merge-recursive: use xstrdup() instead of fixed buffer
[git.git] / merge-recursive.c
blob7ea70c4a9d11dea43ebdad1ba8718410793c263a
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 "config.h"
8 #include "advice.h"
9 #include "lockfile.h"
10 #include "cache-tree.h"
11 #include "commit.h"
12 #include "blob.h"
13 #include "builtin.h"
14 #include "tree-walk.h"
15 #include "diff.h"
16 #include "diffcore.h"
17 #include "tag.h"
18 #include "unpack-trees.h"
19 #include "string-list.h"
20 #include "xdiff-interface.h"
21 #include "ll-merge.h"
22 #include "attr.h"
23 #include "merge-recursive.h"
24 #include "dir.h"
25 #include "submodule.h"
27 struct path_hashmap_entry {
28 struct hashmap_entry e;
29 char path[FLEX_ARRAY];
32 static int path_hashmap_cmp(const void *cmp_data,
33 const void *entry,
34 const void *entry_or_key,
35 const void *keydata)
37 const struct path_hashmap_entry *a = entry;
38 const struct path_hashmap_entry *b = entry_or_key;
39 const char *key = keydata;
41 if (ignore_case)
42 return strcasecmp(a->path, key ? key : b->path);
43 else
44 return strcmp(a->path, key ? key : b->path);
47 static unsigned int path_hash(const char *path)
49 return ignore_case ? strihash(path) : strhash(path);
52 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
53 char *dir)
55 struct dir_rename_entry key;
57 if (dir == NULL)
58 return NULL;
59 hashmap_entry_init(&key, strhash(dir));
60 key.dir = dir;
61 return hashmap_get(hashmap, &key, NULL);
64 static int dir_rename_cmp(const void *unused_cmp_data,
65 const void *entry,
66 const void *entry_or_key,
67 const void *unused_keydata)
69 const struct dir_rename_entry *e1 = entry;
70 const struct dir_rename_entry *e2 = entry_or_key;
72 return strcmp(e1->dir, e2->dir);
75 static void dir_rename_init(struct hashmap *map)
77 hashmap_init(map, dir_rename_cmp, NULL, 0);
80 static void dir_rename_entry_init(struct dir_rename_entry *entry,
81 char *directory)
83 hashmap_entry_init(entry, strhash(directory));
84 entry->dir = directory;
85 entry->non_unique_new_dir = 0;
86 strbuf_init(&entry->new_dir, 0);
87 string_list_init(&entry->possible_new_dirs, 0);
90 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
91 char *target_file)
93 struct collision_entry key;
95 hashmap_entry_init(&key, strhash(target_file));
96 key.target_file = target_file;
97 return hashmap_get(hashmap, &key, NULL);
100 static int collision_cmp(void *unused_cmp_data,
101 const struct collision_entry *e1,
102 const struct collision_entry *e2,
103 const void *unused_keydata)
105 return strcmp(e1->target_file, e2->target_file);
108 static void collision_init(struct hashmap *map)
110 hashmap_init(map, (hashmap_cmp_fn) collision_cmp, NULL, 0);
113 static void flush_output(struct merge_options *o)
115 if (o->buffer_output < 2 && o->obuf.len) {
116 fputs(o->obuf.buf, stdout);
117 strbuf_reset(&o->obuf);
121 static int err(struct merge_options *o, const char *err, ...)
123 va_list params;
125 if (o->buffer_output < 2)
126 flush_output(o);
127 else {
128 strbuf_complete(&o->obuf, '\n');
129 strbuf_addstr(&o->obuf, "error: ");
131 va_start(params, err);
132 strbuf_vaddf(&o->obuf, err, params);
133 va_end(params);
134 if (o->buffer_output > 1)
135 strbuf_addch(&o->obuf, '\n');
136 else {
137 error("%s", o->obuf.buf);
138 strbuf_reset(&o->obuf);
141 return -1;
144 static struct tree *shift_tree_object(struct tree *one, struct tree *two,
145 const char *subtree_shift)
147 struct object_id shifted;
149 if (!*subtree_shift) {
150 shift_tree(&one->object.oid, &two->object.oid, &shifted, 0);
151 } else {
152 shift_tree_by(&one->object.oid, &two->object.oid, &shifted,
153 subtree_shift);
155 if (!oidcmp(&two->object.oid, &shifted))
156 return two;
157 return lookup_tree(&shifted);
160 static struct commit *make_virtual_commit(struct tree *tree, const char *comment)
162 struct commit *commit = alloc_commit_node();
164 set_merge_remote_desc(commit, comment, (struct object *)commit);
165 commit->tree = tree;
166 commit->object.parsed = 1;
167 return commit;
171 * Since we use get_tree_entry(), which does not put the read object into
172 * the object pool, we cannot rely on a == b.
174 static int oid_eq(const struct object_id *a, const struct object_id *b)
176 if (!a && !b)
177 return 2;
178 return a && b && oidcmp(a, b) == 0;
181 enum rename_type {
182 RENAME_NORMAL = 0,
183 RENAME_DIR,
184 RENAME_DELETE,
185 RENAME_ONE_FILE_TO_ONE,
186 RENAME_ONE_FILE_TO_TWO,
187 RENAME_TWO_FILES_TO_ONE
190 struct rename_conflict_info {
191 enum rename_type rename_type;
192 struct diff_filepair *pair1;
193 struct diff_filepair *pair2;
194 const char *branch1;
195 const char *branch2;
196 struct stage_data *dst_entry1;
197 struct stage_data *dst_entry2;
198 struct diff_filespec ren1_other;
199 struct diff_filespec ren2_other;
203 * Since we want to write the index eventually, we cannot reuse the index
204 * for these (temporary) data.
206 struct stage_data {
207 struct {
208 unsigned mode;
209 struct object_id oid;
210 } stages[4];
211 struct rename_conflict_info *rename_conflict_info;
212 unsigned processed:1;
215 static inline void setup_rename_conflict_info(enum rename_type rename_type,
216 struct diff_filepair *pair1,
217 struct diff_filepair *pair2,
218 const char *branch1,
219 const char *branch2,
220 struct stage_data *dst_entry1,
221 struct stage_data *dst_entry2,
222 struct merge_options *o,
223 struct stage_data *src_entry1,
224 struct stage_data *src_entry2)
226 struct rename_conflict_info *ci = xcalloc(1, sizeof(struct rename_conflict_info));
227 ci->rename_type = rename_type;
228 ci->pair1 = pair1;
229 ci->branch1 = branch1;
230 ci->branch2 = branch2;
232 ci->dst_entry1 = dst_entry1;
233 dst_entry1->rename_conflict_info = ci;
234 dst_entry1->processed = 0;
236 assert(!pair2 == !dst_entry2);
237 if (dst_entry2) {
238 ci->dst_entry2 = dst_entry2;
239 ci->pair2 = pair2;
240 dst_entry2->rename_conflict_info = ci;
243 if (rename_type == RENAME_TWO_FILES_TO_ONE) {
245 * For each rename, there could have been
246 * modifications on the side of history where that
247 * file was not renamed.
249 int ostage1 = o->branch1 == branch1 ? 3 : 2;
250 int ostage2 = ostage1 ^ 1;
252 ci->ren1_other.path = pair1->one->path;
253 oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid);
254 ci->ren1_other.mode = src_entry1->stages[ostage1].mode;
256 ci->ren2_other.path = pair2->one->path;
257 oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid);
258 ci->ren2_other.mode = src_entry2->stages[ostage2].mode;
262 static int show(struct merge_options *o, int v)
264 return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5;
267 __attribute__((format (printf, 3, 4)))
268 static void output(struct merge_options *o, int v, const char *fmt, ...)
270 va_list ap;
272 if (!show(o, v))
273 return;
275 strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
277 va_start(ap, fmt);
278 strbuf_vaddf(&o->obuf, fmt, ap);
279 va_end(ap);
281 strbuf_addch(&o->obuf, '\n');
282 if (!o->buffer_output)
283 flush_output(o);
286 static void output_commit_title(struct merge_options *o, struct commit *commit)
288 strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
289 if (commit->util)
290 strbuf_addf(&o->obuf, "virtual %s\n",
291 merge_remote_util(commit)->name);
292 else {
293 strbuf_add_unique_abbrev(&o->obuf, commit->object.oid.hash,
294 DEFAULT_ABBREV);
295 strbuf_addch(&o->obuf, ' ');
296 if (parse_commit(commit) != 0)
297 strbuf_addstr(&o->obuf, _("(bad commit)\n"));
298 else {
299 const char *title;
300 const char *msg = get_commit_buffer(commit, NULL);
301 int len = find_commit_subject(msg, &title);
302 if (len)
303 strbuf_addf(&o->obuf, "%.*s\n", len, title);
304 unuse_commit_buffer(commit, msg);
307 flush_output(o);
310 static int add_cacheinfo(struct merge_options *o,
311 unsigned int mode, const struct object_id *oid,
312 const char *path, int stage, int refresh, int options)
314 struct cache_entry *ce;
315 int ret;
317 ce = make_cache_entry(mode, oid ? oid->hash : null_sha1, path, stage, 0);
318 if (!ce)
319 return err(o, _("addinfo_cache failed for path '%s'"), path);
321 ret = add_cache_entry(ce, options);
322 if (refresh) {
323 struct cache_entry *nce;
325 nce = refresh_cache_entry(ce, CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
326 if (!nce)
327 return err(o, _("addinfo_cache failed for path '%s'"), path);
328 if (nce != ce)
329 ret = add_cache_entry(nce, options);
331 return ret;
334 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
336 parse_tree(tree);
337 init_tree_desc(desc, tree->buffer, tree->size);
340 static int git_merge_trees(struct merge_options *o,
341 struct tree *common,
342 struct tree *head,
343 struct tree *merge)
345 int rc;
346 struct tree_desc t[3];
348 memset(&o->unpack_opts, 0, sizeof(o->unpack_opts));
349 if (o->call_depth)
350 o->unpack_opts.index_only = 1;
351 else
352 o->unpack_opts.update = 1;
353 o->unpack_opts.merge = 1;
354 o->unpack_opts.head_idx = 2;
355 o->unpack_opts.fn = threeway_merge;
356 o->unpack_opts.src_index = &the_index;
357 o->unpack_opts.dst_index = &the_index;
358 setup_unpack_trees_porcelain(&o->unpack_opts, "merge");
360 init_tree_desc_from_tree(t+0, common);
361 init_tree_desc_from_tree(t+1, head);
362 init_tree_desc_from_tree(t+2, merge);
364 rc = unpack_trees(3, t, &o->unpack_opts);
366 * unpack_trees NULLifies src_index, but it's used in verify_uptodate,
367 * so set to the new index which will usually have modification
368 * timestamp info copied over.
370 o->unpack_opts.src_index = &the_index;
371 cache_tree_free(&active_cache_tree);
372 return rc;
375 struct tree *write_tree_from_memory(struct merge_options *o)
377 struct tree *result = NULL;
379 if (unmerged_cache()) {
380 int i;
381 fprintf(stderr, "BUG: There are unmerged index entries:\n");
382 for (i = 0; i < active_nr; i++) {
383 const struct cache_entry *ce = active_cache[i];
384 if (ce_stage(ce))
385 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
386 (int)ce_namelen(ce), ce->name);
388 die("BUG: unmerged index entries in merge-recursive.c");
391 if (!active_cache_tree)
392 active_cache_tree = cache_tree();
394 if (!cache_tree_fully_valid(active_cache_tree) &&
395 cache_tree_update(&the_index, 0) < 0) {
396 err(o, _("error building trees"));
397 return NULL;
400 result = lookup_tree(&active_cache_tree->oid);
402 return result;
405 static int save_files_dirs(const unsigned char *sha1,
406 struct strbuf *base, const char *path,
407 unsigned int mode, int stage, void *context)
409 struct path_hashmap_entry *entry;
410 int baselen = base->len;
411 struct merge_options *o = context;
413 strbuf_addstr(base, path);
415 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
416 hashmap_entry_init(entry, path_hash(entry->path));
417 hashmap_add(&o->current_file_dir_set, entry);
419 strbuf_setlen(base, baselen);
420 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
423 static void get_files_dirs(struct merge_options *o, struct tree *tree)
425 struct pathspec match_all;
426 memset(&match_all, 0, sizeof(match_all));
427 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
430 static int get_tree_entry_if_blob(const unsigned char *tree,
431 const char *path,
432 unsigned char *hashy,
433 unsigned int *mode_o)
435 int ret;
437 ret = get_tree_entry(tree, path, hashy, mode_o);
438 if (S_ISDIR(*mode_o)) {
439 hashcpy(hashy, null_sha1);
440 *mode_o = 0;
442 return ret;
446 * Returns an index_entry instance which doesn't have to correspond to
447 * a real cache entry in Git's index.
