argv-array: return the pushed string from argv_push*()
[git/git-svn.git] / merge-recursive.c
blobddb0fa73698122542e727befa7fbedf47b9226bf
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,
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, _("add_cacheinfo failed for path '%s'; merge aborting."), 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, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), 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 unpack_trees_start(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];
347 struct index_state tmp_index = { NULL };
349 memset(&o->unpack_opts, 0, sizeof(o->unpack_opts));
350 if (o->call_depth)
351 o->unpack_opts.index_only = 1;
352 else
353 o->unpack_opts.update = 1;
354 o->unpack_opts.merge = 1;
355 o->unpack_opts.head_idx = 2;
356 o->unpack_opts.fn = threeway_merge;
357 o->unpack_opts.src_index = &the_index;
358 o->unpack_opts.dst_index = &tmp_index;
359 o->unpack_opts.aggressive = !merge_detect_rename(o);
360 setup_unpack_trees_porcelain(&o->unpack_opts, "merge");
362 init_tree_desc_from_tree(t+0, common);
363 init_tree_desc_from_tree(t+1, head);
364 init_tree_desc_from_tree(t+2, merge);
366 rc = unpack_trees(3, t, &o->unpack_opts);
367 cache_tree_free(&active_cache_tree);
370 * Update the_index to match the new results, AFTER saving a copy
371 * in o->orig_index. Update src_index to point to the saved copy.
372 * (verify_uptodate() checks src_index, and the original index is
373 * the one that had the necessary modification timestamps.)
375 o->orig_index = the_index;
376 the_index = tmp_index;
377 o->unpack_opts.src_index = &o->orig_index;
379 return rc;
382 static void unpack_trees_finish(struct merge_options *o)
384 discard_index(&o->orig_index);
387 struct tree *write_tree_from_memory(struct merge_options *o)
389 struct tree *result = NULL;
391 if (unmerged_cache()) {
392 int i;
393 fprintf(stderr, "BUG: There are unmerged index entries:\n");
394 for (i = 0; i < active_nr; i++) {
395 const struct cache_entry *ce = active_cache[i];
396 if (ce_stage(ce))
397 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
398 (int)ce_namelen(ce), ce->name);
400 die("BUG: unmerged index entries in merge-recursive.c");
403 if (!active_cache_tree)
404 active_cache_tree = cache_tree();
406 if (!cache_tree_fully_valid(active_cache_tree) &&
407 cache_tree_update(&the_index, 0) < 0) {
408 err(o, _("error building trees"));
409 return NULL;
412 result = lookup_tree(&active_cache_tree->oid);
414 return result;
417 static int save_files_dirs(const struct object_id *oid,
418 struct strbuf *base, const char *path,
419 unsigned int mode, int stage, void *context)
421 struct path_hashmap_entry *entry;
422 int baselen = base->len;
423 struct merge_options *o = context;
425 strbuf_addstr(base, path);
427 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
428 hashmap_entry_init(entry, path_hash(entry->path));
429 hashmap_add(&o->current_file_dir_set, entry);
431 strbuf_setlen(base, baselen);
432 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
435 static void get_files_dirs(struct merge_options *o, struct tree *tree)
437 struct pathspec match_all;
438 memset(&match_all, 0, sizeof(match_all));
439 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
442 static int get_tree_entry_if_blob(const struct object_id *tree,
443 const char *path,
444 struct object_id *hashy,
445 unsigned int *mode_o)
447 int ret;
449 ret = get_tree_entry(tree, path, hashy, mode_o);
450 if (S_ISDIR(*mode_o)) {
451 oidcpy(hashy, &null_oid);
452 *mode_o = 0;
454 return ret;
458 * Returns an index_entry instance which doesn't have to correspond to
459 * a real cache entry in Git's index.
461 static struct stage_data *insert_stage_data(const char *path,
462 struct tree *o, struct tree *a, struct tree *b,
463 struct string_list *entries)
465 struct string_list_item *item;
466 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
467 get_tree_entry_if_blob(&o->object.oid, path,
468 &e->stages[1].oid, &e->stages[1].mode);
469 get_tree_entry_if_blob(&a->object.oid, path,
470 &e->stages[2].oid, &e->stages[2].mode);
471 get_tree_entry_if_blob(&b->object.oid, path,
472 &e->stages[3].oid, &e->stages[3].mode);
473 item = string_list_insert(entries, path);
474 item->util = e;
475 return e;
479 * Create a dictionary mapping file names to stage_data objects. The
480 * dictionary contains one entry for every path with a non-zero stage entry.
482 static struct string_list *get_unmerged(void)
484 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
485 int i;
487 unmerged->strdup_strings = 1;
489 for (i = 0; i < active_nr; i++) {
490 struct string_list_item *item;
491 struct stage_data *e;
492 const struct cache_entry *ce = active_cache[i];
493 if (!ce_stage(ce))
494 continue;
496 item = string_list_lookup(unmerged, ce->name);
497 if (!item) {
498 item = string_list_insert(unmerged, ce->name);
499 item->util = xcalloc(1, sizeof(struct stage_data));
501 e = item->util;
502 e->stages[ce_stage(ce)].mode = ce->ce_mode;
503 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
506 return unmerged;
509 static int string_list_df_name_compare(const char *one, const char *two)
511 int onelen = strlen(one);
512 int twolen = strlen(two);
514 * Here we only care that entries for D/F conflicts are
515 * adjacent, in particular with the file of the D/F conflict
516 * appearing before files below the corresponding directory.
517 * The order of the rest of the list is irrelevant for us.
519 * To achieve this, we sort with df_name_compare and provide
520 * the mode S_IFDIR so that D/F conflicts will sort correctly.
521 * We use the mode S_IFDIR for everything else for simplicity,
522 * since in other cases any changes in their order due to
523 * sorting cause no problems for us.
525 int cmp = df_name_compare(one, onelen, S_IFDIR,
526 two, twolen, S_IFDIR);
528 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
529 * that 'foo' comes before 'foo/bar'.
531 if (cmp)
532 return cmp;
533 return onelen - twolen;
536 static void record_df_conflict_files(struct merge_options *o,
537 struct string_list *entries)
539 /* If there is a D/F conflict and the file for such a conflict
540 * currently exist in the working tree, we want to allow it to be
541 * removed to make room for the corresponding directory if needed.
542 * The files underneath the directories of such D/F conflicts will
543 * be processed before the corresponding file involved in the D/F
544 * conflict. If the D/F directory ends up being removed by the
545 * merge, then we won't have to touch the D/F file. If the D/F
546 * directory needs to be written to the working copy, then the D/F
547 * file will simply be removed (in make_room_for_path()) to make
548 * room for the necessary paths. Note that if both the directory
549 * and the file need to be present, then the D/F file will be
550 * reinstated with a new unique name at the time it is processed.
552 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
553 const char *last_file = NULL;
554 int last_len = 0;
555 int i;
558 * If we're merging merge-bases, we don't want to bother with
559 * any working directory changes.
561 if (o->call_depth)
562 return;
564 /* Ensure D/F conflicts are adjacent in the entries list. */
565 for (i = 0; i < entries->nr; i++) {
566 struct string_list_item *next = &entries->items[i];
567 string_list_append(&df_sorted_entries, next->string)->util =
568 next->util;
570 df_sorted_entries.cmp = string_list_df_name_compare;
571 string_list_sort(&df_sorted_entries);
573 string_list_clear(&o->df_conflict_file_set, 1);
574 for (i = 0; i < df_sorted_entries.nr; i++) {
575 const char *path = df_sorted_entries.items[i].string;
576 int len = strlen(path);
577 struct stage_data *e = df_sorted_entries.items[i].util;
580 * Check if last_file & path correspond to a D/F conflict;
581 * i.e. whether path is last_file+'/'+<something>.
582 * If so, record that it's okay to remove last_file to make
583 * room for path and friends if needed.
585 if (last_file &&
586 len > last_len &&
587 memcmp(path, last_file, last_len) == 0 &&
588 path[last_len] == '/') {
589 string_list_insert(&o->df_conflict_file_set, last_file);
593 * Determine whether path could exist as a file in the
594 * working directory as a possible D/F conflict. This
595 * will only occur when it exists in stage 2 as a
596 * file.
598 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
599 last_file = path;
600 last_len = len;
601 } else {
602 last_file = NULL;
605 string_list_clear(&df_sorted_entries, 0);
608 struct rename {
609 struct diff_filepair *pair;
611 * Purpose of src_entry and dst_entry:
613 * If 'before' is renamed to 'after' then src_entry will contain
614 * the versions of 'before' from the merge_base, HEAD, and MERGE in
615 * stages 1, 2, and 3; dst_entry will contain the respective
616 * versions of 'after' in corresponding locations. Thus, we have a
617 * total of six modes and oids, though some will be null. (Stage 0
618 * is ignored; we're interested in handling conflicts.)
620 * Since we don't turn on break-rewrites by default, neither
621 * src_entry nor dst_entry can have all three of their stages have
622 * non-null oids, meaning at most four of the six will be non-null.
623 * Also, since this is a rename, both src_entry and dst_entry will
624 * have at least one non-null oid, meaning at least two will be
625 * non-null. Of the six oids, a typical rename will have three be
626 * non-null. Only two implies a rename/delete, and four implies a
627 * rename/add.
629 struct stage_data *src_entry;
630 struct stage_data *dst_entry;
631 unsigned add_turned_into_rename:1;
632 unsigned processed:1;
635 static int update_stages(struct merge_options *opt, const char *path,
636 const struct diff_filespec *o,
637 const struct diff_filespec *a,
638 const struct diff_filespec *b)
642 * NOTE: It is usually a bad idea to call update_stages on a path
643 * before calling update_file on that same path, since it can
644 * sometimes lead to spurious "refusing to lose untracked file..."
645 * messages from update_file (via make_room_for path via
646 * would_lose_untracked). Instead, reverse the order of the calls
647 * (executing update_file first and then update_stages).
649 int clear = 1;
650 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
651 if (clear)
652 if (remove_file_from_cache(path))
653 return -1;
654 if (o)
655 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
656 return -1;
657 if (a)
658 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
659 return -1;
660 if (b)
661 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
662 return -1;
663 return 0;
666 static int update_stages_for_stage_data(struct merge_options *opt,
667 const char *path,
668 const struct stage_data *stage_data)
670 struct diff_filespec o, a, b;
672 o.mode = stage_data->stages[1].mode;
673 oidcpy(&o.oid, &stage_data->stages[1].oid);
675 a.mode = stage_data->stages[2].mode;
676 oidcpy(&a.oid, &stage_data->stages[2].oid);
678 b.mode = stage_data->stages[3].mode;
679 oidcpy(&b.oid, &stage_data->stages[3].oid);
681 return update_stages(opt, path,
682 is_null_oid(&o.oid) ? NULL : &o,
683 is_null_oid(&a.oid) ? NULL : &a,
684 is_null_oid(&b.oid) ? NULL : &b);
687 static void update_entry(struct stage_data *entry,
688 struct diff_filespec *o,
689 struct diff_filespec *a,
690 struct diff_filespec *b)
692 entry->processed = 0;
693 entry->stages[1].mode = o->mode;
694 entry->stages[2].mode = a->mode;
695 entry->stages[3].mode = b->mode;
696 oidcpy(&entry->stages[1].oid, &o->oid);
697 oidcpy(&entry->stages[2].oid, &a->oid);
698 oidcpy(&entry->stages[3].oid, &b->oid);
701 static int remove_file(struct merge_options *o, int clean,
702 const char *path, int no_wd)
704 int update_cache = o->call_depth || clean;
705 int update_working_directory = !o->call_depth && !no_wd;
707 if (update_cache) {
708 if (remove_file_from_cache(path))
709 return -1;
711 if (update_working_directory) {
712 if (ignore_case) {
713 struct cache_entry *ce;
714 ce = cache_file_exists(path, strlen(path), ignore_case);
715 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
716 return 0;
718 if (remove_path(path))
719 return -1;
721 return 0;
724 /* add a string to a strbuf, but converting "/" to "_" */
725 static void add_flattened_path(struct strbuf *out, const char *s)
727 size_t i = out->len;
728 strbuf_addstr(out, s);
729 for (; i < out->len; i++)
730 if (out->buf[i] == '/')
731 out->buf[i] = '_';
734 static char *unique_path(struct merge_options *o, const char *path, const char *branch)
736 struct path_hashmap_entry *entry;
737 struct strbuf newpath = STRBUF_INIT;
738 int suffix = 0;
739 size_t base_len;
741 strbuf_addf(&newpath, "%s~", path);
742 add_flattened_path(&newpath, branch);
744 base_len = newpath.len;
745 while (hashmap_get_from_hash(&o->current_file_dir_set,
746 path_hash(newpath.buf), newpath.buf) ||
747 (!o->call_depth && file_exists(newpath.buf))) {
748 strbuf_setlen(&newpath, base_len);
749 strbuf_addf(&newpath, "_%d", suffix++);
752 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
753 hashmap_entry_init(entry, path_hash(entry->path));
754 hashmap_add(&o->current_file_dir_set, entry);
755 return strbuf_detach(&newpath, NULL);
759 * Check whether a directory in the index is in the way of an incoming
760 * file. Return 1 if so. If check_working_copy is non-zero, also
761 * check the working directory. If empty_ok is non-zero, also return
762 * 0 in the case where the working-tree dir exists but is empty.
