merge-recursive: fix remainder of was_dirty() to use original index
[git/debian.git] / merge-recursive.c
blob9015be55be9ac67bde9aad7419b5c9d837ecf5e3
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 git_merge_trees(struct merge_options *o,
341 struct tree *common,
342 struct tree *head,
343 struct tree *merge)
345 int rc;
346 struct tree_desc t[3];
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 setup_unpack_trees_porcelain(&o->unpack_opts, "merge");
361 init_tree_desc_from_tree(t+0, common);
362 init_tree_desc_from_tree(t+1, head);
363 init_tree_desc_from_tree(t+2, merge);
365 rc = unpack_trees(3, t, &o->unpack_opts);
366 cache_tree_free(&active_cache_tree);
369 * Update the_index to match the new results, AFTER saving a copy
370 * in o->orig_index. Update src_index to point to the saved copy.
371 * (verify_uptodate() checks src_index, and the original index is
372 * the one that had the necessary modification timestamps.)
374 o->orig_index = the_index;
375 the_index = tmp_index;
376 o->unpack_opts.src_index = &o->orig_index;
378 return rc;
381 struct tree *write_tree_from_memory(struct merge_options *o)
383 struct tree *result = NULL;
385 if (unmerged_cache()) {
386 int i;
387 fprintf(stderr, "BUG: There are unmerged index entries:\n");
388 for (i = 0; i < active_nr; i++) {
389 const struct cache_entry *ce = active_cache[i];
390 if (ce_stage(ce))
391 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
392 (int)ce_namelen(ce), ce->name);
394 die("BUG: unmerged index entries in merge-recursive.c");
397 if (!active_cache_tree)
398 active_cache_tree = cache_tree();
400 if (!cache_tree_fully_valid(active_cache_tree) &&
401 cache_tree_update(&the_index, 0) < 0) {
402 err(o, _("error building trees"));
403 return NULL;
406 result = lookup_tree(&active_cache_tree->oid);
408 return result;
411 static int save_files_dirs(const struct object_id *oid,
412 struct strbuf *base, const char *path,
413 unsigned int mode, int stage, void *context)
415 struct path_hashmap_entry *entry;
416 int baselen = base->len;
417 struct merge_options *o = context;
419 strbuf_addstr(base, path);
421 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
422 hashmap_entry_init(entry, path_hash(entry->path));
423 hashmap_add(&o->current_file_dir_set, entry);
425 strbuf_setlen(base, baselen);
426 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
429 static void get_files_dirs(struct merge_options *o, struct tree *tree)
431 struct pathspec match_all;
432 memset(&match_all, 0, sizeof(match_all));
433 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
436 static int get_tree_entry_if_blob(const struct object_id *tree,
437 const char *path,
438 struct object_id *hashy,
439 unsigned int *mode_o)
441 int ret;
443 ret = get_tree_entry(tree, path, hashy, mode_o);
444 if (S_ISDIR(*mode_o)) {
445 oidcpy(hashy, &null_oid);
446 *mode_o = 0;
448 return ret;
452 * Returns an index_entry instance which doesn't have to correspond to
453 * a real cache entry in Git's index.
455 static struct stage_data *insert_stage_data(const char *path,
456 struct tree *o, struct tree *a, struct tree *b,
457 struct string_list *entries)
459 struct string_list_item *item;
460 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
461 get_tree_entry_if_blob(&o->object.oid, path,
462 &e->stages[1].oid, &e->stages[1].mode);
463 get_tree_entry_if_blob(&a->object.oid, path,
464 &e->stages[2].oid, &e->stages[2].mode);
465 get_tree_entry_if_blob(&b->object.oid, path,
466 &e->stages[3].oid, &e->stages[3].mode);
467 item = string_list_insert(entries, path);
468 item->util = e;
469 return e;
473 * Create a dictionary mapping file names to stage_data objects. The
474 * dictionary contains one entry for every path with a non-zero stage entry.
476 static struct string_list *get_unmerged(void)
478 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
479 int i;
481 unmerged->strdup_strings = 1;
483 for (i = 0; i < active_nr; i++) {
484 struct string_list_item *item;
485 struct stage_data *e;
486 const struct cache_entry *ce = active_cache[i];
487 if (!ce_stage(ce))
488 continue;
490 item = string_list_lookup(unmerged, ce->name);
491 if (!item) {
492 item = string_list_insert(unmerged, ce->name);
493 item->util = xcalloc(1, sizeof(struct stage_data));
495 e = item->util;
496 e->stages[ce_stage(ce)].mode = ce->ce_mode;
497 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
500 return unmerged;
503 static int string_list_df_name_compare(const char *one, const char *two)
505 int onelen = strlen(one);
506 int twolen = strlen(two);
508 * Here we only care that entries for D/F conflicts are
509 * adjacent, in particular with the file of the D/F conflict
510 * appearing before files below the corresponding directory.
511 * The order of the rest of the list is irrelevant for us.
513 * To achieve this, we sort with df_name_compare and provide
514 * the mode S_IFDIR so that D/F conflicts will sort correctly.
515 * We use the mode S_IFDIR for everything else for simplicity,
516 * since in other cases any changes in their order due to
517 * sorting cause no problems for us.
519 int cmp = df_name_compare(one, onelen, S_IFDIR,
520 two, twolen, S_IFDIR);
522 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
523 * that 'foo' comes before 'foo/bar'.
525 if (cmp)
526 return cmp;
527 return onelen - twolen;
530 static void record_df_conflict_files(struct merge_options *o,
531 struct string_list *entries)
533 /* If there is a D/F conflict and the file for such a conflict
534 * currently exist in the working tree, we want to allow it to be
535 * removed to make room for the corresponding directory if needed.
536 * The files underneath the directories of such D/F conflicts will
537 * be processed before the corresponding file involved in the D/F
538 * conflict. If the D/F directory ends up being removed by the
539 * merge, then we won't have to touch the D/F file. If the D/F
540 * directory needs to be written to the working copy, then the D/F
541 * file will simply be removed (in make_room_for_path()) to make
542 * room for the necessary paths. Note that if both the directory
543 * and the file need to be present, then the D/F file will be
544 * reinstated with a new unique name at the time it is processed.
546 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
547 const char *last_file = NULL;
548 int last_len = 0;
549 int i;
552 * If we're merging merge-bases, we don't want to bother with
553 * any working directory changes.
555 if (o->call_depth)
556 return;
558 /* Ensure D/F conflicts are adjacent in the entries list. */
559 for (i = 0; i < entries->nr; i++) {
560 struct string_list_item *next = &entries->items[i];
561 string_list_append(&df_sorted_entries, next->string)->util =
562 next->util;
564 df_sorted_entries.cmp = string_list_df_name_compare;
565 string_list_sort(&df_sorted_entries);
567 string_list_clear(&o->df_conflict_file_set, 1);
568 for (i = 0; i < df_sorted_entries.nr; i++) {
569 const char *path = df_sorted_entries.items[i].string;
570 int len = strlen(path);
571 struct stage_data *e = df_sorted_entries.items[i].util;
574 * Check if last_file & path correspond to a D/F conflict;
575 * i.e. whether path is last_file+'/'+<something>.
576 * If so, record that it's okay to remove last_file to make
577 * room for path and friends if needed.
579 if (last_file &&
580 len > last_len &&
581 memcmp(path, last_file, last_len) == 0 &&
582 path[last_len] == '/') {
583 string_list_insert(&o->df_conflict_file_set, last_file);
587 * Determine whether path could exist as a file in the
588 * working directory as a possible D/F conflict. This
589 * will only occur when it exists in stage 2 as a
590 * file.
592 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
593 last_file = path;
594 last_len = len;
595 } else {
596 last_file = NULL;
599 string_list_clear(&df_sorted_entries, 0);
602 struct rename {
603 struct diff_filepair *pair;
605 * Purpose of src_entry and dst_entry:
607 * If 'before' is renamed to 'after' then src_entry will contain
608 * the versions of 'before' from the merge_base, HEAD, and MERGE in
609 * stages 1, 2, and 3; dst_entry will contain the respective
610 * versions of 'after' in corresponding locations. Thus, we have a
611 * total of six modes and oids, though some will be null. (Stage 0
612 * is ignored; we're interested in handling conflicts.)
614 * Since we don't turn on break-rewrites by default, neither
615 * src_entry nor dst_entry can have all three of their stages have
616 * non-null oids, meaning at most four of the six will be non-null.
617 * Also, since this is a rename, both src_entry and dst_entry will
618 * have at least one non-null oid, meaning at least two will be
619 * non-null. Of the six oids, a typical rename will have three be
620 * non-null. Only two implies a rename/delete, and four implies a
621 * rename/add.
623 struct stage_data *src_entry;
624 struct stage_data *dst_entry;
625 unsigned add_turned_into_rename:1;
626 unsigned processed:1;
629 static int update_stages(struct merge_options *opt, const char *path,
630 const struct diff_filespec *o,
631 const struct diff_filespec *a,
632 const struct diff_filespec *b)
636 * NOTE: It is usually a bad idea to call update_stages on a path
637 * before calling update_file on that same path, since it can
638 * sometimes lead to spurious "refusing to lose untracked file..."
639 * messages from update_file (via make_room_for path via
640 * would_lose_untracked). Instead, reverse the order of the calls
641 * (executing update_file first and then update_stages).
643 int clear = 1;
644 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
645 if (clear)
646 if (remove_file_from_cache(path))
647 return -1;
648 if (o)
649 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
650 return -1;
651 if (a)
652 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
653 return -1;
654 if (b)
655 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
656 return -1;
657 return 0;
660 static int update_stages_for_stage_data(struct merge_options *opt,
661 const char *path,
662 const struct stage_data *stage_data)
664 struct diff_filespec o, a, b;
666 o.mode = stage_data->stages[1].mode;
667 oidcpy(&o.oid, &stage_data->stages[1].oid);
669 a.mode = stage_data->stages[2].mode;
670 oidcpy(&a.oid, &stage_data->stages[2].oid);
672 b.mode = stage_data->stages[3].mode;
673 oidcpy(&b.oid, &stage_data->stages[3].oid);
675 return update_stages(opt, path,
676 is_null_oid(&o.oid) ? NULL : &o,
677 is_null_oid(&a.oid) ? NULL : &a,
678 is_null_oid(&b.oid) ? NULL : &b);
681 static void update_entry(struct stage_data *entry,
682 struct diff_filespec *o,
683 struct diff_filespec *a,
684 struct diff_filespec *b)
686 entry->processed = 0;
687 entry->stages[1].mode = o->mode;
688 entry->stages[2].mode = a->mode;
689 entry->stages[3].mode = b->mode;
690 oidcpy(&entry->stages[1].oid, &o->oid);
691 oidcpy(&entry->stages[2].oid, &a->oid);
692 oidcpy(&entry->stages[3].oid, &b->oid);
695 static int remove_file(struct merge_options *o, int clean,
696 const char *path, int no_wd)
698 int update_cache = o->call_depth || clean;
699 int update_working_directory = !o->call_depth && !no_wd;
701 if (update_cache) {
702 if (remove_file_from_cache(path))
703 return -1;
705 if (update_working_directory) {
706 if (ignore_case) {
707 struct cache_entry *ce;
708 ce = cache_file_exists(path, strlen(path), ignore_case);
709 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
710 return 0;
712 if (remove_path(path))
713 return -1;
715 return 0;
718 /* add a string to a strbuf, but converting "/" to "_" */
719 static void add_flattened_path(struct strbuf *out, const char *s)
721 size_t i = out->len;
722 strbuf_addstr(out, s);
723 for (; i < out->len; i++)
724 if (out->buf[i] == '/')
725 out->buf[i] = '_';
728 static char *unique_path(struct merge_options *o, const char *path, const char *branch)
730 struct path_hashmap_entry *entry;
731 struct strbuf newpath = STRBUF_INIT;
732 int suffix = 0;
733 size_t base_len;
735 strbuf_addf(&newpath, "%s~", path);
736 add_flattened_path(&newpath, branch);
738 base_len = newpath.len;
739 while (hashmap_get_from_hash(&o->current_file_dir_set,
740 path_hash(newpath.buf), newpath.buf) ||
741 (!o->call_depth && file_exists(newpath.buf))) {
742 strbuf_setlen(&newpath, base_len);
743 strbuf_addf(&newpath, "_%d", suffix++);
746 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
747 hashmap_entry_init(entry, path_hash(entry->path));
748 hashmap_add(&o->current_file_dir_set, entry);
749 return strbuf_detach(&newpath, NULL);
753 * Check whether a directory in the index is in the way of an incoming
754 * file. Return 1 if so. If check_working_copy is non-zero, also
755 * check the working directory. If empty_ok is non-zero, also return
756 * 0 in the case where the working-tree dir exists but is empty.