449 static struct stage_data *insert_stage_data(const char *path,
450 struct tree *o, struct tree *a, struct tree *b,
451 struct string_list *entries)
453 struct string_list_item *item;
454 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
455 get_tree_entry_if_blob(o->object.oid.hash, path,
456 e->stages[1].oid.hash, &e->stages[1].mode);
457 get_tree_entry_if_blob(a->object.oid.hash, path,
458 e->stages[2].oid.hash, &e->stages[2].mode);
459 get_tree_entry_if_blob(b->object.oid.hash, path,
460 e->stages[3].oid.hash, &e->stages[3].mode);
461 item = string_list_insert(entries, path);
462 item->util = e;
463 return e;
467 * Create a dictionary mapping file names to stage_data objects. The
468 * dictionary contains one entry for every path with a non-zero stage entry.
470 static struct string_list *get_unmerged(void)
472 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
473 int i;
475 unmerged->strdup_strings = 1;
477 for (i = 0; i < active_nr; i++) {
478 struct string_list_item *item;
479 struct stage_data *e;
480 const struct cache_entry *ce = active_cache[i];
481 if (!ce_stage(ce))
482 continue;
484 item = string_list_lookup(unmerged, ce->name);
485 if (!item) {
486 item = string_list_insert(unmerged, ce->name);
487 item->util = xcalloc(1, sizeof(struct stage_data));
489 e = item->util;
490 e->stages[ce_stage(ce)].mode = ce->ce_mode;
491 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
494 return unmerged;
497 static int string_list_df_name_compare(const char *one, const char *two)
499 int onelen = strlen(one);
500 int twolen = strlen(two);
502 * Here we only care that entries for D/F conflicts are
503 * adjacent, in particular with the file of the D/F conflict
504 * appearing before files below the corresponding directory.
505 * The order of the rest of the list is irrelevant for us.
507 * To achieve this, we sort with df_name_compare and provide
508 * the mode S_IFDIR so that D/F conflicts will sort correctly.
509 * We use the mode S_IFDIR for everything else for simplicity,
510 * since in other cases any changes in their order due to
511 * sorting cause no problems for us.
513 int cmp = df_name_compare(one, onelen, S_IFDIR,
514 two, twolen, S_IFDIR);
516 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
517 * that 'foo' comes before 'foo/bar'.
519 if (cmp)
520 return cmp;
521 return onelen - twolen;
524 static void record_df_conflict_files(struct merge_options *o,
525 struct string_list *entries)
527 /* If there is a D/F conflict and the file for such a conflict
528 * currently exist in the working tree, we want to allow it to be
529 * removed to make room for the corresponding directory if needed.
530 * The files underneath the directories of such D/F conflicts will
531 * be processed before the corresponding file involved in the D/F
532 * conflict. If the D/F directory ends up being removed by the
533 * merge, then we won't have to touch the D/F file. If the D/F
534 * directory needs to be written to the working copy, then the D/F
535 * file will simply be removed (in make_room_for_path()) to make
536 * room for the necessary paths. Note that if both the directory
537 * and the file need to be present, then the D/F file will be
538 * reinstated with a new unique name at the time it is processed.
540 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
541 const char *last_file = NULL;
542 int last_len = 0;
543 int i;
546 * If we're merging merge-bases, we don't want to bother with
547 * any working directory changes.
549 if (o->call_depth)
550 return;
552 /* Ensure D/F conflicts are adjacent in the entries list. */
553 for (i = 0; i < entries->nr; i++) {
554 struct string_list_item *next = &entries->items[i];
555 string_list_append(&df_sorted_entries, next->string)->util =
556 next->util;
558 df_sorted_entries.cmp = string_list_df_name_compare;
559 string_list_sort(&df_sorted_entries);
561 string_list_clear(&o->df_conflict_file_set, 1);
562 for (i = 0; i < df_sorted_entries.nr; i++) {
563 const char *path = df_sorted_entries.items[i].string;
564 int len = strlen(path);
565 struct stage_data *e = df_sorted_entries.items[i].util;
568 * Check if last_file & path correspond to a D/F conflict;
569 * i.e. whether path is last_file+'/'+<something>.
570 * If so, record that it's okay to remove last_file to make
571 * room for path and friends if needed.
573 if (last_file &&
574 len > last_len &&
575 memcmp(path, last_file, last_len) == 0 &&
576 path[last_len] == '/') {
577 string_list_insert(&o->df_conflict_file_set, last_file);
581 * Determine whether path could exist as a file in the
582 * working directory as a possible D/F conflict. This
583 * will only occur when it exists in stage 2 as a
584 * file.
586 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
587 last_file = path;
588 last_len = len;
589 } else {
590 last_file = NULL;
593 string_list_clear(&df_sorted_entries, 0);
596 struct rename {
597 struct diff_filepair *pair;
599 * Purpose of src_entry and dst_entry:
601 * If 'before' is renamed to 'after' then src_entry will contain
602 * the versions of 'before' from the merge_base, HEAD, and MERGE in
603 * stages 1, 2, and 3; dst_entry will contain the respective
604 * versions of 'after' in corresponding locations. Thus, we have a
605 * total of six modes and oids, though some will be null. (Stage 0
606 * is ignored; we're interested in handling conflicts.)
608 * Since we don't turn on break-rewrites by default, neither
609 * src_entry nor dst_entry can have all three of their stages have
610 * non-null oids, meaning at most four of the six will be non-null.
611 * Also, since this is a rename, both src_entry and dst_entry will
612 * have at least one non-null oid, meaning at least two will be
613 * non-null. Of the six oids, a typical rename will have three be
614 * non-null. Only two implies a rename/delete, and four implies a
615 * rename/add.
617 struct stage_data *src_entry;
618 struct stage_data *dst_entry;
619 unsigned add_turned_into_rename:1;
620 unsigned processed:1;
623 static int update_stages(struct merge_options *opt, const char *path,
624 const struct diff_filespec *o,
625 const struct diff_filespec *a,
626 const struct diff_filespec *b)
630 * NOTE: It is usually a bad idea to call update_stages on a path
631 * before calling update_file on that same path, since it can
632 * sometimes lead to spurious "refusing to lose untracked file..."
633 * messages from update_file (via make_room_for path via
634 * would_lose_untracked). Instead, reverse the order of the calls
635 * (executing update_file first and then update_stages).
637 int clear = 1;
638 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
639 if (clear)
640 if (remove_file_from_cache(path))
641 return -1;
642 if (o)
643 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
644 return -1;
645 if (a)
646 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
647 return -1;
648 if (b)
649 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
650 return -1;
651 return 0;
654 static int update_stages_for_stage_data(struct merge_options *opt,
655 const char *path,
656 const struct stage_data *stage_data)
658 struct diff_filespec o, a, b;
660 o.mode = stage_data->stages[1].mode;
661 oidcpy(&o.oid, &stage_data->stages[1].oid);
663 a.mode = stage_data->stages[2].mode;
664 oidcpy(&a.oid, &stage_data->stages[2].oid);
666 b.mode = stage_data->stages[3].mode;
667 oidcpy(&b.oid, &stage_data->stages[3].oid);
669 return update_stages(opt, path,
670 is_null_oid(&o.oid) ? NULL : &o,
671 is_null_oid(&a.oid) ? NULL : &a,
672 is_null_oid(&b.oid) ? NULL : &b);
675 static void update_entry(struct stage_data *entry,
676 struct diff_filespec *o,
677 struct diff_filespec *a,
678 struct diff_filespec *b)
680 entry->processed = 0;
681 entry->stages[1].mode = o->mode;
682 entry->stages[2].mode = a->mode;
683 entry->stages[3].mode = b->mode;
684 oidcpy(&entry->stages[1].oid, &o->oid);
685 oidcpy(&entry->stages[2].oid, &a->oid);
686 oidcpy(&entry->stages[3].oid, &b->oid);
689 static int remove_file(struct merge_options *o, int clean,
690 const char *path, int no_wd)
692 int update_cache = o->call_depth || clean;
693 int update_working_directory = !o->call_depth && !no_wd;
695 if (update_cache) {
696 if (remove_file_from_cache(path))
697 return -1;
699 if (update_working_directory) {
700 if (ignore_case) {
701 struct cache_entry *ce;
702 ce = cache_file_exists(path, strlen(path), ignore_case);
703 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
704 return 0;
706 if (remove_path(path))
707 return -1;
709 return 0;
712 /* add a string to a strbuf, but converting "/" to "_" */
713 static void add_flattened_path(struct strbuf *out, const char *s)
715 size_t i = out->len;
716 strbuf_addstr(out, s);
717 for (; i < out->len; i++)
718 if (out->buf[i] == '/')
719 out->buf[i] = '_';
722 static char *unique_path(struct merge_options *o, const char *path, const char *branch)
724 struct path_hashmap_entry *entry;
725 struct strbuf newpath = STRBUF_INIT;
726 int suffix = 0;
727 size_t base_len;
729 strbuf_addf(&newpath, "%s~", path);
730 add_flattened_path(&newpath, branch);
732 base_len = newpath.len;
733 while (hashmap_get_from_hash(&o->current_file_dir_set,
734 path_hash(newpath.buf), newpath.buf) ||
735 (!o->call_depth && file_exists(newpath.buf))) {
736 strbuf_setlen(&newpath, base_len);
737 strbuf_addf(&newpath, "_%d", suffix++);
740 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
741 hashmap_entry_init(entry, path_hash(entry->path));
742 hashmap_add(&o->current_file_dir_set, entry);
743 return strbuf_detach(&newpath, NULL);
747 * Check whether a directory in the index is in the way of an incoming
748 * file. Return 1 if so. If check_working_copy is non-zero, also
749 * check the working directory. If empty_ok is non-zero, also return
750 * 0 in the case where the working-tree dir exists but is empty.
752 static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
754 int pos;
755 struct strbuf dirpath = STRBUF_INIT;
756 struct stat st;
758 strbuf_addstr(&dirpath, path);
759 strbuf_addch(&dirpath, '/');
761 pos = cache_name_pos(dirpath.buf, dirpath.len);
763 if (pos < 0)
764 pos = -1 - pos;
765 if (pos < active_nr &&
766 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
767 strbuf_release(&dirpath);
768 return 1;
771 strbuf_release(&dirpath);
772 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
773 !(empty_ok && is_empty_dir(path));
776 static int was_tracked(const char *path)
778 int pos = cache_name_pos(path, strlen(path));
780 if (0 <= pos)
781 /* we have been tracking this path */
782 return 1;
785 * Look for an unmerged entry for the path,
786 * specifically stage #2, which would indicate
787 * that "our" side before the merge started
788 * had the path tracked (and resulted in a conflict).
790 for (pos = -1 - pos;
791 pos < active_nr && !strcmp(path, active_cache[pos]->name);
792 pos++)
793 if (ce_stage(active_cache[pos]) == 2)
794 return 1;
795 return 0;
798 static int would_lose_untracked(const char *path)
800 return !was_tracked(path) && file_exists(path);
803 static int was_dirty(struct merge_options *o, const char *path)
805 struct cache_entry *ce;
806 int dirty = 1;
808 if (o->call_depth || !was_tracked(path))
809 return !dirty;
811 ce = cache_file_exists(path, strlen(path), ignore_case);
812 dirty = (ce->ce_stat_data.sd_mtime.sec > 0 &&
813 verify_uptodate(ce, &o->unpack_opts) != 0);
814 return dirty;
817 static int make_room_for_path(struct merge_options *o, const char *path)
819 int status, i;
820 const char *msg = _("failed to create path '%s'%s");
822 /* Unlink any D/F conflict files that are in the way */
823 for (i = 0; i < o->df_conflict_file_set.nr; i++) {
824 const char *df_path = o->df_conflict_file_set.items[i].string;
825 size_t pathlen = strlen(path);
826 size_t df_pathlen = strlen(df_path);
827 if (df_pathlen < pathlen &&
828 path[df_pathlen] == '/' &&
829 strncmp(path, df_path, df_pathlen) == 0) {
830 output(o, 3,
831 _("Removing %s to make room for subdirectory\n"),
832 df_path);
833 unlink(df_path);
834 unsorted_string_list_delete_item(&o->df_conflict_file_set,
835 i, 0);
836 break;
840 /* Make sure leading directories are created */
841 status = safe_create_leading_directories_const(path);
842 if (status) {
843 if (status == SCLD_EXISTS)
844 /* something else exists */
845 return err(o, msg, path, _(": perhaps a D/F conflict?"));
846 return err(o, msg, path, "");
850 * Do not unlink a file in the work tree if we are not
851 * tracking it.
853 if (would_lose_untracked(path))
854 return err(o, _("refusing to lose untracked file at '%s'"),
855 path);
857 /* Successful unlink is good.. */
858 if (!unlink(path))
859 return 0;
860 /* .. and so is no existing file */
861 if (errno == ENOENT)
862 return 0;
863 /* .. but not some other error (who really cares what?) */
864 return err(o, msg, path, _(": perhaps a D/F conflict?"));
867 static int update_file_flags(struct merge_options *o,
868 const struct object_id *oid,
869 unsigned mode,
870 const char *path,
871 int update_cache,
872 int update_wd)
874 int ret = 0;
876 if (o->call_depth)
877 update_wd = 0;
879 if (update_wd) {
880 enum object_type type;
881 void *buf;
882 unsigned long size;
884 if (S_ISGITLINK(mode)) {
886 * We may later decide to recursively descend into
887 * the submodule directory and update its index
888 * and/or work tree, but we do not do that now.
