Merge branch 'jt/partial-clone-proto-v2'
[git/debian.git] / merge-recursive.c
blobe64b4b9d031afff282bf37c7c373d29e11157835
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->maybe_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 o->unpack_opts.aggressive = !merge_detect_rename(o);
360 setup_unpack_trees_porcelain(&o->unpack_opts, "merge");
362 init_tree_desc_from_tree(t+0, common);
363 init_tree_desc_from_tree(t+1, head);
364 init_tree_desc_from_tree(t+2, merge);
366 rc = unpack_trees(3, t, &o->unpack_opts);
367 cache_tree_free(&active_cache_tree);
370 * Update the_index to match the new results, AFTER saving a copy
371 * in o->orig_index. Update src_index to point to the saved copy.
372 * (verify_uptodate() checks src_index, and the original index is
373 * the one that had the necessary modification timestamps.)
375 o->orig_index = the_index;
376 the_index = tmp_index;
377 o->unpack_opts.src_index = &o->orig_index;
379 return rc;
382 struct tree *write_tree_from_memory(struct merge_options *o)
384 struct tree *result = NULL;
386 if (unmerged_cache()) {
387 int i;
388 fprintf(stderr, "BUG: There are unmerged index entries:\n");
389 for (i = 0; i < active_nr; i++) {
390 const struct cache_entry *ce = active_cache[i];
391 if (ce_stage(ce))
392 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
393 (int)ce_namelen(ce), ce->name);
395 BUG("unmerged index entries in merge-recursive.c");
398 if (!active_cache_tree)
399 active_cache_tree = cache_tree();
401 if (!cache_tree_fully_valid(active_cache_tree) &&
402 cache_tree_update(&the_index, 0) < 0) {
403 err(o, _("error building trees"));
404 return NULL;
407 result = lookup_tree(&active_cache_tree->oid);
409 return result;
412 static int save_files_dirs(const struct object_id *oid,
413 struct strbuf *base, const char *path,
414 unsigned int mode, int stage, void *context)
416 struct path_hashmap_entry *entry;
417 int baselen = base->len;
418 struct merge_options *o = context;
420 strbuf_addstr(base, path);
422 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
423 hashmap_entry_init(entry, path_hash(entry->path));
424 hashmap_add(&o->current_file_dir_set, entry);
426 strbuf_setlen(base, baselen);
427 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
430 static void get_files_dirs(struct merge_options *o, struct tree *tree)
432 struct pathspec match_all;
433 memset(&match_all, 0, sizeof(match_all));
434 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
437 static int get_tree_entry_if_blob(const struct object_id *tree,
438 const char *path,
439 struct object_id *hashy,
440 unsigned int *mode_o)
442 int ret;
444 ret = get_tree_entry(tree, path, hashy, mode_o);
445 if (S_ISDIR(*mode_o)) {
446 oidcpy(hashy, &null_oid);
447 *mode_o = 0;
449 return ret;
453 * Returns an index_entry instance which doesn't have to correspond to
454 * a real cache entry in Git's index.
456 static struct stage_data *insert_stage_data(const char *path,
457 struct tree *o, struct tree *a, struct tree *b,
458 struct string_list *entries)
460 struct string_list_item *item;
461 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
462 get_tree_entry_if_blob(&o->object.oid, path,
463 &e->stages[1].oid, &e->stages[1].mode);
464 get_tree_entry_if_blob(&a->object.oid, path,
465 &e->stages[2].oid, &e->stages[2].mode);
466 get_tree_entry_if_blob(&b->object.oid, path,
467 &e->stages[3].oid, &e->stages[3].mode);
468 item = string_list_insert(entries, path);
469 item->util = e;
470 return e;
474 * Create a dictionary mapping file names to stage_data objects. The
475 * dictionary contains one entry for every path with a non-zero stage entry.
477 static struct string_list *get_unmerged(void)
479 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
480 int i;
482 unmerged->strdup_strings = 1;
484 for (i = 0; i < active_nr; i++) {
485 struct string_list_item *item;
486 struct stage_data *e;
487 const struct cache_entry *ce = active_cache[i];
488 if (!ce_stage(ce))
489 continue;
491 item = string_list_lookup(unmerged, ce->name);
492 if (!item) {
493 item = string_list_insert(unmerged, ce->name);
494 item->util = xcalloc(1, sizeof(struct stage_data));
496 e = item->util;
497 e->stages[ce_stage(ce)].mode = ce->ce_mode;
498 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
501 return unmerged;
504 static int string_list_df_name_compare(const char *one, const char *two)
506 int onelen = strlen(one);
507 int twolen = strlen(two);
509 * Here we only care that entries for D/F conflicts are
510 * adjacent, in particular with the file of the D/F conflict
511 * appearing before files below the corresponding directory.
512 * The order of the rest of the list is irrelevant for us.
514 * To achieve this, we sort with df_name_compare and provide
515 * the mode S_IFDIR so that D/F conflicts will sort correctly.
516 * We use the mode S_IFDIR for everything else for simplicity,
517 * since in other cases any changes in their order due to
518 * sorting cause no problems for us.
520 int cmp = df_name_compare(one, onelen, S_IFDIR,
521 two, twolen, S_IFDIR);
523 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
524 * that 'foo' comes before 'foo/bar'.
526 if (cmp)
527 return cmp;
528 return onelen - twolen;
531 static void record_df_conflict_files(struct merge_options *o,
532 struct string_list *entries)
534 /* If there is a D/F conflict and the file for such a conflict
535 * currently exist in the working tree, we want to allow it to be
536 * removed to make room for the corresponding directory if needed.
537 * The files underneath the directories of such D/F conflicts will
538 * be processed before the corresponding file involved in the D/F
539 * conflict. If the D/F directory ends up being removed by the
540 * merge, then we won't have to touch the D/F file. If the D/F
541 * directory needs to be written to the working copy, then the D/F
542 * file will simply be removed (in make_room_for_path()) to make
543 * room for the necessary paths. Note that if both the directory
544 * and the file need to be present, then the D/F file will be
545 * reinstated with a new unique name at the time it is processed.
547 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
548 const char *last_file = NULL;
549 int last_len = 0;
550 int i;
553 * If we're merging merge-bases, we don't want to bother with
554 * any working directory changes.