764 static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
766 int pos;
767 struct strbuf dirpath = STRBUF_INIT;
768 struct stat st;
770 strbuf_addstr(&dirpath, path);
771 strbuf_addch(&dirpath, '/');
773 pos = cache_name_pos(dirpath.buf, dirpath.len);
775 if (pos < 0)
776 pos = -1 - pos;
777 if (pos < active_nr &&
778 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
779 strbuf_release(&dirpath);
780 return 1;
783 strbuf_release(&dirpath);
784 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
785 !(empty_ok && is_empty_dir(path));
789 * Returns whether path was tracked in the index before the merge started,
790 * and its oid and mode match the specified values
792 static int was_tracked_and_matches(struct merge_options *o, const char *path,
793 const struct object_id *oid, unsigned mode)
795 int pos = index_name_pos(&o->orig_index, path, strlen(path));
796 struct cache_entry *ce;
798 if (0 > pos)
799 /* we were not tracking this path before the merge */
800 return 0;
802 /* See if the file we were tracking before matches */
803 ce = o->orig_index.cache[pos];
804 return (oid_eq(&ce->oid, oid) && ce->ce_mode == mode);
808 * Returns whether path was tracked in the index before the merge started
810 static int was_tracked(struct merge_options *o, const char *path)
812 int pos = index_name_pos(&o->orig_index, path, strlen(path));
814 if (0 <= pos)
815 /* we were tracking this path before the merge */
816 return 1;
818 return 0;
821 static int would_lose_untracked(const char *path)
824 * This may look like it can be simplified to:
825 * return !was_tracked(o, path) && file_exists(path)
826 * but it can't. This function needs to know whether path was in
827 * the working tree due to EITHER having been tracked in the index
828 * before the merge OR having been put into the working copy and
829 * index by unpack_trees(). Due to that either-or requirement, we
830 * check the current index instead of the original one.
832 * Note that we do not need to worry about merge-recursive itself
833 * updating the index after unpack_trees() and before calling this
834 * function, because we strictly require all code paths in
835 * merge-recursive to update the working tree first and the index
836 * second. Doing otherwise would break
837 * update_file()/would_lose_untracked(); see every comment in this
838 * file which mentions "update_stages".
840 int pos = cache_name_pos(path, strlen(path));
842 if (pos < 0)
843 pos = -1 - pos;
844 while (pos < active_nr &&
845 !strcmp(path, active_cache[pos]->name)) {
847 * If stage #0, it is definitely tracked.
848 * If it has stage #2 then it was tracked
849 * before this merge started. All other
850 * cases the path was not tracked.
852 switch (ce_stage(active_cache[pos])) {
853 case 0:
854 case 2:
855 return 0;
857 pos++;
859 return file_exists(path);
862 static int was_dirty(struct merge_options *o, const char *path)
864 struct cache_entry *ce;
865 int dirty = 1;
867 if (o->call_depth || !was_tracked(o, path))
868 return !dirty;
870 ce = index_file_exists(o->unpack_opts.src_index,
871 path, strlen(path), ignore_case);
872 dirty = verify_uptodate(ce, &o->unpack_opts) != 0;
873 return dirty;
876 static int make_room_for_path(struct merge_options *o, const char *path)
878 int status, i;
879 const char *msg = _("failed to create path '%s'%s");
881 /* Unlink any D/F conflict files that are in the way */
882 for (i = 0; i < o->df_conflict_file_set.nr; i++) {
883 const char *df_path = o->df_conflict_file_set.items[i].string;
884 size_t pathlen = strlen(path);
885 size_t df_pathlen = strlen(df_path);
886 if (df_pathlen < pathlen &&
887 path[df_pathlen] == '/' &&
888 strncmp(path, df_path, df_pathlen) == 0) {
889 output(o, 3,
890 _("Removing %s to make room for subdirectory\n"),
891 df_path);
892 unlink(df_path);
893 unsorted_string_list_delete_item(&o->df_conflict_file_set,
894 i, 0);
895 break;
899 /* Make sure leading directories are created */
900 status = safe_create_leading_directories_const(path);
901 if (status) {
902 if (status == SCLD_EXISTS)
903 /* something else exists */
904 return err(o, msg, path, _(": perhaps a D/F conflict?"));
905 return err(o, msg, path, "");
909 * Do not unlink a file in the work tree if we are not
910 * tracking it.
912 if (would_lose_untracked(path))
913 return err(o, _("refusing to lose untracked file at '%s'"),
914 path);
916 /* Successful unlink is good.. */
917 if (!unlink(path))
918 return 0;
919 /* .. and so is no existing file */
920 if (errno == ENOENT)
921 return 0;
922 /* .. but not some other error (who really cares what?) */
923 return err(o, msg, path, _(": perhaps a D/F conflict?"));
926 static int update_file_flags(struct merge_options *o,
927 const struct object_id *oid,
928 unsigned mode,
929 const char *path,
930 int update_cache,
931 int update_wd)
933 int ret = 0;
935 if (o->call_depth)
936 update_wd = 0;
938 if (update_wd) {
939 enum object_type type;
940 void *buf;
941 unsigned long size;
943 if (S_ISGITLINK(mode)) {
945 * We may later decide to recursively descend into
946 * the submodule directory and update its index
947 * and/or work tree, but we do not do that now.
949 update_wd = 0;
950 goto update_index;
953 buf = read_object_file(oid, &type, &size);
954 if (!buf)
955 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
956 if (type != OBJ_BLOB) {
957 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
958 goto free_buf;
960 if (S_ISREG(mode)) {
961 struct strbuf strbuf = STRBUF_INIT;
962 if (convert_to_working_tree(path, buf, size, &strbuf)) {
963 free(buf);
964 size = strbuf.len;
965 buf = strbuf_detach(&strbuf, NULL);
969 if (make_room_for_path(o, path) < 0) {
970 update_wd = 0;
971 goto free_buf;
973 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
974 int fd;
975 if (mode & 0100)
976 mode = 0777;
977 else
978 mode = 0666;
979 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
980 if (fd < 0) {
981 ret = err(o, _("failed to open '%s': %s"),
982 path, strerror(errno));
983 goto free_buf;
985 write_in_full(fd, buf, size);
986 close(fd);
987 } else if (S_ISLNK(mode)) {
988 char *lnk = xmemdupz(buf, size);
989 safe_create_leading_directories_const(path);
990 unlink(path);
991 if (symlink(lnk, path))
992 ret = err(o, _("failed to symlink '%s': %s"),
993 path, strerror(errno));
994 free(lnk);
995 } else
996 ret = err(o,
997 _("do not know what to do with %06o %s '%s'"),
998 mode, oid_to_hex(oid), path);
999 free_buf:
1000 free(buf);
1002 update_index:
1003 if (!ret && update_cache)
1004 if (add_cacheinfo(o, mode, oid, path, 0, update_wd,
1005 ADD_CACHE_OK_TO_ADD))
1006 return -1;
1007 return ret;
1010 static int update_file(struct merge_options *o,
1011 int clean,
1012 const struct object_id *oid,
1013 unsigned mode,
1014 const char *path)
1016 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
1019 /* Low level file merging, update and removal */
1021 struct merge_file_info {
1022 struct object_id oid;
1023 unsigned mode;
1024 unsigned clean:1,
1025 merge:1;
1028 static int merge_3way(struct merge_options *o,
1029 mmbuffer_t *result_buf,
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)
1036 mmfile_t orig, src1, src2;
1037 struct ll_merge_options ll_opts = {0};
1038 char *base_name, *name1, *name2;
1039 int merge_status;
1041 ll_opts.renormalize = o->renormalize;
1042 ll_opts.xdl_opts = o->xdl_opts;
1044 if (o->call_depth) {
1045 ll_opts.virtual_ancestor = 1;
1046 ll_opts.variant = 0;
1047 } else {
1048 switch (o->recursive_variant) {
1049 case MERGE_RECURSIVE_OURS:
1050 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1051 break;
1052 case MERGE_RECURSIVE_THEIRS:
1053 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1054 break;
1055 default:
1056 ll_opts.variant = 0;
1057 break;
1061 if (strcmp(a->path, b->path) ||
1062 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1063 base_name = o->ancestor == NULL ? NULL :
1064 mkpathdup("%s:%s", o->ancestor, one->path);
1065 name1 = mkpathdup("%s:%s", branch1, a->path);
1066 name2 = mkpathdup("%s:%s", branch2, b->path);
1067 } else {
1068 base_name = o->ancestor == NULL ? NULL :
1069 mkpathdup("%s", o->ancestor);
1070 name1 = mkpathdup("%s", branch1);
1071 name2 = mkpathdup("%s", branch2);
1074 read_mmblob(&orig, &one->oid);
1075 read_mmblob(&src1, &a->oid);
1076 read_mmblob(&src2, &b->oid);
1078 merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1079 &src1, name1, &src2, name2, &ll_opts);
1081 free(base_name);
1082 free(name1);
1083 free(name2);
1084 free(orig.ptr);
1085 free(src1.ptr);
1086 free(src2.ptr);
1087 return merge_status;
1090 static int merge_file_1(struct merge_options *o,
1091 const struct diff_filespec *one,
1092 const struct diff_filespec *a,
1093 const struct diff_filespec *b,
1094 const char *filename,
1095 const char *branch1,
1096 const char *branch2,
1097 struct merge_file_info *result)
1099 result->merge = 0;
1100 result->clean = 1;
1102 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1103 result->clean = 0;
1104 if (S_ISREG(a->mode)) {
1105 result->mode = a->mode;
1106 oidcpy(&result->oid, &a->oid);
1107 } else {
1108 result->mode = b->mode;
1109 oidcpy(&result->oid, &b->oid);
1111 } else {
1112 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1113 result->merge = 1;
1116 * Merge modes
1118 if (a->mode == b->mode || a->mode == one->mode)
1119 result->mode = b->mode;
1120 else {
1121 result->mode = a->mode;
1122 if (b->mode != one->mode) {
1123 result->clean = 0;
1124 result->merge = 1;
1128 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1129 oidcpy(&result->oid, &b->oid);
1130 else if (oid_eq(&b->oid, &one->oid))
1131 oidcpy(&result->oid, &a->oid);
1132 else if (S_ISREG(a->mode)) {
1133 mmbuffer_t result_buf;
1134 int ret = 0, merge_status;
1136 merge_status = merge_3way(o, &result_buf, one, a, b,
1137 branch1, branch2);
1139 if ((merge_status < 0) || !result_buf.ptr)
1140 ret = err(o, _("Failed to execute internal merge"));
1142 if (!ret &&
1143 write_object_file(result_buf.ptr, result_buf.size,
1144 blob_type, &result->oid))
1145 ret = err(o, _("Unable to add %s to database"),
1146 a->path);
1148 free(result_buf.ptr);
1149 if (ret)
1150 return ret;
1151 result->clean = (merge_status == 0);
1152 } else if (S_ISGITLINK(a->mode)) {
1153 result->clean = merge_submodule(&result->oid,
1154 one->path,
1155 &one->oid,
1156 &a->oid,
1157 &b->oid,
1158 !o->call_depth);
1159 } else if (S_ISLNK(a->mode)) {
1160 switch (o->recursive_variant) {
1161 case MERGE_RECURSIVE_NORMAL:
1162 oidcpy(&result->oid, &a->oid);
1163 if (!oid_eq(&a->oid, &b->oid))
1164 result->clean = 0;
1165 break;
1166 case MERGE_RECURSIVE_OURS:
1167 oidcpy(&result->oid, &a->oid);
1168 break;
1169 case MERGE_RECURSIVE_THEIRS:
1170 oidcpy(&result->oid, &b->oid);
1171 break;
1173 } else
1174 die("BUG: unsupported object type in the tree");
1177 if (result->merge)
1178 output(o, 2, _("Auto-merging %s"), filename);
1180 return 0;
1183 static int merge_file_special_markers(struct merge_options *o,
1184 const struct diff_filespec *one,
1185 const struct diff_filespec *a,
1186 const struct diff_filespec *b,
1187 const char *target_filename,
1188 const char *branch1,
1189 const char *filename1,
1190 const char *branch2,
1191 const char *filename2,
1192 struct merge_file_info *mfi)
1194 char *side1 = NULL;
1195 char *side2 = NULL;
1196 int ret;
1198 if (filename1)
1199 side1 = xstrfmt("%s:%s", branch1, filename1);
1200 if (filename2)
1201 side2 = xstrfmt("%s:%s", branch2, filename2);
1203 ret = merge_file_1(o, one, a, b, target_filename,
1204 side1 ? side1 : branch1,
1205 side2 ? side2 : branch2, mfi);
1207 free(side1);
1208 free(side2);
1209 return ret;
1212 static int merge_file_one(struct merge_options *o,
1213 const char *path,
1214 const struct object_id *o_oid, int o_mode,
1215 const struct object_id *a_oid, int a_mode,
1216 const struct object_id *b_oid, int b_mode,
1217 const char *branch1,
1218 const char *branch2,
1219 struct merge_file_info *mfi)
1221 struct diff_filespec one, a, b;
1223 one.path = a.path = b.path = (char *)path;
1224 oidcpy(&one.oid, o_oid);
1225 one.mode = o_mode;
1226 oidcpy(&a.oid, a_oid);
1227 a.mode = a_mode;
1228 oidcpy(&b.oid, b_oid);
1229 b.mode = b_mode;
1230 return merge_file_1(o, &one, &a, &b, path, branch1, branch2, mfi);
1233 static int conflict_rename_dir(struct merge_options *o,
1234 struct diff_filepair *pair,
1235 const char *rename_branch,
1236 const char *other_branch)
1238 const struct diff_filespec *dest = pair->two;
1240 if (!o->call_depth && would_lose_untracked(dest->path)) {
1241 char *alt_path = unique_path(o, dest->path, rename_branch);
1243 output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1244 "writing to %s instead."),
1245 dest->path, alt_path);
1247 * Write the file in worktree at alt_path, but not in the
1248 * index. Instead, write to dest->path for the index but
1249 * only at the higher appropriate stage.