890 update_wd = 0;
891 goto update_index;
894 buf = read_sha1_file(oid->hash, &type, &size);
895 if (!buf)
896 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
897 if (type != OBJ_BLOB) {
898 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
899 goto free_buf;
901 if (S_ISREG(mode)) {
902 struct strbuf strbuf = STRBUF_INIT;
903 if (convert_to_working_tree(path, buf, size, &strbuf)) {
904 free(buf);
905 size = strbuf.len;
906 buf = strbuf_detach(&strbuf, NULL);
910 if (make_room_for_path(o, path) < 0) {
911 update_wd = 0;
912 goto free_buf;
914 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
915 int fd;
916 if (mode & 0100)
917 mode = 0777;
918 else
919 mode = 0666;
920 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
921 if (fd < 0) {
922 ret = err(o, _("failed to open '%s': %s"),
923 path, strerror(errno));
924 goto free_buf;
926 write_in_full(fd, buf, size);
927 close(fd);
928 } else if (S_ISLNK(mode)) {
929 char *lnk = xmemdupz(buf, size);
930 safe_create_leading_directories_const(path);
931 unlink(path);
932 if (symlink(lnk, path))
933 ret = err(o, _("failed to symlink '%s': %s"),
934 path, strerror(errno));
935 free(lnk);
936 } else
937 ret = err(o,
938 _("do not know what to do with %06o %s '%s'"),
939 mode, oid_to_hex(oid), path);
940 free_buf:
941 free(buf);
943 update_index:
944 if (!ret && update_cache)
945 add_cacheinfo(o, mode, oid, path, 0, update_wd, ADD_CACHE_OK_TO_ADD);
946 return ret;
949 static int update_file(struct merge_options *o,
950 int clean,
951 const struct object_id *oid,
952 unsigned mode,
953 const char *path)
955 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
958 /* Low level file merging, update and removal */
960 struct merge_file_info {
961 struct object_id oid;
962 unsigned mode;
963 unsigned clean:1,
964 merge:1;
967 static int merge_3way(struct merge_options *o,
968 mmbuffer_t *result_buf,
969 const struct diff_filespec *one,
970 const struct diff_filespec *a,
971 const struct diff_filespec *b,
972 const char *branch1,
973 const char *branch2)
975 mmfile_t orig, src1, src2;
976 struct ll_merge_options ll_opts = {0};
977 char *base_name, *name1, *name2;
978 int merge_status;
980 ll_opts.renormalize = o->renormalize;
981 ll_opts.xdl_opts = o->xdl_opts;
983 if (o->call_depth) {
984 ll_opts.virtual_ancestor = 1;
985 ll_opts.variant = 0;
986 } else {
987 switch (o->recursive_variant) {
988 case MERGE_RECURSIVE_OURS:
989 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
990 break;
991 case MERGE_RECURSIVE_THEIRS:
992 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
993 break;
994 default:
995 ll_opts.variant = 0;
996 break;
1000 if (strcmp(a->path, b->path) ||
1001 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1002 base_name = o->ancestor == NULL ? NULL :
1003 mkpathdup("%s:%s", o->ancestor, one->path);
1004 name1 = mkpathdup("%s:%s", branch1, a->path);
1005 name2 = mkpathdup("%s:%s", branch2, b->path);
1006 } else {
1007 base_name = o->ancestor == NULL ? NULL :
1008 mkpathdup("%s", o->ancestor);
1009 name1 = mkpathdup("%s", branch1);
1010 name2 = mkpathdup("%s", branch2);
1013 read_mmblob(&orig, &one->oid);
1014 read_mmblob(&src1, &a->oid);
1015 read_mmblob(&src2, &b->oid);
1017 merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1018 &src1, name1, &src2, name2, &ll_opts);
1020 free(base_name);
1021 free(name1);
1022 free(name2);
1023 free(orig.ptr);
1024 free(src1.ptr);
1025 free(src2.ptr);
1026 return merge_status;
1029 static int merge_file_1(struct merge_options *o,
1030 const struct diff_filespec *one,
1031 const struct diff_filespec *a,
1032 const struct diff_filespec *b,
1033 const char *branch1,
1034 const char *branch2,
1035 struct merge_file_info *result)
1037 result->merge = 0;
1038 result->clean = 1;
1040 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1041 result->clean = 0;
1042 if (S_ISREG(a->mode)) {
1043 result->mode = a->mode;
1044 oidcpy(&result->oid, &a->oid);
1045 } else {
1046 result->mode = b->mode;
1047 oidcpy(&result->oid, &b->oid);
1049 } else {
1050 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1051 result->merge = 1;
1054 * Merge modes
1056 if (a->mode == b->mode || a->mode == one->mode)
1057 result->mode = b->mode;
1058 else {
1059 result->mode = a->mode;
1060 if (b->mode != one->mode) {
1061 result->clean = 0;
1062 result->merge = 1;
1066 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1067 oidcpy(&result->oid, &b->oid);
1068 else if (oid_eq(&b->oid, &one->oid))
1069 oidcpy(&result->oid, &a->oid);
1070 else if (S_ISREG(a->mode)) {
1071 mmbuffer_t result_buf;
1072 int ret = 0, merge_status;
1074 merge_status = merge_3way(o, &result_buf, one, a, b,
1075 branch1, branch2);
1077 if ((merge_status < 0) || !result_buf.ptr)
1078 ret = err(o, _("Failed to execute internal merge"));
1080 if (!ret && write_sha1_file(result_buf.ptr, result_buf.size,
1081 blob_type, result->oid.hash))
1082 ret = err(o, _("Unable to add %s to database"),
1083 a->path);
1085 free(result_buf.ptr);
1086 if (ret)
1087 return ret;
1088 result->clean = (merge_status == 0);
1089 } else if (S_ISGITLINK(a->mode)) {
1090 result->clean = merge_submodule(&result->oid,
1091 one->path,
1092 &one->oid,
1093 &a->oid,
1094 &b->oid,
1095 !o->call_depth);
1096 } else if (S_ISLNK(a->mode)) {
1097 oidcpy(&result->oid, &a->oid);
1099 if (!oid_eq(&a->oid, &b->oid))
1100 result->clean = 0;
1101 } else
1102 die("BUG: unsupported object type in the tree");
1105 return 0;
1108 static int merge_file_special_markers(struct merge_options *o,
1109 const struct diff_filespec *one,
1110 const struct diff_filespec *a,
1111 const struct diff_filespec *b,
1112 const char *branch1,
1113 const char *filename1,
1114 const char *branch2,
1115 const char *filename2,
1116 struct merge_file_info *mfi)
1118 char *side1 = NULL;
1119 char *side2 = NULL;
1120 int ret;
1122 if (filename1)
1123 side1 = xstrfmt("%s:%s", branch1, filename1);
1124 if (filename2)
1125 side2 = xstrfmt("%s:%s", branch2, filename2);
1127 ret = merge_file_1(o, one, a, b,
1128 side1 ? side1 : branch1,
1129 side2 ? side2 : branch2, mfi);
1130 free(side1);
1131 free(side2);
1132 return ret;
1135 static int merge_file_one(struct merge_options *o,
1136 const char *path,
1137 const struct object_id *o_oid, int o_mode,
1138 const struct object_id *a_oid, int a_mode,
1139 const struct object_id *b_oid, int b_mode,
1140 const char *branch1,
1141 const char *branch2,
1142 struct merge_file_info *mfi)
1144 struct diff_filespec one, a, b;
1146 one.path = a.path = b.path = (char *)path;
1147 oidcpy(&one.oid, o_oid);
1148 one.mode = o_mode;
1149 oidcpy(&a.oid, a_oid);
1150 a.mode = a_mode;
1151 oidcpy(&b.oid, b_oid);
1152 b.mode = b_mode;
1153 return merge_file_1(o, &one, &a, &b, branch1, branch2, mfi);
1156 static int conflict_rename_dir(struct merge_options *o,
1157 struct diff_filepair *pair,
1158 const char *rename_branch,
1159 const char *other_branch)
1161 const struct diff_filespec *dest = pair->two;
1163 if (!o->call_depth && would_lose_untracked(dest->path)) {
1164 char *alt_path = unique_path(o, dest->path, rename_branch);
1166 output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1167 "writing to %s instead."),
1168 dest->path, alt_path);
1170 * Write the file in worktree at alt_path, but not in the
1171 * index. Instead, write to dest->path for the index but
1172 * only at the higher appropriate stage.
1174 if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1175 return -1;
1176 free(alt_path);
1177 return update_stages(o, dest->path, NULL,
1178 rename_branch == o->branch1 ? dest : NULL,
1179 rename_branch == o->branch1 ? NULL : dest);
1182 /* Update dest->path both in index and in worktree */
1183 if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1184 return -1;
1185 return 0;
1188 static int handle_change_delete(struct merge_options *o,
1189 const char *path, const char *old_path,
1190 const struct object_id *o_oid, int o_mode,
1191 const struct object_id *changed_oid,
1192 int changed_mode,
1193 const char *change_branch,
1194 const char *delete_branch,
1195 const char *change, const char *change_past)
1197 char *alt_path = NULL;
1198 const char *update_path = path;
1199 int ret = 0;
1201 if (dir_in_way(path, !o->call_depth, 0) ||
1202 (!o->call_depth && would_lose_untracked(path))) {
1203 update_path = alt_path = unique_path(o, path, change_branch);
1206 if (o->call_depth) {
1208 * We cannot arbitrarily accept either a_sha or b_sha as
1209 * correct; since there is no true "middle point" between
1210 * them, simply reuse the base version for virtual merge base.
1212 ret = remove_file_from_cache(path);
1213 if (!ret)
1214 ret = update_file(o, 0, o_oid, o_mode, update_path);
1215 } else {
1216 if (!alt_path) {
1217 if (!old_path) {
1218 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1219 "and %s in %s. Version %s of %s left in tree."),
1220 change, path, delete_branch, change_past,
1221 change_branch, change_branch, path);
1222 } else {
1223 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1224 "and %s to %s in %s. Version %s of %s left in tree."),
1225 change, old_path, delete_branch, change_past, path,
1226 change_branch, change_branch, path);
1228 } else {
1229 if (!old_path) {
1230 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1231 "and %s in %s. Version %s of %s left in tree at %s."),
1232 change, path, delete_branch, change_past,
1233 change_branch, change_branch, path, alt_path);
1234 } else {
1235 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1236 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1237 change, old_path, delete_branch, change_past, path,
1238 change_branch, change_branch, path, alt_path);
1242 * No need to call update_file() on path when change_branch ==
1243 * o->branch1 && !alt_path, since that would needlessly touch
1244 * path. We could call update_file_flags() with update_cache=0
1245 * and update_wd=0, but that's a no-op.
1247 if (change_branch != o->branch1 || alt_path)
1248 ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1250 free(alt_path);
1252 return ret;
1255 static int conflict_rename_delete(struct merge_options *o,
1256 struct diff_filepair *pair,
1257 const char *rename_branch,
1258 const char *delete_branch)
1260 const struct diff_filespec *orig = pair->one;
1261 const struct diff_filespec *dest = pair->two;
1263 if (handle_change_delete(o,
1264 o->call_depth ? orig->path : dest->path,
1265 o->call_depth ? NULL : orig->path,
1266 &orig->oid, orig->mode,
1267 &dest->oid, dest->mode,
1268 rename_branch, delete_branch,
1269 _("rename"), _("renamed")))
1270 return -1;
1272 if (o->call_depth)
1273 return remove_file_from_cache(dest->path);
1274 else
1275 return update_stages(o, dest->path, NULL,
1276 rename_branch == o->branch1 ? dest : NULL,
1277 rename_branch == o->branch1 ? NULL : dest);
1280 static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1281 struct stage_data *entry,
1282 int stage)
1284 struct object_id *oid = &entry->stages[stage].oid;
1285 unsigned mode = entry->stages[stage].mode;
1286 if (mode == 0 || is_null_oid(oid))
1287 return NULL;
1288 oidcpy(&target->oid, oid);
1289 target->mode = mode;
1290 return target;
1293 static int handle_file(struct merge_options *o,
1294 struct diff_filespec *rename,
1295 int stage,
1296 struct rename_conflict_info *ci)
1298 char *dst_name = rename->path;
1299 struct stage_data *dst_entry;
1300 const char *cur_branch, *other_branch;
1301 struct diff_filespec other;
1302 struct diff_filespec *add;
1303 int ret;
1305 if (stage == 2) {
1306 dst_entry = ci->dst_entry1;
1307 cur_branch = ci->branch1;
1308 other_branch = ci->branch2;
1309 } else {
1310 dst_entry = ci->dst_entry2;
1311 cur_branch = ci->branch2;
1312 other_branch = ci->branch1;
1315 add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1316 if (add) {
1317 int ren_src_was_dirty = was_dirty(o, rename->path);
1318 char *add_name = unique_path(o, rename->path, other_branch);
1319 if (update_file(o, 0, &add->oid, add->mode, add_name))
1320 return -1;
1322 if (ren_src_was_dirty) {
1323 output(o, 1, _("Refusing to lose dirty file at %s"),
1324 rename->path);
1327 * Because the double negatives somehow keep confusing me...