758 static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
760 int pos;
761 struct strbuf dirpath = STRBUF_INIT;
762 struct stat st;
764 strbuf_addstr(&dirpath, path);
765 strbuf_addch(&dirpath, '/');
767 pos = cache_name_pos(dirpath.buf, dirpath.len);
769 if (pos < 0)
770 pos = -1 - pos;
771 if (pos < active_nr &&
772 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
773 strbuf_release(&dirpath);
774 return 1;
777 strbuf_release(&dirpath);
778 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
779 !(empty_ok && is_empty_dir(path));
783 * Returns whether path was tracked in the index before the merge started
785 static int was_tracked(struct merge_options *o, const char *path)
787 int pos = index_name_pos(&o->orig_index, path, strlen(path));
789 if (0 <= pos)
790 /* we were tracking this path before the merge */
791 return 1;
793 return 0;
796 static int would_lose_untracked(const char *path)
799 * This may look like it can be simplified to:
800 * return !was_tracked(o, path) && file_exists(path)
801 * but it can't. This function needs to know whether path was in
802 * the working tree due to EITHER having been tracked in the index
803 * before the merge OR having been put into the working copy and
804 * index by unpack_trees(). Due to that either-or requirement, we
805 * check the current index instead of the original one.
807 * Note that we do not need to worry about merge-recursive itself
808 * updating the index after unpack_trees() and before calling this
809 * function, because we strictly require all code paths in
810 * merge-recursive to update the working tree first and the index
811 * second. Doing otherwise would break
812 * update_file()/would_lose_untracked(); see every comment in this
813 * file which mentions "update_stages".
815 int pos = cache_name_pos(path, strlen(path));
817 if (pos < 0)
818 pos = -1 - pos;
819 while (pos < active_nr &&
820 !strcmp(path, active_cache[pos]->name)) {
822 * If stage #0, it is definitely tracked.
823 * If it has stage #2 then it was tracked
824 * before this merge started. All other
825 * cases the path was not tracked.
827 switch (ce_stage(active_cache[pos])) {
828 case 0:
829 case 2:
830 return 0;
832 pos++;
834 return file_exists(path);
837 static int was_dirty(struct merge_options *o, const char *path)
839 struct cache_entry *ce;
840 int dirty = 1;
842 if (o->call_depth || !was_tracked(o, path))
843 return !dirty;
845 ce = index_file_exists(o->unpack_opts.src_index,
846 path, strlen(path), ignore_case);
847 dirty = verify_uptodate(ce, &o->unpack_opts) != 0;
848 return dirty;
851 static int make_room_for_path(struct merge_options *o, const char *path)
853 int status, i;
854 const char *msg = _("failed to create path '%s'%s");
856 /* Unlink any D/F conflict files that are in the way */
857 for (i = 0; i < o->df_conflict_file_set.nr; i++) {
858 const char *df_path = o->df_conflict_file_set.items[i].string;
859 size_t pathlen = strlen(path);
860 size_t df_pathlen = strlen(df_path);
861 if (df_pathlen < pathlen &&
862 path[df_pathlen] == '/' &&
863 strncmp(path, df_path, df_pathlen) == 0) {
864 output(o, 3,
865 _("Removing %s to make room for subdirectory\n"),
866 df_path);
867 unlink(df_path);
868 unsorted_string_list_delete_item(&o->df_conflict_file_set,
869 i, 0);
870 break;
874 /* Make sure leading directories are created */
875 status = safe_create_leading_directories_const(path);
876 if (status) {
877 if (status == SCLD_EXISTS)
878 /* something else exists */
879 return err(o, msg, path, _(": perhaps a D/F conflict?"));
880 return err(o, msg, path, "");
884 * Do not unlink a file in the work tree if we are not
885 * tracking it.
887 if (would_lose_untracked(path))
888 return err(o, _("refusing to lose untracked file at '%s'"),
889 path);
891 /* Successful unlink is good.. */
892 if (!unlink(path))
893 return 0;
894 /* .. and so is no existing file */
895 if (errno == ENOENT)
896 return 0;
897 /* .. but not some other error (who really cares what?) */
898 return err(o, msg, path, _(": perhaps a D/F conflict?"));
901 static int update_file_flags(struct merge_options *o,
902 const struct object_id *oid,
903 unsigned mode,
904 const char *path,
905 int update_cache,
906 int update_wd)
908 int ret = 0;
910 if (o->call_depth)
911 update_wd = 0;
913 if (update_wd) {
914 enum object_type type;
915 void *buf;
916 unsigned long size;
918 if (S_ISGITLINK(mode)) {
920 * We may later decide to recursively descend into
921 * the submodule directory and update its index
922 * and/or work tree, but we do not do that now.
924 update_wd = 0;
925 goto update_index;
928 buf = read_object_file(oid, &type, &size);
929 if (!buf)
930 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
931 if (type != OBJ_BLOB) {
932 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
933 goto free_buf;
935 if (S_ISREG(mode)) {
936 struct strbuf strbuf = STRBUF_INIT;
937 if (convert_to_working_tree(path, buf, size, &strbuf)) {
938 free(buf);
939 size = strbuf.len;
940 buf = strbuf_detach(&strbuf, NULL);
944 if (make_room_for_path(o, path) < 0) {
945 update_wd = 0;
946 goto free_buf;
948 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
949 int fd;
950 if (mode & 0100)
951 mode = 0777;
952 else
953 mode = 0666;
954 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
955 if (fd < 0) {
956 ret = err(o, _("failed to open '%s': %s"),
957 path, strerror(errno));
958 goto free_buf;
960 write_in_full(fd, buf, size);
961 close(fd);
962 } else if (S_ISLNK(mode)) {
963 char *lnk = xmemdupz(buf, size);
964 safe_create_leading_directories_const(path);
965 unlink(path);
966 if (symlink(lnk, path))
967 ret = err(o, _("failed to symlink '%s': %s"),
968 path, strerror(errno));
969 free(lnk);
970 } else
971 ret = err(o,
972 _("do not know what to do with %06o %s '%s'"),
973 mode, oid_to_hex(oid), path);
974 free_buf:
975 free(buf);
977 update_index:
978 if (!ret && update_cache)
979 if (add_cacheinfo(o, mode, oid, path, 0, update_wd,
980 ADD_CACHE_OK_TO_ADD))
981 return -1;
982 return ret;
985 static int update_file(struct merge_options *o,
986 int clean,
987 const struct object_id *oid,
988 unsigned mode,
989 const char *path)
991 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
994 /* Low level file merging, update and removal */
996 struct merge_file_info {
997 struct object_id oid;
998 unsigned mode;
999 unsigned clean:1,
1000 merge:1;
1003 static int merge_3way(struct merge_options *o,
1004 mmbuffer_t *result_buf,
1005 const struct diff_filespec *one,
1006 const struct diff_filespec *a,
1007 const struct diff_filespec *b,
1008 const char *branch1,
1009 const char *branch2)
1011 mmfile_t orig, src1, src2;
1012 struct ll_merge_options ll_opts = {0};
1013 char *base_name, *name1, *name2;
1014 int merge_status;
1016 ll_opts.renormalize = o->renormalize;
1017 ll_opts.xdl_opts = o->xdl_opts;
1019 if (o->call_depth) {
1020 ll_opts.virtual_ancestor = 1;
1021 ll_opts.variant = 0;
1022 } else {
1023 switch (o->recursive_variant) {
1024 case MERGE_RECURSIVE_OURS:
1025 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1026 break;
1027 case MERGE_RECURSIVE_THEIRS:
1028 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1029 break;
1030 default:
1031 ll_opts.variant = 0;
1032 break;
1036 if (strcmp(a->path, b->path) ||
1037 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1038 base_name = o->ancestor == NULL ? NULL :
1039 mkpathdup("%s:%s", o->ancestor, one->path);
1040 name1 = mkpathdup("%s:%s", branch1, a->path);
1041 name2 = mkpathdup("%s:%s", branch2, b->path);
1042 } else {
1043 base_name = o->ancestor == NULL ? NULL :
1044 mkpathdup("%s", o->ancestor);
1045 name1 = mkpathdup("%s", branch1);
1046 name2 = mkpathdup("%s", branch2);
1049 read_mmblob(&orig, &one->oid);
1050 read_mmblob(&src1, &a->oid);
1051 read_mmblob(&src2, &b->oid);
1053 merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1054 &src1, name1, &src2, name2, &ll_opts);
1056 free(base_name);
1057 free(name1);
1058 free(name2);
1059 free(orig.ptr);
1060 free(src1.ptr);
1061 free(src2.ptr);
1062 return merge_status;
1065 static int merge_file_1(struct merge_options *o,
1066 const struct diff_filespec *one,
1067 const struct diff_filespec *a,
1068 const struct diff_filespec *b,
1069 const char *branch1,
1070 const char *branch2,
1071 struct merge_file_info *result)
1073 result->merge = 0;
1074 result->clean = 1;
1076 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1077 result->clean = 0;
1078 if (S_ISREG(a->mode)) {
1079 result->mode = a->mode;
1080 oidcpy(&result->oid, &a->oid);
1081 } else {
1082 result->mode = b->mode;
1083 oidcpy(&result->oid, &b->oid);
1085 } else {
1086 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1087 result->merge = 1;
1090 * Merge modes
1092 if (a->mode == b->mode || a->mode == one->mode)
1093 result->mode = b->mode;
1094 else {
1095 result->mode = a->mode;
1096 if (b->mode != one->mode) {
1097 result->clean = 0;
1098 result->merge = 1;
1102 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1103 oidcpy(&result->oid, &b->oid);
1104 else if (oid_eq(&b->oid, &one->oid))
1105 oidcpy(&result->oid, &a->oid);
1106 else if (S_ISREG(a->mode)) {
1107 mmbuffer_t result_buf;
1108 int ret = 0, merge_status;
1110 merge_status = merge_3way(o, &result_buf, one, a, b,
1111 branch1, branch2);
1113 if ((merge_status < 0) || !result_buf.ptr)
1114 ret = err(o, _("Failed to execute internal merge"));
1116 if (!ret &&
1117 write_object_file(result_buf.ptr, result_buf.size,
1118 blob_type, &result->oid))
1119 ret = err(o, _("Unable to add %s to database"),
1120 a->path);
1122 free(result_buf.ptr);
1123 if (ret)
1124 return ret;
1125 result->clean = (merge_status == 0);
1126 } else if (S_ISGITLINK(a->mode)) {
1127 result->clean = merge_submodule(&result->oid,
1128 one->path,
1129 &one->oid,
1130 &a->oid,
1131 &b->oid,
1132 !o->call_depth);
1133 } else if (S_ISLNK(a->mode)) {
1134 switch (o->recursive_variant) {
1135 case MERGE_RECURSIVE_NORMAL:
1136 oidcpy(&result->oid, &a->oid);
1137 if (!oid_eq(&a->oid, &b->oid))
1138 result->clean = 0;
1139 break;
1140 case MERGE_RECURSIVE_OURS:
1141 oidcpy(&result->oid, &a->oid);
1142 break;
1143 case MERGE_RECURSIVE_THEIRS:
1144 oidcpy(&result->oid, &b->oid);
1145 break;
1147 } else
1148 die("BUG: unsupported object type in the tree");
1151 return 0;
1154 static int merge_file_special_markers(struct merge_options *o,
1155 const struct diff_filespec *one,
1156 const struct diff_filespec *a,
1157 const struct diff_filespec *b,
1158 const char *branch1,
1159 const char *filename1,
1160 const char *branch2,
1161 const char *filename2,
1162 struct merge_file_info *mfi)
1164 char *side1 = NULL;
1165 char *side2 = NULL;
1166 int ret;
1168 if (filename1)
1169 side1 = xstrfmt("%s:%s", branch1, filename1);
1170 if (filename2)
1171 side2 = xstrfmt("%s:%s", branch2, filename2);
1173 ret = merge_file_1(o, one, a, b,
1174 side1 ? side1 : branch1,
1175 side2 ? side2 : branch2, mfi);
1176 free(side1);
1177 free(side2);
1178 return ret;
1181 static int merge_file_one(struct merge_options *o,
1182 const char *path,
1183 const struct object_id *o_oid, int o_mode,
1184 const struct object_id *a_oid, int a_mode,
1185 const struct object_id *b_oid, int b_mode,
1186 const char *branch1,
1187 const char *branch2,
1188 struct merge_file_info *mfi)
1190 struct diff_filespec one, a, b;
1192 one.path = a.path = b.path = (char *)path;
1193 oidcpy(&one.oid, o_oid);
1194 one.mode = o_mode;
1195 oidcpy(&a.oid, a_oid);
1196 a.mode = a_mode;
1197 oidcpy(&b.oid, b_oid);
1198 b.mode = b_mode;
1199 return merge_file_1(o, &one, &a, &b, branch1, branch2, mfi);
1202 static int conflict_rename_dir(struct merge_options *o,
1203 struct diff_filepair *pair,
1204 const char *rename_branch,
1205 const char *other_branch)
1207 const struct diff_filespec *dest = pair->two;
1209 if (!o->call_depth && would_lose_untracked(dest->path)) {
1210 char *alt_path = unique_path(o, dest->path, rename_branch);
1212 output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1213 "writing to %s instead."),
1214 dest->path, alt_path);
1216 * Write the file in worktree at alt_path, but not in the
1217 * index. Instead, write to dest->path for the index but
1218 * only at the higher appropriate stage.