1251 if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1252 return -1;
1253 free(alt_path);
1254 return update_stages(o, dest->path, NULL,
1255 rename_branch == o->branch1 ? dest : NULL,
1256 rename_branch == o->branch1 ? NULL : dest);
1259 /* Update dest->path both in index and in worktree */
1260 if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1261 return -1;
1262 return 0;
1265 static int handle_change_delete(struct merge_options *o,
1266 const char *path, const char *old_path,
1267 const struct object_id *o_oid, int o_mode,
1268 const struct object_id *changed_oid,
1269 int changed_mode,
1270 const char *change_branch,
1271 const char *delete_branch,
1272 const char *change, const char *change_past)
1274 char *alt_path = NULL;
1275 const char *update_path = path;
1276 int ret = 0;
1278 if (dir_in_way(path, !o->call_depth, 0) ||
1279 (!o->call_depth && would_lose_untracked(path))) {
1280 update_path = alt_path = unique_path(o, path, change_branch);
1283 if (o->call_depth) {
1285 * We cannot arbitrarily accept either a_sha or b_sha as
1286 * correct; since there is no true "middle point" between
1287 * them, simply reuse the base version for virtual merge base.
1289 ret = remove_file_from_cache(path);
1290 if (!ret)
1291 ret = update_file(o, 0, o_oid, o_mode, update_path);
1292 } else {
1293 if (!alt_path) {
1294 if (!old_path) {
1295 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1296 "and %s in %s. Version %s of %s left in tree."),
1297 change, path, delete_branch, change_past,
1298 change_branch, change_branch, path);
1299 } else {
1300 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1301 "and %s to %s in %s. Version %s of %s left in tree."),
1302 change, old_path, delete_branch, change_past, path,
1303 change_branch, change_branch, path);
1305 } else {
1306 if (!old_path) {
1307 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1308 "and %s in %s. Version %s of %s left in tree at %s."),
1309 change, path, delete_branch, change_past,
1310 change_branch, change_branch, path, alt_path);
1311 } else {
1312 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1313 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1314 change, old_path, delete_branch, change_past, path,
1315 change_branch, change_branch, path, alt_path);
1319 * No need to call update_file() on path when change_branch ==
1320 * o->branch1 && !alt_path, since that would needlessly touch
1321 * path. We could call update_file_flags() with update_cache=0
1322 * and update_wd=0, but that's a no-op.
1324 if (change_branch != o->branch1 || alt_path)
1325 ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1327 free(alt_path);
1329 return ret;
1332 static int conflict_rename_delete(struct merge_options *o,
1333 struct diff_filepair *pair,
1334 const char *rename_branch,
1335 const char *delete_branch)
1337 const struct diff_filespec *orig = pair->one;
1338 const struct diff_filespec *dest = pair->two;
1340 if (handle_change_delete(o,
1341 o->call_depth ? orig->path : dest->path,
1342 o->call_depth ? NULL : orig->path,
1343 &orig->oid, orig->mode,
1344 &dest->oid, dest->mode,
1345 rename_branch, delete_branch,
1346 _("rename"), _("renamed")))
1347 return -1;
1349 if (o->call_depth)
1350 return remove_file_from_cache(dest->path);
1351 else
1352 return update_stages(o, dest->path, NULL,
1353 rename_branch == o->branch1 ? dest : NULL,
1354 rename_branch == o->branch1 ? NULL : dest);
1357 static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1358 struct stage_data *entry,
1359 int stage)
1361 struct object_id *oid = &entry->stages[stage].oid;
1362 unsigned mode = entry->stages[stage].mode;
1363 if (mode == 0 || is_null_oid(oid))
1364 return NULL;
1365 oidcpy(&target->oid, oid);
1366 target->mode = mode;
1367 return target;
1370 static int handle_file(struct merge_options *o,
1371 struct diff_filespec *rename,
1372 int stage,
1373 struct rename_conflict_info *ci)
1375 char *dst_name = rename->path;
1376 struct stage_data *dst_entry;
1377 const char *cur_branch, *other_branch;
1378 struct diff_filespec other;
1379 struct diff_filespec *add;
1380 int ret;
1382 if (stage == 2) {
1383 dst_entry = ci->dst_entry1;
1384 cur_branch = ci->branch1;
1385 other_branch = ci->branch2;
1386 } else {
1387 dst_entry = ci->dst_entry2;
1388 cur_branch = ci->branch2;
1389 other_branch = ci->branch1;
1392 add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1393 if (add) {
1394 int ren_src_was_dirty = was_dirty(o, rename->path);
1395 char *add_name = unique_path(o, rename->path, other_branch);
1396 if (update_file(o, 0, &add->oid, add->mode, add_name))
1397 return -1;
1399 if (ren_src_was_dirty) {
1400 output(o, 1, _("Refusing to lose dirty file at %s"),
1401 rename->path);
1404 * Because the double negatives somehow keep confusing me...
1405 * 1) update_wd iff !ren_src_was_dirty.
1406 * 2) no_wd iff !update_wd
1407 * 3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty
1409 remove_file(o, 0, rename->path, ren_src_was_dirty);
1410 dst_name = unique_path(o, rename->path, cur_branch);
1411 } else {
1412 if (dir_in_way(rename->path, !o->call_depth, 0)) {
1413 dst_name = unique_path(o, rename->path, cur_branch);
1414 output(o, 1, _("%s is a directory in %s adding as %s instead"),
1415 rename->path, other_branch, dst_name);
1416 } else if (!o->call_depth &&
1417 would_lose_untracked(rename->path)) {
1418 dst_name = unique_path(o, rename->path, cur_branch);
1419 output(o, 1, _("Refusing to lose untracked file at %s; "
1420 "adding as %s instead"),
1421 rename->path, dst_name);
1424 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1425 ; /* fall through, do allow dst_name to be released */
1426 else if (stage == 2)
1427 ret = update_stages(o, rename->path, NULL, rename, add);
1428 else
1429 ret = update_stages(o, rename->path, NULL, add, rename);
1431 if (dst_name != rename->path)
1432 free(dst_name);
1434 return ret;
1437 static int conflict_rename_rename_1to2(struct merge_options *o,
1438 struct rename_conflict_info *ci)
1440 /* One file was renamed in both branches, but to different names. */
1441 struct diff_filespec *one = ci->pair1->one;
1442 struct diff_filespec *a = ci->pair1->two;
1443 struct diff_filespec *b = ci->pair2->two;
1445 output(o, 1, _("CONFLICT (rename/rename): "
1446 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1447 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1448 one->path, a->path, ci->branch1,
1449 one->path, b->path, ci->branch2,
1450 o->call_depth ? _(" (left unresolved)") : "");
1451 if (o->call_depth) {
1452 struct merge_file_info mfi;
1453 struct diff_filespec other;
1454 struct diff_filespec *add;
1455 if (merge_file_one(o, one->path,
1456 &one->oid, one->mode,
1457 &a->oid, a->mode,
1458 &b->oid, b->mode,
1459 ci->branch1, ci->branch2, &mfi))
1460 return -1;
1463 * FIXME: For rename/add-source conflicts (if we could detect
1464 * such), this is wrong. We should instead find a unique
1465 * pathname and then either rename the add-source file to that
1466 * unique path, or use that unique path instead of src here.
1468 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1469 return -1;
1472 * Above, we put the merged content at the merge-base's
1473 * path. Now we usually need to delete both a->path and
1474 * b->path. However, the rename on each side of the merge
1475 * could also be involved in a rename/add conflict. In
1476 * such cases, we should keep the added file around,
1477 * resolving the conflict at that path in its favor.
1479 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1480 if (add) {
1481 if (update_file(o, 0, &add->oid, add->mode, a->path))
1482 return -1;
1484 else
1485 remove_file_from_cache(a->path);
1486 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1487 if (add) {
1488 if (update_file(o, 0, &add->oid, add->mode, b->path))
1489 return -1;
1491 else
1492 remove_file_from_cache(b->path);
1493 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1494 return -1;
1496 return 0;
1499 static int conflict_rename_rename_2to1(struct merge_options *o,
1500 struct rename_conflict_info *ci)
1502 /* Two files, a & b, were renamed to the same thing, c. */
1503 struct diff_filespec *a = ci->pair1->one;
1504 struct diff_filespec *b = ci->pair2->one;
1505 struct diff_filespec *c1 = ci->pair1->two;
1506 struct diff_filespec *c2 = ci->pair2->two;
1507 char *path = c1->path; /* == c2->path */
1508 char *path_side_1_desc;
1509 char *path_side_2_desc;
1510 struct merge_file_info mfi_c1;
1511 struct merge_file_info mfi_c2;
1512 int ret;
1514 output(o, 1, _("CONFLICT (rename/rename): "
1515 "Rename %s->%s in %s. "
1516 "Rename %s->%s in %s"),
1517 a->path, c1->path, ci->branch1,
1518 b->path, c2->path, ci->branch2);
1520 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1521 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1523 path_side_1_desc = xstrfmt("%s (was %s)", path, a->path);
1524 path_side_2_desc = xstrfmt("%s (was %s)", path, b->path);
1525 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1526 path_side_1_desc,
1527 o->branch1, c1->path,
1528 o->branch2, ci->ren1_other.path, &mfi_c1) ||
1529 merge_file_special_markers(o, b, &ci->ren2_other, c2,
1530 path_side_2_desc,
1531 o->branch1, ci->ren2_other.path,
1532 o->branch2, c2->path, &mfi_c2))
1533 return -1;
1534 free(path_side_1_desc);
1535 free(path_side_2_desc);
1537 if (o->call_depth) {
1539 * If mfi_c1.clean && mfi_c2.clean, then it might make
1540 * sense to do a two-way merge of those results. But, I
1541 * think in all cases, it makes sense to have the virtual
1542 * merge base just undo the renames; they can be detected
1543 * again later for the non-recursive merge.