1328 * 1) update_wd iff !ren_src_was_dirty.
1329 * 2) no_wd iff !update_wd
1330 * 3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty
1332 remove_file(o, 0, rename->path, ren_src_was_dirty);
1333 dst_name = unique_path(o, rename->path, cur_branch);
1334 } else {
1335 if (dir_in_way(rename->path, !o->call_depth, 0)) {
1336 dst_name = unique_path(o, rename->path, cur_branch);
1337 output(o, 1, _("%s is a directory in %s adding as %s instead"),
1338 rename->path, other_branch, dst_name);
1339 } else if (!o->call_depth &&
1340 would_lose_untracked(rename->path)) {
1341 dst_name = unique_path(o, rename->path, cur_branch);
1342 output(o, 1, _("Refusing to lose untracked file at %s; "
1343 "adding as %s instead"),
1344 rename->path, dst_name);
1347 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1348 ; /* fall through, do allow dst_name to be released */
1349 else if (stage == 2)
1350 ret = update_stages(o, rename->path, NULL, rename, add);
1351 else
1352 ret = update_stages(o, rename->path, NULL, add, rename);
1354 if (dst_name != rename->path)
1355 free(dst_name);
1357 return ret;
1360 static int conflict_rename_rename_1to2(struct merge_options *o,
1361 struct rename_conflict_info *ci)
1363 /* One file was renamed in both branches, but to different names. */
1364 struct diff_filespec *one = ci->pair1->one;
1365 struct diff_filespec *a = ci->pair1->two;
1366 struct diff_filespec *b = ci->pair2->two;
1368 output(o, 1, _("CONFLICT (rename/rename): "
1369 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1370 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1371 one->path, a->path, ci->branch1,
1372 one->path, b->path, ci->branch2,
1373 o->call_depth ? _(" (left unresolved)") : "");
1374 if (o->call_depth) {
1375 struct merge_file_info mfi;
1376 struct diff_filespec other;
1377 struct diff_filespec *add;
1378 if (merge_file_one(o, one->path,
1379 &one->oid, one->mode,
1380 &a->oid, a->mode,
1381 &b->oid, b->mode,
1382 ci->branch1, ci->branch2, &mfi))
1383 return -1;
1386 * FIXME: For rename/add-source conflicts (if we could detect
1387 * such), this is wrong. We should instead find a unique
1388 * pathname and then either rename the add-source file to that
1389 * unique path, or use that unique path instead of src here.
1391 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1392 return -1;
1395 * Above, we put the merged content at the merge-base's
1396 * path. Now we usually need to delete both a->path and
1397 * b->path. However, the rename on each side of the merge
1398 * could also be involved in a rename/add conflict. In
1399 * such cases, we should keep the added file around,
1400 * resolving the conflict at that path in its favor.
1402 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1403 if (add) {
1404 if (update_file(o, 0, &add->oid, add->mode, a->path))
1405 return -1;
1407 else
1408 remove_file_from_cache(a->path);
1409 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1410 if (add) {
1411 if (update_file(o, 0, &add->oid, add->mode, b->path))
1412 return -1;
1414 else
1415 remove_file_from_cache(b->path);
1416 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1417 return -1;
1419 return 0;
1422 static int conflict_rename_rename_2to1(struct merge_options *o,
1423 struct rename_conflict_info *ci)
1425 /* Two files, a & b, were renamed to the same thing, c. */
1426 struct diff_filespec *a = ci->pair1->one;
1427 struct diff_filespec *b = ci->pair2->one;
1428 struct diff_filespec *c1 = ci->pair1->two;
1429 struct diff_filespec *c2 = ci->pair2->two;
1430 char *path = c1->path; /* == c2->path */
1431 struct merge_file_info mfi_c1;
1432 struct merge_file_info mfi_c2;
1433 int ret;
1435 output(o, 1, _("CONFLICT (rename/rename): "
1436 "Rename %s->%s in %s. "
1437 "Rename %s->%s in %s"),
1438 a->path, c1->path, ci->branch1,
1439 b->path, c2->path, ci->branch2);
1441 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1442 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1444 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1445 o->branch1, c1->path,
1446 o->branch2, ci->ren1_other.path, &mfi_c1) ||
1447 merge_file_special_markers(o, b, &ci->ren2_other, c2,
1448 o->branch1, ci->ren2_other.path,
1449 o->branch2, c2->path, &mfi_c2))
1450 return -1;
1452 if (o->call_depth) {
1454 * If mfi_c1.clean && mfi_c2.clean, then it might make
1455 * sense to do a two-way merge of those results. But, I
1456 * think in all cases, it makes sense to have the virtual
1457 * merge base just undo the renames; they can be detected
1458 * again later for the non-recursive merge.
1460 remove_file(o, 0, path, 0);
1461 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1462 if (!ret)
1463 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1464 b->path);
1465 } else {
1466 char *new_path1 = unique_path(o, path, ci->branch1);
1467 char *new_path2 = unique_path(o, path, ci->branch2);
1468 output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1469 a->path, new_path1, b->path, new_path2);
1470 if (was_dirty(o, path))
1471 output(o, 1, _("Refusing to lose dirty file at %s"),
1472 path);
1473 else if (would_lose_untracked(path))
1475 * Only way we get here is if both renames were from
1476 * a directory rename AND user had an untracked file
1477 * at the location where both files end up after the
1478 * two directory renames. See testcase 10d of t6043.
1480 output(o, 1, _("Refusing to lose untracked file at "
1481 "%s, even though it's in the way."),
1482 path);
1483 else
1484 remove_file(o, 0, path, 0);
1485 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1486 if (!ret)
1487 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1488 new_path2);
1490 * unpack_trees() actually populates the index for us for
1491 * "normal" rename/rename(2to1) situtations so that the
1492 * correct entries are at the higher stages, which would
1493 * make the call below to update_stages_for_stage_data
1494 * unnecessary. However, if either of the renames came
1495 * from a directory rename, then unpack_trees() will not
1496 * have gotten the right data loaded into the index, so we
1497 * need to do so now. (While it'd be tempting to move this
1498 * call to update_stages_for_stage_data() to
1499 * apply_directory_rename_modifications(), that would break
1500 * our intermediate calls to would_lose_untracked() since
1501 * those rely on the current in-memory index. See also the
1502 * big "NOTE" in update_stages()).
1504 if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1505 ret = -1;
1507 free(new_path2);
1508 free(new_path1);
1511 return ret;
1515 * Get the diff_filepairs changed between o_tree and tree.
1517 static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1518 struct tree *o_tree,
1519 struct tree *tree)
1521 struct diff_queue_struct *ret;
1522 struct diff_options opts;
1524 diff_setup(&opts);
1525 opts.flags.recursive = 1;
1526 opts.flags.rename_empty = 0;
1527 opts.detect_rename = DIFF_DETECT_RENAME;
1528 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1529 o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1530 1000;
1531 opts.rename_score = o->rename_score;
1532 opts.show_rename_progress = o->show_rename_progress;
1533 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1534 diff_setup_done(&opts);
1535 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1536 diffcore_std(&opts);
1537 if (opts.needed_rename_limit > o->needed_rename_limit)
1538 o->needed_rename_limit = opts.needed_rename_limit;
1540 ret = xmalloc(sizeof(*ret));
1541 *ret = diff_queued_diff;
1543 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1544 diff_queued_diff.nr = 0;
1545 diff_queued_diff.queue = NULL;
1546 diff_flush(&opts);
1547 return ret;
1550 static int tree_has_path(struct tree *tree, const char *path)
1552 unsigned char hashy[GIT_MAX_RAWSZ];
1553 unsigned int mode_o;
1555 return !get_tree_entry(tree->object.oid.hash, path,
1556 hashy, &mode_o);
1560 * Return a new string that replaces the beginning portion (which matches
1561 * entry->dir), with entry->new_dir. In perl-speak:
1562 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1563 * NOTE:
1564 * Caller must ensure that old_path starts with entry->dir + '/'.
1566 static char *apply_dir_rename(struct dir_rename_entry *entry,
1567 const char *old_path)
1569 struct strbuf new_path = STRBUF_INIT;
1570 int oldlen, newlen;
1572 if (entry->non_unique_new_dir)
1573 return NULL;
1575 oldlen = strlen(entry->dir);
1576 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1577 strbuf_grow(&new_path, newlen);
1578 strbuf_addbuf(&new_path, &entry->new_dir);
1579 strbuf_addstr(&new_path, &old_path[oldlen]);
1581 return strbuf_detach(&new_path, NULL);
1584 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1585 char **old_dir, char **new_dir)
1587 char *end_of_old, *end_of_new;
1588 int old_len, new_len;
1590 *old_dir = NULL;
1591 *new_dir = NULL;
1594 * For
1595 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1596 * the "e/foo.c" part is the same, we just want to know that
1597 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1598 * so, for this example, this function returns "a/b/c/d" in
1599 * *old_dir and "a/b/some/thing/else" in *new_dir.
1601 * Also, if the basename of the file changed, we don't care. We
1602 * want to know which portion of the directory, if any, changed.
1604 end_of_old = strrchr(old_path, '/');
1605 end_of_new = strrchr(new_path, '/');
1607 if (end_of_old == NULL || end_of_new == NULL)
1608 return;
1609 while (*--end_of_new == *--end_of_old &&
1610 end_of_old != old_path &&
1611 end_of_new != new_path)
1612 ; /* Do nothing; all in the while loop */
1614 * We've found the first non-matching character in the directory
1615 * paths. That means the current directory we were comparing
1616 * represents the rename. Move end_of_old and end_of_new back
1617 * to the full directory name.
1619 if (*end_of_old == '/')
1620 end_of_old++;
1621 if (*end_of_old != '/')
1622 end_of_new++;
1623 end_of_old = strchr(end_of_old, '/');
1624 end_of_new = strchr(end_of_new, '/');
1627 * It may have been the case that old_path and new_path were the same
1628 * directory all along. Don't claim a rename if they're the same.
1630 old_len = end_of_old - old_path;
1631 new_len = end_of_new - new_path;
1633 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1634 *old_dir = xstrndup(old_path, old_len);
1635 *new_dir = xstrndup(new_path, new_len);
1639 static void remove_hashmap_entries(struct hashmap *dir_renames,
1640 struct string_list *items_to_remove)
1642 int i;
1643 struct dir_rename_entry *entry;
1645 for (i = 0; i < items_to_remove->nr; i++) {
1646 entry = items_to_remove->items[i].util;
1647 hashmap_remove(dir_renames, entry, NULL);
1649 string_list_clear(items_to_remove, 0);
1653 * See if there is a directory rename for path, and if there are any file
1654 * level conflicts for the renamed location. If there is a rename and
1655 * there are no conflicts, return the new name. Otherwise, return NULL.
1657 static char *handle_path_level_conflicts(struct merge_options *o,
1658 const char *path,
1659 struct dir_rename_entry *entry,
1660 struct hashmap *collisions,
1661 struct tree *tree)
1663 char *new_path = NULL;
1664 struct collision_entry *collision_ent;
1665 int clean = 1;
1666 struct strbuf collision_paths = STRBUF_INIT;
1669 * entry has the mapping of old directory name to new directory name
1670 * that we want to apply to path.
1672 new_path = apply_dir_rename(entry, path);
1674 if (!new_path) {
1675 /* This should only happen when entry->non_unique_new_dir set */
1676 if (!entry->non_unique_new_dir)
1677 BUG("entry->non_unqiue_dir not set and !new_path");
1678 output(o, 1, _("CONFLICT (directory rename split): "
1679 "Unclear where to place %s because directory "
1680 "%s was renamed to multiple other directories, "
1681 "with no destination getting a majority of the "
1682 "files."),
1683 path, entry->dir);
1684 clean = 0;
1685 return NULL;
1689 * The caller needs to have ensured that it has pre-populated
1690 * collisions with all paths that map to new_path. Do a quick check
1691 * to ensure that's the case.
1693 collision_ent = collision_find_entry(collisions, new_path);
1694 if (collision_ent == NULL)
1695 BUG("collision_ent is NULL");
1698 * Check for one-sided add/add/.../add conflicts, i.e.
1699 * where implicit renames from the other side doing
1700 * directory rename(s) can affect this side of history
1701 * to put multiple paths into the same location. Warn
1702 * and bail on directory renames for such paths.