1220 if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1221 return -1;
1222 free(alt_path);
1223 return update_stages(o, dest->path, NULL,
1224 rename_branch == o->branch1 ? dest : NULL,
1225 rename_branch == o->branch1 ? NULL : dest);
1228 /* Update dest->path both in index and in worktree */
1229 if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1230 return -1;
1231 return 0;
1234 static int handle_change_delete(struct merge_options *o,
1235 const char *path, const char *old_path,
1236 const struct object_id *o_oid, int o_mode,
1237 const struct object_id *changed_oid,
1238 int changed_mode,
1239 const char *change_branch,
1240 const char *delete_branch,
1241 const char *change, const char *change_past)
1243 char *alt_path = NULL;
1244 const char *update_path = path;
1245 int ret = 0;
1247 if (dir_in_way(path, !o->call_depth, 0) ||
1248 (!o->call_depth && would_lose_untracked(path))) {
1249 update_path = alt_path = unique_path(o, path, change_branch);
1252 if (o->call_depth) {
1254 * We cannot arbitrarily accept either a_sha or b_sha as
1255 * correct; since there is no true "middle point" between
1256 * them, simply reuse the base version for virtual merge base.
1258 ret = remove_file_from_cache(path);
1259 if (!ret)
1260 ret = update_file(o, 0, o_oid, o_mode, update_path);
1261 } else {
1262 if (!alt_path) {
1263 if (!old_path) {
1264 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1265 "and %s in %s. Version %s of %s left in tree."),
1266 change, path, delete_branch, change_past,
1267 change_branch, change_branch, path);
1268 } else {
1269 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1270 "and %s to %s in %s. Version %s of %s left in tree."),
1271 change, old_path, delete_branch, change_past, path,
1272 change_branch, change_branch, path);
1274 } else {
1275 if (!old_path) {
1276 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1277 "and %s in %s. Version %s of %s left in tree at %s."),
1278 change, path, delete_branch, change_past,
1279 change_branch, change_branch, path, alt_path);
1280 } else {
1281 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1282 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1283 change, old_path, delete_branch, change_past, path,
1284 change_branch, change_branch, path, alt_path);
1288 * No need to call update_file() on path when change_branch ==
1289 * o->branch1 && !alt_path, since that would needlessly touch
1290 * path. We could call update_file_flags() with update_cache=0
1291 * and update_wd=0, but that's a no-op.
1293 if (change_branch != o->branch1 || alt_path)
1294 ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1296 free(alt_path);
1298 return ret;
1301 static int conflict_rename_delete(struct merge_options *o,
1302 struct diff_filepair *pair,
1303 const char *rename_branch,
1304 const char *delete_branch)
1306 const struct diff_filespec *orig = pair->one;
1307 const struct diff_filespec *dest = pair->two;
1309 if (handle_change_delete(o,
1310 o->call_depth ? orig->path : dest->path,
1311 o->call_depth ? NULL : orig->path,
1312 &orig->oid, orig->mode,
1313 &dest->oid, dest->mode,
1314 rename_branch, delete_branch,
1315 _("rename"), _("renamed")))
1316 return -1;
1318 if (o->call_depth)
1319 return remove_file_from_cache(dest->path);
1320 else
1321 return update_stages(o, dest->path, NULL,
1322 rename_branch == o->branch1 ? dest : NULL,
1323 rename_branch == o->branch1 ? NULL : dest);
1326 static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1327 struct stage_data *entry,
1328 int stage)
1330 struct object_id *oid = &entry->stages[stage].oid;
1331 unsigned mode = entry->stages[stage].mode;
1332 if (mode == 0 || is_null_oid(oid))
1333 return NULL;
1334 oidcpy(&target->oid, oid);
1335 target->mode = mode;
1336 return target;
1339 static int handle_file(struct merge_options *o,
1340 struct diff_filespec *rename,
1341 int stage,
1342 struct rename_conflict_info *ci)
1344 char *dst_name = rename->path;
1345 struct stage_data *dst_entry;
1346 const char *cur_branch, *other_branch;
1347 struct diff_filespec other;
1348 struct diff_filespec *add;
1349 int ret;
1351 if (stage == 2) {
1352 dst_entry = ci->dst_entry1;
1353 cur_branch = ci->branch1;
1354 other_branch = ci->branch2;
1355 } else {
1356 dst_entry = ci->dst_entry2;
1357 cur_branch = ci->branch2;
1358 other_branch = ci->branch1;
1361 add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1362 if (add) {
1363 int ren_src_was_dirty = was_dirty(o, rename->path);
1364 char *add_name = unique_path(o, rename->path, other_branch);
1365 if (update_file(o, 0, &add->oid, add->mode, add_name))
1366 return -1;
1368 if (ren_src_was_dirty) {
1369 output(o, 1, _("Refusing to lose dirty file at %s"),
1370 rename->path);
1373 * Because the double negatives somehow keep confusing me...
1374 * 1) update_wd iff !ren_src_was_dirty.
1375 * 2) no_wd iff !update_wd
1376 * 3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty
1378 remove_file(o, 0, rename->path, ren_src_was_dirty);
1379 dst_name = unique_path(o, rename->path, cur_branch);
1380 } else {
1381 if (dir_in_way(rename->path, !o->call_depth, 0)) {
1382 dst_name = unique_path(o, rename->path, cur_branch);
1383 output(o, 1, _("%s is a directory in %s adding as %s instead"),
1384 rename->path, other_branch, dst_name);
1385 } else if (!o->call_depth &&
1386 would_lose_untracked(rename->path)) {
1387 dst_name = unique_path(o, rename->path, cur_branch);
1388 output(o, 1, _("Refusing to lose untracked file at %s; "
1389 "adding as %s instead"),
1390 rename->path, dst_name);
1393 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1394 ; /* fall through, do allow dst_name to be released */
1395 else if (stage == 2)
1396 ret = update_stages(o, rename->path, NULL, rename, add);
1397 else
1398 ret = update_stages(o, rename->path, NULL, add, rename);
1400 if (dst_name != rename->path)
1401 free(dst_name);
1403 return ret;
1406 static int conflict_rename_rename_1to2(struct merge_options *o,
1407 struct rename_conflict_info *ci)
1409 /* One file was renamed in both branches, but to different names. */
1410 struct diff_filespec *one = ci->pair1->one;
1411 struct diff_filespec *a = ci->pair1->two;
1412 struct diff_filespec *b = ci->pair2->two;
1414 output(o, 1, _("CONFLICT (rename/rename): "
1415 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1416 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1417 one->path, a->path, ci->branch1,
1418 one->path, b->path, ci->branch2,
1419 o->call_depth ? _(" (left unresolved)") : "");
1420 if (o->call_depth) {
1421 struct merge_file_info mfi;
1422 struct diff_filespec other;
1423 struct diff_filespec *add;
1424 if (merge_file_one(o, one->path,
1425 &one->oid, one->mode,
1426 &a->oid, a->mode,
1427 &b->oid, b->mode,
1428 ci->branch1, ci->branch2, &mfi))
1429 return -1;
1432 * FIXME: For rename/add-source conflicts (if we could detect
1433 * such), this is wrong. We should instead find a unique
1434 * pathname and then either rename the add-source file to that
1435 * unique path, or use that unique path instead of src here.
1437 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1438 return -1;
1441 * Above, we put the merged content at the merge-base's
1442 * path. Now we usually need to delete both a->path and
1443 * b->path. However, the rename on each side of the merge
1444 * could also be involved in a rename/add conflict. In
1445 * such cases, we should keep the added file around,
1446 * resolving the conflict at that path in its favor.
1448 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1449 if (add) {
1450 if (update_file(o, 0, &add->oid, add->mode, a->path))
1451 return -1;
1453 else
1454 remove_file_from_cache(a->path);
1455 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1456 if (add) {
1457 if (update_file(o, 0, &add->oid, add->mode, b->path))
1458 return -1;
1460 else
1461 remove_file_from_cache(b->path);
1462 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1463 return -1;
1465 return 0;
1468 static int conflict_rename_rename_2to1(struct merge_options *o,
1469 struct rename_conflict_info *ci)
1471 /* Two files, a & b, were renamed to the same thing, c. */
1472 struct diff_filespec *a = ci->pair1->one;
1473 struct diff_filespec *b = ci->pair2->one;
1474 struct diff_filespec *c1 = ci->pair1->two;
1475 struct diff_filespec *c2 = ci->pair2->two;
1476 char *path = c1->path; /* == c2->path */
1477 struct merge_file_info mfi_c1;
1478 struct merge_file_info mfi_c2;
1479 int ret;
1481 output(o, 1, _("CONFLICT (rename/rename): "
1482 "Rename %s->%s in %s. "
1483 "Rename %s->%s in %s"),
1484 a->path, c1->path, ci->branch1,
1485 b->path, c2->path, ci->branch2);
1487 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1488 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1490 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1491 o->branch1, c1->path,
1492 o->branch2, ci->ren1_other.path, &mfi_c1) ||
1493 merge_file_special_markers(o, b, &ci->ren2_other, c2,
1494 o->branch1, ci->ren2_other.path,
1495 o->branch2, c2->path, &mfi_c2))
1496 return -1;
1498 if (o->call_depth) {
1500 * If mfi_c1.clean && mfi_c2.clean, then it might make
1501 * sense to do a two-way merge of those results. But, I
1502 * think in all cases, it makes sense to have the virtual
1503 * merge base just undo the renames; they can be detected
1504 * again later for the non-recursive merge.
1506 remove_file(o, 0, path, 0);
1507 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1508 if (!ret)
1509 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1510 b->path);
1511 } else {
1512 char *new_path1 = unique_path(o, path, ci->branch1);
1513 char *new_path2 = unique_path(o, path, ci->branch2);
1514 output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1515 a->path, new_path1, b->path, new_path2);
1516 if (was_dirty(o, path))
1517 output(o, 1, _("Refusing to lose dirty file at %s"),
1518 path);
1519 else if (would_lose_untracked(path))
1521 * Only way we get here is if both renames were from
1522 * a directory rename AND user had an untracked file
1523 * at the location where both files end up after the
1524 * two directory renames. See testcase 10d of t6043.