1545 remove_file(o, 0, path, 0);
1546 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1547 if (!ret)
1548 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1549 b->path);
1550 } else {
1551 char *new_path1 = unique_path(o, path, ci->branch1);
1552 char *new_path2 = unique_path(o, path, ci->branch2);
1553 output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1554 a->path, new_path1, b->path, new_path2);
1555 if (was_dirty(o, path))
1556 output(o, 1, _("Refusing to lose dirty file at %s"),
1557 path);
1558 else if (would_lose_untracked(path))
1560 * Only way we get here is if both renames were from
1561 * a directory rename AND user had an untracked file
1562 * at the location where both files end up after the
1563 * two directory renames. See testcase 10d of t6043.
1565 output(o, 1, _("Refusing to lose untracked file at "
1566 "%s, even though it's in the way."),
1567 path);
1568 else
1569 remove_file(o, 0, path, 0);
1570 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1571 if (!ret)
1572 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1573 new_path2);
1575 * unpack_trees() actually populates the index for us for
1576 * "normal" rename/rename(2to1) situtations so that the
1577 * correct entries are at the higher stages, which would
1578 * make the call below to update_stages_for_stage_data
1579 * unnecessary. However, if either of the renames came
1580 * from a directory rename, then unpack_trees() will not
1581 * have gotten the right data loaded into the index, so we
1582 * need to do so now. (While it'd be tempting to move this
1583 * call to update_stages_for_stage_data() to
1584 * apply_directory_rename_modifications(), that would break
1585 * our intermediate calls to would_lose_untracked() since
1586 * those rely on the current in-memory index. See also the
1587 * big "NOTE" in update_stages()).
1589 if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1590 ret = -1;
1592 free(new_path2);
1593 free(new_path1);
1596 return ret;
1600 * Get the diff_filepairs changed between o_tree and tree.
1602 static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1603 struct tree *o_tree,
1604 struct tree *tree)
1606 struct diff_queue_struct *ret;
1607 struct diff_options opts;
1609 diff_setup(&opts);
1610 opts.flags.recursive = 1;
1611 opts.flags.rename_empty = 0;
1612 opts.detect_rename = merge_detect_rename(o);
1614 * We do not have logic to handle the detection of copies. In
1615 * fact, it may not even make sense to add such logic: would we
1616 * really want a change to a base file to be propagated through
1617 * multiple other files by a merge?
1619 if (opts.detect_rename > DIFF_DETECT_RENAME)
1620 opts.detect_rename = DIFF_DETECT_RENAME;
1621 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1622 o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1623 1000;
1624 opts.rename_score = o->rename_score;
1625 opts.show_rename_progress = o->show_rename_progress;
1626 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1627 diff_setup_done(&opts);
1628 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1629 diffcore_std(&opts);
1630 if (opts.needed_rename_limit > o->needed_rename_limit)
1631 o->needed_rename_limit = opts.needed_rename_limit;
1633 ret = xmalloc(sizeof(*ret));
1634 *ret = diff_queued_diff;
1636 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1637 diff_queued_diff.nr = 0;
1638 diff_queued_diff.queue = NULL;
1639 diff_flush(&opts);
1640 return ret;
1643 static int tree_has_path(struct tree *tree, const char *path)
1645 struct object_id hashy;
1646 unsigned int mode_o;
1648 return !get_tree_entry(&tree->object.oid, path,
1649 &hashy, &mode_o);
1653 * Return a new string that replaces the beginning portion (which matches
1654 * entry->dir), with entry->new_dir. In perl-speak:
1655 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1656 * NOTE:
1657 * Caller must ensure that old_path starts with entry->dir + '/'.
1659 static char *apply_dir_rename(struct dir_rename_entry *entry,
1660 const char *old_path)
1662 struct strbuf new_path = STRBUF_INIT;
1663 int oldlen, newlen;
1665 if (entry->non_unique_new_dir)
1666 return NULL;
1668 oldlen = strlen(entry->dir);
1669 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1670 strbuf_grow(&new_path, newlen);
1671 strbuf_addbuf(&new_path, &entry->new_dir);
1672 strbuf_addstr(&new_path, &old_path[oldlen]);
1674 return strbuf_detach(&new_path, NULL);
1677 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1678 char **old_dir, char **new_dir)
1680 char *end_of_old, *end_of_new;
1681 int old_len, new_len;
1683 *old_dir = NULL;
1684 *new_dir = NULL;
1687 * For
1688 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1689 * the "e/foo.c" part is the same, we just want to know that
1690 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1691 * so, for this example, this function returns "a/b/c/d" in
1692 * *old_dir and "a/b/some/thing/else" in *new_dir.
1694 * Also, if the basename of the file changed, we don't care. We
1695 * want to know which portion of the directory, if any, changed.
1697 end_of_old = strrchr(old_path, '/');
1698 end_of_new = strrchr(new_path, '/');
1700 if (end_of_old == NULL || end_of_new == NULL)
1701 return;
1702 while (*--end_of_new == *--end_of_old &&
1703 end_of_old != old_path &&
1704 end_of_new != new_path)
1705 ; /* Do nothing; all in the while loop */
1707 * We've found the first non-matching character in the directory
1708 * paths. That means the current directory we were comparing
1709 * represents the rename. Move end_of_old and end_of_new back
1710 * to the full directory name.
1712 if (*end_of_old == '/')
1713 end_of_old++;
1714 if (*end_of_old != '/')
1715 end_of_new++;
1716 end_of_old = strchr(end_of_old, '/');
1717 end_of_new = strchr(end_of_new, '/');
1720 * It may have been the case that old_path and new_path were the same
1721 * directory all along. Don't claim a rename if they're the same.
1723 old_len = end_of_old - old_path;
1724 new_len = end_of_new - new_path;
1726 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1727 *old_dir = xstrndup(old_path, old_len);
1728 *new_dir = xstrndup(new_path, new_len);
1732 static void remove_hashmap_entries(struct hashmap *dir_renames,
1733 struct string_list *items_to_remove)
1735 int i;
1736 struct dir_rename_entry *entry;
1738 for (i = 0; i < items_to_remove->nr; i++) {
1739 entry = items_to_remove->items[i].util;
1740 hashmap_remove(dir_renames, entry, NULL);
1742 string_list_clear(items_to_remove, 0);
1746 * See if there is a directory rename for path, and if there are any file
1747 * level conflicts for the renamed location. If there is a rename and
1748 * there are no conflicts, return the new name. Otherwise, return NULL.
1750 static char *handle_path_level_conflicts(struct merge_options *o,
1751 const char *path,
1752 struct dir_rename_entry *entry,
1753 struct hashmap *collisions,
1754 struct tree *tree)
1756 char *new_path = NULL;
1757 struct collision_entry *collision_ent;
1758 int clean = 1;
1759 struct strbuf collision_paths = STRBUF_INIT;
1762 * entry has the mapping of old directory name to new directory name
1763 * that we want to apply to path.
1765 new_path = apply_dir_rename(entry, path);
1767 if (!new_path) {
1768 /* This should only happen when entry->non_unique_new_dir set */
1769 if (!entry->non_unique_new_dir)
1770 BUG("entry->non_unqiue_dir not set and !new_path");
1771 output(o, 1, _("CONFLICT (directory rename split): "
1772 "Unclear where to place %s because directory "
1773 "%s was renamed to multiple other directories, "
1774 "with no destination getting a majority of the "
1775 "files."),
1776 path, entry->dir);
1777 clean = 0;
1778 return NULL;
1782 * The caller needs to have ensured that it has pre-populated
1783 * collisions with all paths that map to new_path. Do a quick check
1784 * to ensure that's the case.
1786 collision_ent = collision_find_entry(collisions, new_path);
1787 if (collision_ent == NULL)
1788 BUG("collision_ent is NULL");
1791 * Check for one-sided add/add/.../add conflicts, i.e.
1792 * where implicit renames from the other side doing
1793 * directory rename(s) can affect this side of history
1794 * to put multiple paths into the same location. Warn
1795 * and bail on directory renames for such paths.
1797 if (collision_ent->reported_already) {
1798 clean = 0;
1799 } else if (tree_has_path(tree, new_path)) {
1800 collision_ent->reported_already = 1;
1801 strbuf_add_separated_string_list(&collision_paths, ", ",
1802 &collision_ent->source_files);
1803 output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1804 "file/dir at %s in the way of implicit "
1805 "directory rename(s) putting the following "
1806 "path(s) there: %s."),
1807 new_path, collision_paths.buf);
1808 clean = 0;
1809 } else if (collision_ent->source_files.nr > 1) {
1810 collision_ent->reported_already = 1;
1811 strbuf_add_separated_string_list(&collision_paths, ", ",
1812 &collision_ent->source_files);
1813 output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1814 "more than one path to %s; implicit directory "
1815 "renames tried to put these paths there: %s"),
1816 new_path, collision_paths.buf);
1817 clean = 0;
1820 /* Free memory we no longer need */
1821 strbuf_release(&collision_paths);
1822 if (!clean && new_path) {
1823 free(new_path);
1824 return NULL;
1827 return new_path;
1831 * There are a couple things we want to do at the directory level:
1832 * 1. Check for both sides renaming to the same thing, in order to avoid
1833 * implicit renaming of files that should be left in place. (See
1834 * testcase 6b in t6043 for details.)
1835 * 2. Prune directory renames if there are still files left in the
1836 * the original directory. These represent a partial directory rename,
1837 * i.e. a rename where only some of the files within the directory
1838 * were renamed elsewhere. (Technically, this could be done earlier
1839 * in get_directory_renames(), except that would prevent us from
1840 * doing the previous check and thus failing testcase 6b.)
1841 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
1842 * In the future, we could potentially record this info as well and
1843 * omit reporting rename/rename(1to2) conflicts for each path within
1844 * the affected directories, thus cleaning up the merge output.
1845 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1846 * directory level, because merging directories is fine. If it
1847 * causes conflicts for files within those merged directories, then
1848 * that should be detected at the individual path level.
1850 static void handle_directory_level_conflicts(struct merge_options *o,
1851 struct hashmap *dir_re_head,
1852 struct tree *head,
1853 struct hashmap *dir_re_merge,
1854 struct tree *merge)
1856 struct hashmap_iter iter;
1857 struct dir_rename_entry *head_ent;
1858 struct dir_rename_entry *merge_ent;
1860 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1861 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1863 hashmap_iter_init(dir_re_head, &iter);
1864 while ((head_ent = hashmap_iter_next(&iter))) {
1865 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1866 if (merge_ent &&
1867 !head_ent->non_unique_new_dir &&
1868 !merge_ent->non_unique_new_dir &&
1869 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1870 /* 1. Renamed identically; remove it from both sides */
1871 string_list_append(&remove_from_head,
1872 head_ent->dir)->util = head_ent;
1873 strbuf_release(&head_ent->new_dir);
1874 string_list_append(&remove_from_merge,
1875 merge_ent->dir)->util = merge_ent;
1876 strbuf_release(&merge_ent->new_dir);
1877 } else if (tree_has_path(head, head_ent->dir)) {
1878 /* 2. This wasn't a directory rename after all */
1879 string_list_append(&remove_from_head,
1880 head_ent->dir)->util = head_ent;
1881 strbuf_release(&head_ent->new_dir);
1885 remove_hashmap_entries(dir_re_head, &remove_from_head);
1886 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1888 hashmap_iter_init(dir_re_merge, &iter);
1889 while ((merge_ent = hashmap_iter_next(&iter))) {
1890 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1891 if (tree_has_path(merge, merge_ent->dir)) {
1892 /* 2. This wasn't a directory rename after all */
1893 string_list_append(&remove_from_merge,
1894 merge_ent->dir)->util = merge_ent;
1895 } else if (head_ent &&
1896 !head_ent->non_unique_new_dir &&
1897 !merge_ent->non_unique_new_dir) {
1898 /* 3. rename/rename(1to2) */
1900 * We can assume it's not rename/rename(1to1) because
1901 * that was case (1), already checked above. So we
1902 * know that head_ent->new_dir and merge_ent->new_dir
1903 * are different strings.