1704 if (collision_ent->reported_already) {
1705 clean = 0;
1706 } else if (tree_has_path(tree, new_path)) {
1707 collision_ent->reported_already = 1;
1708 strbuf_add_separated_string_list(&collision_paths, ", ",
1709 &collision_ent->source_files);
1710 output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1711 "file/dir at %s in the way of implicit "
1712 "directory rename(s) putting the following "
1713 "path(s) there: %s."),
1714 new_path, collision_paths.buf);
1715 clean = 0;
1716 } else if (collision_ent->source_files.nr > 1) {
1717 collision_ent->reported_already = 1;
1718 strbuf_add_separated_string_list(&collision_paths, ", ",
1719 &collision_ent->source_files);
1720 output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1721 "more than one path to %s; implicit directory "
1722 "renames tried to put these paths there: %s"),
1723 new_path, collision_paths.buf);
1724 clean = 0;
1727 /* Free memory we no longer need */
1728 strbuf_release(&collision_paths);
1729 if (!clean && new_path) {
1730 free(new_path);
1731 return NULL;
1734 return new_path;
1738 * There are a couple things we want to do at the directory level:
1739 * 1. Check for both sides renaming to the same thing, in order to avoid
1740 * implicit renaming of files that should be left in place. (See
1741 * testcase 6b in t6043 for details.)
1742 * 2. Prune directory renames if there are still files left in the
1743 * the original directory. These represent a partial directory rename,
1744 * i.e. a rename where only some of the files within the directory
1745 * were renamed elsewhere. (Technically, this could be done earlier
1746 * in get_directory_renames(), except that would prevent us from
1747 * doing the previous check and thus failing testcase 6b.)
1748 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
1749 * In the future, we could potentially record this info as well and
1750 * omit reporting rename/rename(1to2) conflicts for each path within
1751 * the affected directories, thus cleaning up the merge output.
1752 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1753 * directory level, because merging directories is fine. If it
1754 * causes conflicts for files within those merged directories, then
1755 * that should be detected at the individual path level.
1757 static void handle_directory_level_conflicts(struct merge_options *o,
1758 struct hashmap *dir_re_head,
1759 struct tree *head,
1760 struct hashmap *dir_re_merge,
1761 struct tree *merge)
1763 struct hashmap_iter iter;
1764 struct dir_rename_entry *head_ent;
1765 struct dir_rename_entry *merge_ent;
1767 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1768 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1770 hashmap_iter_init(dir_re_head, &iter);
1771 while ((head_ent = hashmap_iter_next(&iter))) {
1772 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1773 if (merge_ent &&
1774 !head_ent->non_unique_new_dir &&
1775 !merge_ent->non_unique_new_dir &&
1776 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1777 /* 1. Renamed identically; remove it from both sides */
1778 string_list_append(&remove_from_head,
1779 head_ent->dir)->util = head_ent;
1780 strbuf_release(&head_ent->new_dir);
1781 string_list_append(&remove_from_merge,
1782 merge_ent->dir)->util = merge_ent;
1783 strbuf_release(&merge_ent->new_dir);
1784 } else if (tree_has_path(head, head_ent->dir)) {
1785 /* 2. This wasn't a directory rename after all */
1786 string_list_append(&remove_from_head,
1787 head_ent->dir)->util = head_ent;
1788 strbuf_release(&head_ent->new_dir);
1792 remove_hashmap_entries(dir_re_head, &remove_from_head);
1793 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1795 hashmap_iter_init(dir_re_merge, &iter);
1796 while ((merge_ent = hashmap_iter_next(&iter))) {
1797 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1798 if (tree_has_path(merge, merge_ent->dir)) {
1799 /* 2. This wasn't a directory rename after all */
1800 string_list_append(&remove_from_merge,
1801 merge_ent->dir)->util = merge_ent;
1802 } else if (head_ent &&
1803 !head_ent->non_unique_new_dir &&
1804 !merge_ent->non_unique_new_dir) {
1805 /* 3. rename/rename(1to2) */
1807 * We can assume it's not rename/rename(1to1) because
1808 * that was case (1), already checked above. So we
1809 * know that head_ent->new_dir and merge_ent->new_dir
1810 * are different strings.
1812 output(o, 1, _("CONFLICT (rename/rename): "
1813 "Rename directory %s->%s in %s. "
1814 "Rename directory %s->%s in %s"),
1815 head_ent->dir, head_ent->new_dir.buf, o->branch1,
1816 head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1817 string_list_append(&remove_from_head,
1818 head_ent->dir)->util = head_ent;
1819 strbuf_release(&head_ent->new_dir);
1820 string_list_append(&remove_from_merge,
1821 merge_ent->dir)->util = merge_ent;
1822 strbuf_release(&merge_ent->new_dir);
1826 remove_hashmap_entries(dir_re_head, &remove_from_head);
1827 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1830 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1831 struct tree *tree)
1833 struct hashmap *dir_renames;
1834 struct hashmap_iter iter;
1835 struct dir_rename_entry *entry;
1836 int i;
1839 * Typically, we think of a directory rename as all files from a
1840 * certain directory being moved to a target directory. However,
1841 * what if someone first moved two files from the original
1842 * directory in one commit, and then renamed the directory
1843 * somewhere else in a later commit? At merge time, we just know
1844 * that files from the original directory went to two different
1845 * places, and that the bulk of them ended up in the same place.
1846 * We want each directory rename to represent where the bulk of the
1847 * files from that directory end up; this function exists to find
1848 * where the bulk of the files went.
1850 * The first loop below simply iterates through the list of file
1851 * renames, finding out how often each directory rename pair
1852 * possibility occurs.
1854 dir_renames = xmalloc(sizeof(struct hashmap));
1855 dir_rename_init(dir_renames);
1856 for (i = 0; i < pairs->nr; ++i) {
1857 struct string_list_item *item;
1858 int *count;
1859 struct diff_filepair *pair = pairs->queue[i];
1860 char *old_dir, *new_dir;
1862 /* File not part of directory rename if it wasn't renamed */
1863 if (pair->status != 'R')
1864 continue;
1866 get_renamed_dir_portion(pair->one->path, pair->two->path,
1867 &old_dir, &new_dir);
1868 if (!old_dir)
1869 /* Directory didn't change at all; ignore this one. */
1870 continue;
1872 entry = dir_rename_find_entry(dir_renames, old_dir);
1873 if (!entry) {
1874 entry = xmalloc(sizeof(struct dir_rename_entry));
1875 dir_rename_entry_init(entry, old_dir);
1876 hashmap_put(dir_renames, entry);
1877 } else {
1878 free(old_dir);
1880 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1881 if (!item) {
1882 item = string_list_insert(&entry->possible_new_dirs,
1883 new_dir);
1884 item->util = xcalloc(1, sizeof(int));
1885 } else {
1886 free(new_dir);
1888 count = item->util;
1889 *count += 1;
1893 * For each directory with files moved out of it, we find out which
1894 * target directory received the most files so we can declare it to
1895 * be the "winning" target location for the directory rename. This
1896 * winner gets recorded in new_dir. If there is no winner
1897 * (multiple target directories received the same number of files),
1898 * we set non_unique_new_dir. Once we've determined the winner (or
1899 * that there is no winner), we no longer need possible_new_dirs.
1901 hashmap_iter_init(dir_renames, &iter);
1902 while ((entry = hashmap_iter_next(&iter))) {
1903 int max = 0;
1904 int bad_max = 0;
1905 char *best = NULL;
1907 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1908 int *count = entry->possible_new_dirs.items[i].util;
1910 if (*count == max)
1911 bad_max = max;
1912 else if (*count > max) {
1913 max = *count;
1914 best = entry->possible_new_dirs.items[i].string;
1917 if (bad_max == max)
1918 entry->non_unique_new_dir = 1;
1919 else {
1920 assert(entry->new_dir.len == 0);
1921 strbuf_addstr(&entry->new_dir, best);
1924 * The relevant directory sub-portion of the original full
1925 * filepaths were xstrndup'ed before inserting into
1926 * possible_new_dirs, and instead of manually iterating the
1927 * list and free'ing each, just lie and tell
1928 * possible_new_dirs that it did the strdup'ing so that it
1929 * will free them for us.
1931 entry->possible_new_dirs.strdup_strings = 1;
1932 string_list_clear(&entry->possible_new_dirs, 1);
1935 return dir_renames;
1938 static struct dir_rename_entry *check_dir_renamed(const char *path,
1939 struct hashmap *dir_renames)
1941 char *temp = xstrdup(path);
1942 char *end;
1943 struct dir_rename_entry *entry = NULL;;
1945 while ((end = strrchr(temp, '/'))) {
1946 *end = '\0';
1947 entry = dir_rename_find_entry(dir_renames, temp);
1948 if (entry)
1949 break;
1951 free(temp);
1952 return entry;
1955 static void compute_collisions(struct hashmap *collisions,
1956 struct hashmap *dir_renames,
1957 struct diff_queue_struct *pairs)
1959 int i;
1962 * Multiple files can be mapped to the same path due to directory
1963 * renames done by the other side of history. Since that other
1964 * side of history could have merged multiple directories into one,
1965 * if our side of history added the same file basename to each of
1966 * those directories, then all N of them would get implicitly
1967 * renamed by the directory rename detection into the same path,
1968 * and we'd get an add/add/.../add conflict, and all those adds
1969 * from *this* side of history. This is not representable in the
1970 * index, and users aren't going to easily be able to make sense of
1971 * it. So we need to provide a good warning about what's
1972 * happening, and fall back to no-directory-rename detection
1973 * behavior for those paths.
1975 * See testcases 9e and all of section 5 from t6043 for examples.
1977 collision_init(collisions);
1979 for (i = 0; i < pairs->nr; ++i) {
1980 struct dir_rename_entry *dir_rename_ent;
1981 struct collision_entry *collision_ent;
1982 char *new_path;
1983 struct diff_filepair *pair = pairs->queue[i];
1985 if (pair->status != 'A' && pair->status != 'R')
1986 continue;
1987 dir_rename_ent = check_dir_renamed(pair->two->path,
1988 dir_renames);
1989 if (!dir_rename_ent)
1990 continue;
1992 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
1993 if (!new_path)
1995 * dir_rename_ent->non_unique_new_path is true, which
1996 * means there is no directory rename for us to use,
1997 * which means it won't cause us any additional
1998 * collisions.
2000 continue;
2001 collision_ent = collision_find_entry(collisions, new_path);
2002 if (!collision_ent) {
2003 collision_ent = xcalloc(1,
2004 sizeof(struct collision_entry));
2005 hashmap_entry_init(collision_ent, strhash(new_path));
2006 hashmap_put(collisions, collision_ent);
2007 collision_ent->target_file = new_path;
2008 } else {
2009 free(new_path);
2011 string_list_insert(&collision_ent->source_files,
2012 pair->two->path);
2016 static char *check_for_directory_rename(struct merge_options *o,
2017 const char *path,
2018 struct tree *tree,
2019 struct hashmap *dir_renames,
2020 struct hashmap *dir_rename_exclusions,
2021 struct hashmap *collisions,
2022 int *clean_merge)
2024 char *new_path = NULL;
2025 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2026 struct dir_rename_entry *oentry = NULL;
2028 if (!entry)
2029 return new_path;
2032 * This next part is a little weird. We do not want to do an
2033 * implicit rename into a directory we renamed on our side, because
2034 * that will result in a spurious rename/rename(1to2) conflict. An
2035 * example:
2036 * Base commit: dumbdir/afile, otherdir/bfile
2037 * Side 1: smrtdir/afile, otherdir/bfile
2038 * Side 2: dumbdir/afile, dumbdir/bfile
2039 * Here, while working on Side 1, we could notice that otherdir was
2040 * renamed/merged to dumbdir, and change the diff_filepair for
2041 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2042 * 2 will notice the rename from dumbdir to smrtdir, and do the
2043 * transitive rename to move it from dumbdir/bfile to
2044 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2045 * smrtdir, a rename/rename(1to2) conflict. We really just want
2046 * the file to end up in smrtdir. And the way to achieve that is
2047 * to not let Side1 do the rename to dumbdir, since we know that is
2048 * the source of one of our directory renames.
2050 * That's why oentry and dir_rename_exclusions is here.
2052 * As it turns out, this also prevents N-way transient rename
2053 * confusion; See testcases 9c and 9d of t6043.
2055 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2056 if (oentry) {
2057 output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2058 "to %s, because %s itself was renamed."),
2059 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2060 } else {
2061 new_path = handle_path_level_conflicts(o, path, entry,
2062 collisions, tree);
2063 *clean_merge &= (new_path != NULL);
2066 return new_path;
2069 static void apply_directory_rename_modifications(struct merge_options *o,
2070 struct diff_filepair *pair,
2071 char *new_path,
2072 struct rename *re,
2073 struct tree *tree,
2074 struct tree *o_tree,
2075 struct tree *a_tree,
2076 struct tree *b_tree,
2077 struct string_list *entries,
2078 int *clean)
2080 struct string_list_item *item;
2081 int stage = (tree == a_tree ? 2 : 3);
2082 int update_wd;
2085 * In all cases where we can do directory rename detection,
2086 * unpack_trees() will have read pair->two->path into the
2087 * index and the working copy. We need to remove it so that
2088 * we can instead place it at new_path. It is guaranteed to
2089 * not be untracked (unpack_trees() would have errored out
2090 * saying the file would have been overwritten), but it might
2091 * be dirty, though.