556 if (o->call_depth)
557 return;
559 /* Ensure D/F conflicts are adjacent in the entries list. */
560 for (i = 0; i < entries->nr; i++) {
561 struct string_list_item *next = &entries->items[i];
562 string_list_append(&df_sorted_entries, next->string)->util =
563 next->util;
565 df_sorted_entries.cmp = string_list_df_name_compare;
566 string_list_sort(&df_sorted_entries);
568 string_list_clear(&o->df_conflict_file_set, 1);
569 for (i = 0; i < df_sorted_entries.nr; i++) {
570 const char *path = df_sorted_entries.items[i].string;
571 int len = strlen(path);
572 struct stage_data *e = df_sorted_entries.items[i].util;
575 * Check if last_file & path correspond to a D/F conflict;
576 * i.e. whether path is last_file+'/'+<something>.
577 * If so, record that it's okay to remove last_file to make
578 * room for path and friends if needed.
580 if (last_file &&
581 len > last_len &&
582 memcmp(path, last_file, last_len) == 0 &&
583 path[last_len] == '/') {
584 string_list_insert(&o->df_conflict_file_set, last_file);
588 * Determine whether path could exist as a file in the
589 * working directory as a possible D/F conflict. This
590 * will only occur when it exists in stage 2 as a
591 * file.
593 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
594 last_file = path;
595 last_len = len;
596 } else {
597 last_file = NULL;
600 string_list_clear(&df_sorted_entries, 0);
603 struct rename {
604 struct diff_filepair *pair;
606 * Purpose of src_entry and dst_entry:
608 * If 'before' is renamed to 'after' then src_entry will contain
609 * the versions of 'before' from the merge_base, HEAD, and MERGE in
610 * stages 1, 2, and 3; dst_entry will contain the respective
611 * versions of 'after' in corresponding locations. Thus, we have a
612 * total of six modes and oids, though some will be null. (Stage 0
613 * is ignored; we're interested in handling conflicts.)
615 * Since we don't turn on break-rewrites by default, neither
616 * src_entry nor dst_entry can have all three of their stages have
617 * non-null oids, meaning at most four of the six will be non-null.
618 * Also, since this is a rename, both src_entry and dst_entry will
619 * have at least one non-null oid, meaning at least two will be
620 * non-null. Of the six oids, a typical rename will have three be
621 * non-null. Only two implies a rename/delete, and four implies a
622 * rename/add.
624 struct stage_data *src_entry;
625 struct stage_data *dst_entry;
626 unsigned add_turned_into_rename:1;
627 unsigned processed:1;
630 static int update_stages(struct merge_options *opt, const char *path,
631 const struct diff_filespec *o,
632 const struct diff_filespec *a,
633 const struct diff_filespec *b)
637 * NOTE: It is usually a bad idea to call update_stages on a path
638 * before calling update_file on that same path, since it can
639 * sometimes lead to spurious "refusing to lose untracked file..."
640 * messages from update_file (via make_room_for path via
641 * would_lose_untracked). Instead, reverse the order of the calls
642 * (executing update_file first and then update_stages).
644 int clear = 1;
645 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
646 if (clear)
647 if (remove_file_from_cache(path))
648 return -1;
649 if (o)
650 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
651 return -1;
652 if (a)
653 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
654 return -1;
655 if (b)
656 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
657 return -1;
658 return 0;
661 static int update_stages_for_stage_data(struct merge_options *opt,
662 const char *path,
663 const struct stage_data *stage_data)
665 struct diff_filespec o, a, b;
667 o.mode = stage_data->stages[1].mode;
668 oidcpy(&o.oid, &stage_data->stages[1].oid);
670 a.mode = stage_data->stages[2].mode;
671 oidcpy(&a.oid, &stage_data->stages[2].oid);
673 b.mode = stage_data->stages[3].mode;
674 oidcpy(&b.oid, &stage_data->stages[3].oid);
676 return update_stages(opt, path,
677 is_null_oid(&o.oid) ? NULL : &o,
678 is_null_oid(&a.oid) ? NULL : &a,
679 is_null_oid(&b.oid) ? NULL : &b);
682 static void update_entry(struct stage_data *entry,
683 struct diff_filespec *o,
684 struct diff_filespec *a,
685 struct diff_filespec *b)
687 entry->processed = 0;
688 entry->stages[1].mode = o->mode;
689 entry->stages[2].mode = a->mode;
690 entry->stages[3].mode = b->mode;
691 oidcpy(&entry->stages[1].oid, &o->oid);
692 oidcpy(&entry->stages[2].oid, &a->oid);
693 oidcpy(&entry->stages[3].oid, &b->oid);
696 static int remove_file(struct merge_options *o, int clean,
697 const char *path, int no_wd)
699 int update_cache = o->call_depth || clean;
700 int update_working_directory = !o->call_depth && !no_wd;
702 if (update_cache) {
703 if (remove_file_from_cache(path))
704 return -1;
706 if (update_working_directory) {
707 if (ignore_case) {
708 struct cache_entry *ce;
709 ce = cache_file_exists(path, strlen(path), ignore_case);
710 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
711 return 0;
713 if (remove_path(path))
714 return -1;
716 return 0;
719 /* add a string to a strbuf, but converting "/" to "_" */
720 static void add_flattened_path(struct strbuf *out, const char *s)
722 size_t i = out->len;
723 strbuf_addstr(out, s);
724 for (; i < out->len; i++)
725 if (out->buf[i] == '/')
726 out->buf[i] = '_';
729 static char *unique_path(struct merge_options *o, const char *path, const char *branch)
731 struct path_hashmap_entry *entry;
732 struct strbuf newpath = STRBUF_INIT;
733 int suffix = 0;
734 size_t base_len;
736 strbuf_addf(&newpath, "%s~", path);
737 add_flattened_path(&newpath, branch);
739 base_len = newpath.len;
740 while (hashmap_get_from_hash(&o->current_file_dir_set,
741 path_hash(newpath.buf), newpath.buf) ||
742 (!o->call_depth && file_exists(newpath.buf))) {
743 strbuf_setlen(&newpath, base_len);
744 strbuf_addf(&newpath, "_%d", suffix++);
747 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
748 hashmap_entry_init(entry, path_hash(entry->path));
749 hashmap_add(&o->current_file_dir_set, entry);
750 return strbuf_detach(&newpath, NULL);
754 * Check whether a directory in the index is in the way of an incoming
755 * file. Return 1 if so. If check_working_copy is non-zero, also
756 * check the working directory. If empty_ok is non-zero, also return
757 * 0 in the case where the working-tree dir exists but is empty.