1526 output(o, 1, _("Refusing to lose untracked file at "
1527 "%s, even though it's in the way."),
1528 path);
1529 else
1530 remove_file(o, 0, path, 0);
1531 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1532 if (!ret)
1533 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1534 new_path2);
1536 * unpack_trees() actually populates the index for us for
1537 * "normal" rename/rename(2to1) situtations so that the
1538 * correct entries are at the higher stages, which would
1539 * make the call below to update_stages_for_stage_data
1540 * unnecessary. However, if either of the renames came
1541 * from a directory rename, then unpack_trees() will not
1542 * have gotten the right data loaded into the index, so we
1543 * need to do so now. (While it'd be tempting to move this
1544 * call to update_stages_for_stage_data() to
1545 * apply_directory_rename_modifications(), that would break
1546 * our intermediate calls to would_lose_untracked() since
1547 * those rely on the current in-memory index. See also the
1548 * big "NOTE" in update_stages()).
1550 if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1551 ret = -1;
1553 free(new_path2);
1554 free(new_path1);
1557 return ret;
1561 * Get the diff_filepairs changed between o_tree and tree.
1563 static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1564 struct tree *o_tree,
1565 struct tree *tree)
1567 struct diff_queue_struct *ret;
1568 struct diff_options opts;
1570 diff_setup(&opts);
1571 opts.flags.recursive = 1;
1572 opts.flags.rename_empty = 0;
1573 opts.detect_rename = DIFF_DETECT_RENAME;
1574 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1575 o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1576 1000;
1577 opts.rename_score = o->rename_score;
1578 opts.show_rename_progress = o->show_rename_progress;
1579 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1580 diff_setup_done(&opts);
1581 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1582 diffcore_std(&opts);
1583 if (opts.needed_rename_limit > o->needed_rename_limit)
1584 o->needed_rename_limit = opts.needed_rename_limit;
1586 ret = xmalloc(sizeof(*ret));
1587 *ret = diff_queued_diff;
1589 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1590 diff_queued_diff.nr = 0;
1591 diff_queued_diff.queue = NULL;
1592 diff_flush(&opts);
1593 return ret;
1596 static int tree_has_path(struct tree *tree, const char *path)
1598 struct object_id hashy;
1599 unsigned int mode_o;
1601 return !get_tree_entry(&tree->object.oid, path,
1602 &hashy, &mode_o);
1606 * Return a new string that replaces the beginning portion (which matches
1607 * entry->dir), with entry->new_dir. In perl-speak:
1608 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1609 * NOTE:
1610 * Caller must ensure that old_path starts with entry->dir + '/'.
1612 static char *apply_dir_rename(struct dir_rename_entry *entry,
1613 const char *old_path)
1615 struct strbuf new_path = STRBUF_INIT;
1616 int oldlen, newlen;
1618 if (entry->non_unique_new_dir)
1619 return NULL;
1621 oldlen = strlen(entry->dir);
1622 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1623 strbuf_grow(&new_path, newlen);
1624 strbuf_addbuf(&new_path, &entry->new_dir);
1625 strbuf_addstr(&new_path, &old_path[oldlen]);
1627 return strbuf_detach(&new_path, NULL);
1630 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1631 char **old_dir, char **new_dir)
1633 char *end_of_old, *end_of_new;
1634 int old_len, new_len;
1636 *old_dir = NULL;
1637 *new_dir = NULL;
1640 * For
1641 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1642 * the "e/foo.c" part is the same, we just want to know that
1643 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1644 * so, for this example, this function returns "a/b/c/d" in
1645 * *old_dir and "a/b/some/thing/else" in *new_dir.
1647 * Also, if the basename of the file changed, we don't care. We
1648 * want to know which portion of the directory, if any, changed.
1650 end_of_old = strrchr(old_path, '/');
1651 end_of_new = strrchr(new_path, '/');
1653 if (end_of_old == NULL || end_of_new == NULL)
1654 return;
1655 while (*--end_of_new == *--end_of_old &&
1656 end_of_old != old_path &&
1657 end_of_new != new_path)
1658 ; /* Do nothing; all in the while loop */
1660 * We've found the first non-matching character in the directory
1661 * paths. That means the current directory we were comparing
1662 * represents the rename. Move end_of_old and end_of_new back
1663 * to the full directory name.
1665 if (*end_of_old == '/')
1666 end_of_old++;
1667 if (*end_of_old != '/')
1668 end_of_new++;
1669 end_of_old = strchr(end_of_old, '/');
1670 end_of_new = strchr(end_of_new, '/');
1673 * It may have been the case that old_path and new_path were the same
1674 * directory all along. Don't claim a rename if they're the same.
1676 old_len = end_of_old - old_path;
1677 new_len = end_of_new - new_path;
1679 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1680 *old_dir = xstrndup(old_path, old_len);
1681 *new_dir = xstrndup(new_path, new_len);
1685 static void remove_hashmap_entries(struct hashmap *dir_renames,
1686 struct string_list *items_to_remove)
1688 int i;
1689 struct dir_rename_entry *entry;
1691 for (i = 0; i < items_to_remove->nr; i++) {
1692 entry = items_to_remove->items[i].util;
1693 hashmap_remove(dir_renames, entry, NULL);
1695 string_list_clear(items_to_remove, 0);
1699 * See if there is a directory rename for path, and if there are any file
1700 * level conflicts for the renamed location. If there is a rename and
1701 * there are no conflicts, return the new name. Otherwise, return NULL.
1703 static char *handle_path_level_conflicts(struct merge_options *o,
1704 const char *path,
1705 struct dir_rename_entry *entry,
1706 struct hashmap *collisions,
1707 struct tree *tree)
1709 char *new_path = NULL;
1710 struct collision_entry *collision_ent;
1711 int clean = 1;
1712 struct strbuf collision_paths = STRBUF_INIT;
1715 * entry has the mapping of old directory name to new directory name
1716 * that we want to apply to path.
1718 new_path = apply_dir_rename(entry, path);
1720 if (!new_path) {
1721 /* This should only happen when entry->non_unique_new_dir set */
1722 if (!entry->non_unique_new_dir)
1723 BUG("entry->non_unqiue_dir not set and !new_path");
1724 output(o, 1, _("CONFLICT (directory rename split): "
1725 "Unclear where to place %s because directory "
1726 "%s was renamed to multiple other directories, "
1727 "with no destination getting a majority of the "
1728 "files."),
1729 path, entry->dir);
1730 clean = 0;
1731 return NULL;
1735 * The caller needs to have ensured that it has pre-populated
1736 * collisions with all paths that map to new_path. Do a quick check
1737 * to ensure that's the case.
1739 collision_ent = collision_find_entry(collisions, new_path);
1740 if (collision_ent == NULL)
1741 BUG("collision_ent is NULL");
1744 * Check for one-sided add/add/.../add conflicts, i.e.
1745 * where implicit renames from the other side doing
1746 * directory rename(s) can affect this side of history
1747 * to put multiple paths into the same location. Warn
1748 * and bail on directory renames for such paths.
1750 if (collision_ent->reported_already) {
1751 clean = 0;
1752 } else if (tree_has_path(tree, new_path)) {
1753 collision_ent->reported_already = 1;
1754 strbuf_add_separated_string_list(&collision_paths, ", ",
1755 &collision_ent->source_files);
1756 output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1757 "file/dir at %s in the way of implicit "
1758 "directory rename(s) putting the following "
1759 "path(s) there: %s."),
1760 new_path, collision_paths.buf);
1761 clean = 0;
1762 } else if (collision_ent->source_files.nr > 1) {
1763 collision_ent->reported_already = 1;
1764 strbuf_add_separated_string_list(&collision_paths, ", ",
1765 &collision_ent->source_files);
1766 output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1767 "more than one path to %s; implicit directory "
1768 "renames tried to put these paths there: %s"),
1769 new_path, collision_paths.buf);
1770 clean = 0;
1773 /* Free memory we no longer need */
1774 strbuf_release(&collision_paths);
1775 if (!clean && new_path) {
1776 free(new_path);
1777 return NULL;
1780 return new_path;
1784 * There are a couple things we want to do at the directory level:
1785 * 1. Check for both sides renaming to the same thing, in order to avoid
1786 * implicit renaming of files that should be left in place. (See
1787 * testcase 6b in t6043 for details.)
1788 * 2. Prune directory renames if there are still files left in the
1789 * the original directory. These represent a partial directory rename,
1790 * i.e. a rename where only some of the files within the directory
1791 * were renamed elsewhere. (Technically, this could be done earlier
1792 * in get_directory_renames(), except that would prevent us from
1793 * doing the previous check and thus failing testcase 6b.)
1794 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
1795 * In the future, we could potentially record this info as well and
1796 * omit reporting rename/rename(1to2) conflicts for each path within
1797 * the affected directories, thus cleaning up the merge output.
1798 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1799 * directory level, because merging directories is fine. If it
1800 * causes conflicts for files within those merged directories, then
1801 * that should be detected at the individual path level.
1803 static void handle_directory_level_conflicts(struct merge_options *o,
1804 struct hashmap *dir_re_head,
1805 struct tree *head,
1806 struct hashmap *dir_re_merge,
1807 struct tree *merge)
1809 struct hashmap_iter iter;
1810 struct dir_rename_entry *head_ent;
1811 struct dir_rename_entry *merge_ent;
1813 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1814 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1816 hashmap_iter_init(dir_re_head, &iter);
1817 while ((head_ent = hashmap_iter_next(&iter))) {
1818 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1819 if (merge_ent &&
1820 !head_ent->non_unique_new_dir &&
1821 !merge_ent->non_unique_new_dir &&
1822 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1823 /* 1. Renamed identically; remove it from both sides */
1824 string_list_append(&remove_from_head,
1825 head_ent->dir)->util = head_ent;
1826 strbuf_release(&head_ent->new_dir);
1827 string_list_append(&remove_from_merge,
1828 merge_ent->dir)->util = merge_ent;
1829 strbuf_release(&merge_ent->new_dir);
1830 } else if (tree_has_path(head, head_ent->dir)) {
1831 /* 2. This wasn't a directory rename after all */
1832 string_list_append(&remove_from_head,
1833 head_ent->dir)->util = head_ent;
1834 strbuf_release(&head_ent->new_dir);
1838 remove_hashmap_entries(dir_re_head, &remove_from_head);
1839 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1841 hashmap_iter_init(dir_re_merge, &iter);
1842 while ((merge_ent = hashmap_iter_next(&iter))) {
1843 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1844 if (tree_has_path(merge, merge_ent->dir)) {
1845 /* 2. This wasn't a directory rename after all */
1846 string_list_append(&remove_from_merge,
1847 merge_ent->dir)->util = merge_ent;
1848 } else if (head_ent &&
1849 !head_ent->non_unique_new_dir &&
1850 !merge_ent->non_unique_new_dir) {
1851 /* 3. rename/rename(1to2) */
1853 * We can assume it's not rename/rename(1to1) because
1854 * that was case (1), already checked above. So we
1855 * know that head_ent->new_dir and merge_ent->new_dir
1856 * are different strings.
1858 output(o, 1, _("CONFLICT (rename/rename): "
1859 "Rename directory %s->%s in %s. "
1860 "Rename directory %s->%s in %s"),
1861 head_ent->dir, head_ent->new_dir.buf, o->branch1,
1862 head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1863 string_list_append(&remove_from_head,
1864 head_ent->dir)->util = head_ent;
1865 strbuf_release(&head_ent->new_dir);
1866 string_list_append(&remove_from_merge,
1867 merge_ent->dir)->util = merge_ent;
1868 strbuf_release(&merge_ent->new_dir);
1872 remove_hashmap_entries(dir_re_head, &remove_from_head);
1873 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1876 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1877 struct tree *tree)
1879 struct hashmap *dir_renames;
1880 struct hashmap_iter iter;
1881 struct dir_rename_entry *entry;
1882 int i;
1885 * Typically, we think of a directory rename as all files from a
1886 * certain directory being moved to a target directory. However,
1887 * what if someone first moved two files from the original
1888 * directory in one commit, and then renamed the directory
1889 * somewhere else in a later commit? At merge time, we just know
1890 * that files from the original directory went to two different
1891 * places, and that the bulk of them ended up in the same place.