1905 output(o, 1, _("CONFLICT (rename/rename): "
1906 "Rename directory %s->%s in %s. "
1907 "Rename directory %s->%s in %s"),
1908 head_ent->dir, head_ent->new_dir.buf, o->branch1,
1909 head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1910 string_list_append(&remove_from_head,
1911 head_ent->dir)->util = head_ent;
1912 strbuf_release(&head_ent->new_dir);
1913 string_list_append(&remove_from_merge,
1914 merge_ent->dir)->util = merge_ent;
1915 strbuf_release(&merge_ent->new_dir);
1919 remove_hashmap_entries(dir_re_head, &remove_from_head);
1920 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1923 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1924 struct tree *tree)
1926 struct hashmap *dir_renames;
1927 struct hashmap_iter iter;
1928 struct dir_rename_entry *entry;
1929 int i;
1932 * Typically, we think of a directory rename as all files from a
1933 * certain directory being moved to a target directory. However,
1934 * what if someone first moved two files from the original
1935 * directory in one commit, and then renamed the directory
1936 * somewhere else in a later commit? At merge time, we just know
1937 * that files from the original directory went to two different
1938 * places, and that the bulk of them ended up in the same place.
1939 * We want each directory rename to represent where the bulk of the
1940 * files from that directory end up; this function exists to find
1941 * where the bulk of the files went.
1943 * The first loop below simply iterates through the list of file
1944 * renames, finding out how often each directory rename pair
1945 * possibility occurs.
1947 dir_renames = xmalloc(sizeof(*dir_renames));
1948 dir_rename_init(dir_renames);
1949 for (i = 0; i < pairs->nr; ++i) {
1950 struct string_list_item *item;
1951 int *count;
1952 struct diff_filepair *pair = pairs->queue[i];
1953 char *old_dir, *new_dir;
1955 /* File not part of directory rename if it wasn't renamed */
1956 if (pair->status != 'R')
1957 continue;
1959 get_renamed_dir_portion(pair->one->path, pair->two->path,
1960 &old_dir, &new_dir);
1961 if (!old_dir)
1962 /* Directory didn't change at all; ignore this one. */
1963 continue;
1965 entry = dir_rename_find_entry(dir_renames, old_dir);
1966 if (!entry) {
1967 entry = xmalloc(sizeof(*entry));
1968 dir_rename_entry_init(entry, old_dir);
1969 hashmap_put(dir_renames, entry);
1970 } else {
1971 free(old_dir);
1973 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1974 if (!item) {
1975 item = string_list_insert(&entry->possible_new_dirs,
1976 new_dir);
1977 item->util = xcalloc(1, sizeof(int));
1978 } else {
1979 free(new_dir);
1981 count = item->util;
1982 *count += 1;
1986 * For each directory with files moved out of it, we find out which
1987 * target directory received the most files so we can declare it to
1988 * be the "winning" target location for the directory rename. This
1989 * winner gets recorded in new_dir. If there is no winner
1990 * (multiple target directories received the same number of files),
1991 * we set non_unique_new_dir. Once we've determined the winner (or
1992 * that there is no winner), we no longer need possible_new_dirs.
1994 hashmap_iter_init(dir_renames, &iter);
1995 while ((entry = hashmap_iter_next(&iter))) {
1996 int max = 0;
1997 int bad_max = 0;
1998 char *best = NULL;
2000 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2001 int *count = entry->possible_new_dirs.items[i].util;
2003 if (*count == max)
2004 bad_max = max;
2005 else if (*count > max) {
2006 max = *count;
2007 best = entry->possible_new_dirs.items[i].string;
2010 if (bad_max == max)
2011 entry->non_unique_new_dir = 1;
2012 else {
2013 assert(entry->new_dir.len == 0);
2014 strbuf_addstr(&entry->new_dir, best);
2017 * The relevant directory sub-portion of the original full
2018 * filepaths were xstrndup'ed before inserting into
2019 * possible_new_dirs, and instead of manually iterating the
2020 * list and free'ing each, just lie and tell
2021 * possible_new_dirs that it did the strdup'ing so that it
2022 * will free them for us.
2024 entry->possible_new_dirs.strdup_strings = 1;
2025 string_list_clear(&entry->possible_new_dirs, 1);
2028 return dir_renames;
2031 static struct dir_rename_entry *check_dir_renamed(const char *path,
2032 struct hashmap *dir_renames)
2034 char temp[PATH_MAX];
2035 char *end;
2036 struct dir_rename_entry *entry;
2038 strcpy(temp, path);
2039 while ((end = strrchr(temp, '/'))) {
2040 *end = '\0';
2041 entry = dir_rename_find_entry(dir_renames, temp);
2042 if (entry)
2043 return entry;
2045 return NULL;
2048 static void compute_collisions(struct hashmap *collisions,
2049 struct hashmap *dir_renames,
2050 struct diff_queue_struct *pairs)
2052 int i;
2055 * Multiple files can be mapped to the same path due to directory
2056 * renames done by the other side of history. Since that other
2057 * side of history could have merged multiple directories into one,
2058 * if our side of history added the same file basename to each of
2059 * those directories, then all N of them would get implicitly
2060 * renamed by the directory rename detection into the same path,
2061 * and we'd get an add/add/.../add conflict, and all those adds
2062 * from *this* side of history. This is not representable in the
2063 * index, and users aren't going to easily be able to make sense of
2064 * it. So we need to provide a good warning about what's
2065 * happening, and fall back to no-directory-rename detection
2066 * behavior for those paths.
2068 * See testcases 9e and all of section 5 from t6043 for examples.
2070 collision_init(collisions);
2072 for (i = 0; i < pairs->nr; ++i) {
2073 struct dir_rename_entry *dir_rename_ent;
2074 struct collision_entry *collision_ent;
2075 char *new_path;
2076 struct diff_filepair *pair = pairs->queue[i];
2078 if (pair->status != 'A' && pair->status != 'R')
2079 continue;
2080 dir_rename_ent = check_dir_renamed(pair->two->path,
2081 dir_renames);
2082 if (!dir_rename_ent)
2083 continue;
2085 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2086 if (!new_path)
2088 * dir_rename_ent->non_unique_new_path is true, which
2089 * means there is no directory rename for us to use,
2090 * which means it won't cause us any additional
2091 * collisions.
2093 continue;
2094 collision_ent = collision_find_entry(collisions, new_path);
2095 if (!collision_ent) {
2096 collision_ent = xcalloc(1,
2097 sizeof(struct collision_entry));
2098 hashmap_entry_init(collision_ent, strhash(new_path));
2099 hashmap_put(collisions, collision_ent);
2100 collision_ent->target_file = new_path;
2101 } else {
2102 free(new_path);
2104 string_list_insert(&collision_ent->source_files,
2105 pair->two->path);
2109 static char *check_for_directory_rename(struct merge_options *o,
2110 const char *path,
2111 struct tree *tree,
2112 struct hashmap *dir_renames,
2113 struct hashmap *dir_rename_exclusions,
2114 struct hashmap *collisions,
2115 int *clean_merge)
2117 char *new_path = NULL;
2118 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2119 struct dir_rename_entry *oentry = NULL;
2121 if (!entry)
2122 return new_path;
2125 * This next part is a little weird. We do not want to do an
2126 * implicit rename into a directory we renamed on our side, because
2127 * that will result in a spurious rename/rename(1to2) conflict. An
2128 * example:
2129 * Base commit: dumbdir/afile, otherdir/bfile
2130 * Side 1: smrtdir/afile, otherdir/bfile
2131 * Side 2: dumbdir/afile, dumbdir/bfile
2132 * Here, while working on Side 1, we could notice that otherdir was
2133 * renamed/merged to dumbdir, and change the diff_filepair for
2134 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2135 * 2 will notice the rename from dumbdir to smrtdir, and do the
2136 * transitive rename to move it from dumbdir/bfile to
2137 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2138 * smrtdir, a rename/rename(1to2) conflict. We really just want
2139 * the file to end up in smrtdir. And the way to achieve that is
2140 * to not let Side1 do the rename to dumbdir, since we know that is
2141 * the source of one of our directory renames.
2143 * That's why oentry and dir_rename_exclusions is here.
2145 * As it turns out, this also prevents N-way transient rename
2146 * confusion; See testcases 9c and 9d of t6043.
2148 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2149 if (oentry) {
2150 output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2151 "to %s, because %s itself was renamed."),
2152 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2153 } else {
2154 new_path = handle_path_level_conflicts(o, path, entry,
2155 collisions, tree);
2156 *clean_merge &= (new_path != NULL);
2159 return new_path;
2162 static void apply_directory_rename_modifications(struct merge_options *o,
2163 struct diff_filepair *pair,
2164 char *new_path,
2165 struct rename *re,
2166 struct tree *tree,
2167 struct tree *o_tree,
2168 struct tree *a_tree,
2169 struct tree *b_tree,
2170 struct string_list *entries,
2171 int *clean)
2173 struct string_list_item *item;
2174 int stage = (tree == a_tree ? 2 : 3);
2175 int update_wd;
2178 * In all cases where we can do directory rename detection,
2179 * unpack_trees() will have read pair->two->path into the
2180 * index and the working copy. We need to remove it so that
2181 * we can instead place it at new_path. It is guaranteed to
2182 * not be untracked (unpack_trees() would have errored out
2183 * saying the file would have been overwritten), but it might
2184 * be dirty, though.
2186 update_wd = !was_dirty(o, pair->two->path);
2187 if (!update_wd)
2188 output(o, 1, _("Refusing to lose dirty file at %s"),
2189 pair->two->path);
2190 remove_file(o, 1, pair->two->path, !update_wd);
2192 /* Find or create a new re->dst_entry */
2193 item = string_list_lookup(entries, new_path);
2194 if (item) {
2196 * Since we're renaming on this side of history, and it's
2197 * due to a directory rename on the other side of history
2198 * (which we only allow when the directory in question no
2199 * longer exists on the other side of history), the
2200 * original entry for re->dst_entry is no longer
2201 * necessary...
2203 re->dst_entry->processed = 1;
2206 * ...because we'll be using this new one.
2208 re->dst_entry = item->util;
2209 } else {
2211 * re->dst_entry is for the before-dir-rename path, and we
2212 * need it to hold information for the after-dir-rename
2213 * path. Before creating a new entry, we need to mark the
2214 * old one as unnecessary (...unless it is shared by
2215 * src_entry, i.e. this didn't use to be a rename, in which
2216 * case we can just allow the normal processing to happen
2217 * for it).
2219 if (pair->status == 'R')
2220 re->dst_entry->processed = 1;
2222 re->dst_entry = insert_stage_data(new_path,
2223 o_tree, a_tree, b_tree,
2224 entries);
2225 item = string_list_insert(entries, new_path);
2226 item->util = re->dst_entry;
2230 * Update the stage_data with the information about the path we are
2231 * moving into place. That slot will be empty and available for us
2232 * to write to because of the collision checks in
2233 * handle_path_level_conflicts(). In other words,
2234 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2235 * open for us to write to.
2237 * It may be tempting to actually update the index at this point as
2238 * well, using update_stages_for_stage_data(), but as per the big
2239 * "NOTE" in update_stages(), doing so will modify the current
2240 * in-memory index which will break calls to would_lose_untracked()
2241 * that we need to make. Instead, we need to just make sure that
2242 * the various conflict_rename_*() functions update the index
2243 * explicitly rather than relying on unpack_trees() to have done it.
2245 get_tree_entry(&tree->object.oid,
2246 pair->two->path,
2247 &re->dst_entry->stages[stage].oid,
2248 &re->dst_entry->stages[stage].mode);
2250 /* Update pair status */
2251 if (pair->status == 'A') {
2253 * Recording rename information for this add makes it look
2254 * like a rename/delete conflict. Make sure we can
2255 * correctly handle this as an add that was moved to a new
2256 * directory instead of reporting a rename/delete conflict.