759 static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
761 int pos;
762 struct strbuf dirpath = STRBUF_INIT;
763 struct stat st;
765 strbuf_addstr(&dirpath, path);
766 strbuf_addch(&dirpath, '/');
768 pos = cache_name_pos(dirpath.buf, dirpath.len);
770 if (pos < 0)
771 pos = -1 - pos;
772 if (pos < active_nr &&
773 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
774 strbuf_release(&dirpath);
775 return 1;
778 strbuf_release(&dirpath);
779 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
780 !(empty_ok && is_empty_dir(path));
784 * Returns whether path was tracked in the index before the merge started,
785 * and its oid and mode match the specified values
787 static int was_tracked_and_matches(struct merge_options *o, const char *path,
788 const struct object_id *oid, unsigned mode)
790 int pos = index_name_pos(&o->orig_index, path, strlen(path));
791 struct cache_entry *ce;
793 if (0 > pos)
794 /* we were not tracking this path before the merge */
795 return 0;
797 /* See if the file we were tracking before matches */
798 ce = o->orig_index.cache[pos];
799 return (oid_eq(&ce->oid, oid) && ce->ce_mode == mode);
803 * Returns whether path was tracked in the index before the merge started
805 static int was_tracked(struct merge_options *o, const char *path)
807 int pos = index_name_pos(&o->orig_index, path, strlen(path));
809 if (0 <= pos)
810 /* we were tracking this path before the merge */
811 return 1;
813 return 0;
816 static int would_lose_untracked(const char *path)
819 * This may look like it can be simplified to:
820 * return !was_tracked(o, path) && file_exists(path)
821 * but it can't. This function needs to know whether path was in
822 * the working tree due to EITHER having been tracked in the index
823 * before the merge OR having been put into the working copy and
824 * index by unpack_trees(). Due to that either-or requirement, we
825 * check the current index instead of the original one.
827 * Note that we do not need to worry about merge-recursive itself
828 * updating the index after unpack_trees() and before calling this
829 * function, because we strictly require all code paths in
830 * merge-recursive to update the working tree first and the index
831 * second. Doing otherwise would break
832 * update_file()/would_lose_untracked(); see every comment in this
833 * file which mentions "update_stages".
835 int pos = cache_name_pos(path, strlen(path));
837 if (pos < 0)
838 pos = -1 - pos;
839 while (pos < active_nr &&
840 !strcmp(path, active_cache[pos]->name)) {
842 * If stage #0, it is definitely tracked.
843 * If it has stage #2 then it was tracked
844 * before this merge started. All other
845 * cases the path was not tracked.
847 switch (ce_stage(active_cache[pos])) {
848 case 0:
849 case 2:
850 return 0;
852 pos++;
854 return file_exists(path);
857 static int was_dirty(struct merge_options *o, const char *path)
859 struct cache_entry *ce;
860 int dirty = 1;
862 if (o->call_depth || !was_tracked(o, path))
863 return !dirty;
865 ce = index_file_exists(o->unpack_opts.src_index,
866 path, strlen(path), ignore_case);
867 dirty = verify_uptodate(ce, &o->unpack_opts) != 0;
868 return dirty;
871 static int make_room_for_path(struct merge_options *o, const char *path)
873 int status, i;
874 const char *msg = _("failed to create path '%s'%s");
876 /* Unlink any D/F conflict files that are in the way */
877 for (i = 0; i < o->df_conflict_file_set.nr; i++) {
878 const char *df_path = o->df_conflict_file_set.items[i].string;
879 size_t pathlen = strlen(path);
880 size_t df_pathlen = strlen(df_path);
881 if (df_pathlen < pathlen &&
882 path[df_pathlen] == '/' &&
883 strncmp(path, df_path, df_pathlen) == 0) {
884 output(o, 3,
885 _("Removing %s to make room for subdirectory\n"),
886 df_path);
887 unlink(df_path);
888 unsorted_string_list_delete_item(&o->df_conflict_file_set,
889 i, 0);
890 break;
894 /* Make sure leading directories are created */
895 status = safe_create_leading_directories_const(path);
896 if (status) {
897 if (status == SCLD_EXISTS)
898 /* something else exists */
899 return err(o, msg, path, _(": perhaps a D/F conflict?"));
900 return err(o, msg, path, "");
904 * Do not unlink a file in the work tree if we are not
905 * tracking it.
907 if (would_lose_untracked(path))
908 return err(o, _("refusing to lose untracked file at '%s'"),
909 path);
911 /* Successful unlink is good.. */
912 if (!unlink(path))
913 return 0;
914 /* .. and so is no existing file */
915 if (errno == ENOENT)
916 return 0;
917 /* .. but not some other error (who really cares what?) */
918 return err(o, msg, path, _(": perhaps a D/F conflict?"));
921 static int update_file_flags(struct merge_options *o,
922 const struct object_id *oid,
923 unsigned mode,
924 const char *path,
925 int update_cache,
926 int update_wd)
928 int ret = 0;
930 if (o->call_depth)
931 update_wd = 0;
933 if (update_wd) {
934 enum object_type type;
935 void *buf;
936 unsigned long size;
938 if (S_ISGITLINK(mode)) {
940 * We may later decide to recursively descend into
941 * the submodule directory and update its index
942 * and/or work tree, but we do not do that now.