1892 * We want each directory rename to represent where the bulk of the
1893 * files from that directory end up; this function exists to find
1894 * where the bulk of the files went.
1896 * The first loop below simply iterates through the list of file
1897 * renames, finding out how often each directory rename pair
1898 * possibility occurs.
1900 dir_renames = xmalloc(sizeof(*dir_renames));
1901 dir_rename_init(dir_renames);
1902 for (i = 0; i < pairs->nr; ++i) {
1903 struct string_list_item *item;
1904 int *count;
1905 struct diff_filepair *pair = pairs->queue[i];
1906 char *old_dir, *new_dir;
1908 /* File not part of directory rename if it wasn't renamed */
1909 if (pair->status != 'R')
1910 continue;
1912 get_renamed_dir_portion(pair->one->path, pair->two->path,
1913 &old_dir, &new_dir);
1914 if (!old_dir)
1915 /* Directory didn't change at all; ignore this one. */
1916 continue;
1918 entry = dir_rename_find_entry(dir_renames, old_dir);
1919 if (!entry) {
1920 entry = xmalloc(sizeof(*entry));
1921 dir_rename_entry_init(entry, old_dir);
1922 hashmap_put(dir_renames, entry);
1923 } else {
1924 free(old_dir);
1926 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1927 if (!item) {
1928 item = string_list_insert(&entry->possible_new_dirs,
1929 new_dir);
1930 item->util = xcalloc(1, sizeof(int));
1931 } else {
1932 free(new_dir);
1934 count = item->util;
1935 *count += 1;
1939 * For each directory with files moved out of it, we find out which
1940 * target directory received the most files so we can declare it to
1941 * be the "winning" target location for the directory rename. This
1942 * winner gets recorded in new_dir. If there is no winner
1943 * (multiple target directories received the same number of files),
1944 * we set non_unique_new_dir. Once we've determined the winner (or
1945 * that there is no winner), we no longer need possible_new_dirs.
1947 hashmap_iter_init(dir_renames, &iter);
1948 while ((entry = hashmap_iter_next(&iter))) {
1949 int max = 0;
1950 int bad_max = 0;
1951 char *best = NULL;
1953 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1954 int *count = entry->possible_new_dirs.items[i].util;
1956 if (*count == max)
1957 bad_max = max;
1958 else if (*count > max) {
1959 max = *count;
1960 best = entry->possible_new_dirs.items[i].string;
1963 if (bad_max == max)
1964 entry->non_unique_new_dir = 1;
1965 else {
1966 assert(entry->new_dir.len == 0);
1967 strbuf_addstr(&entry->new_dir, best);
1970 * The relevant directory sub-portion of the original full
1971 * filepaths were xstrndup'ed before inserting into
1972 * possible_new_dirs, and instead of manually iterating the
1973 * list and free'ing each, just lie and tell
1974 * possible_new_dirs that it did the strdup'ing so that it
1975 * will free them for us.
1977 entry->possible_new_dirs.strdup_strings = 1;
1978 string_list_clear(&entry->possible_new_dirs, 1);
1981 return dir_renames;
1984 static struct dir_rename_entry *check_dir_renamed(const char *path,
1985 struct hashmap *dir_renames)
1987 char temp[PATH_MAX];
1988 char *end;
1989 struct dir_rename_entry *entry;
1991 strcpy(temp, path);
1992 while ((end = strrchr(temp, '/'))) {
1993 *end = '\0';
1994 entry = dir_rename_find_entry(dir_renames, temp);
1995 if (entry)
1996 return entry;
1998 return NULL;
2001 static void compute_collisions(struct hashmap *collisions,
2002 struct hashmap *dir_renames,
2003 struct diff_queue_struct *pairs)
2005 int i;
2008 * Multiple files can be mapped to the same path due to directory
2009 * renames done by the other side of history. Since that other
2010 * side of history could have merged multiple directories into one,
2011 * if our side of history added the same file basename to each of
2012 * those directories, then all N of them would get implicitly
2013 * renamed by the directory rename detection into the same path,
2014 * and we'd get an add/add/.../add conflict, and all those adds
2015 * from *this* side of history. This is not representable in the
2016 * index, and users aren't going to easily be able to make sense of
2017 * it. So we need to provide a good warning about what's
2018 * happening, and fall back to no-directory-rename detection
2019 * behavior for those paths.
2021 * See testcases 9e and all of section 5 from t6043 for examples.
2023 collision_init(collisions);
2025 for (i = 0; i < pairs->nr; ++i) {
2026 struct dir_rename_entry *dir_rename_ent;
2027 struct collision_entry *collision_ent;
2028 char *new_path;
2029 struct diff_filepair *pair = pairs->queue[i];
2031 if (pair->status != 'A' && pair->status != 'R')
2032 continue;
2033 dir_rename_ent = check_dir_renamed(pair->two->path,
2034 dir_renames);
2035 if (!dir_rename_ent)
2036 continue;
2038 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2039 if (!new_path)
2041 * dir_rename_ent->non_unique_new_path is true, which
2042 * means there is no directory rename for us to use,
2043 * which means it won't cause us any additional
2044 * collisions.
2046 continue;
2047 collision_ent = collision_find_entry(collisions, new_path);
2048 if (!collision_ent) {
2049 collision_ent = xcalloc(1,
2050 sizeof(struct collision_entry));
2051 hashmap_entry_init(collision_ent, strhash(new_path));
2052 hashmap_put(collisions, collision_ent);
2053 collision_ent->target_file = new_path;
2054 } else {
2055 free(new_path);
2057 string_list_insert(&collision_ent->source_files,
2058 pair->two->path);
2062 static char *check_for_directory_rename(struct merge_options *o,
2063 const char *path,
2064 struct tree *tree,
2065 struct hashmap *dir_renames,
2066 struct hashmap *dir_rename_exclusions,
2067 struct hashmap *collisions,
2068 int *clean_merge)
2070 char *new_path = NULL;
2071 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2072 struct dir_rename_entry *oentry = NULL;
2074 if (!entry)
2075 return new_path;
2078 * This next part is a little weird. We do not want to do an
2079 * implicit rename into a directory we renamed on our side, because
2080 * that will result in a spurious rename/rename(1to2) conflict. An
2081 * example:
2082 * Base commit: dumbdir/afile, otherdir/bfile
2083 * Side 1: smrtdir/afile, otherdir/bfile
2084 * Side 2: dumbdir/afile, dumbdir/bfile
2085 * Here, while working on Side 1, we could notice that otherdir was
2086 * renamed/merged to dumbdir, and change the diff_filepair for
2087 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2088 * 2 will notice the rename from dumbdir to smrtdir, and do the
2089 * transitive rename to move it from dumbdir/bfile to
2090 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2091 * smrtdir, a rename/rename(1to2) conflict. We really just want
2092 * the file to end up in smrtdir. And the way to achieve that is
2093 * to not let Side1 do the rename to dumbdir, since we know that is
2094 * the source of one of our directory renames.
2096 * That's why oentry and dir_rename_exclusions is here.
2098 * As it turns out, this also prevents N-way transient rename
2099 * confusion; See testcases 9c and 9d of t6043.
2101 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2102 if (oentry) {
2103 output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2104 "to %s, because %s itself was renamed."),
2105 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2106 } else {
2107 new_path = handle_path_level_conflicts(o, path, entry,
2108 collisions, tree);
2109 *clean_merge &= (new_path != NULL);
2112 return new_path;
2115 static void apply_directory_rename_modifications(struct merge_options *o,
2116 struct diff_filepair *pair,
2117 char *new_path,
2118 struct rename *re,
2119 struct tree *tree,
2120 struct tree *o_tree,
2121 struct tree *a_tree,
2122 struct tree *b_tree,
2123 struct string_list *entries,
2124 int *clean)
2126 struct string_list_item *item;
2127 int stage = (tree == a_tree ? 2 : 3);
2128 int update_wd;
2131 * In all cases where we can do directory rename detection,
2132 * unpack_trees() will have read pair->two->path into the
2133 * index and the working copy. We need to remove it so that
2134 * we can instead place it at new_path. It is guaranteed to
2135 * not be untracked (unpack_trees() would have errored out
2136 * saying the file would have been overwritten), but it might
2137 * be dirty, though.
2139 update_wd = !was_dirty(o, pair->two->path);
2140 if (!update_wd)
2141 output(o, 1, _("Refusing to lose dirty file at %s"),
2142 pair->two->path);
2143 remove_file(o, 1, pair->two->path, !update_wd);
2145 /* Find or create a new re->dst_entry */
2146 item = string_list_lookup(entries, new_path);
2147 if (item) {
2149 * Since we're renaming on this side of history, and it's
2150 * due to a directory rename on the other side of history
2151 * (which we only allow when the directory in question no
2152 * longer exists on the other side of history), the
2153 * original entry for re->dst_entry is no longer
2154 * necessary...
2156 re->dst_entry->processed = 1;
2159 * ...because we'll be using this new one.
2161 re->dst_entry = item->util;
2162 } else {
2164 * re->dst_entry is for the before-dir-rename path, and we
2165 * need it to hold information for the after-dir-rename
2166 * path. Before creating a new entry, we need to mark the
2167 * old one as unnecessary (...unless it is shared by
2168 * src_entry, i.e. this didn't use to be a rename, in which
2169 * case we can just allow the normal processing to happen
2170 * for it).
2172 if (pair->status == 'R')
2173 re->dst_entry->processed = 1;
2175 re->dst_entry = insert_stage_data(new_path,
2176 o_tree, a_tree, b_tree,
2177 entries);
2178 item = string_list_insert(entries, new_path);
2179 item->util = re->dst_entry;
2183 * Update the stage_data with the information about the path we are
2184 * moving into place. That slot will be empty and available for us
2185 * to write to because of the collision checks in
2186 * handle_path_level_conflicts(). In other words,
2187 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2188 * open for us to write to.
2190 * It may be tempting to actually update the index at this point as
2191 * well, using update_stages_for_stage_data(), but as per the big
2192 * "NOTE" in update_stages(), doing so will modify the current
2193 * in-memory index which will break calls to would_lose_untracked()
2194 * that we need to make. Instead, we need to just make sure that
2195 * the various conflict_rename_*() functions update the index
2196 * explicitly rather than relying on unpack_trees() to have done it.
2198 get_tree_entry(&tree->object.oid,
2199 pair->two->path,
2200 &re->dst_entry->stages[stage].oid,
2201 &re->dst_entry->stages[stage].mode);
2203 /* Update pair status */
2204 if (pair->status == 'A') {
2206 * Recording rename information for this add makes it look
2207 * like a rename/delete conflict. Make sure we can
2208 * correctly handle this as an add that was moved to a new
2209 * directory instead of reporting a rename/delete conflict.
2211 re->add_turned_into_rename = 1;
2214 * We don't actually look at pair->status again, but it seems
2215 * pedagogically correct to adjust it.
2217 pair->status = 'R';
2220 * Finally, record the new location.
2222 pair->two->path = new_path;
2226 * Get information of all renames which occurred in 'pairs', making use of
2227 * any implicit directory renames inferred from the other side of history.
2228 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2229 * to be able to associate the correct cache entries with the rename
2230 * information; tree is always equal to either a_tree or b_tree.
