name = opt->ancestor;
3571 } else if (merge_bases) {
3572 ancestor_name = "merged common ancestors";
3573 } else {
3574 strbuf_add_unique_abbrev(&merge_base_abbrev,
3575 &merged_merge_bases->object.oid,
3576 DEFAULT_ABBREV);
3577 ancestor_name = merge_base_abbrev.buf;
3580 for (iter = merge_bases; iter; iter = iter->next) {
3581 const char *saved_b1, *saved_b2;
3582 opt->priv->call_depth++;
3584 * When the merge fails, the result contains files
3585 * with conflict markers. The cleanness flag is
3586 * ignored (unless indicating an error), it was never
3587 * actually used, as result of merge_trees has always
3588 * overwritten it: the committed "conflicts" were
3589 * already resolved.
3591 discard_index(opt->repo->index);
3592 saved_b1 = opt->branch1;
3593 saved_b2 = opt->branch2;
3594 opt->branch1 = "Temporary merge branch 1";
3595 opt->branch2 = "Temporary merge branch 2";
3596 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3597 NULL, &merged_merge_bases) < 0)
3598 return -1;
3599 opt->branch1 = saved_b1;
3600 opt->branch2 = saved_b2;
3601 opt->priv->call_depth--;
3603 if (!merged_merge_bases)
3604 return err(opt, _("merge returned no commit"));
3607 discard_index(opt->repo->index);
3608 if (!opt->priv->call_depth)
3609 repo_read_index(opt->repo);
3611 opt->ancestor = ancestor_name;
3612 clean = merge_trees_internal(opt,
3613 repo_get_commit_tree(opt->repo, h1),
3614 repo_get_commit_tree(opt->repo, h2),
3615 repo_get_commit_tree(opt->repo,
3616 merged_merge_bases),
3617 &result_tree);
3618 strbuf_release(&merge_base_abbrev);
3619 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3620 if (clean < 0) {
3621 flush_output(opt);
3622 return clean;
3625 if (opt->priv->call_depth) {
3626 *result = make_virtual_commit(opt->repo, result_tree,
3627 "merged tree");
3628 commit_list_insert(h1, &(*result)->parents);
3629 commit_list_insert(h2, &(*result)->parents->next);
3631 return clean;
3634 static int merge_start(struct merge_options *opt, struct tree *head)
3636 struct strbuf sb = STRBUF_INIT;
3638 /* Sanity checks on opt */
3639 assert(opt->repo);
3641 assert(opt->branch1 && opt->branch2);
3643 assert(opt->detect_renames >= -1 &&
3644 opt->detect_renames <= DIFF_DETECT_COPY);
3645 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3646 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3647 assert(opt->rename_limit >= -1);
3648 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3649 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3651 assert(opt->xdl_opts >= 0);
3652 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3653 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3655 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3656 assert(opt->buffer_output <= 2);
3657 assert(opt->obuf.len == 0);
3659 assert(opt->priv == NULL);
3661 /* Sanity check on repo state; index must match head */
3662 if (repo_index_has_changes(opt->repo, head, &sb)) {
3663 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3664 sb.buf);
3665 strbuf_release(&sb);
3666 return -1;
3669 opt->priv = xcalloc(1, sizeof(*opt->priv));
3670 string_list_init(&opt->priv->df_conflict_file_set, 1);
3671 return 0;
3674 static void merge_finalize(struct merge_options *opt)
3676 flush_output(opt);
3677 if (!opt->priv->call_depth && opt->buffer_output < 2)
3678 strbuf_release(&opt->obuf);
3679 if (show(opt, 2))
3680 diff_warn_rename_limit("merge.renamelimit",
3681 opt->priv->needed_rename_limit, 0);
3682 FREE_AND_NULL(opt->priv);
3685 int merge_trees(struct merge_options *opt,
3686 struct tree *head,
3687 struct tree *merge,
3688 struct tree *merge_base)
3690 int clean;
3691 struct tree *ignored;
3693 assert(opt->ancestor != NULL);
3695 if (merge_start(opt, head))
3696 return -1;
3697 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3698 merge_finalize(opt);
3700 return clean;
3703 int merge_recursive(struct merge_options *opt,
3704 struct commit *h1,
3705 struct commit *h2,
3706 struct commit_list *merge_bases,
3707 struct commit **result)
3709 int clean;
3711 assert(opt->ancestor == NULL ||
3712 !strcmp(opt->ancestor, "constructed merge base"));
3714 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3715 return -1;
3716 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3717 merge_finalize(opt);
3719 return clean;
3722 static struct commit *get_ref(struct repository *repo,
3723 const struct object_id *oid,
3724 const char *name)
3726 struct object *object;
3728 object = deref_tag(repo, parse_object(repo, oid),
3729 name, strlen(name));
3730 if (!object)
3731 return NULL;
3732 if (object->type == OBJ_TREE)
3733 return make_virtual_commit(repo, (struct tree*)object, name);
3734 if (object->type != OBJ_COMMIT)
3735 return NULL;
3736 if (parse_commit((struct commit *)object))
3737 return NULL;
3738 return (struct commit *)object;
3741 int merge_recursive_generic(struct merge_options *opt,
3742 const struct object_id *head,
3743 const struct object_id *merge,
3744 int num_merge_bases,