944 update_wd = 0;
945 goto update_index;
948 buf = read_object_file(oid, &type, &size);
949 if (!buf)
950 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
951 if (type != OBJ_BLOB) {
952 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
953 goto free_buf;
955 if (S_ISREG(mode)) {
956 struct strbuf strbuf = STRBUF_INIT;
957 if (convert_to_working_tree(path, buf, size, &strbuf)) {
958 free(buf);
959 size = strbuf.len;
960 buf = strbuf_detach(&strbuf, NULL);
964 if (make_room_for_path(o, path) < 0) {
965 update_wd = 0;
966 goto free_buf;
968 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
969 int fd;
970 if (mode & 0100)
971 mode = 0777;
972 else
973 mode = 0666;
974 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
975 if (fd < 0) {
976 ret = err(o, _("failed to open '%s': %s"),
977 path, strerror(errno));
978 goto free_buf;
980 write_in_full(fd, buf, size);
981 close(fd);
982 } else if (S_ISLNK(mode)) {
983 char *lnk = xmemdupz(buf, size);
984 safe_create_leading_directories_const(path);
985 unlink(path);
986 if (symlink(lnk, path))
987 ret = err(o, _("failed to symlink '%s': %s"),
988 path, strerror(errno));
989 free(lnk);
990 } else
991 ret = err(o,
992 _("do not know what to do with %06o %s '%s'"),
993 mode, oid_to_hex(oid), path);
994 free_buf:
995 free(buf);
997 update_index:
998 if (!ret && update_cache)
999 if (add_cacheinfo(o, mode, oid, path, 0, update_wd,
1000 ADD_CACHE_OK_TO_ADD))
1001 return -1;
1002 return ret;
1005 static int update_file(struct merge_options *o,
1006 int clean,
1007 const struct object_id *oid,
1008 unsigned mode,
1009 const char *path)
1011 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
1014 /* Low level file merging, update and removal */
1016 struct merge_file_info {
1017 struct object_id oid;
1018 unsigned mode;
1019 unsigned clean:1,
1020 merge:1;
1023 static int merge_3way(struct merge_options *o,
1024 mmbuffer_t *result_buf,
1025 const struct diff_filespec *one,
1026 const struct diff_filespec *a,
1027 const struct diff_filespec *b,
1028 const char *branch1,
1029 const char *branch2)
1031 mmfile_t orig, src1, src2;
1032 struct ll_merge_options ll_opts = {0};
1033 char *base_name, *name1, *name2;
1034 int merge_status;
1036 ll_opts.renormalize = o->renormalize;
1037 ll_opts.xdl_opts = o->xdl_opts;
1039 if (o->call_depth) {
1040 ll_opts.virtual_ancestor = 1;
1041 ll_opts.variant = 0;
1042 } else {
1043 switch (o->recursive_variant) {
1044 case MERGE_RECURSIVE_OURS:
1045 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1046 break;
1047 case MERGE_RECURSIVE_THEIRS:
1048 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1049 break;
1050 default:
1051 ll_opts.variant = 0;
1052 break;
1056 if (strcmp(a->path, b->path) ||
1057 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1058 base_name = o->ancestor == NULL ? NULL :
1059 mkpathdup("%s:%s", o->ancestor, one->path);
1060 name1 = mkpathdup("%s:%s", branch1, a->path);
1061 name2 = mkpathdup("%s:%s", branch2, b->path);
1062 } else {
1063 base_name = o->ancestor == NULL ? NULL :
1064 mkpathdup("%s", o->ancestor);
1065 name1 = mkpathdup("%s", branch1);
1066 name2 = mkpathdup("%s", branch2);
1069 read_mmblob(&orig, &one->oid);
1070 read_mmblob(&src1, &a->oid);
1071 read_mmblob(&src2, &b->oid);
1073 merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1074 &src1, name1, &src2, name2, &ll_opts);
1076 free(base_name);
1077 free(name1);
1078 free(name2);
1079 free(orig.ptr);
1080 free(src1.ptr);
1081 free(src2.ptr);
1082 return merge_status;
1085 static int merge_file_1(struct merge_options *o,
1086 const struct diff_filespec *one,
1087 const struct diff_filespec *a,
1088 const struct diff_filespec *b,
1089 const char *filename,
1090 const char *branch1,
1091 const char *branch2,
1092 struct merge_file_info *result)
1094 result->merge = 0;
1095 result->clean = 1;
1097 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1098 result->clean = 0;
1099 if (S_ISREG(a->mode)) {
1100 result->mode = a->mode;
1101 oidcpy(&result->oid, &a->oid);
1102 } else {
1103 result->mode = b->mode;
1104 oidcpy(&result->oid, &b->oid);
1106 } else {
1107 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1108 result->merge = 1;
1111 * Merge modes
1113 if (a->mode == b->mode || a->mode == one->mode)
1114 result->mode = b->mode;
1115 else {
1116 result->mode = a->mode;
1117 if (b->mode != one->mode) {
1118 result->clean = 0;
1119 result->merge = 1;
1123 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1124 oidcpy(&result->oid, &b->oid);
1125 else if (oid_eq(&b->oid, &one->oid))
1126 oidcpy(&result->oid, &a->oid);
1127 else if (S_ISREG(a->mode)) {
1128 mmbuffer_t result_buf;
1129 int ret = 0, merge_status;
1131 merge_status = merge_3way(o, &result_buf, one, a, b,
1132 branch1, branch2);
1134 if ((merge_status < 0) || !result_buf.ptr)
1135 ret = err(o, _("Failed to execute internal merge"));
1137 if (!ret &&
1138 write_object_file(result_buf.ptr, result_buf.size,
1139 blob_type, &result->oid))
1140 ret = err(o, _("Unable to add %s to database"),
1141 a->path);
1143 free(result_buf.ptr);
1144 if (ret)
1145 return ret;
1146 result->clean = (merge_status == 0);
1147 } else if (S_ISGITLINK(a->mode)) {
1148 result->clean = merge_submodule(&result->oid,
1149 one->path,
1150 &one->oid,
1151 &a->oid,
1152 &b->oid,
1153 !o->call_depth);
1154 } else if (S_ISLNK(a->mode)) {
1155 switch (o->recursive_variant) {
1156 case MERGE_RECURSIVE_NORMAL:
1157 oidcpy(&result->oid, &a->oid);
1158 if (!oid_eq(&a->oid, &b->oid))
1159 result->clean = 0;
1160 break;
1161 case MERGE_RECURSIVE_OURS:
1162 oidcpy(&result->oid, &a->oid);
1163 break;
1164 case MERGE_RECURSIVE_THEIRS:
1165 oidcpy(&result->oid, &b->oid);
1166 break;
1168 } else
1169 BUG("unsupported object type in the tree");
1172 if (result->merge)
1173 output(o, 2, _("Auto-merging %s"), filename);
1175 return 0;
1178 static int merge_file_special_markers(struct merge_options *o,
1179 const struct diff_filespec *one,
1180 const struct diff_filespec *a,
1181 const struct diff_filespec *b,
1182 const char *target_filename,
1183 const char *branch1,
1184 const char *filename1,
1185 const char *branch2,
1186 const char *filename2,
1187 struct merge_file_info *mfi)
1189 char *side1 = NULL;
1190 char *side2 = NULL;
1191 int ret;
1193 if (filename1)
1194 side1 = xstrfmt("%s:%s", branch1, filename1);
1195 if (filename2)
1196 side2 = xstrfmt("%s:%s", branch2, filename2);
1198 ret = merge_file_1(o, one, a, b, target_filename,
1199 side1 ? side1 : branch1,
1200 side2 ? side2 : branch2, mfi);
1202 free(side1);
1203 free(side2);
1204 return ret;
1207 static int merge_file_one(struct merge_options *o,
1208 const char *path,
1209 const struct object_id *o_oid, int o_mode,
1210 const struct object_id *a_oid, int a_mode,
1211 const struct object_id *b_oid, int b_mode,
1212 const char *branch1,
1213 const char *branch2,
1214 struct merge_file_info *mfi)
1216 struct diff_filespec one, a, b;
1218 one.path = a.path = b.path = (char *)path;
1219 oidcpy(&one.oid, o_oid);
1220 one.mode = o_mode;
1221 oidcpy(&a.oid, a_oid);
1222 a.mode = a_mode;
1223 oidcpy(&b.oid, b_oid);
1224 b.mode = b_mode;
1225 return merge_file_1(o, &one, &a, &b, path, branch1, branch2, mfi);
1228 static int conflict_rename_dir(struct merge_options *o,
1229 struct diff_filepair *pair,
1230 const char *rename_branch,
1231 const char *other_branch)
1233 const struct diff_filespec *dest = pair->two;
1235 if (!o->call_depth && would_lose_untracked(dest->path)) {
1236 char *alt_path = unique_path(o, dest->path, rename_branch);
1238 output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1239 "writing to %s instead."),
1240 dest->path, alt_path);
1242 * Write the file in worktree at alt_path, but not in the
1243 * index. Instead, write to dest->path for the index but
1244 * only at the higher appropriate stage.