2232 static struct string_list *get_renames(struct merge_options *o,
2233 struct diff_queue_struct *pairs,
2234 struct hashmap *dir_renames,
2235 struct hashmap *dir_rename_exclusions,
2236 struct tree *tree,
2237 struct tree *o_tree,
2238 struct tree *a_tree,
2239 struct tree *b_tree,
2240 struct string_list *entries,
2241 int *clean_merge)
2243 int i;
2244 struct hashmap collisions;
2245 struct hashmap_iter iter;
2246 struct collision_entry *e;
2247 struct string_list *renames;
2249 compute_collisions(&collisions, dir_renames, pairs);
2250 renames = xcalloc(1, sizeof(struct string_list));
2252 for (i = 0; i < pairs->nr; ++i) {
2253 struct string_list_item *item;
2254 struct rename *re;
2255 struct diff_filepair *pair = pairs->queue[i];
2256 char *new_path; /* non-NULL only with directory renames */
2258 if (pair->status != 'A' && pair->status != 'R') {
2259 diff_free_filepair(pair);
2260 continue;
2262 new_path = check_for_directory_rename(o, pair->two->path, tree,
2263 dir_renames,
2264 dir_rename_exclusions,
2265 &collisions,
2266 clean_merge);
2267 if (pair->status != 'R' && !new_path) {
2268 diff_free_filepair(pair);
2269 continue;
2272 re = xmalloc(sizeof(*re));
2273 re->processed = 0;
2274 re->add_turned_into_rename = 0;
2275 re->pair = pair;
2276 item = string_list_lookup(entries, re->pair->one->path);
2277 if (!item)
2278 re->src_entry = insert_stage_data(re->pair->one->path,
2279 o_tree, a_tree, b_tree, entries);
2280 else
2281 re->src_entry = item->util;
2283 item = string_list_lookup(entries, re->pair->two->path);
2284 if (!item)
2285 re->dst_entry = insert_stage_data(re->pair->two->path,
2286 o_tree, a_tree, b_tree, entries);
2287 else
2288 re->dst_entry = item->util;
2289 item = string_list_insert(renames, pair->one->path);
2290 item->util = re;
2291 if (new_path)
2292 apply_directory_rename_modifications(o, pair, new_path,
2293 re, tree, o_tree,
2294 a_tree, b_tree,
2295 entries,
2296 clean_merge);
2299 hashmap_iter_init(&collisions, &iter);
2300 while ((e = hashmap_iter_next(&iter))) {
2301 free(e->target_file);
2302 string_list_clear(&e->source_files, 0);
2304 hashmap_free(&collisions, 1);
2305 return renames;
2308 static int process_renames(struct merge_options *o,
2309 struct string_list *a_renames,
2310 struct string_list *b_renames)
2312 int clean_merge = 1, i, j;
2313 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2314 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2315 const struct rename *sre;
2317 for (i = 0; i < a_renames->nr; i++) {
2318 sre = a_renames->items[i].util;
2319 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2320 = (void *)sre;
2322 for (i = 0; i < b_renames->nr; i++) {
2323 sre = b_renames->items[i].util;
2324 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2325 = (void *)sre;
2328 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2329 struct string_list *renames1, *renames2Dst;
2330 struct rename *ren1 = NULL, *ren2 = NULL;
2331 const char *branch1, *branch2;
2332 const char *ren1_src, *ren1_dst;
2333 struct string_list_item *lookup;
2335 if (i >= a_renames->nr) {
2336 ren2 = b_renames->items[j++].util;
2337 } else if (j >= b_renames->nr) {
2338 ren1 = a_renames->items[i++].util;
2339 } else {
2340 int compare = strcmp(a_renames->items[i].string,
2341 b_renames->items[j].string);
2342 if (compare <= 0)
2343 ren1 = a_renames->items[i++].util;
2344 if (compare >= 0)
2345 ren2 = b_renames->items[j++].util;
2348 /* TODO: refactor, so that 1/2 are not needed */
2349 if (ren1) {
2350 renames1 = a_renames;
2351 renames2Dst = &b_by_dst;
2352 branch1 = o->branch1;
2353 branch2 = o->branch2;
2354 } else {
2355 renames1 = b_renames;
2356 renames2Dst = &a_by_dst;
2357 branch1 = o->branch2;
2358 branch2 = o->branch1;
2359 SWAP(ren2, ren1);
2362 if (ren1->processed)
2363 continue;
2364 ren1->processed = 1;
2365 ren1->dst_entry->processed = 1;
2366 /* BUG: We should only mark src_entry as processed if we
2367 * are not dealing with a rename + add-source case.
2369 ren1->src_entry->processed = 1;
2371 ren1_src = ren1->pair->one->path;
2372 ren1_dst = ren1->pair->two->path;
2374 if (ren2) {
2375 /* One file renamed on both sides */
2376 const char *ren2_src = ren2->pair->one->path;
2377 const char *ren2_dst = ren2->pair->two->path;
2378 enum rename_type rename_type;
2379 if (strcmp(ren1_src, ren2_src) != 0)
2380 die("BUG: ren1_src != ren2_src");
2381 ren2->dst_entry->processed = 1;
2382 ren2->processed = 1;
2383 if (strcmp(ren1_dst, ren2_dst) != 0) {
2384 rename_type = RENAME_ONE_FILE_TO_TWO;
2385 clean_merge = 0;
2386 } else {
2387 rename_type = RENAME_ONE_FILE_TO_ONE;
2388 /* BUG: We should only remove ren1_src in
2389 * the base stage (think of rename +
2390 * add-source cases).
2392 remove_file(o, 1, ren1_src, 1);
2393 update_entry(ren1->dst_entry,
2394 ren1->pair->one,
2395 ren1->pair->two,
2396 ren2->pair->two);
2398 setup_rename_conflict_info(rename_type,
2399 ren1->pair,
2400 ren2->pair,
2401 branch1,
2402 branch2,
2403 ren1->dst_entry,
2404 ren2->dst_entry,
2406 NULL,
2407 NULL);
2408 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2409 /* Two different files renamed to the same thing */
2410 char *ren2_dst;
2411 ren2 = lookup->util;
2412 ren2_dst = ren2->pair->two->path;
2413 if (strcmp(ren1_dst, ren2_dst) != 0)
2414 die("BUG: ren1_dst != ren2_dst");
2416 clean_merge = 0;
2417 ren2->processed = 1;
2419 * BUG: We should only mark src_entry as processed
2420 * if we are not dealing with a rename + add-source
2421 * case.
2423 ren2->src_entry->processed = 1;
2425 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2426 ren1->pair,
2427 ren2->pair,
2428 branch1,
2429 branch2,
2430 ren1->dst_entry,
2431 ren2->dst_entry,
2433 ren1->src_entry,
2434 ren2->src_entry);
2436 } else {
2437 /* Renamed in 1, maybe changed in 2 */
2438 /* we only use sha1 and mode of these */
2439 struct diff_filespec src_other, dst_other;
2440 int try_merge;
2443 * unpack_trees loads entries from common-commit
2444 * into stage 1, from head-commit into stage 2, and
2445 * from merge-commit into stage 3. We keep track
2446 * of which side corresponds to the rename.
2448 int renamed_stage = a_renames == renames1 ? 2 : 3;
2449 int other_stage = a_renames == renames1 ? 3 : 2;
2451 /* BUG: We should only remove ren1_src in the base
2452 * stage and in other_stage (think of rename +
2453 * add-source case).
2455 remove_file(o, 1, ren1_src,
2456 renamed_stage == 2 || !was_tracked(o, ren1_src));
2458 oidcpy(&src_other.oid,
2459 &ren1->src_entry->stages[other_stage].oid);
2460 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2461 oidcpy(&dst_other.oid,
2462 &ren1->dst_entry->stages[other_stage].oid);
2463 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2464 try_merge = 0;
2466 if (oid_eq(&src_other.oid, &null_oid) &&
2467 ren1->add_turned_into_rename) {
2468 setup_rename_conflict_info(RENAME_DIR,
2469 ren1->pair,
2470 NULL,
2471 branch1,
2472 branch2,
2473 ren1->dst_entry,
2474 NULL,
2476 NULL,
2477 NULL);
2478 } else if (oid_eq(&src_other.oid, &null_oid)) {
2479 setup_rename_conflict_info(RENAME_DELETE,
2480 ren1->pair,
2481 NULL,
2482 branch1,
2483 branch2,
2484 ren1->dst_entry,
2485 NULL,
2487 NULL,
2488 NULL);
2489 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2490 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2492 * Added file on the other side identical to
2493 * the file being renamed: clean merge.
2494 * Also, there is no need to overwrite the
2495 * file already in the working copy, so call
2496 * update_file_flags() instead of
2497 * update_file().
2499 if (update_file_flags(o,
2500 &ren1->pair->two->oid,
2501 ren1->pair->two->mode,
2502 ren1_dst,
2503 1, /* update_cache */
2504 0 /* update_wd */))
2505 clean_merge = -1;
2506 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2507 clean_merge = 0;
2508 try_merge = 1;
2509 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2510 "%s added in %s"),
2511 ren1_src, ren1_dst, branch1,
2512 ren1_dst, branch2);
2513 if (o->call_depth) {
2514 struct merge_file_info mfi;
2515 if (merge_file_one(o, ren1_dst, &null_oid, 0,
2516 &ren1->pair->two->oid,
2517 ren1->pair->two->mode,
2518 &dst_other.oid,
2519 dst_other.mode,
2520 branch1, branch2, &mfi)) {
2521 clean_merge = -1;
2522 goto cleanup_and_return;
2524 output(o, 1, _("Adding merged %s"), ren1_dst);
2525 if (update_file(o, 0, &mfi.oid,
2526 mfi.mode, ren1_dst))
2527 clean_merge = -1;
2528 try_merge = 0;
2529 } else {
2530 char *new_path = unique_path(o, ren1_dst, branch2);
2531 output(o, 1, _("Adding as %s instead"), new_path);
2532 if (update_file(o, 0, &dst_other.oid,
2533 dst_other.mode, new_path))
2534 clean_merge = -1;
2535 free(new_path);
2537 } else
2538 try_merge = 1;
2540 if (clean_merge < 0)
2541 goto cleanup_and_return;
2542 if (try_merge) {
2543 struct diff_filespec *one, *a, *b;
2544 src_other.path = (char *)ren1_src;
2546 one = ren1->pair->one;
2547 if (a_renames == renames1) {
2548 a = ren1->pair->two;
2549 b = &src_other;
2550 } else {
2551 b = ren1->pair->two;
2552 a = &src_other;
2554 update_entry(ren1->dst_entry, one, a, b);
2555 setup_rename_conflict_info(RENAME_NORMAL,
2556 ren1->pair,
2557 NULL,
2558 branch1,
2559 NULL,
2560 ren1->dst_entry,
2561 NULL,
2563 NULL,
2564 NULL);
2568 cleanup_and_return:
2569 string_list_clear(&a_by_dst, 0);
2570 string_list_clear(&b_by_dst, 0);
2572 return clean_merge;
2575 struct rename_info {
2576 struct string_list *head_renames;
2577 struct string_list *merge_renames;
2580 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2581 struct hashmap *dir_renames)
2583 struct hashmap_iter iter;
2584 struct dir_rename_entry *e;
2586 hashmap_iter_init(dir_renames, &iter);
2587 while ((e = hashmap_iter_next(&iter))) {
2588 free(e->dir);
2589 strbuf_release(&e->new_dir);
2590 /* possible_new_dirs already cleared in get_directory_renames */
2592 hashmap_free(dir_renames, 1);
2593 free(dir_renames);
2595 free(pairs->queue);
2596 free(pairs);
2599 static int handle_renames(struct merge_options *o,
2600 struct tree *common,
2601 struct tree *head,
2602 struct tree *merge,
2603 struct string_list *entries,
2604 struct rename_info *ri)
2606 struct diff_queue_struct *head_pairs, *merge_pairs;
2607 struct hashmap *dir_re_head, *dir_re_merge;
2608 int clean = 1;
2610 ri->head_renames = NULL;
2611 ri->merge_renames = NULL;
2613 if (!o->detect_rename)
2614 return 1;
2616 head_pairs = get_diffpairs(o, common, head);
2617 merge_pairs = get_diffpairs(o, common, merge);
2619 dir_re_head = get_directory_renames(head_pairs, head);
2620 dir_re_merge = get_directory_renames(merge_pairs, merge);
2622 handle_directory_level_conflicts(o,
2623 dir_re_head, head,
2624 dir_re_merge, merge);
2626 ri->head_renames = get_renames(o, head_pairs,
2627 dir_re_merge, dir_re_head, head,
2628 common, head, merge, entries,
2629 &clean);
2630 if (clean < 0)
2631 goto cleanup;
2632 ri->merge_renames = get_renames(o, merge_pairs,
2633 dir_re_head, dir_re_merge, merge,
2634 common, head, merge, entries,
2635 &clean);
2636 if (clean < 0)
2637 goto cleanup;
2638 clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2640 cleanup:
2642 * Some cleanup is deferred until cleanup_renames() because the
2643 * data structures are still needed and referenced in
2644 * process_entry(). But there are a few things we can free now.