2258 re->add_turned_into_rename = 1;
2261 * We don't actually look at pair->status again, but it seems
2262 * pedagogically correct to adjust it.
2264 pair->status = 'R';
2267 * Finally, record the new location.
2269 pair->two->path = new_path;
2273 * Get information of all renames which occurred in 'pairs', making use of
2274 * any implicit directory renames inferred from the other side of history.
2275 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2276 * to be able to associate the correct cache entries with the rename
2277 * information; tree is always equal to either a_tree or b_tree.
2279 static struct string_list *get_renames(struct merge_options *o,
2280 struct diff_queue_struct *pairs,
2281 struct hashmap *dir_renames,
2282 struct hashmap *dir_rename_exclusions,
2283 struct tree *tree,
2284 struct tree *o_tree,
2285 struct tree *a_tree,
2286 struct tree *b_tree,
2287 struct string_list *entries,
2288 int *clean_merge)
2290 int i;
2291 struct hashmap collisions;
2292 struct hashmap_iter iter;
2293 struct collision_entry *e;
2294 struct string_list *renames;
2296 compute_collisions(&collisions, dir_renames, pairs);
2297 renames = xcalloc(1, sizeof(struct string_list));
2299 for (i = 0; i < pairs->nr; ++i) {
2300 struct string_list_item *item;
2301 struct rename *re;
2302 struct diff_filepair *pair = pairs->queue[i];
2303 char *new_path; /* non-NULL only with directory renames */
2305 if (pair->status != 'A' && pair->status != 'R') {
2306 diff_free_filepair(pair);
2307 continue;
2309 new_path = check_for_directory_rename(o, pair->two->path, tree,
2310 dir_renames,
2311 dir_rename_exclusions,
2312 &collisions,
2313 clean_merge);
2314 if (pair->status != 'R' && !new_path) {
2315 diff_free_filepair(pair);
2316 continue;
2319 re = xmalloc(sizeof(*re));
2320 re->processed = 0;
2321 re->add_turned_into_rename = 0;
2322 re->pair = pair;
2323 item = string_list_lookup(entries, re->pair->one->path);
2324 if (!item)
2325 re->src_entry = insert_stage_data(re->pair->one->path,
2326 o_tree, a_tree, b_tree, entries);
2327 else
2328 re->src_entry = item->util;
2330 item = string_list_lookup(entries, re->pair->two->path);
2331 if (!item)
2332 re->dst_entry = insert_stage_data(re->pair->two->path,
2333 o_tree, a_tree, b_tree, entries);
2334 else
2335 re->dst_entry = item->util;
2336 item = string_list_insert(renames, pair->one->path);
2337 item->util = re;
2338 if (new_path)
2339 apply_directory_rename_modifications(o, pair, new_path,
2340 re, tree, o_tree,
2341 a_tree, b_tree,
2342 entries,
2343 clean_merge);
2346 hashmap_iter_init(&collisions, &iter);
2347 while ((e = hashmap_iter_next(&iter))) {
2348 free(e->target_file);
2349 string_list_clear(&e->source_files, 0);
2351 hashmap_free(&collisions, 1);
2352 return renames;
2355 static int process_renames(struct merge_options *o,
2356 struct string_list *a_renames,
2357 struct string_list *b_renames)
2359 int clean_merge = 1, i, j;
2360 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2361 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2362 const struct rename *sre;
2364 for (i = 0; i < a_renames->nr; i++) {
2365 sre = a_renames->items[i].util;
2366 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2367 = (void *)sre;
2369 for (i = 0; i < b_renames->nr; i++) {
2370 sre = b_renames->items[i].util;
2371 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2372 = (void *)sre;
2375 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2376 struct string_list *renames1, *renames2Dst;
2377 struct rename *ren1 = NULL, *ren2 = NULL;
2378 const char *branch1, *branch2;
2379 const char *ren1_src, *ren1_dst;
2380 struct string_list_item *lookup;
2382 if (i >= a_renames->nr) {
2383 ren2 = b_renames->items[j++].util;
2384 } else if (j >= b_renames->nr) {
2385 ren1 = a_renames->items[i++].util;
2386 } else {
2387 int compare = strcmp(a_renames->items[i].string,
2388 b_renames->items[j].string);
2389 if (compare <= 0)
2390 ren1 = a_renames->items[i++].util;
2391 if (compare >= 0)
2392 ren2 = b_renames->items[j++].util;
2395 /* TODO: refactor, so that 1/2 are not needed */
2396 if (ren1) {
2397 renames1 = a_renames;
2398 renames2Dst = &b_by_dst;
2399 branch1 = o->branch1;
2400 branch2 = o->branch2;
2401 } else {
2402 renames1 = b_renames;
2403 renames2Dst = &a_by_dst;
2404 branch1 = o->branch2;
2405 branch2 = o->branch1;
2406 SWAP(ren2, ren1);
2409 if (ren1->processed)
2410 continue;
2411 ren1->processed = 1;
2412 ren1->dst_entry->processed = 1;
2413 /* BUG: We should only mark src_entry as processed if we
2414 * are not dealing with a rename + add-source case.
2416 ren1->src_entry->processed = 1;
2418 ren1_src = ren1->pair->one->path;
2419 ren1_dst = ren1->pair->two->path;
2421 if (ren2) {
2422 /* One file renamed on both sides */
2423 const char *ren2_src = ren2->pair->one->path;
2424 const char *ren2_dst = ren2->pair->two->path;
2425 enum rename_type rename_type;
2426 if (strcmp(ren1_src, ren2_src) != 0)
2427 die("BUG: ren1_src != ren2_src");
2428 ren2->dst_entry->processed = 1;
2429 ren2->processed = 1;
2430 if (strcmp(ren1_dst, ren2_dst) != 0) {
2431 rename_type = RENAME_ONE_FILE_TO_TWO;
2432 clean_merge = 0;
2433 } else {
2434 rename_type = RENAME_ONE_FILE_TO_ONE;
2435 /* BUG: We should only remove ren1_src in
2436 * the base stage (think of rename +
2437 * add-source cases).
2439 remove_file(o, 1, ren1_src, 1);
2440 update_entry(ren1->dst_entry,
2441 ren1->pair->one,
2442 ren1->pair->two,
2443 ren2->pair->two);
2445 setup_rename_conflict_info(rename_type,
2446 ren1->pair,
2447 ren2->pair,
2448 branch1,
2449 branch2,
2450 ren1->dst_entry,
2451 ren2->dst_entry,
2453 NULL,
2454 NULL);
2455 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2456 /* Two different files renamed to the same thing */
2457 char *ren2_dst;
2458 ren2 = lookup->util;
2459 ren2_dst = ren2->pair->two->path;
2460 if (strcmp(ren1_dst, ren2_dst) != 0)
2461 die("BUG: ren1_dst != ren2_dst");
2463 clean_merge = 0;
2464 ren2->processed = 1;
2466 * BUG: We should only mark src_entry as processed
2467 * if we are not dealing with a rename + add-source
2468 * case.
2470 ren2->src_entry->processed = 1;
2472 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2473 ren1->pair,
2474 ren2->pair,
2475 branch1,
2476 branch2,
2477 ren1->dst_entry,
2478 ren2->dst_entry,
2480 ren1->src_entry,
2481 ren2->src_entry);
2483 } else {
2484 /* Renamed in 1, maybe changed in 2 */
2485 /* we only use sha1 and mode of these */
2486 struct diff_filespec src_other, dst_other;
2487 int try_merge;
2490 * unpack_trees loads entries from common-commit
2491 * into stage 1, from head-commit into stage 2, and
2492 * from merge-commit into stage 3. We keep track
2493 * of which side corresponds to the rename.
2495 int renamed_stage = a_renames == renames1 ? 2 : 3;
2496 int other_stage = a_renames == renames1 ? 3 : 2;
2498 /* BUG: We should only remove ren1_src in the base
2499 * stage and in other_stage (think of rename +
2500 * add-source case).
2502 remove_file(o, 1, ren1_src,
2503 renamed_stage == 2 || !was_tracked(o, ren1_src));
2505 oidcpy(&src_other.oid,
2506 &ren1->src_entry->stages[other_stage].oid);
2507 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2508 oidcpy(&dst_other.oid,
2509 &ren1->dst_entry->stages[other_stage].oid);
2510 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2511 try_merge = 0;
2513 if (oid_eq(&src_other.oid, &null_oid) &&
2514 ren1->add_turned_into_rename) {
2515 setup_rename_conflict_info(RENAME_DIR,
2516 ren1->pair,
2517 NULL,
2518 branch1,
2519 branch2,
2520 ren1->dst_entry,
2521 NULL,
2523 NULL,
2524 NULL);
2525 } else if (oid_eq(&src_other.oid, &null_oid)) {
2526 setup_rename_conflict_info(RENAME_DELETE,
2527 ren1->pair,
2528 NULL,
2529 branch1,
2530 branch2,
2531 ren1->dst_entry,
2532 NULL,
2534 NULL,
2535 NULL);
2536 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2537 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2539 * Added file on the other side identical to
2540 * the file being renamed: clean merge.
2541 * Also, there is no need to overwrite the
2542 * file already in the working copy, so call
2543 * update_file_flags() instead of
2544 * update_file().
2546 if (update_file_flags(o,
2547 &ren1->pair->two->oid,
2548 ren1->pair->two->mode,
2549 ren1_dst,
2550 1, /* update_cache */
2551 0 /* update_wd */))
2552 clean_merge = -1;
2553 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2554 clean_merge = 0;
2555 try_merge = 1;
2556 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2557 "%s added in %s"),
2558 ren1_src, ren1_dst, branch1,
2559 ren1_dst, branch2);
2560 if (o->call_depth) {
2561 struct merge_file_info mfi;
2562 if (merge_file_one(o, ren1_dst, &null_oid, 0,
2563 &ren1->pair->two->oid,
2564 ren1->pair->two->mode,
2565 &dst_other.oid,
2566 dst_other.mode,
2567 branch1, branch2, &mfi)) {
2568 clean_merge = -1;
2569 goto cleanup_and_return;
2571 output(o, 1, _("Adding merged %s"), ren1_dst);
2572 if (update_file(o, 0, &mfi.oid,
2573 mfi.mode, ren1_dst))
2574 clean_merge = -1;
2575 try_merge = 0;
2576 } else {
2577 char *new_path = unique_path(o, ren1_dst, branch2);
2578 output(o, 1, _("Adding as %s instead"), new_path);
2579 if (update_file(o, 0, &dst_other.oid,
2580 dst_other.mode, new_path))
2581 clean_merge = -1;
2582 free(new_path);
2584 } else
2585 try_merge = 1;
2587 if (clean_merge < 0)
2588 goto cleanup_and_return;
2589 if (try_merge) {
2590 struct diff_filespec *one, *a, *b;
2591 src_other.path = (char *)ren1_src;
2593 one = ren1->pair->one;
2594 if (a_renames == renames1) {
2595 a = ren1->pair->two;
2596 b = &src_other;
2597 } else {
2598 b = ren1->pair->two;
2599 a = &src_other;
2601 update_entry(ren1->dst_entry, one, a, b);
2602 setup_rename_conflict_info(RENAME_NORMAL,
2603 ren1->pair,
2604 NULL,
2605 branch1,
2606 NULL,
2607 ren1->dst_entry,
2608 NULL,
2610 NULL,
2611 NULL);
2615 cleanup_and_return:
2616 string_list_clear(&a_by_dst, 0);
2617 string_list_clear(&b_by_dst, 0);
2619 return clean_merge;
2622 struct rename_info {
2623 struct string_list *head_renames;
2624 struct string_list *merge_renames;
2627 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2628 struct hashmap *dir_renames)
2630 struct hashmap_iter iter;
2631 struct dir_rename_entry *e;
2633 hashmap_iter_init(dir_renames, &iter);
2634 while ((e = hashmap_iter_next(&iter))) {
2635 free(e->dir);
2636 strbuf_release(&e->new_dir);
2637 /* possible_new_dirs already cleared in get_directory_renames */
2639 hashmap_free(dir_renames, 1);
2640 free(dir_renames);
2642 free(pairs->queue);
2643 free(pairs);
2646 static int handle_renames(struct merge_options *o,
2647 struct tree *common,
2648 struct tree *head,
2649 struct tree *merge,
2650 struct string_list *entries,
2651 struct rename_info *ri)
2653 struct diff_queue_struct *head_pairs, *merge_pairs;
2654 struct hashmap *dir_re_head, *dir_re_merge;
2655 int clean = 1;
2657 ri->head_renames = NULL;
2658 ri->merge_renames = NULL;
2660 if (!merge_detect_rename(o))
2661 return 1;
2663 head_pairs = get_diffpairs(o, common, head);
2664 merge_pairs = get_diffpairs(o, common, merge);
2666 dir_re_head = get_directory_renames(head_pairs, head);
2667 dir_re_merge = get_directory_renames(merge_pairs, merge);
2669 handle_directory_level_conflicts(o,
2670 dir_re_head, head,
2671 dir_re_merge, merge);
2673 ri->head_renames = get_renames(o, head_pairs,
2674 dir_re_merge, dir_re_head, head,
2675 common, head, merge, entries,
2676 &clean);
2677 if (clean < 0)
2678 goto cleanup;
2679 ri->merge_renames = get_renames(o, merge_pairs,
2680 dir_re_head, dir_re_merge, merge,
2681 common, head, merge, entries,
2682 &clean);
2683 if (clean < 0)
2684 goto cleanup;
2685 clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2687 cleanup:
2689 * Some cleanup is deferred until cleanup_renames() because the
2690 * data structures are still needed and referenced in
2691 * process_entry(). But there are a few things we can free now.