1246 if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1247 return -1;
1248 free(alt_path);
1249 return update_stages(o, dest->path, NULL,
1250 rename_branch == o->branch1 ? dest : NULL,
1251 rename_branch == o->branch1 ? NULL : dest);
1254 /* Update dest->path both in index and in worktree */
1255 if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1256 return -1;
1257 return 0;
1260 static int handle_change_delete(struct merge_options *o,
1261 const char *path, const char *old_path,
1262 const struct object_id *o_oid, int o_mode,
1263 const struct object_id *changed_oid,
1264 int changed_mode,
1265 const char *change_branch,
1266 const char *delete_branch,
1267 const char *change, const char *change_past)
1269 char *alt_path = NULL;
1270 const char *update_path = path;
1271 int ret = 0;
1273 if (dir_in_way(path, !o->call_depth, 0) ||
1274 (!o->call_depth && would_lose_untracked(path))) {
1275 update_path = alt_path = unique_path(o, path, change_branch);
1278 if (o->call_depth) {
1280 * We cannot arbitrarily accept either a_sha or b_sha as
1281 * correct; since there is no true "middle point" between
1282 * them, simply reuse the base version for virtual merge base.
1284 ret = remove_file_from_cache(path);
1285 if (!ret)
1286 ret = update_file(o, 0, o_oid, o_mode, update_path);
1287 } else {
1288 if (!alt_path) {
1289 if (!old_path) {
1290 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1291 "and %s in %s. Version %s of %s left in tree."),
1292 change, path, delete_branch, change_past,
1293 change_branch, change_branch, path);
1294 } else {
1295 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1296 "and %s to %s in %s. Version %s of %s left in tree."),
1297 change, old_path, delete_branch, change_past, path,
1298 change_branch, change_branch, path);
1300 } else {
1301 if (!old_path) {
1302 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1303 "and %s in %s. Version %s of %s left in tree at %s."),
1304 change, path, delete_branch, change_past,
1305 change_branch, change_branch, path, alt_path);
1306 } else {
1307 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1308 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1309 change, old_path, delete_branch, change_past, path,
1310 change_branch, change_branch, path, alt_path);
1314 * No need to call update_file() on path when change_branch ==
1315 * o->branch1 && !alt_path, since that would needlessly touch
1316 * path. We could call update_file_flags() with update_cache=0
1317 * and update_wd=0, but that's a no-op.
1319 if (change_branch != o->branch1 || alt_path)
1320 ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1322 free(alt_path);
1324 return ret;
1327 static int conflict_rename_delete(struct merge_options *o,
1328 struct diff_filepair *pair,
1329 const char *rename_branch,
1330 const char *delete_branch)
1332 const struct diff_filespec *orig = pair->one;
1333 const struct diff_filespec *dest = pair->two;
1335 if (handle_change_delete(o,
1336 o->call_depth ? orig->path : dest->path,
1337 o->call_depth ? NULL : orig->path,
1338 &orig->oid, orig->mode,
1339 &dest->oid, dest->mode,
1340 rename_branch, delete_branch,
1341 _("rename"), _("renamed")))
1342 return -1;
1344 if (o->call_depth)
1345 return remove_file_from_cache(dest->path);
1346 else
1347 return update_stages(o, dest->path, NULL,
1348 rename_branch == o->branch1 ? dest : NULL,
1349 rename_branch == o->branch1 ? NULL : dest);
1352 static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1353 struct stage_data *entry,
1354 int stage)
1356 struct object_id *oid = &entry->stages[stage].oid;
1357 unsigned mode = entry->stages[stage].mode;
1358 if (mode == 0 || is_null_oid(oid))
1359 return NULL;
1360 oidcpy(&target->oid, oid);
1361 target->mode = mode;
1362 return target;
1365 static int handle_file(struct merge_options *o,
1366 struct diff_filespec *rename,
1367 int stage,
1368 struct rename_conflict_info *ci)
1370 char *dst_name = rename->path;
1371 struct stage_data *dst_entry;
1372 const char *cur_branch, *other_branch;
1373 struct diff_filespec other;
1374 struct diff_filespec *add;
1375 int ret;
1377 if (stage == 2) {
1378 dst_entry = ci->dst_entry1;
1379 cur_branch = ci->branch1;
1380 other_branch = ci->branch2;
1381 } else {
1382 dst_entry = ci->dst_entry2;
1383 cur_branch = ci->branch2;
1384 other_branch = ci->branch1;
1387 add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1388 if (add) {
1389 int ren_src_was_dirty = was_dirty(o, rename->path);
1390 char *add_name = unique_path(o, rename->path, other_branch);
1391 if (update_file(o, 0, &add->oid, add->mode, add_name))
1392 return -1;
1394 if (ren_src_was_dirty) {
1395 output(o, 1, _("Refusing to lose dirty file at %s"),
1396 rename->path);
1399 * Because the double negatives somehow keep confusing me...