3745 const struct object_id **merge_bases,
3746 struct commit **result)
3748 int clean;
3749 struct lock_file lock = LOCK_INIT;
3750 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3751 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3752 struct commit_list *ca = NULL;
3754 if (merge_bases) {
3755 int i;
3756 for (i = 0; i < num_merge_bases; ++i) {
3757 struct commit *base;
3758 if (!(base = get_ref(opt->repo, merge_bases[i],
3759 oid_to_hex(merge_bases[i]))))
3760 return err(opt, _("Could not parse object '%s'"),
3761 oid_to_hex(merge_bases[i]));
3762 commit_list_insert(base, &ca);
3764 if (num_merge_bases == 1)
3765 opt->ancestor = "constructed merge base";
3768 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3769 clean = merge_recursive(opt, head_commit, next_commit, ca,
3770 result);
3771 if (clean < 0) {
3772 rollback_lock_file(&lock);
3773 return clean;
3776 if (write_locked_index(opt->repo->index, &lock,
3777 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3778 return err(opt, _("Unable to write index."));
3780 return clean ? 0 : 1;
3783 static void merge_recursive_config(struct merge_options *opt)
3785 char *value = NULL;
3786 int renormalize = 0;
3787 git_config_get_int("merge.verbosity", &opt->verbosity);
3788 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3789 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3790 git_config_get_bool("merge.renormalize", &renormalize);
3791 opt->renormalize = renormalize;
3792 if (!git_config_get_string("diff.renames", &value)) {
3793 opt->detect_renames = git_config_rename("diff.renames", value);
3794 free(value);
3796 if (!git_config_get_string("merge.renames", &value)) {
3797 opt->detect_renames = git_config_rename("merge.renames", value);
3798 free(value);
3800 if (!git_config_get_string("merge.directoryrenames", &value)) {
3801 int boolval = git_parse_maybe_bool(value);
3802 if (0 <= boolval) {
3803 opt->detect_directory_renames = boolval ?
3804 MERGE_DIRECTORY_RENAMES_TRUE :
3805 MERGE_DIRECTORY_RENAMES_NONE;
3806 } else if (!strcasecmp(value, "conflict")) {
3807 opt->detect_directory_renames =
3808 MERGE_DIRECTORY_RENAMES_CONFLICT;
3809 } /* avoid erroring on values from future versions of git */
3810 free(value);
3812 git_config(git_xmerge_config, NULL);
3815 void init_merge_options(struct merge_options *opt,
3816 struct repository *repo)
3818 const char *merge_verbosity;
3819 memset(opt, 0, sizeof(struct merge_options));
3821 opt->repo = repo;
3823 opt->detect_renames = -1;
3824 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3825 opt->rename_limit = -1;
3827 opt->verbosity = 2;
3828 opt->buffer_output = 1;
3829 strbuf_init(&opt->obuf, 0);
3831 opt->renormalize = 0;
3833 merge_recursive_config(opt);
3834 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3835 if (merge_verbosity)
3836 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3837 if (opt->verbosity >= 5)
3838 opt->buffer_output = 0;
3841 int parse_merge_opt(struct merge_options *opt, const char *s)
3843 const char *arg;
3845 if (!s || !*s)
3846 return -1;
3847 if (!strcmp(s, "ours"))
3848 opt->recursive_variant = MERGE_VARIANT_OURS;
3849 else if (!strcmp(s, "theirs"))
3850 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3851 else if (!strcmp(s, "subtree"))
3852 opt->subtree_shift = "";
3853 else if (skip_prefix(s, "subtree=", &arg))
3854 opt->subtree_shift = arg;
3855 else if (!strcmp(s, "patience"))
3856 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3857 else if (!strcmp(s, "histogram"))
3858 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3859 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3860 long value = parse_algorithm_value(arg);
3861 if (value < 0)
3862 return -1;
3863 /* clear out previous settings */
3864 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3865 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3866 opt->xdl_opts |= value;
3868 else if (!strcmp(s, "ignore-space-change"))
3869 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3870 else if (!strcmp(s, "ignore-all-space"))
3871 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3872 else if (!strcmp(s, "ignore-space-at-eol"))
3873 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3874 else if (!strcmp(s, "ignore-cr-at-eol"))
3875 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3876 else if (!strcmp(s, "renormalize"))
3877 opt->renormalize = 1;
3878 else if (!strcmp(s, "no-renormalize"))
3879 opt->renormalize = 0;
3880 else if (!strcmp(s, "no-renames"))
3881 opt->detect_renames = 0;
3882 else if (!strcmp(s, "find-renames")) {
3883 opt->detect_renames = 1;
3884 opt->rename_score = 0;
3886 else if (skip_prefix(s, "find-renames=", &arg) ||
3887 skip_prefix(s, "rename-threshold=", &arg)) {
3888 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3889 return -1;
3890 opt->detect_renames = 1;
3893 * Please update $__git_merge_strategy_options in
3894 * git-completion.bash when you add new options
3896 else
3897 return -1;
3898 return 0;