2093 update_wd = !was_dirty(o, pair->two->path);
2094 if (!update_wd)
2095 output(o, 1, _("Refusing to lose dirty file at %s"),
2096 pair->two->path);
2097 remove_file(o, 1, pair->two->path, !update_wd);
2099 /* Find or create a new re->dst_entry */
2100 item = string_list_lookup(entries, new_path);
2101 if (item) {
2103 * Since we're renaming on this side of history, and it's
2104 * due to a directory rename on the other side of history
2105 * (which we only allow when the directory in question no
2106 * longer exists on the other side of history), the
2107 * original entry for re->dst_entry is no longer
2108 * necessary...
2110 re->dst_entry->processed = 1;
2113 * ...because we'll be using this new one.
2115 re->dst_entry = item->util;
2116 } else {
2118 * re->dst_entry is for the before-dir-rename path, and we
2119 * need it to hold information for the after-dir-rename
2120 * path. Before creating a new entry, we need to mark the
2121 * old one as unnecessary (...unless it is shared by
2122 * src_entry, i.e. this didn't use to be a rename, in which
2123 * case we can just allow the normal processing to happen
2124 * for it).
2126 if (pair->status == 'R')
2127 re->dst_entry->processed = 1;
2129 re->dst_entry = insert_stage_data(new_path,
2130 o_tree, a_tree, b_tree,
2131 entries);
2132 item = string_list_insert(entries, new_path);
2133 item->util = re->dst_entry;
2137 * Update the stage_data with the information about the path we are
2138 * moving into place. That slot will be empty and available for us
2139 * to write to because of the collision checks in
2140 * handle_path_level_conflicts(). In other words,
2141 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2142 * open for us to write to.
2144 * It may be tempting to actually update the index at this point as
2145 * well, using update_stages_for_stage_data(), but as per the big
2146 * "NOTE" in update_stages(), doing so will modify the current
2147 * in-memory index which will break calls to would_lose_untracked()
2148 * that we need to make. Instead, we need to just make sure that
2149 * the various conflict_rename_*() functions update the index
2150 * explicitly rather than relying on unpack_trees() to have done it.
2152 get_tree_entry(tree->object.oid.hash,
2153 pair->two->path,
2154 re->dst_entry->stages[stage].oid.hash,
2155 &re->dst_entry->stages[stage].mode);
2157 /* Update pair status */
2158 if (pair->status == 'A') {
2160 * Recording rename information for this add makes it look
2161 * like a rename/delete conflict. Make sure we can
2162 * correctly handle this as an add that was moved to a new
2163 * directory instead of reporting a rename/delete conflict.
2165 re->add_turned_into_rename = 1;
2168 * We don't actually look at pair->status again, but it seems
2169 * pedagogically correct to adjust it.
2171 pair->status = 'R';
2174 * Finally, record the new location.
2176 pair->two->path = new_path;
2180 * Get information of all renames which occurred in 'pairs', making use of
2181 * any implicit directory renames inferred from the other side of history.
2182 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2183 * to be able to associate the correct cache entries with the rename
2184 * information; tree is always equal to either a_tree or b_tree.
2186 static struct string_list *get_renames(struct merge_options *o,
2187 struct diff_queue_struct *pairs,
2188 struct hashmap *dir_renames,
2189 struct hashmap *dir_rename_exclusions,
2190 struct tree *tree,
2191 struct tree *o_tree,
2192 struct tree *a_tree,
2193 struct tree *b_tree,
2194 struct string_list *entries,
2195 int *clean_merge)
2197 int i;
2198 struct hashmap collisions;
2199 struct hashmap_iter iter;
2200 struct collision_entry *e;
2201 struct string_list *renames;
2203 compute_collisions(&collisions, dir_renames, pairs);
2204 renames = xcalloc(1, sizeof(struct string_list));
2206 for (i = 0; i < pairs->nr; ++i) {
2207 struct string_list_item *item;
2208 struct rename *re;
2209 struct diff_filepair *pair = pairs->queue[i];
2210 char *new_path; /* non-NULL only with directory renames */
2212 if (pair->status != 'A' && pair->status != 'R') {
2213 diff_free_filepair(pair);
2214 continue;
2216 new_path = check_for_directory_rename(o, pair->two->path, tree,
2217 dir_renames,
2218 dir_rename_exclusions,
2219 &collisions,
2220 clean_merge);
2221 if (pair->status != 'R' && !new_path) {
2222 diff_free_filepair(pair);
2223 continue;
2226 re = xmalloc(sizeof(*re));
2227 re->processed = 0;
2228 re->add_turned_into_rename = 0;
2229 re->pair = pair;
2230 item = string_list_lookup(entries, re->pair->one->path);
2231 if (!item)
2232 re->src_entry = insert_stage_data(re->pair->one->path,
2233 o_tree, a_tree, b_tree, entries);
2234 else
2235 re->src_entry = item->util;
2237 item = string_list_lookup(entries, re->pair->two->path);
2238 if (!item)
2239 re->dst_entry = insert_stage_data(re->pair->two->path,
2240 o_tree, a_tree, b_tree, entries);
2241 else
2242 re->dst_entry = item->util;
2243 item = string_list_insert(renames, pair->one->path);
2244 item->util = re;
2245 if (new_path)
2246 apply_directory_rename_modifications(o, pair, new_path,
2247 re, tree, o_tree,
2248 a_tree, b_tree,
2249 entries,
2250 clean_merge);
2253 hashmap_iter_init(&collisions, &iter);
2254 while ((e = hashmap_iter_next(&iter))) {
2255 free(e->target_file);
2256 string_list_clear(&e->source_files, 0);
2258 hashmap_free(&collisions, 1);
2259 return renames;
2262 static int process_renames(struct merge_options *o,
2263 struct string_list *a_renames,
2264 struct string_list *b_renames)
2266 int clean_merge = 1, i, j;
2267 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2268 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2269 const struct rename *sre;
2271 for (i = 0; i < a_renames->nr; i++) {
2272 sre = a_renames->items[i].util;
2273 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2274 = (void *)sre;
2276 for (i = 0; i < b_renames->nr; i++) {
2277 sre = b_renames->items[i].util;
2278 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2279 = (void *)sre;
2282 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2283 struct string_list *renames1, *renames2Dst;
2284 struct rename *ren1 = NULL, *ren2 = NULL;
2285 const char *branch1, *branch2;
2286 const char *ren1_src, *ren1_dst;
2287 struct string_list_item *lookup;
2289 if (i >= a_renames->nr) {
2290 ren2 = b_renames->items[j++].util;
2291 } else if (j >= b_renames->nr) {
2292 ren1 = a_renames->items[i++].util;
2293 } else {
2294 int compare = strcmp(a_renames->items[i].string,
2295 b_renames->items[j].string);
2296 if (compare <= 0)
2297 ren1 = a_renames->items[i++].util;
2298 if (compare >= 0)
2299 ren2 = b_renames->items[j++].util;
2302 /* TODO: refactor, so that 1/2 are not needed */
2303 if (ren1) {
2304 renames1 = a_renames;
2305 renames2Dst = &b_by_dst;
2306 branch1 = o->branch1;
2307 branch2 = o->branch2;
2308 } else {
2309 renames1 = b_renames;
2310 renames2Dst = &a_by_dst;
2311 branch1 = o->branch2;
2312 branch2 = o->branch1;
2313 SWAP(ren2, ren1);
2316 if (ren1->processed)
2317 continue;
2318 ren1->processed = 1;
2319 ren1->dst_entry->processed = 1;
2320 /* BUG: We should only mark src_entry as processed if we
2321 * are not dealing with a rename + add-source case.
2323 ren1->src_entry->processed = 1;
2325 ren1_src = ren1->pair->one->path;
2326 ren1_dst = ren1->pair->two->path;
2328 if (ren2) {
2329 /* One file renamed on both sides */
2330 const char *ren2_src = ren2->pair->one->path;
2331 const char *ren2_dst = ren2->pair->two->path;
2332 enum rename_type rename_type;
2333 if (strcmp(ren1_src, ren2_src) != 0)
2334 die("BUG: ren1_src != ren2_src");
2335 ren2->dst_entry->processed = 1;
2336 ren2->processed = 1;
2337 if (strcmp(ren1_dst, ren2_dst) != 0) {
2338 rename_type = RENAME_ONE_FILE_TO_TWO;
2339 clean_merge = 0;
2340 } else {
2341 rename_type = RENAME_ONE_FILE_TO_ONE;
2342 /* BUG: We should only remove ren1_src in
2343 * the base stage (think of rename +
2344 * add-source cases).
2346 remove_file(o, 1, ren1_src, 1);
2347 update_entry(ren1->dst_entry,
2348 ren1->pair->one,
2349 ren1->pair->two,
2350 ren2->pair->two);
2352 setup_rename_conflict_info(rename_type,
2353 ren1->pair,
2354 ren2->pair,
2355 branch1,
2356 branch2,
2357 ren1->dst_entry,
2358 ren2->dst_entry,
2360 NULL,
2361 NULL);
2362 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2363 /* Two different files renamed to the same thing */
2364 char *ren2_dst;
2365 ren2 = lookup->util;
2366 ren2_dst = ren2->pair->two->path;
2367 if (strcmp(ren1_dst, ren2_dst) != 0)
2368 die("BUG: ren1_dst != ren2_dst");
2370 clean_merge = 0;
2371 ren2->processed = 1;
2373 * BUG: We should only mark src_entry as processed
2374 * if we are not dealing with a rename + add-source
2375 * case.
2377 ren2->src_entry->processed = 1;
2379 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2380 ren1->pair,
2381 ren2->pair,
2382 branch1,
2383 branch2,
2384 ren1->dst_entry,
2385 ren2->dst_entry,
2387 ren1->src_entry,
2388 ren2->src_entry);
2390 } else {
2391 /* Renamed in 1, maybe changed in 2 */
2392 /* we only use sha1 and mode of these */
2393 struct diff_filespec src_other, dst_other;
2394 int try_merge;
2397 * unpack_trees loads entries from common-commit
2398 * into stage 1, from head-commit into stage 2, and
2399 * from merge-commit into stage 3. We keep track
2400 * of which side corresponds to the rename.
2402 int renamed_stage = a_renames == renames1 ? 2 : 3;
2403 int other_stage = a_renames == renames1 ? 3 : 2;
2405 /* BUG: We should only remove ren1_src in the base
2406 * stage and in other_stage (think of rename +
2407 * add-source case).
2409 remove_file(o, 1, ren1_src,
2410 renamed_stage == 2 || !was_tracked(ren1_src));
2412 oidcpy(&src_other.oid,
2413 &ren1->src_entry->stages[other_stage].oid);
2414 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2415 oidcpy(&dst_other.oid,
2416 &ren1->dst_entry->stages[other_stage].oid);
2417 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2418 try_merge = 0;
2420 if (oid_eq(&src_other.oid, &null_oid) &&
2421 ren1->add_turned_into_rename) {
2422 setup_rename_conflict_info(RENAME_DIR,
2423 ren1->pair,
2424 NULL,
2425 branch1,
2426 branch2,
2427 ren1->dst_entry,
2428 NULL,
2430 NULL,
2431 NULL);
2432 } else if (oid_eq(&src_other.oid, &null_oid)) {
2433 setup_rename_conflict_info(RENAME_DELETE,
2434 ren1->pair,
2435 NULL,
2436 branch1,
2437 branch2,
2438 ren1->dst_entry,
2439 NULL,
2441 NULL,
2442 NULL);
2443 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2444 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2446 * Added file on the other side identical to
2447 * the file being renamed: clean merge.
2448 * Also, there is no need to overwrite the
2449 * file already in the working copy, so call
2450 * update_file_flags() instead of
2451 * update_file().