1400 * 1) update_wd iff !ren_src_was_dirty.
1401 * 2) no_wd iff !update_wd
1402 * 3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty
1404 remove_file(o, 0, rename->path, ren_src_was_dirty);
1405 dst_name = unique_path(o, rename->path, cur_branch);
1406 } else {
1407 if (dir_in_way(rename->path, !o->call_depth, 0)) {
1408 dst_name = unique_path(o, rename->path, cur_branch);
1409 output(o, 1, _("%s is a directory in %s adding as %s instead"),
1410 rename->path, other_branch, dst_name);
1411 } else if (!o->call_depth &&
1412 would_lose_untracked(rename->path)) {
1413 dst_name = unique_path(o, rename->path, cur_branch);
1414 output(o, 1, _("Refusing to lose untracked file at %s; "
1415 "adding as %s instead"),
1416 rename->path, dst_name);
1419 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1420 ; /* fall through, do allow dst_name to be released */
1421 else if (stage == 2)
1422 ret = update_stages(o, rename->path, NULL, rename, add);
1423 else
1424 ret = update_stages(o, rename->path, NULL, add, rename);
1426 if (dst_name != rename->path)
1427 free(dst_name);
1429 return ret;
1432 static int conflict_rename_rename_1to2(struct merge_options *o,
1433 struct rename_conflict_info *ci)
1435 /* One file was renamed in both branches, but to different names. */
1436 struct diff_filespec *one = ci->pair1->one;
1437 struct diff_filespec *a = ci->pair1->two;
1438 struct diff_filespec *b = ci->pair2->two;
1440 output(o, 1, _("CONFLICT (rename/rename): "
1441 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1442 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1443 one->path, a->path, ci->branch1,
1444 one->path, b->path, ci->branch2,
1445 o->call_depth ? _(" (left unresolved)") : "");
1446 if (o->call_depth) {
1447 struct merge_file_info mfi;
1448 struct diff_filespec other;
1449 struct diff_filespec *add;
1450 if (merge_file_one(o, one->path,
1451 &one->oid, one->mode,
1452 &a->oid, a->mode,
1453 &b->oid, b->mode,
1454 ci->branch1, ci->branch2, &mfi))
1455 return -1;
1458 * FIXME: For rename/add-source conflicts (if we could detect
1459 * such), this is wrong. We should instead find a unique
1460 * pathname and then either rename the add-source file to that
1461 * unique path, or use that unique path instead of src here.
1463 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1464 return -1;
1467 * Above, we put the merged content at the merge-base's
1468 * path. Now we usually need to delete both a->path and
1469 * b->path. However, the rename on each side of the merge
1470 * could also be involved in a rename/add conflict. In
1471 * such cases, we should keep the added file around,
1472 * resolving the conflict at that path in its favor.
1474 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1475 if (add) {
1476 if (update_file(o, 0, &add->oid, add->mode, a->path))
1477 return -1;
1479 else
1480 remove_file_from_cache(a->path);
1481 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1482 if (add) {
1483 if (update_file(o, 0, &add->oid, add->mode, b->path))
1484 return -1;
1486 else
1487 remove_file_from_cache(b->path);
1488 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1489 return -1;
1491 return 0;
1494 static int conflict_rename_rename_2to1(struct merge_options *o,
1495 struct rename_conflict_info *ci)
1497 /* Two files, a & b, were renamed to the same thing, c. */
1498 struct diff_filespec *a = ci->pair1->one;
1499 struct diff_filespec *b = ci->pair2->one;
1500 struct diff_filespec *c1 = ci->pair1->two;
1501 struct diff_filespec *c2 = ci->pair2->two;
1502 char *path = c1->path; /* == c2->path */
1503 char *path_side_1_desc;
1504 char *path_side_2_desc;
1505 struct merge_file_info mfi_c1;
1506 struct merge_file_info mfi_c2;
1507 int ret;
1509 output(o, 1, _("CONFLICT (rename/rename): "
1510 "Rename %s->%s in %s. "
1511 "Rename %s->%s in %s"),
1512 a->path, c1->path, ci->branch1,
1513 b->path, c2->path, ci->branch2);
1515 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1516 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1518 path_side_1_desc = xstrfmt("%s (was %s)", path, a->path);
1519 path_side_2_desc = xstrfmt("%s (was %s)", path, b->path);
1520 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1521 path_side_1_desc,
1522 o->branch1, c1->path,
1523 o->branch2, ci->ren1_other.path, &mfi_c1) ||
1524 merge_file_special_markers(o, b, &ci->ren2_other, c2,
1525 path_side_2_desc,
1526 o->branch1, ci->ren2_other.path,
1527 o->branch2, c2->path, &mfi_c2))
1528 return -1;
1529 free(path_side_1_desc);
1530 free(path_side_2_desc);
1532 if (o->call_depth) {
1534 * If mfi_c1.clean && mfi_c2.clean, then it might make
1535 * sense to do a two-way merge of those results. But, I
1536 * think in all cases, it makes sense to have the virtual
1537 * merge base just undo the renames; they can be detected
1538 * again later for the non-recursive merge.
1540 remove_file(o, 0, path, 0);
1541 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1542 if (!ret)
1543 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1544 b->path);
1545 } else {
1546 char *new_path1 = unique_path(o, path, ci->branch1);
1547 char *new_path2 = unique_path(o, path, ci->branch2);
1548 output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1549 a->path, new_path1, b->path, new_path2);
1550 if (was_dirty(o, path))
1551 output(o, 1, _("Refusing to lose dirty file at %s"),
1552 path);
1553 else if (would_lose_untracked(path))
1555 * Only way we get here is if both renames were from
1556 * a directory rename AND user had an untracked file
1557 * at the location where both files end up after the
1558 * two directory renames. See testcase 10d of t6043.
1560 output(o, 1, _("Refusing to lose untracked file at "
1561 "%s, even though it's in the way."),
1562 path);
1563 else
1564 remove_file(o, 0, path, 0);
1565 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1566 if (!ret)
1567 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1568 new_path2);
1570 * unpack_trees() actually populates the index for us for
1571 * "normal" rename/rename(2to1) situtations so that the
1572 * correct entries are at the higher stages, which would
1573 * make the call below to update_stages_for_stage_data
1574 * unnecessary. However, if either of the renames came
1575 * from a directory rename, then unpack_trees() will not
1576 * have gotten the right data loaded into the index, so we
1577 * need to do so now. (While it'd be tempting to move this
1578 * call to update_stages_for_stage_data() to
1579 * apply_directory_rename_modifications(), that would break
1580 * our intermediate calls to would_lose_untracked() since
1581 * those rely on the current in-memory index. See also the
1582 * big "NOTE" in update_stages()).