2693 initial_cleanup_rename(head_pairs, dir_re_head);
2694 initial_cleanup_rename(merge_pairs, dir_re_merge);
2696 return clean;
2699 static void final_cleanup_rename(struct string_list *rename)
2701 const struct rename *re;
2702 int i;
2704 if (rename == NULL)
2705 return;
2707 for (i = 0; i < rename->nr; i++) {
2708 re = rename->items[i].util;
2709 diff_free_filepair(re->pair);
2711 string_list_clear(rename, 1);
2712 free(rename);
2715 static void final_cleanup_renames(struct rename_info *re_info)
2717 final_cleanup_rename(re_info->head_renames);
2718 final_cleanup_rename(re_info->merge_renames);
2721 static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2723 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2726 static int read_oid_strbuf(struct merge_options *o,
2727 const struct object_id *oid, struct strbuf *dst)
2729 void *buf;
2730 enum object_type type;
2731 unsigned long size;
2732 buf = read_object_file(oid, &type, &size);
2733 if (!buf)
2734 return err(o, _("cannot read object %s"), oid_to_hex(oid));
2735 if (type != OBJ_BLOB) {
2736 free(buf);
2737 return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2739 strbuf_attach(dst, buf, size, size + 1);
2740 return 0;
2743 static int blob_unchanged(struct merge_options *opt,
2744 const struct object_id *o_oid,
2745 unsigned o_mode,
2746 const struct object_id *a_oid,
2747 unsigned a_mode,
2748 int renormalize, const char *path)
2750 struct strbuf o = STRBUF_INIT;
2751 struct strbuf a = STRBUF_INIT;
2752 int ret = 0; /* assume changed for safety */
2754 if (a_mode != o_mode)
2755 return 0;
2756 if (oid_eq(o_oid, a_oid))
2757 return 1;
2758 if (!renormalize)
2759 return 0;
2761 assert(o_oid && a_oid);
2762 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2763 goto error_return;
2765 * Note: binary | is used so that both renormalizations are
2766 * performed. Comparison can be skipped if both files are
2767 * unchanged since their sha1s have already been compared.
2769 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2770 renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2771 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2773 error_return:
2774 strbuf_release(&o);
2775 strbuf_release(&a);
2776 return ret;
2779 static int handle_modify_delete(struct merge_options *o,
2780 const char *path,
2781 struct object_id *o_oid, int o_mode,
2782 struct object_id *a_oid, int a_mode,
2783 struct object_id *b_oid, int b_mode)
2785 const char *modify_branch, *delete_branch;
2786 struct object_id *changed_oid;
2787 int changed_mode;
2789 if (a_oid) {
2790 modify_branch = o->branch1;
2791 delete_branch = o->branch2;
2792 changed_oid = a_oid;
2793 changed_mode = a_mode;
2794 } else {
2795 modify_branch = o->branch2;
2796 delete_branch = o->branch1;
2797 changed_oid = b_oid;
2798 changed_mode = b_mode;
2801 return handle_change_delete(o,
2802 path, NULL,
2803 o_oid, o_mode,
2804 changed_oid, changed_mode,
2805 modify_branch, delete_branch,
2806 _("modify"), _("modified"));
2809 static int merge_content(struct merge_options *o,
2810 const char *path,
2811 int is_dirty,
2812 struct object_id *o_oid, int o_mode,
2813 struct object_id *a_oid, int a_mode,
2814 struct object_id *b_oid, int b_mode,
2815 struct rename_conflict_info *rename_conflict_info)
2817 const char *reason = _("content");
2818 const char *path1 = NULL, *path2 = NULL;
2819 struct merge_file_info mfi;
2820 struct diff_filespec one, a, b;
2821 unsigned df_conflict_remains = 0;
2823 if (!o_oid) {
2824 reason = _("add/add");
2825 o_oid = (struct object_id *)&null_oid;
2827 one.path = a.path = b.path = (char *)path;
2828 oidcpy(&one.oid, o_oid);
2829 one.mode = o_mode;
2830 oidcpy(&a.oid, a_oid);
2831 a.mode = a_mode;
2832 oidcpy(&b.oid, b_oid);
2833 b.mode = b_mode;
2835 if (rename_conflict_info) {
2836 struct diff_filepair *pair1 = rename_conflict_info->pair1;
2838 path1 = (o->branch1 == rename_conflict_info->branch1) ?
2839 pair1->two->path : pair1->one->path;
2840 /* If rename_conflict_info->pair2 != NULL, we are in
2841 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
2842 * normal rename.
2844 path2 = (rename_conflict_info->pair2 ||
2845 o->branch2 == rename_conflict_info->branch1) ?
2846 pair1->two->path : pair1->one->path;
2848 if (dir_in_way(path, !o->call_depth,
2849 S_ISGITLINK(pair1->two->mode)))
2850 df_conflict_remains = 1;
2852 if (merge_file_special_markers(o, &one, &a, &b, path,
2853 o->branch1, path1,
2854 o->branch2, path2, &mfi))
2855 return -1;
2858 * We can skip updating the working tree file iff:
2859 * a) The merge is clean
2860 * b) The merge matches what was in HEAD (content, mode, pathname)
2861 * c) The target path is usable (i.e. not involved in D/F conflict)
2863 if (mfi.clean &&
2864 was_tracked_and_matches(o, path, &mfi.oid, mfi.mode) &&
2865 !df_conflict_remains) {
2866 output(o, 3, _("Skipped %s (merged same as existing)"), path);
2867 if (add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2868 0, (!o->call_depth && !is_dirty), 0))
2869 return -1;
2870 return mfi.clean;
2873 if (!mfi.clean) {
2874 if (S_ISGITLINK(mfi.mode))
2875 reason = _("submodule");
2876 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2877 reason, path);
2878 if (rename_conflict_info && !df_conflict_remains)
2879 if (update_stages(o, path, &one, &a, &b))
2880 return -1;
2883 if (df_conflict_remains || is_dirty) {
2884 char *new_path;
2885 if (o->call_depth) {
2886 remove_file_from_cache(path);
2887 } else {
2888 if (!mfi.clean) {
2889 if (update_stages(o, path, &one, &a, &b))
2890 return -1;
2891 } else {
2892 int file_from_stage2 = was_tracked(o, path);
2893 struct diff_filespec merged;
2894 oidcpy(&merged.oid, &mfi.oid);
2895 merged.mode = mfi.mode;
2897 if (update_stages(o, path, NULL,
2898 file_from_stage2 ? &merged : NULL,
2899 file_from_stage2 ? NULL : &merged))
2900 return -1;
2904 new_path = unique_path(o, path, rename_conflict_info->branch1);
2905 if (is_dirty) {
2906 output(o, 1, _("Refusing to lose dirty file at %s"),
2907 path);
2909 output(o, 1, _("Adding as %s instead"), new_path);
2910 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2911 free(new_path);
2912 return -1;
2914 free(new_path);
2915 mfi.clean = 0;
2916 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2917 return -1;
2918 return !is_dirty && mfi.clean;
2921 static int conflict_rename_normal(struct merge_options *o,
2922 const char *path,
2923 struct object_id *o_oid, unsigned int o_mode,
2924 struct object_id *a_oid, unsigned int a_mode,
2925 struct object_id *b_oid, unsigned int b_mode,
2926 struct rename_conflict_info *ci)
2928 /* Merge the content and write it out */
2929 return merge_content(o, path, was_dirty(o, path),
2930 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2931 ci);
2934 /* Per entry merge function */
2935 static int process_entry(struct merge_options *o,
2936 const char *path, struct stage_data *entry)
2938 int clean_merge = 1;
2939 int normalize = o->renormalize;
2940 unsigned o_mode = entry->stages[1].mode;
2941 unsigned a_mode = entry->stages[2].mode;
2942 unsigned b_mode = entry->stages[3].mode;
2943 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2944 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2945 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2947 entry->processed = 1;
2948 if (entry->rename_conflict_info) {
2949 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2950 switch (conflict_info->rename_type) {
2951 case RENAME_NORMAL:
2952 case RENAME_ONE_FILE_TO_ONE:
2953 clean_merge = conflict_rename_normal(o,
2954 path,
2955 o_oid, o_mode,
2956 a_oid, a_mode,
2957 b_oid, b_mode,
2958 conflict_info);
2959 break;
2960 case RENAME_DIR:
2961 clean_merge = 1;
2962 if (conflict_rename_dir(o,
2963 conflict_info->pair1,
2964 conflict_info->branch1,
2965 conflict_info->branch2))
2966 clean_merge = -1;
2967 break;
2968 case RENAME_DELETE:
2969 clean_merge = 0;
2970 if (conflict_rename_delete(o,
2971 conflict_info->pair1,
2972 conflict_info->branch1,
2973 conflict_info->branch2))
2974 clean_merge = -1;
2975 break;
2976 case RENAME_ONE_FILE_TO_TWO:
2977 clean_merge = 0;
2978 if (conflict_rename_rename_1to2(o, conflict_info))
2979 clean_merge = -1;
2980 break;
2981 case RENAME_TWO_FILES_TO_ONE:
2982 clean_merge = 0;
2983 if (conflict_rename_rename_2to1(o, conflict_info))
2984 clean_merge = -1;
2985 break;
2986 default:
2987 entry->processed = 0;
2988 break;
2990 } else if (o_oid && (!a_oid || !b_oid)) {
2991 /* Case A: Deleted in one */
2992 if ((!a_oid && !b_oid) ||
2993 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2994 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2995 /* Deleted in both or deleted in one and
2996 * unchanged in the other */
2997 if (a_oid)
2998 output(o, 2, _("Removing %s"), path);
2999 /* do not touch working file if it did not exist */
3000 remove_file(o, 1, path, !a_oid);
3001 } else {
3002 /* Modify/delete; deleted side may have put a directory in the way */
3003 clean_merge = 0;
3004 if (handle_modify_delete(o, path, o_oid, o_mode,
3005 a_oid, a_mode, b_oid, b_mode))
3006 clean_merge = -1;
3008 } else if ((!o_oid && a_oid && !b_oid) ||
3009 (!o_oid && !a_oid && b_oid)) {
3010 /* Case B: Added in one. */
3011 /* [nothing|directory] -> ([nothing|directory], file) */
3013 const char *add_branch;
3014 const char *other_branch;
3015 unsigned mode;
3016 const struct object_id *oid;
3017 const char *conf;
3019 if (a_oid) {
3020 add_branch = o->branch1;
3021 other_branch = o->branch2;
3022 mode = a_mode;
3023 oid = a_oid;
3024 conf = _("file/directory");
3025 } else {
3026 add_branch = o->branch2;
3027 other_branch = o->branch1;
3028 mode = b_mode;
3029 oid = b_oid;
3030 conf = _("directory/file");
3032 if (dir_in_way(path,
3033 !o->call_depth && !S_ISGITLINK(a_mode),
3034 0)) {
3035 char *new_path = unique_path(o, path, add_branch);
3036 clean_merge = 0;
3037 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3038 "Adding %s as %s"),
3039 conf, path, other_branch, path, new_path);
3040 if (update_file(o, 0, oid, mode, new_path))
3041 clean_merge = -1;
3042 else if (o->call_depth)
3043 remove_file_from_cache(path);
3044 free(new_path);
3045 } else {
3046 output(o, 2, _("Adding %s"), path);
3047 /* do not overwrite file if already present */
3048 if (update_file_flags(o, oid, mode, path, 1, !a_oid))
3049 clean_merge = -1;
3051 } else if (a_oid && b_oid) {
3052 /* Case C: Added in both (check for same permissions) and */
3053 /* case D: Modified in both, but differently. */
3054 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3055 clean_merge = merge_content(o, path, is_dirty,
3056 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
3057 NULL);
3058 } else if (!o_oid && !a_oid && !b_oid) {
3060 * this entry was deleted altogether. a_mode == 0 means
3061 * we had that path and want to actively remove it.