2453 if (update_file_flags(o,
2454 &ren1->pair->two->oid,
2455 ren1->pair->two->mode,
2456 ren1_dst,
2457 1, /* update_cache */
2458 0 /* update_wd */))
2459 clean_merge = -1;
2460 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2461 clean_merge = 0;
2462 try_merge = 1;
2463 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2464 "%s added in %s"),
2465 ren1_src, ren1_dst, branch1,
2466 ren1_dst, branch2);
2467 if (o->call_depth) {
2468 struct merge_file_info mfi;
2469 if (merge_file_one(o, ren1_dst, &null_oid, 0,
2470 &ren1->pair->two->oid,
2471 ren1->pair->two->mode,
2472 &dst_other.oid,
2473 dst_other.mode,
2474 branch1, branch2, &mfi)) {
2475 clean_merge = -1;
2476 goto cleanup_and_return;
2478 output(o, 1, _("Adding merged %s"), ren1_dst);
2479 if (update_file(o, 0, &mfi.oid,
2480 mfi.mode, ren1_dst))
2481 clean_merge = -1;
2482 try_merge = 0;
2483 } else {
2484 char *new_path = unique_path(o, ren1_dst, branch2);
2485 output(o, 1, _("Adding as %s instead"), new_path);
2486 if (update_file(o, 0, &dst_other.oid,
2487 dst_other.mode, new_path))
2488 clean_merge = -1;
2489 free(new_path);
2491 } else
2492 try_merge = 1;
2494 if (clean_merge < 0)
2495 goto cleanup_and_return;
2496 if (try_merge) {
2497 struct diff_filespec *one, *a, *b;
2498 src_other.path = (char *)ren1_src;
2500 one = ren1->pair->one;
2501 if (a_renames == renames1) {
2502 a = ren1->pair->two;
2503 b = &src_other;
2504 } else {
2505 b = ren1->pair->two;
2506 a = &src_other;
2508 update_entry(ren1->dst_entry, one, a, b);
2509 setup_rename_conflict_info(RENAME_NORMAL,
2510 ren1->pair,
2511 NULL,
2512 branch1,
2513 NULL,
2514 ren1->dst_entry,
2515 NULL,
2517 NULL,
2518 NULL);
2522 cleanup_and_return:
2523 string_list_clear(&a_by_dst, 0);
2524 string_list_clear(&b_by_dst, 0);
2526 return clean_merge;
2529 struct rename_info {
2530 struct string_list *head_renames;
2531 struct string_list *merge_renames;
2534 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2535 struct hashmap *dir_renames)
2537 struct hashmap_iter iter;
2538 struct dir_rename_entry *e;
2540 hashmap_iter_init(dir_renames, &iter);
2541 while ((e = hashmap_iter_next(&iter))) {
2542 free(e->dir);
2543 strbuf_release(&e->new_dir);
2544 /* possible_new_dirs already cleared in get_directory_renames */
2546 hashmap_free(dir_renames, 1);
2547 free(dir_renames);
2549 free(pairs->queue);
2550 free(pairs);
2553 static int handle_renames(struct merge_options *o,
2554 struct tree *common,
2555 struct tree *head,
2556 struct tree *merge,
2557 struct string_list *entries,
2558 struct rename_info *ri)
2560 struct diff_queue_struct *head_pairs, *merge_pairs;
2561 struct hashmap *dir_re_head, *dir_re_merge;
2562 int clean = 1;
2564 ri->head_renames = NULL;
2565 ri->merge_renames = NULL;
2567 if (!o->detect_rename)
2568 return 1;
2570 head_pairs = get_diffpairs(o, common, head);
2571 merge_pairs = get_diffpairs(o, common, merge);
2573 dir_re_head = get_directory_renames(head_pairs, head);
2574 dir_re_merge = get_directory_renames(merge_pairs, merge);
2576 handle_directory_level_conflicts(o,
2577 dir_re_head, head,
2578 dir_re_merge, merge);
2580 ri->head_renames = get_renames(o, head_pairs,
2581 dir_re_merge, dir_re_head, head,
2582 common, head, merge, entries,
2583 &clean);
2584 if (clean < 0)
2585 goto cleanup;
2586 ri->merge_renames = get_renames(o, merge_pairs,
2587 dir_re_head, dir_re_merge, merge,
2588 common, head, merge, entries,
2589 &clean);
2590 if (clean < 0)
2591 goto cleanup;
2592 clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2594 cleanup:
2596 * Some cleanup is deferred until cleanup_renames() because the
2597 * data structures are still needed and referenced in
2598 * process_entry(). But there are a few things we can free now.
2600 initial_cleanup_rename(head_pairs, dir_re_head);
2601 initial_cleanup_rename(merge_pairs, dir_re_merge);
2603 return clean;
2606 static void final_cleanup_rename(struct string_list *rename)
2608 const struct rename *re;
2609 int i;
2611 if (rename == NULL)
2612 return;
2614 for (i = 0; i < rename->nr; i++) {
2615 re = rename->items[i].util;
2616 diff_free_filepair(re->pair);
2618 string_list_clear(rename, 1);
2619 free(rename);
2622 static void final_cleanup_renames(struct rename_info *re_info)
2624 final_cleanup_rename(re_info->head_renames);
2625 final_cleanup_rename(re_info->merge_renames);
2628 static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2630 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2633 static int read_oid_strbuf(struct merge_options *o,
2634 const struct object_id *oid, struct strbuf *dst)
2636 void *buf;
2637 enum object_type type;
2638 unsigned long size;
2639 buf = read_sha1_file(oid->hash, &type, &size);
2640 if (!buf)
2641 return err(o, _("cannot read object %s"), oid_to_hex(oid));
2642 if (type != OBJ_BLOB) {
2643 free(buf);
2644 return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2646 strbuf_attach(dst, buf, size, size + 1);
2647 return 0;
2650 static int blob_unchanged(struct merge_options *opt,
2651 const struct object_id *o_oid,
2652 unsigned o_mode,
2653 const struct object_id *a_oid,
2654 unsigned a_mode,
2655 int renormalize, const char *path)
2657 struct strbuf o = STRBUF_INIT;
2658 struct strbuf a = STRBUF_INIT;
2659 int ret = 0; /* assume changed for safety */
2661 if (a_mode != o_mode)
2662 return 0;
2663 if (oid_eq(o_oid, a_oid))
2664 return 1;
2665 if (!renormalize)
2666 return 0;
2668 assert(o_oid && a_oid);
2669 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2670 goto error_return;
2672 * Note: binary | is used so that both renormalizations are
2673 * performed. Comparison can be skipped if both files are
2674 * unchanged since their sha1s have already been compared.
2676 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2677 renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2678 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2680 error_return:
2681 strbuf_release(&o);
2682 strbuf_release(&a);
2683 return ret;
2686 static int handle_modify_delete(struct merge_options *o,
2687 const char *path,
2688 struct object_id *o_oid, int o_mode,
2689 struct object_id *a_oid, int a_mode,
2690 struct object_id *b_oid, int b_mode)
2692 const char *modify_branch, *delete_branch;
2693 struct object_id *changed_oid;
2694 int changed_mode;
2696 if (a_oid) {
2697 modify_branch = o->branch1;
2698 delete_branch = o->branch2;
2699 changed_oid = a_oid;
2700 changed_mode = a_mode;
2701 } else {
2702 modify_branch = o->branch2;
2703 delete_branch = o->branch1;
2704 changed_oid = b_oid;
2705 changed_mode = b_mode;
2708 return handle_change_delete(o,
2709 path, NULL,
2710 o_oid, o_mode,
2711 changed_oid, changed_mode,
2712 modify_branch, delete_branch,
2713 _("modify"), _("modified"));
2716 static int merge_content(struct merge_options *o,
2717 const char *path,
2718 int file_in_way,
2719 struct object_id *o_oid, int o_mode,
2720 struct object_id *a_oid, int a_mode,
2721 struct object_id *b_oid, int b_mode,
2722 struct rename_conflict_info *rename_conflict_info)
2724 const char *reason = _("content");
2725 const char *path1 = NULL, *path2 = NULL;
2726 struct merge_file_info mfi;
2727 struct diff_filespec one, a, b;
2728 unsigned df_conflict_remains = 0;
2730 if (!o_oid) {
2731 reason = _("add/add");
2732 o_oid = (struct object_id *)&null_oid;
2734 one.path = a.path = b.path = (char *)path;
2735 oidcpy(&one.oid, o_oid);
2736 one.mode = o_mode;
2737 oidcpy(&a.oid, a_oid);
2738 a.mode = a_mode;
2739 oidcpy(&b.oid, b_oid);
2740 b.mode = b_mode;
2742 if (rename_conflict_info) {
2743 struct diff_filepair *pair1 = rename_conflict_info->pair1;
2745 path1 = (o->branch1 == rename_conflict_info->branch1) ?
2746 pair1->two->path : pair1->one->path;
2747 /* If rename_conflict_info->pair2 != NULL, we are in
2748 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
2749 * normal rename.
2751 path2 = (rename_conflict_info->pair2 ||
2752 o->branch2 == rename_conflict_info->branch1) ?
2753 pair1->two->path : pair1->one->path;
2755 if (dir_in_way(path, !o->call_depth,
2756 S_ISGITLINK(pair1->two->mode)))
2757 df_conflict_remains = 1;
2759 if (merge_file_special_markers(o, &one, &a, &b,
2760 o->branch1, path1,
2761 o->branch2, path2, &mfi))
2762 return -1;
2764 if (mfi.clean && !df_conflict_remains &&
2765 oid_eq(&mfi.oid, a_oid) && mfi.mode == a_mode) {
2766 output(o, 3, _("Skipped %s (merged same as existing)"), path);
2768 * The content merge resulted in the same file contents we
2769 * already had. We can return early if those file contents
2770 * are recorded at the correct path (which may not be true
2771 * if the merge involves a rename).
2773 if (was_tracked(path)) {
2774 add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2775 0, (!o->call_depth), 0);
2776 return mfi.clean;
2778 } else
2779 output(o, 2, _("Auto-merging %s"), path);
2781 if (!mfi.clean) {
2782 if (S_ISGITLINK(mfi.mode))
2783 reason = _("submodule");
2784 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2785 reason, path);
2786 if (rename_conflict_info && !df_conflict_remains)
2787 if (update_stages(o, path, &one, &a, &b))
2788 return -1;
2791 if (df_conflict_remains || file_in_way) {
2792 char *new_path;
2793 if (o->call_depth) {
2794 remove_file_from_cache(path);
2795 } else {
2796 if (!mfi.clean) {
2797 if (update_stages(o, path, &one, &a, &b))
2798 return -1;
2799 } else {
2800 int file_from_stage2 = was_tracked(path);
2801 struct diff_filespec merged;
2802 oidcpy(&merged.oid, &mfi.oid);
2803 merged.mode = mfi.mode;
2805 if (update_stages(o, path, NULL,
2806 file_from_stage2 ? &merged : NULL,
2807 file_from_stage2 ? NULL : &merged))
2808 return -1;
2812 new_path = unique_path(o, path, rename_conflict_info->branch1);
2813 output(o, 1, _("Adding as %s instead"), new_path);
2814 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2815 free(new_path);
2816 return -1;
2818 free(new_path);
2819 mfi.clean = 0;
2820 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2821 return -1;
2822 return mfi.clean;
2825 static int conflict_rename_normal(struct merge_options *o,
2826 const char *path,
2827 struct object_id *o_oid, unsigned int o_mode,
2828 struct object_id *a_oid, unsigned int a_mode,
2829 struct object_id *b_oid, unsigned int b_mode,
2830 struct rename_conflict_info *ci)
2832 int clean_merge;
2833 int file_in_the_way = 0;
2835 if (was_dirty(o, path)) {
2836 file_in_the_way = 1;
2837 output(o, 1, _("Refusing to lose dirty file at %s"), path);
2840 /* Merge the content and write it out */
2841 clean_merge = merge_content(o, path, file_in_the_way,
2842 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2843 ci);
2844 if (clean_merge > 0 && file_in_the_way)
2845 clean_merge = 0;
2846 return clean_merge;
2849 /* Per entry merge function */
2850 static int process_entry(struct merge_options *o,
2851 const char *path, struct stage_data *entry)
2853 int clean_merge = 1;
2854 int normalize = o->renormalize;
2855 unsigned o_mode = entry->stages[1].mode;
2856 unsigned a_mode = entry->stages[2].mode;
2857 unsigned b_mode = entry->stages[3].mode;
2858 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2859 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2860 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2862 entry->processed = 1;
2863 if (entry->rename_conflict_info) {
2864 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2865 switch (conflict_info->rename_type) {
2866 case RENAME_NORMAL:
2867 case RENAME_ONE_FILE_TO_ONE:
2868 clean_merge = conflict_rename_normal(o,
2869 path,
2870 o_oid, o_mode,
2871 a_oid, a_mode,
2872 b_oid, b_mode,
2873 conflict_info);
2874 break;
2875 case RENAME_DIR:
2876 clean_merge = 1;
2877 if (conflict_rename_dir(o,
2878 conflict_info->pair1,
2879 conflict_info->branch1,
2880 conflict_info->branch2))
2881 clean_merge = -1;
2882 break;
2883 case RENAME_DELETE:
2884 clean_merge = 0;
2885 if (conflict_rename_delete(o,
2886 conflict_info->pair1,
2887 conflict_info->branch1,
2888 conflict_info->branch2))
2889 clean_merge = -1;
2890 break;
2891 case RENAME_ONE_FILE_TO_TWO:
2892 clean_merge = 0;
2893 if (conflict_rename_rename_1to2(o, conflict_info))
2894 clean_merge = -1;
2895 break;
2896 case RENAME_TWO_FILES_TO_ONE:
2897 clean_merge = 0;
2898 if (conflict_rename_rename_2to1(o, conflict_info))
2899 clean_merge = -1;
2900 break;
2901 default:
2902 entry->processed = 0;
2903 break;
2905 } else if (o_oid && (!a_oid || !b_oid)) {
2906 /* Case A: Deleted in one */
2907 if ((!a_oid && !b_oid) ||
2908 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2909 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2910 /* Deleted in both or deleted in one and
2911 * unchanged in the other */
2912 if (a_oid)
2913 output(o, 2, _("Removing %s"), path);
2914 /* do not touch working file if it did not exist */
2915 remove_file(o, 1, path, !a_oid);
2916 } else {
2917 /* Modify/delete; deleted side may have put a directory in the way */
2918 clean_merge = 0;
2919 if (handle_modify_delete(o, path, o_oid, o_mode,
2920 a_oid, a_mode, b_oid, b_mode))
2921 clean_merge = -1;
2923 } else if ((!o_oid && a_oid && !b_oid) ||
2924 (!o_oid && !a_oid && b_oid)) {
2925 /* Case B: Added in one. */
2926 /* [nothing|directory] -> ([nothing|directory], file) */
2928 const char *add_branch;
2929 const char *other_branch;
2930 unsigned mode;
2931 const struct object_id *oid;
2932 const char *conf;
2934 if (a_oid) {
2935 add_branch = o->branch1;
2936 other_branch = o->branch2;
2937 mode = a_mode;
2938 oid = a_oid;
2939 conf = _("file/directory");
2940 } else {
2941 add_branch = o->branch2;
2942 other_branch = o->branch1;
2943 mode = b_mode;
2944 oid = b_oid;
2945 conf = _("directory/file");
2947 if (dir_in_way(path,
2948 !o->call_depth && !S_ISGITLINK(a_mode),
2949 0)) {
2950 char *new_path = unique_path(o, path, add_branch);
2951 clean_merge = 0;
2952 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
2953 "Adding %s as %s"),
2954 conf, path, other_branch, path, new_path);
2955 if (update_file(o, 0, oid, mode, new_path))
2956 clean_merge = -1;
2957 else if (o->call_depth)
2958 remove_file_from_cache(path);
2959 free(new_path);
2960 } else {
2961 output(o, 2, _("Adding %s"), path);
2962 /* do not overwrite file if already present */
2963 if (update_file_flags(o, oid, mode, path, 1, !a_oid))
2964 clean_merge = -1;
2966 } else if (a_oid && b_oid) {
2967 /* Case C: Added in both (check for same permissions) and */
2968 /* case D: Modified in both, but differently. */
2969 clean_merge = merge_content(o, path, 0 /* file_in_way */,
2970 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2971 NULL);
2972 } else if (!o_oid && !a_oid && !b_oid) {
2974 * this entry was deleted altogether. a_mode == 0 means
2975 * we had that path and want to actively remove it.