1584 if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1585 ret = -1;
1587 free(new_path2);
1588 free(new_path1);
1591 return ret;
1595 * Get the diff_filepairs changed between o_tree and tree.
1597 static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1598 struct tree *o_tree,
1599 struct tree *tree)
1601 struct diff_queue_struct *ret;
1602 struct diff_options opts;
1604 diff_setup(&opts);
1605 opts.flags.recursive = 1;
1606 opts.flags.rename_empty = 0;
1607 opts.detect_rename = merge_detect_rename(o);
1609 * We do not have logic to handle the detection of copies. In
1610 * fact, it may not even make sense to add such logic: would we
1611 * really want a change to a base file to be propagated through
1612 * multiple other files by a merge?
1614 if (opts.detect_rename > DIFF_DETECT_RENAME)
1615 opts.detect_rename = DIFF_DETECT_RENAME;
1616 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1617 o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1618 1000;
1619 opts.rename_score = o->rename_score;
1620 opts.show_rename_progress = o->show_rename_progress;
1621 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1622 diff_setup_done(&opts);
1623 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1624 diffcore_std(&opts);
1625 if (opts.needed_rename_limit > o->needed_rename_limit)
1626 o->needed_rename_limit = opts.needed_rename_limit;
1628 ret = xmalloc(sizeof(*ret));
1629 *ret = diff_queued_diff;
1631 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1632 diff_queued_diff.nr = 0;
1633 diff_queued_diff.queue = NULL;
1634 diff_flush(&opts);
1635 return ret;
1638 static int tree_has_path(struct tree *tree, const char *path)
1640 struct object_id hashy;
1641 unsigned int mode_o;
1643 return !get_tree_entry(&tree->object.oid, path,
1644 &hashy, &mode_o);
1648 * Return a new string that replaces the beginning portion (which matches
1649 * entry->dir), with entry->new_dir. In perl-speak:
1650 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1651 * NOTE:
1652 * Caller must ensure that old_path starts with entry->dir + '/'.
1654 static char *apply_dir_rename(struct dir_rename_entry *entry,
1655 const char *old_path)
1657 struct strbuf new_path = STRBUF_INIT;
1658 int oldlen, newlen;
1660 if (entry->non_unique_new_dir)
1661 return NULL;
1663 oldlen = strlen(entry->dir);
1664 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1665 strbuf_grow(&new_path, newlen);
1666 strbuf_addbuf(&new_path, &entry->new_dir);
1667 strbuf_addstr(&new_path, &old_path[oldlen]);
1669 return strbuf_detach(&new_path, NULL);
1672 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1673 char **old_dir, char **new_dir)
1675 char *end_of_old, *end_of_new;
1676 int old_len, new_len;
1678 *old_dir = NULL;
1679 *new_dir = NULL;
1682 * For
1683 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1684 * the "e/foo.c" part is the same, we just want to know that
1685 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1686 * so, for this example, this function returns "a/b/c/d" in
1687 * *old_dir and "a/b/some/thing/else" in *new_dir.
1689 * Also, if the basename of the file changed, we don't care. We
1690 * want to know which portion of the directory, if any, changed.
1692 end_of_old = strrchr(old_path, '/');
1693 end_of_new = strrchr(new_path, '/');
1695 if (end_of_old == NULL || end_of_new == NULL)
1696 return;
1697 while (*--end_of_new == *--end_of_old &&
1698 end_of_old != old_path &&
1699 end_of_new != new_path)
1700 ; /* Do nothing; all in the while loop */
1702 * We've found the first non-matching character in the directory
1703 * paths. That means the current directory we were comparing
1704 * represents the rename. Move end_of_old and end_of_new back
1705 * to the full directory name.
1707 if (*end_of_old == '/')
1708 end_of_old++;
1709 if (*end_of_old != '/')
1710 end_of_new++;
1711 end_of_old = strchr(end_of_old, '/');
1712 end_of_new = strchr(end_of_new, '/');
1715 * It may have been the case that old_path and new_path were the same
1716 * directory all along. Don't claim a rename if they're the same.
1718 old_len = end_of_old - old_path;
1719 new_len = end_of_new - new_path;
1721 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1722 *old_dir = xstrndup(old_path, old_len);
1723 *new_dir = xstrndup(new_path, new_len);
1727 static void remove_hashmap_entries(struct hashmap *dir_renames,
1728 struct string_list *items_to_remove)
1730 int i;
1731 struct dir_rename_entry *entry;
1733 for (i = 0; i < items_to_remove->nr; i++) {
1734 entry = items_to_remove->items[i].util;
1735 hashmap_remove(dir_renames, entry, NULL);
1737 string_list_clear(items_to_remove, 0);
1741 * See if there is a directory rename for path, and if there are any file
1742 * level conflicts for the renamed location. If there is a rename and
1743 * there are no conflicts, return the new name. Otherwise, return NULL.
1745 static char *handle_path_level_conflicts(struct merge_options *o,
1746 const char *path,
1747 struct dir_rename_entry *entry,
1748 struct hashmap *collisions,
1749 struct tree *tree)
1751 char *new_path = NULL;
1752 struct collision_entry *collision_ent;
1753 int clean = 1;
1754 struct strbuf collision_paths = STRBUF_INIT;
1757 * entry has the mapping of old directory name to new directory name
1758 * that we want to apply to path.
1760 new_path = apply_dir_rename(entry, path);
1762 if (!new_path) {
1763 /* This should only happen when entry->non_unique_new_dir set */
1764 if (!entry->non_unique_new_dir)
1765 BUG("entry->non_unqiue_dir not set and !new_path");
1766 output(o, 1, _("CONFLICT (directory rename split): "
1767 "Unclear where to place %s because directory "
1768 "%s was renamed to multiple other directories, "
1769 "with no destination getting a majority of the "
1770 "files."),
1771 path, entry->dir);
1772 clean = 0;
1773 return NULL;
1777 * The caller needs to have ensured that it has pre-populated
1778 * collisions with all paths that map to new_path. Do a quick check
1779 * to ensure that's the case.
1781 collision_ent = collision_find_entry(collisions, new_path);
1782 if (collision_ent == NULL)
1783 BUG("collision_ent is NULL");
1786 * Check for one-sided add/add/.../add conflicts, i.e.