3063 remove_file(o, 1, path, !a_mode);
3064 } else
3065 die("BUG: fatal merge failure, shouldn't happen.");
3067 return clean_merge;
3070 int merge_trees(struct merge_options *o,
3071 struct tree *head,
3072 struct tree *merge,
3073 struct tree *common,
3074 struct tree **result)
3076 int code, clean;
3078 if (o->subtree_shift) {
3079 merge = shift_tree_object(head, merge, o->subtree_shift);
3080 common = shift_tree_object(head, common, o->subtree_shift);
3083 if (oid_eq(&common->object.oid, &merge->object.oid)) {
3084 struct strbuf sb = STRBUF_INIT;
3086 if (!o->call_depth && index_has_changes(&sb)) {
3087 err(o, _("Dirty index: cannot merge (dirty: %s)"),
3088 sb.buf);
3089 return 0;
3091 output(o, 0, _("Already up to date!"));
3092 *result = head;
3093 return 1;
3096 code = unpack_trees_start(o, common, head, merge);
3098 if (code != 0) {
3099 if (show(o, 4) || o->call_depth)
3100 err(o, _("merging of trees %s and %s failed"),
3101 oid_to_hex(&head->object.oid),
3102 oid_to_hex(&merge->object.oid));
3103 unpack_trees_finish(o);
3104 return -1;
3107 if (unmerged_cache()) {
3108 struct string_list *entries;
3109 struct rename_info re_info;
3110 int i;
3112 * Only need the hashmap while processing entries, so
3113 * initialize it here and free it when we are done running
3114 * through the entries. Keeping it in the merge_options as
3115 * opposed to decaring a local hashmap is for convenience
3116 * so that we don't have to pass it to around.
3118 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3119 get_files_dirs(o, head);
3120 get_files_dirs(o, merge);
3122 entries = get_unmerged();
3123 clean = handle_renames(o, common, head, merge, entries,
3124 &re_info);
3125 record_df_conflict_files(o, entries);
3126 if (clean < 0)
3127 goto cleanup;
3128 for (i = entries->nr-1; 0 <= i; i--) {
3129 const char *path = entries->items[i].string;
3130 struct stage_data *e = entries->items[i].util;
3131 if (!e->processed) {
3132 int ret = process_entry(o, path, e);
3133 if (!ret)
3134 clean = 0;
3135 else if (ret < 0) {
3136 clean = ret;
3137 goto cleanup;
3141 for (i = 0; i < entries->nr; i++) {
3142 struct stage_data *e = entries->items[i].util;
3143 if (!e->processed)
3144 die("BUG: unprocessed path??? %s",
3145 entries->items[i].string);
3148 cleanup:
3149 final_cleanup_renames(&re_info);
3151 string_list_clear(entries, 1);
3152 free(entries);
3154 hashmap_free(&o->current_file_dir_set, 1);
3156 if (clean < 0) {
3157 unpack_trees_finish(o);
3158 return clean;
3161 else
3162 clean = 1;
3164 unpack_trees_finish(o);
3166 if (o->call_depth && !(*result = write_tree_from_memory(o)))
3167 return -1;
3169 return clean;
3172 static struct commit_list *reverse_commit_list(struct commit_list *list)
3174 struct commit_list *next = NULL, *current, *backup;
3175 for (current = list; current; current = backup) {
3176 backup = current->next;
3177 current->next = next;
3178 next = current;
3180 return next;
3184 * Merge the commits h1 and h2, return the resulting virtual
3185 * commit object and a flag indicating the cleanness of the merge.
3187 int merge_recursive(struct merge_options *o,
3188 struct commit *h1,
3189 struct commit *h2,
3190 struct commit_list *ca,
3191 struct commit **result)
3193 struct commit_list *iter;
3194 struct commit *merged_common_ancestors;
3195 struct tree *mrtree;
3196 int clean;
3198 if (show(o, 4)) {
3199 output(o, 4, _("Merging:"));
3200 output_commit_title(o, h1);
3201 output_commit_title(o, h2);
3204 if (!ca) {
3205 ca = get_merge_bases(h1, h2);
3206 ca = reverse_commit_list(ca);
3209 if (show(o, 5)) {
3210 unsigned cnt = commit_list_count(ca);
3212 output(o, 5, Q_("found %u common ancestor:",
3213 "found %u common ancestors:", cnt), cnt);
3214 for (iter = ca; iter; iter = iter->next)
3215 output_commit_title(o, iter->item);
3218 merged_common_ancestors = pop_commit(&ca);
3219 if (merged_common_ancestors == NULL) {
3220 /* if there is no common ancestor, use an empty tree */
3221 struct tree *tree;
3223 tree = lookup_tree(the_hash_algo->empty_tree);
3224 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3227 for (iter = ca; iter; iter = iter->next) {
3228 const char *saved_b1, *saved_b2;
3229 o->call_depth++;
3231 * When the merge fails, the result contains files
3232 * with conflict markers. The cleanness flag is
3233 * ignored (unless indicating an error), it was never
3234 * actually used, as result of merge_trees has always
3235 * overwritten it: the committed "conflicts" were
3236 * already resolved.
3238 discard_cache();
3239 saved_b1 = o->branch1;
3240 saved_b2 = o->branch2;
3241 o->branch1 = "Temporary merge branch 1";
3242 o->branch2 = "Temporary merge branch 2";
3243 if (merge_recursive(o, merged_common_ancestors, iter->item,
3244 NULL, &merged_common_ancestors) < 0)
3245 return -1;
3246 o->branch1 = saved_b1;
3247 o->branch2 = saved_b2;
3248 o->call_depth--;
3250 if (!merged_common_ancestors)
3251 return err(o, _("merge returned no commit"));
3254 discard_cache();
3255 if (!o->call_depth)
3256 read_cache();
3258 o->ancestor = "merged common ancestors";
3259 clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,
3260 &mrtree);
3261 if (clean < 0) {
3262 flush_output(o);
3263 return clean;
3266 if (o->call_depth) {
3267 *result = make_virtual_commit(mrtree, "merged tree");
3268 commit_list_insert(h1, &(*result)->parents);
3269 commit_list_insert(h2, &(*result)->parents->next);
3271 flush_output(o);
3272 if (!o->call_depth && o->buffer_output < 2)
3273 strbuf_release(&o->obuf);
3274 if (show(o, 2))
3275 diff_warn_rename_limit("merge.renamelimit",
3276 o->needed_rename_limit, 0);
3277 return clean;
3280 static struct commit *get_ref(const struct object_id *oid, const char *name)
3282 struct object *object;
3284 object = deref_tag(parse_object(oid), name, strlen(name));
3285 if (!object)
3286 return NULL;
3287 if (object->type == OBJ_TREE)
3288 return make_virtual_commit((struct tree*)object, name);
3289 if (object->type != OBJ_COMMIT)
3290 return NULL;
3291 if (parse_commit((struct commit *)object))
3292 return NULL;
3293 return (struct commit *)object;
3296 int merge_recursive_generic(struct merge_options *o,
3297 const struct object_id *head,
3298 const struct object_id *merge,
3299 int num_base_list,
3300 const struct object_id **base_list,
3301 struct commit **result)
3303 int clean;
3304 struct lock_file lock = LOCK_INIT;
3305 struct commit *head_commit = get_ref(head, o->branch1);
3306 struct commit *next_commit = get_ref(merge, o->branch2);
3307 struct commit_list *ca = NULL;
3309 if (base_list) {
3310 int i;
3311 for (i = 0; i < num_base_list; ++i) {
3312 struct commit *base;
3313 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3314 return err(o, _("Could not parse object '%s'"),
3315 oid_to_hex(base_list[i]));
3316 commit_list_insert(base, &ca);
3320 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3321 clean = merge_recursive(o, head_commit, next_commit, ca,
3322 result);
3323 if (clean < 0) {
3324 rollback_lock_file(&lock);
3325 return clean;
3328 if (write_locked_index(&the_index, &lock,
3329 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3330 return err(o, _("Unable to write index."));
3332 return clean ? 0 : 1;
3335 static void merge_recursive_config(struct merge_options *o)
3337 char *value = NULL;
3338 git_config_get_int("merge.verbosity", &o->verbosity);
3339 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3340 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3341 if (!git_config_get_string("diff.renames", &value)) {
3342 o->diff_detect_rename = git_config_rename("diff.renames", value);
3343 free(value);
3345 if (!git_config_get_string("merge.renames", &value)) {
3346 o->merge_detect_rename = git_config_rename("merge.renames", value);
3347 free(value);
3349 git_config(git_xmerge_config, NULL);
3352 void init_merge_options(struct merge_options *o)
3354 const char *merge_verbosity;
3355 memset(o, 0, sizeof(struct merge_options));
3356 o->verbosity = 2;
3357 o->buffer_output = 1;
3358 o->diff_rename_limit = -1;
3359 o->merge_rename_limit = -1;
3360 o->renormalize = 0;
3361 o->diff_detect_rename = -1;
3362 o->merge_detect_rename = -1;
3363 merge_recursive_config(o);
3364 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3365 if (merge_verbosity)
3366 o->verbosity = strtol(merge_verbosity, NULL, 10);
3367 if (o->verbosity >= 5)
3368 o->buffer_output = 0;
3369 strbuf_init(&o->obuf, 0);
3370 string_list_init(&o->df_conflict_file_set, 1);
3373 int parse_merge_opt(struct merge_options *o, const char *s)
3375 const char *arg;
3377 if (!s || !*s)
3378 return -1;
3379 if (!strcmp(s, "ours"))
3380 o->recursive_variant = MERGE_RECURSIVE_OURS;
3381 else if (!strcmp(s, "theirs"))
3382 o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3383 else if (!strcmp(s, "subtree"))
3384 o->subtree_shift = "";
3385 else if (skip_prefix(s, "subtree=", &arg))
3386 o->subtree_shift = arg;
3387 else if (!strcmp(s, "patience"))
3388 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3389 else if (!strcmp(s, "histogram"))
3390 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3391 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3392 long value = parse_algorithm_value(arg);
3393 if (value < 0)
3394 return -1;
3395 /* clear out previous settings */
3396 DIFF_XDL_CLR(o, NEED_MINIMAL);
3397 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3398 o->xdl_opts |= value;
3400 else if (!strcmp(s, "ignore-space-change"))
3401 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3402 else if (!strcmp(s, "ignore-all-space"))
3403 DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3404 else if (!strcmp(s, "ignore-space-at-eol"))
3405 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3406 else if (!strcmp(s, "ignore-cr-at-eol"))
3407 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3408 else if (!strcmp(s, "renormalize"))
3409 o->renormalize = 1;
3410 else if (!strcmp(s, "no-renormalize"))
3411 o->renormalize = 0;
3412 else if (!strcmp(s, "no-renames"))
3413 o->merge_detect_rename = 0;
3414 else if (!strcmp(s, "find-renames")) {
3415 o->merge_detect_rename = 1;
3416 o->rename_score = 0;
3418 else if (skip_prefix(s, "find-renames=", &arg) ||
3419 skip_prefix(s, "rename-threshold=", &arg)) {
3420 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3421 return -1;
3422 o->merge_detect_rename = 1;
3424 else
3425 return -1;
3426 return 0;