2646 initial_cleanup_rename(head_pairs, dir_re_head);
2647 initial_cleanup_rename(merge_pairs, dir_re_merge);
2649 return clean;
2652 static void final_cleanup_rename(struct string_list *rename)
2654 const struct rename *re;
2655 int i;
2657 if (rename == NULL)
2658 return;
2660 for (i = 0; i < rename->nr; i++) {
2661 re = rename->items[i].util;
2662 diff_free_filepair(re->pair);
2664 string_list_clear(rename, 1);
2665 free(rename);
2668 static void final_cleanup_renames(struct rename_info *re_info)
2670 final_cleanup_rename(re_info->head_renames);
2671 final_cleanup_rename(re_info->merge_renames);
2674 static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2676 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2679 static int read_oid_strbuf(struct merge_options *o,
2680 const struct object_id *oid, struct strbuf *dst)
2682 void *buf;
2683 enum object_type type;
2684 unsigned long size;
2685 buf = read_object_file(oid, &type, &size);
2686 if (!buf)
2687 return err(o, _("cannot read object %s"), oid_to_hex(oid));
2688 if (type != OBJ_BLOB) {
2689 free(buf);
2690 return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2692 strbuf_attach(dst, buf, size, size + 1);
2693 return 0;
2696 static int blob_unchanged(struct merge_options *opt,
2697 const struct object_id *o_oid,
2698 unsigned o_mode,
2699 const struct object_id *a_oid,
2700 unsigned a_mode,
2701 int renormalize, const char *path)
2703 struct strbuf o = STRBUF_INIT;
2704 struct strbuf a = STRBUF_INIT;
2705 int ret = 0; /* assume changed for safety */
2707 if (a_mode != o_mode)
2708 return 0;
2709 if (oid_eq(o_oid, a_oid))
2710 return 1;
2711 if (!renormalize)
2712 return 0;
2714 assert(o_oid && a_oid);
2715 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2716 goto error_return;
2718 * Note: binary | is used so that both renormalizations are
2719 * performed. Comparison can be skipped if both files are
2720 * unchanged since their sha1s have already been compared.
2722 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2723 renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2724 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2726 error_return:
2727 strbuf_release(&o);
2728 strbuf_release(&a);
2729 return ret;
2732 static int handle_modify_delete(struct merge_options *o,
2733 const char *path,
2734 struct object_id *o_oid, int o_mode,
2735 struct object_id *a_oid, int a_mode,
2736 struct object_id *b_oid, int b_mode)
2738 const char *modify_branch, *delete_branch;
2739 struct object_id *changed_oid;
2740 int changed_mode;
2742 if (a_oid) {
2743 modify_branch = o->branch1;
2744 delete_branch = o->branch2;
2745 changed_oid = a_oid;
2746 changed_mode = a_mode;
2747 } else {
2748 modify_branch = o->branch2;
2749 delete_branch = o->branch1;
2750 changed_oid = b_oid;
2751 changed_mode = b_mode;
2754 return handle_change_delete(o,
2755 path, NULL,
2756 o_oid, o_mode,
2757 changed_oid, changed_mode,
2758 modify_branch, delete_branch,
2759 _("modify"), _("modified"));
2762 static int merge_content(struct merge_options *o,
2763 const char *path,
2764 int is_dirty,
2765 struct object_id *o_oid, int o_mode,
2766 struct object_id *a_oid, int a_mode,
2767 struct object_id *b_oid, int b_mode,
2768 struct rename_conflict_info *rename_conflict_info)
2770 const char *reason = _("content");
2771 const char *path1 = NULL, *path2 = NULL;
2772 struct merge_file_info mfi;
2773 struct diff_filespec one, a, b;
2774 unsigned df_conflict_remains = 0;
2776 if (!o_oid) {
2777 reason = _("add/add");
2778 o_oid = (struct object_id *)&null_oid;
2780 one.path = a.path = b.path = (char *)path;
2781 oidcpy(&one.oid, o_oid);
2782 one.mode = o_mode;
2783 oidcpy(&a.oid, a_oid);
2784 a.mode = a_mode;
2785 oidcpy(&b.oid, b_oid);
2786 b.mode = b_mode;
2788 if (rename_conflict_info) {
2789 struct diff_filepair *pair1 = rename_conflict_info->pair1;
2791 path1 = (o->branch1 == rename_conflict_info->branch1) ?
2792 pair1->two->path : pair1->one->path;
2793 /* If rename_conflict_info->pair2 != NULL, we are in
2794 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
2795 * normal rename.
2797 path2 = (rename_conflict_info->pair2 ||
2798 o->branch2 == rename_conflict_info->branch1) ?
2799 pair1->two->path : pair1->one->path;
2801 if (dir_in_way(path, !o->call_depth,
2802 S_ISGITLINK(pair1->two->mode)))
2803 df_conflict_remains = 1;
2805 if (merge_file_special_markers(o, &one, &a, &b,
2806 o->branch1, path1,
2807 o->branch2, path2, &mfi))
2808 return -1;
2810 if (mfi.clean && !df_conflict_remains &&
2811 oid_eq(&mfi.oid, a_oid) && mfi.mode == a_mode) {
2812 int path_renamed_outside_HEAD;
2813 output(o, 3, _("Skipped %s (merged same as existing)"), path);
2815 * The content merge resulted in the same file contents we
2816 * already had. We can return early if those file contents
2817 * are recorded at the correct path (which may not be true
2818 * if the merge involves a rename).
2820 path_renamed_outside_HEAD = !path2 || !strcmp(path, path2);
2821 if (!path_renamed_outside_HEAD) {
2822 if (add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2823 0, (!o->call_depth && !is_dirty), 0))
2824 return -1;
2825 return mfi.clean;
2827 } else
2828 output(o, 2, _("Auto-merging %s"), path);
2830 if (!mfi.clean) {
2831 if (S_ISGITLINK(mfi.mode))
2832 reason = _("submodule");
2833 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2834 reason, path);
2835 if (rename_conflict_info && !df_conflict_remains)
2836 if (update_stages(o, path, &one, &a, &b))
2837 return -1;
2840 if (df_conflict_remains || is_dirty) {
2841 char *new_path;
2842 if (o->call_depth) {
2843 remove_file_from_cache(path);
2844 } else {
2845 if (!mfi.clean) {
2846 if (update_stages(o, path, &one, &a, &b))
2847 return -1;
2848 } else {
2849 int file_from_stage2 = was_tracked(o, path);
2850 struct diff_filespec merged;
2851 oidcpy(&merged.oid, &mfi.oid);
2852 merged.mode = mfi.mode;
2854 if (update_stages(o, path, NULL,
2855 file_from_stage2 ? &merged : NULL,
2856 file_from_stage2 ? NULL : &merged))
2857 return -1;
2861 new_path = unique_path(o, path, rename_conflict_info->branch1);
2862 if (is_dirty) {
2863 output(o, 1, _("Refusing to lose dirty file at %s"),
2864 path);
2866 output(o, 1, _("Adding as %s instead"), new_path);
2867 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2868 free(new_path);
2869 return -1;
2871 free(new_path);
2872 mfi.clean = 0;
2873 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2874 return -1;
2875 return !is_dirty && mfi.clean;
2878 static int conflict_rename_normal(struct merge_options *o,
2879 const char *path,
2880 struct object_id *o_oid, unsigned int o_mode,
2881 struct object_id *a_oid, unsigned int a_mode,
2882 struct object_id *b_oid, unsigned int b_mode,
2883 struct rename_conflict_info *ci)
2885 /* Merge the content and write it out */
2886 return merge_content(o, path, was_dirty(o, path),
2887 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2888 ci);
2891 /* Per entry merge function */
2892 static int process_entry(struct merge_options *o,
2893 const char *path, struct stage_data *entry)
2895 int clean_merge = 1;
2896 int normalize = o->renormalize;
2897 unsigned o_mode = entry->stages[1].mode;
2898 unsigned a_mode = entry->stages[2].mode;
2899 unsigned b_mode = entry->stages[3].mode;
2900 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2901 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2902 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2904 entry->processed = 1;
2905 if (entry->rename_conflict_info) {
2906 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2907 switch (conflict_info->rename_type) {
2908 case RENAME_NORMAL:
2909 case RENAME_ONE_FILE_TO_ONE:
2910 clean_merge = conflict_rename_normal(o,
2911 path,
2912 o_oid, o_mode,
2913 a_oid, a_mode,
2914 b_oid, b_mode,
2915 conflict_info);
2916 break;
2917 case RENAME_DIR:
2918 clean_merge = 1;
2919 if (conflict_rename_dir(o,
2920 conflict_info->pair1,
2921 conflict_info->branch1,
2922 conflict_info->branch2))
2923 clean_merge = -1;
2924 break;
2925 case RENAME_DELETE:
2926 clean_merge = 0;
2927 if (conflict_rename_delete(o,
2928 conflict_info->pair1,
2929 conflict_info->branch1,
2930 conflict_info->branch2))
2931 clean_merge = -1;
2932 break;
2933 case RENAME_ONE_FILE_TO_TWO:
2934 clean_merge = 0;
2935 if (conflict_rename_rename_1to2(o, conflict_info))
2936 clean_merge = -1;
2937 break;
2938 case RENAME_TWO_FILES_TO_ONE:
2939 clean_merge = 0;
2940 if (conflict_rename_rename_2to1(o, conflict_info))
2941 clean_merge = -1;
2942 break;
2943 default:
2944 entry->processed = 0;
2945 break;
2947 } else if (o_oid && (!a_oid || !b_oid)) {
2948 /* Case A: Deleted in one */
2949 if ((!a_oid && !b_oid) ||
2950 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2951 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2952 /* Deleted in both or deleted in one and
2953 * unchanged in the other */
2954 if (a_oid)
2955 output(o, 2, _("Removing %s"), path);
2956 /* do not touch working file if it did not exist */
2957 remove_file(o, 1, path, !a_oid);
2958 } else {
2959 /* Modify/delete; deleted side may have put a directory in the way */
2960 clean_merge = 0;
2961 if (handle_modify_delete(o, path, o_oid, o_mode,
2962 a_oid, a_mode, b_oid, b_mode))
2963 clean_merge = -1;
2965 } else if ((!o_oid && a_oid && !b_oid) ||
2966 (!o_oid && !a_oid && b_oid)) {
2967 /* Case B: Added in one. */
2968 /* [nothing|directory] -> ([nothing|directory], file) */
2970 const char *add_branch;
2971 const char *other_branch;
2972 unsigned mode;
2973 const struct object_id *oid;
2974 const char *conf;
2976 if (a_oid) {
2977 add_branch = o->branch1;
2978 other_branch = o->branch2;
2979 mode = a_mode;
2980 oid = a_oid;
2981 conf = _("file/directory");
2982 } else {
2983 add_branch = o->branch2;
2984 other_branch = o->branch1;
2985 mode = b_mode;
2986 oid = b_oid;
2987 conf = _("directory/file");
2989 if (dir_in_way(path,
2990 !o->call_depth && !S_ISGITLINK(a_mode),
2991 0)) {
2992 char *new_path = unique_path(o, path, add_branch);
2993 clean_merge = 0;
2994 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
2995 "Adding %s as %s"),
2996 conf, path, other_branch, path, new_path);
2997 if (update_file(o, 0, oid, mode, new_path))
2998 clean_merge = -1;
2999 else if (o->call_depth)
3000 remove_file_from_cache(path);
3001 free(new_path);
3002 } else {
3003 output(o, 2, _("Adding %s"), path);
3004 /* do not overwrite file if already present */
3005 if (update_file_flags(o, oid, mode, path, 1, !a_oid))
3006 clean_merge = -1;
3008 } else if (a_oid && b_oid) {
3009 /* Case C: Added in both (check for same permissions) and */
3010 /* case D: Modified in both, but differently. */
3011 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3012 clean_merge = merge_content(o, path, is_dirty,
3013 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
3014 NULL);
3015 } else if (!o_oid && !a_oid && !b_oid) {
3017 * this entry was deleted altogether. a_mode == 0 means
3018 * we had that path and want to actively remove it.