2977 remove_file(o, 1, path, !a_mode);
2978 } else
2979 die("BUG: fatal merge failure, shouldn't happen.");
2981 return clean_merge;
2984 int merge_trees(struct merge_options *o,
2985 struct tree *head,
2986 struct tree *merge,
2987 struct tree *common,
2988 struct tree **result)
2990 int code, clean;
2992 if (o->subtree_shift) {
2993 merge = shift_tree_object(head, merge, o->subtree_shift);
2994 common = shift_tree_object(head, common, o->subtree_shift);
2997 if (oid_eq(&common->object.oid, &merge->object.oid)) {
2998 struct strbuf sb = STRBUF_INIT;
3000 if (!o->call_depth && index_has_changes(&sb)) {
3001 err(o, _("Dirty index: cannot merge (dirty: %s)"),
3002 sb.buf);
3003 return 0;
3005 output(o, 0, _("Already up to date!"));
3006 *result = head;
3007 return 1;
3010 code = git_merge_trees(o, common, head, merge);
3012 if (code != 0) {
3013 if (show(o, 4) || o->call_depth)
3014 err(o, _("merging of trees %s and %s failed"),
3015 oid_to_hex(&head->object.oid),
3016 oid_to_hex(&merge->object.oid));
3017 return -1;
3020 if (unmerged_cache()) {
3021 struct string_list *entries;
3022 struct rename_info re_info;
3023 int i;
3025 * Only need the hashmap while processing entries, so
3026 * initialize it here and free it when we are done running
3027 * through the entries. Keeping it in the merge_options as
3028 * opposed to decaring a local hashmap is for convenience
3029 * so that we don't have to pass it to around.
3031 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3032 get_files_dirs(o, head);
3033 get_files_dirs(o, merge);
3035 entries = get_unmerged();
3036 clean = handle_renames(o, common, head, merge, entries,
3037 &re_info);
3038 record_df_conflict_files(o, entries);
3039 if (clean < 0)
3040 goto cleanup;
3041 for (i = entries->nr-1; 0 <= i; i--) {
3042 const char *path = entries->items[i].string;
3043 struct stage_data *e = entries->items[i].util;
3044 if (!e->processed) {
3045 int ret = process_entry(o, path, e);
3046 if (!ret)
3047 clean = 0;
3048 else if (ret < 0) {
3049 clean = ret;
3050 goto cleanup;
3054 for (i = 0; i < entries->nr; i++) {
3055 struct stage_data *e = entries->items[i].util;
3056 if (!e->processed)
3057 die("BUG: unprocessed path??? %s",
3058 entries->items[i].string);
3061 cleanup:
3062 final_cleanup_renames(&re_info);
3064 string_list_clear(entries, 1);
3065 free(entries);
3067 hashmap_free(&o->current_file_dir_set, 1);
3069 if (clean < 0)
3070 return clean;
3072 else
3073 clean = 1;
3075 if (o->call_depth && !(*result = write_tree_from_memory(o)))
3076 return -1;
3078 return clean;
3081 static struct commit_list *reverse_commit_list(struct commit_list *list)
3083 struct commit_list *next = NULL, *current, *backup;
3084 for (current = list; current; current = backup) {
3085 backup = current->next;
3086 current->next = next;
3087 next = current;
3089 return next;
3093 * Merge the commits h1 and h2, return the resulting virtual
3094 * commit object and a flag indicating the cleanness of the merge.
3096 int merge_recursive(struct merge_options *o,
3097 struct commit *h1,
3098 struct commit *h2,
3099 struct commit_list *ca,
3100 struct commit **result)
3102 struct commit_list *iter;
3103 struct commit *merged_common_ancestors;
3104 struct tree *mrtree = mrtree;
3105 int clean;
3107 if (show(o, 4)) {
3108 output(o, 4, _("Merging:"));
3109 output_commit_title(o, h1);
3110 output_commit_title(o, h2);
3113 if (!ca) {
3114 ca = get_merge_bases(h1, h2);
3115 ca = reverse_commit_list(ca);
3118 if (show(o, 5)) {
3119 unsigned cnt = commit_list_count(ca);
3121 output(o, 5, Q_("found %u common ancestor:",
3122 "found %u common ancestors:", cnt), cnt);
3123 for (iter = ca; iter; iter = iter->next)
3124 output_commit_title(o, iter->item);
3127 merged_common_ancestors = pop_commit(&ca);
3128 if (merged_common_ancestors == NULL) {
3129 /* if there is no common ancestor, use an empty tree */
3130 struct tree *tree;
3132 tree = lookup_tree(the_hash_algo->empty_tree);
3133 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3136 for (iter = ca; iter; iter = iter->next) {
3137 const char *saved_b1, *saved_b2;
3138 o->call_depth++;
3140 * When the merge fails, the result contains files
3141 * with conflict markers. The cleanness flag is
3142 * ignored (unless indicating an error), it was never
3143 * actually used, as result of merge_trees has always
3144 * overwritten it: the committed "conflicts" were
3145 * already resolved.
3147 discard_cache();
3148 saved_b1 = o->branch1;
3149 saved_b2 = o->branch2;
3150 o->branch1 = "Temporary merge branch 1";
3151 o->branch2 = "Temporary merge branch 2";
3152 if (merge_recursive(o, merged_common_ancestors, iter->item,
3153 NULL, &merged_common_ancestors) < 0)
3154 return -1;
3155 o->branch1 = saved_b1;
3156 o->branch2 = saved_b2;
3157 o->call_depth--;
3159 if (!merged_common_ancestors)
3160 return err(o, _("merge returned no commit"));
3163 discard_cache();
3164 if (!o->call_depth)
3165 read_cache();
3167 o->ancestor = "merged common ancestors";
3168 clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,
3169 &mrtree);
3170 if (clean < 0) {
3171 flush_output(o);
3172 return clean;
3175 if (o->call_depth) {
3176 *result = make_virtual_commit(mrtree, "merged tree");
3177 commit_list_insert(h1, &(*result)->parents);
3178 commit_list_insert(h2, &(*result)->parents->next);
3180 flush_output(o);
3181 if (!o->call_depth && o->buffer_output < 2)
3182 strbuf_release(&o->obuf);
3183 if (show(o, 2))
3184 diff_warn_rename_limit("merge.renamelimit",
3185 o->needed_rename_limit, 0);
3186 return clean;
3189 static struct commit *get_ref(const struct object_id *oid, const char *name)
3191 struct object *object;
3193 object = deref_tag(parse_object(oid), name, strlen(name));
3194 if (!object)
3195 return NULL;
3196 if (object->type == OBJ_TREE)
3197 return make_virtual_commit((struct tree*)object, name);
3198 if (object->type != OBJ_COMMIT)
3199 return NULL;
3200 if (parse_commit((struct commit *)object))
3201 return NULL;
3202 return (struct commit *)object;
3205 int merge_recursive_generic(struct merge_options *o,
3206 const struct object_id *head,
3207 const struct object_id *merge,
3208 int num_base_list,
3209 const struct object_id **base_list,
3210 struct commit **result)
3212 int clean;
3213 struct lock_file lock = LOCK_INIT;
3214 struct commit *head_commit = get_ref(head, o->branch1);
3215 struct commit *next_commit = get_ref(merge, o->branch2);
3216 struct commit_list *ca = NULL;
3218 if (base_list) {
3219 int i;
3220 for (i = 0; i < num_base_list; ++i) {
3221 struct commit *base;
3222 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3223 return err(o, _("Could not parse object '%s'"),
3224 oid_to_hex(base_list[i]));
3225 commit_list_insert(base, &ca);
3229 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3230 clean = merge_recursive(o, head_commit, next_commit, ca,
3231 result);
3232 if (clean < 0)
3233 return clean;
3235 if (active_cache_changed &&
3236 write_locked_index(&the_index, &lock, COMMIT_LOCK))
3237 return err(o, _("Unable to write index."));
3239 return clean ? 0 : 1;
3242 static void merge_recursive_config(struct merge_options *o)
3244 git_config_get_int("merge.verbosity", &o->verbosity);
3245 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3246 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3247 git_config(git_xmerge_config, NULL);
3250 void init_merge_options(struct merge_options *o)
3252 const char *merge_verbosity;
3253 memset(o, 0, sizeof(struct merge_options));
3254 o->verbosity = 2;
3255 o->buffer_output = 1;
3256 o->diff_rename_limit = -1;
3257 o->merge_rename_limit = -1;
3258 o->renormalize = 0;
3259 o->detect_rename = 1;
3260 merge_recursive_config(o);
3261 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3262 if (merge_verbosity)
3263 o->verbosity = strtol(merge_verbosity, NULL, 10);
3264 if (o->verbosity >= 5)
3265 o->buffer_output = 0;
3266 strbuf_init(&o->obuf, 0);
3267 string_list_init(&o->df_conflict_file_set, 1);
3270 int parse_merge_opt(struct merge_options *o, const char *s)
3272 const char *arg;
3274 if (!s || !*s)
3275 return -1;
3276 if (!strcmp(s, "ours"))
3277 o->recursive_variant = MERGE_RECURSIVE_OURS;
3278 else if (!strcmp(s, "theirs"))
3279 o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3280 else if (!strcmp(s, "subtree"))
3281 o->subtree_shift = "";
3282 else if (skip_prefix(s, "subtree=", &arg))
3283 o->subtree_shift = arg;
3284 else if (!strcmp(s, "patience"))
3285 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3286 else if (!strcmp(s, "histogram"))
3287 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3288 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3289 long value = parse_algorithm_value(arg);
3290 if (value < 0)
3291 return -1;
3292 /* clear out previous settings */
3293 DIFF_XDL_CLR(o, NEED_MINIMAL);
3294 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3295 o->xdl_opts |= value;
3297 else if (!strcmp(s, "ignore-space-change"))
3298 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3299 else if (!strcmp(s, "ignore-all-space"))
3300 DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3301 else if (!strcmp(s, "ignore-space-at-eol"))
3302 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3303 else if (!strcmp(s, "ignore-cr-at-eol"))
3304 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3305 else if (!strcmp(s, "renormalize"))
3306 o->renormalize = 1;
3307 else if (!strcmp(s, "no-renormalize"))
3308 o->renormalize = 0;
3309 else if (!strcmp(s, "no-renames"))
3310 o->detect_rename = 0;
3311 else if (!strcmp(s, "find-renames")) {
3312 o->detect_rename = 1;
3313 o->rename_score = 0;
3315 else if (skip_prefix(s, "find-renames=", &arg) ||
3316 skip_prefix(s, "rename-threshold=", &arg)) {
3317 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3318 return -1;
3319 o->detect_rename = 1;
3321 else
3322 return -1;
3323 return 0;