3020 remove_file(o, 1, path, !a_mode);
3021 } else
3022 die("BUG: fatal merge failure, shouldn't happen.");
3024 return clean_merge;
3027 int merge_trees(struct merge_options *o,
3028 struct tree *head,
3029 struct tree *merge,
3030 struct tree *common,
3031 struct tree **result)
3033 int code, clean;
3035 if (o->subtree_shift) {
3036 merge = shift_tree_object(head, merge, o->subtree_shift);
3037 common = shift_tree_object(head, common, o->subtree_shift);
3040 if (oid_eq(&common->object.oid, &merge->object.oid)) {
3041 struct strbuf sb = STRBUF_INIT;
3043 if (!o->call_depth && index_has_changes(&sb)) {
3044 err(o, _("Dirty index: cannot merge (dirty: %s)"),
3045 sb.buf);
3046 return 0;
3048 output(o, 0, _("Already up to date!"));
3049 *result = head;
3050 return 1;
3053 code = git_merge_trees(o, common, head, merge);
3055 if (code != 0) {
3056 if (show(o, 4) || o->call_depth)
3057 err(o, _("merging of trees %s and %s failed"),
3058 oid_to_hex(&head->object.oid),
3059 oid_to_hex(&merge->object.oid));
3060 return -1;
3063 if (unmerged_cache()) {
3064 struct string_list *entries;
3065 struct rename_info re_info;
3066 int i;
3068 * Only need the hashmap while processing entries, so
3069 * initialize it here and free it when we are done running
3070 * through the entries. Keeping it in the merge_options as
3071 * opposed to decaring a local hashmap is for convenience
3072 * so that we don't have to pass it to around.
3074 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3075 get_files_dirs(o, head);
3076 get_files_dirs(o, merge);
3078 entries = get_unmerged();
3079 clean = handle_renames(o, common, head, merge, entries,
3080 &re_info);
3081 record_df_conflict_files(o, entries);
3082 if (clean < 0)
3083 goto cleanup;
3084 for (i = entries->nr-1; 0 <= i; i--) {
3085 const char *path = entries->items[i].string;
3086 struct stage_data *e = entries->items[i].util;
3087 if (!e->processed) {
3088 int ret = process_entry(o, path, e);
3089 if (!ret)
3090 clean = 0;
3091 else if (ret < 0) {
3092 clean = ret;
3093 goto cleanup;
3097 for (i = 0; i < entries->nr; i++) {
3098 struct stage_data *e = entries->items[i].util;
3099 if (!e->processed)
3100 die("BUG: unprocessed path??? %s",
3101 entries->items[i].string);
3104 cleanup:
3105 final_cleanup_renames(&re_info);
3107 string_list_clear(entries, 1);
3108 free(entries);
3110 hashmap_free(&o->current_file_dir_set, 1);
3112 if (clean < 0)
3113 return clean;
3115 else
3116 clean = 1;
3118 /* Free the extra index left from git_merge_trees() */
3120 * FIXME: Need to also free data allocated by
3121 * setup_unpack_trees_porcelain() tucked away in o->unpack_opts.msgs,
3122 * but the problem is that only half of it refers to dynamically
3123 * allocated data, while the other half points at static strings.
3125 discard_index(&o->orig_index);
3127 if (o->call_depth && !(*result = write_tree_from_memory(o)))
3128 return -1;
3130 return clean;
3133 static struct commit_list *reverse_commit_list(struct commit_list *list)
3135 struct commit_list *next = NULL, *current, *backup;
3136 for (current = list; current; current = backup) {
3137 backup = current->next;
3138 current->next = next;
3139 next = current;
3141 return next;
3145 * Merge the commits h1 and h2, return the resulting virtual
3146 * commit object and a flag indicating the cleanness of the merge.
3148 int merge_recursive(struct merge_options *o,
3149 struct commit *h1,
3150 struct commit *h2,
3151 struct commit_list *ca,
3152 struct commit **result)
3154 struct commit_list *iter;
3155 struct commit *merged_common_ancestors;
3156 struct tree *mrtree;
3157 int clean;
3159 if (show(o, 4)) {
3160 output(o, 4, _("Merging:"));
3161 output_commit_title(o, h1);
3162 output_commit_title(o, h2);
3165 if (!ca) {
3166 ca = get_merge_bases(h1, h2);
3167 ca = reverse_commit_list(ca);
3170 if (show(o, 5)) {
3171 unsigned cnt = commit_list_count(ca);
3173 output(o, 5, Q_("found %u common ancestor:",
3174 "found %u common ancestors:", cnt), cnt);
3175 for (iter = ca; iter; iter = iter->next)
3176 output_commit_title(o, iter->item);
3179 merged_common_ancestors = pop_commit(&ca);
3180 if (merged_common_ancestors == NULL) {
3181 /* if there is no common ancestor, use an empty tree */
3182 struct tree *tree;
3184 tree = lookup_tree(the_hash_algo->empty_tree);
3185 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3188 for (iter = ca; iter; iter = iter->next) {
3189 const char *saved_b1, *saved_b2;
3190 o->call_depth++;
3192 * When the merge fails, the result contains files
3193 * with conflict markers. The cleanness flag is
3194 * ignored (unless indicating an error), it was never
3195 * actually used, as result of merge_trees has always
3196 * overwritten it: the committed "conflicts" were
3197 * already resolved.
3199 discard_cache();
3200 saved_b1 = o->branch1;
3201 saved_b2 = o->branch2;
3202 o->branch1 = "Temporary merge branch 1";
3203 o->branch2 = "Temporary merge branch 2";
3204 if (merge_recursive(o, merged_common_ancestors, iter->item,
3205 NULL, &merged_common_ancestors) < 0)
3206 return -1;
3207 o->branch1 = saved_b1;
3208 o->branch2 = saved_b2;
3209 o->call_depth--;
3211 if (!merged_common_ancestors)
3212 return err(o, _("merge returned no commit"));
3215 discard_cache();
3216 if (!o->call_depth)
3217 read_cache();
3219 o->ancestor = "merged common ancestors";
3220 clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,
3221 &mrtree);
3222 if (clean < 0) {
3223 flush_output(o);
3224 return clean;
3227 if (o->call_depth) {
3228 *result = make_virtual_commit(mrtree, "merged tree");
3229 commit_list_insert(h1, &(*result)->parents);
3230 commit_list_insert(h2, &(*result)->parents->next);
3232 flush_output(o);
3233 if (!o->call_depth && o->buffer_output < 2)
3234 strbuf_release(&o->obuf);
3235 if (show(o, 2))
3236 diff_warn_rename_limit("merge.renamelimit",
3237 o->needed_rename_limit, 0);
3238 return clean;
3241 static struct commit *get_ref(const struct object_id *oid, const char *name)
3243 struct object *object;
3245 object = deref_tag(parse_object(oid), name, strlen(name));
3246 if (!object)
3247 return NULL;
3248 if (object->type == OBJ_TREE)
3249 return make_virtual_commit((struct tree*)object, name);
3250 if (object->type != OBJ_COMMIT)
3251 return NULL;
3252 if (parse_commit((struct commit *)object))
3253 return NULL;
3254 return (struct commit *)object;
3257 int merge_recursive_generic(struct merge_options *o,
3258 const struct object_id *head,
3259 const struct object_id *merge,
3260 int num_base_list,
3261 const struct object_id **base_list,
3262 struct commit **result)
3264 int clean;
3265 struct lock_file lock = LOCK_INIT;
3266 struct commit *head_commit = get_ref(head, o->branch1);
3267 struct commit *next_commit = get_ref(merge, o->branch2);
3268 struct commit_list *ca = NULL;
3270 if (base_list) {
3271 int i;
3272 for (i = 0; i < num_base_list; ++i) {
3273 struct commit *base;
3274 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3275 return err(o, _("Could not parse object '%s'"),
3276 oid_to_hex(base_list[i]));
3277 commit_list_insert(base, &ca);
3281 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3282 clean = merge_recursive(o, head_commit, next_commit, ca,
3283 result);
3284 if (clean < 0) {
3285 rollback_lock_file(&lock);
3286 return clean;
3289 if (write_locked_index(&the_index, &lock,
3290 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3291 return err(o, _("Unable to write index."));
3293 return clean ? 0 : 1;
3296 static void merge_recursive_config(struct merge_options *o)
3298 git_config_get_int("merge.verbosity", &o->verbosity);
3299 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3300 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3301 git_config(git_xmerge_config, NULL);
3304 void init_merge_options(struct merge_options *o)
3306 const char *merge_verbosity;
3307 memset(o, 0, sizeof(struct merge_options));
3308 o->verbosity = 2;
3309 o->buffer_output = 1;
3310 o->diff_rename_limit = -1;
3311 o->merge_rename_limit = -1;
3312 o->renormalize = 0;
3313 o->detect_rename = 1;
3314 merge_recursive_config(o);
3315 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3316 if (merge_verbosity)
3317 o->verbosity = strtol(merge_verbosity, NULL, 10);
3318 if (o->verbosity >= 5)
3319 o->buffer_output = 0;
3320 strbuf_init(&o->obuf, 0);
3321 string_list_init(&o->df_conflict_file_set, 1);
3324 int parse_merge_opt(struct merge_options *o, const char *s)
3326 const char *arg;
3328 if (!s || !*s)
3329 return -1;
3330 if (!strcmp(s, "ours"))
3331 o->recursive_variant = MERGE_RECURSIVE_OURS;
3332 else if (!strcmp(s, "theirs"))
3333 o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3334 else if (!strcmp(s, "subtree"))
3335 o->subtree_shift = "";
3336 else if (skip_prefix(s, "subtree=", &arg))
3337 o->subtree_shift = arg;
3338 else if (!strcmp(s, "patience"))
3339 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3340 else if (!strcmp(s, "histogram"))
3341 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3342 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3343 long value = parse_algorithm_value(arg);
3344 if (value < 0)
3345 return -1;
3346 /* clear out previous settings */
3347 DIFF_XDL_CLR(o, NEED_MINIMAL);
3348 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3349 o->xdl_opts |= value;
3351 else if (!strcmp(s, "ignore-space-change"))
3352 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3353 else if (!strcmp(s, "ignore-all-space"))
3354 DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3355 else if (!strcmp(s, "ignore-space-at-eol"))
3356 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3357 else if (!strcmp(s, "ignore-cr-at-eol"))
3358 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3359 else if (!strcmp(s, "renormalize"))
3360 o->renormalize = 1;
3361 else if (!strcmp(s, "no-renormalize"))
3362 o->renormalize = 0;
3363 else if (!strcmp(s, "no-renames"))
3364 o->detect_rename = 0;
3365 else if (!strcmp(s, "find-renames")) {
3366 o->detect_rename = 1;
3367 o->rename_score = 0;
3369 else if (skip_prefix(s, "find-renames=", &arg) ||
3370 skip_prefix(s, "rename-threshold=", &arg)) {
3371 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3372 return -1;
3373 o->detect_rename = 1;
3375 else
3376 return -1;
3377 return 0;