1787 * where implicit renames from the other side doing
1788 * directory rename(s) can affect this side of history
1789 * to put multiple paths into the same location. Warn
1790 * and bail on directory renames for such paths.
1792 if (collision_ent->reported_already) {
1793 clean = 0;
1794 } else if (tree_has_path(tree, new_path)) {
1795 collision_ent->reported_already = 1;
1796 strbuf_add_separated_string_list(&collision_paths, ", ",
1797 &collision_ent->source_files);
1798 output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1799 "file/dir at %s in the way of implicit "
1800 "directory rename(s) putting the following "
1801 "path(s) there: %s."),
1802 new_path, collision_paths.buf);
1803 clean = 0;
1804 } else if (collision_ent->source_files.nr > 1) {
1805 collision_ent->reported_already = 1;
1806 strbuf_add_separated_string_list(&collision_paths, ", ",
1807 &collision_ent->source_files);
1808 output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1809 "more than one path to %s; implicit directory "
1810 "renames tried to put these paths there: %s"),
1811 new_path, collision_paths.buf);
1812 clean = 0;
1815 /* Free memory we no longer need */
1816 strbuf_release(&collision_paths);
1817 if (!clean && new_path) {
1818 free(new_path);
1819 return NULL;
1822 return new_path;
1826 * There are a couple things we want to do at the directory level:
1827 * 1. Check for both sides renaming to the same thing, in order to avoid
1828 * implicit renaming of files that should be left in place. (See
1829 * testcase 6b in t6043 for details.)
1830 * 2. Prune directory renames if there are still files left in the
1831 * the original directory. These represent a partial directory rename,
1832 * i.e. a rename where only some of the files within the directory
1833 * were renamed elsewhere. (Technically, this could be done earlier
1834 * in get_directory_renames(), except that would prevent us from
1835 * doing the previous check and thus failing testcase 6b.)
1836 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
1837 * In the future, we could potentially record this info as well and
1838 * omit reporting rename/rename(1to2) conflicts for each path within
1839 * the affected directories, thus cleaning up the merge output.
1840 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1841 * directory level, because merging directories is fine. If it
1842 * causes conflicts for files within those merged directories, then
1843 * that should be detected at the individual path level.
1845 static void handle_directory_level_conflicts(struct merge_options *o,
1846 struct hashmap *dir_re_head,
1847 struct tree *head,
1848 struct hashmap *dir_re_merge,
1849 struct tree *merge)
1851 struct hashmap_iter iter;
1852 struct dir_rename_entry *head_ent;
1853 struct dir_rename_entry *merge_ent;
1855 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1856 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1858 hashmap_iter_init(dir_re_head, &iter);
1859 while ((head_ent = hashmap_iter_next(&iter))) {
1860 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1861 if (merge_ent &&
1862 !head_ent->non_unique_new_dir &&
1863 !merge_ent->non_unique_new_dir &&
1864 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1865 /* 1. Renamed identically; remove it from both sides */
1866 string_list_append(&remove_from_head,
1867 head_ent->dir)->util = head_ent;
1868 strbuf_release(&head_ent->new_dir);
1869 string_list_append(&remove_from_merge,
1870 merge_ent->dir)->util = merge_ent;
1871 strbuf_release(&merge_ent->new_dir);
1872 } else if (tree_has_path(head, head_ent->dir)) {
1873 /* 2. This wasn't a directory rename after all */
1874 string_list_append(&remove_from_head,
1875 head_ent->dir)->util = head_ent;
1876 strbuf_release(&head_ent->new_dir);
1880 remove_hashmap_entries(dir_re_head, &remove_from_head);
1881 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1883 hashmap_iter_init(dir_re_merge, &iter);
1884 while ((merge_ent = hashmap_iter_next(&iter))) {
1885 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1886 if (tree_has_path(merge, merge_ent->dir)) {
1887 /* 2. This wasn't a directory rename after all */
1888 string_list_append(&remove_from_merge,
1889 merge_ent->dir)->util = merge_ent;
1890 } else if (head_ent &&
1891 !head_ent->non_unique_new_dir &&
1892 !merge_ent->non_unique_new_dir) {
1893 /* 3. rename/rename(1to2) */
1895 * We can assume it's not rename/rename(1to1) because
1896 * that was case (1), already checked above. So we
1897 * know that head_ent->new_dir and merge_ent->new_dir
1898 * are different strings.
1900 output(o, 1, _("CONFLICT (rename/rename): "
1901 "Rename directory %s->%s in %s. "
1902 "Rename directory %s->%s in %s"),
1903 head_ent->dir, head_ent->new_dir.buf, o->branch1,
1904 head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1905 string_list_append(&remove_from_head,
1906 head_ent->dir)->util = head_ent;
1907 strbuf_release(&head_ent->new_dir);
1908 string_list_append(&remove_from_merge,
1909 merge_ent->dir)->util = merge_ent;
1910 strbuf_release(&merge_ent->new_dir);
1914 remove_hashmap_entries(dir_re_head, &remove_from_head);
1915 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1918 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1919 struct tree *tree)
1921 struct hashmap *dir_renames;
1922 struct hashmap_iter iter;
1923 struct dir_rename_entry *entry;
1924 int i;
1927 * Typically, we think of a directory rename as all files from a
1928 * certain directory being moved to a target directory. However,
1929 * what if someone first moved two files from the original
1930 * directory in one commit, and then renamed the directory
1931 * somewhere else in a later commit? At merge time, we just know
1932 * that files from the original directory went to two different
1933 * places, and that the bulk of them ended up in the same place.
1934 * We want each directory rename to represent where the bulk of the
1935 * files from that directory end up; this function exists to find
1936 * where the bulk of the files went.
1938 * The first loop below simply iterates through the list of file
1939 * renames, finding out how often each directory rename pair
1940 * possibility occurs.
1942 dir_renames = xmalloc(sizeof(*dir_renames));
1943 dir_rename_init(dir_renames);
1944 for (i = 0; i < pairs->nr; ++i) {
1945 struct string_list_item *item;
1946 int *count;
1947 struct diff_filepair *pair = pairs->queue[i];
1948 char *old_dir, *new_dir;
1950 /* File not part of directory rename if it wasn't renamed */
1951 if (pair->status != 'R')
1952 continue;
1954 get_renamed_dir_portion(pair->one->path, pair->two->path,
1955 &old_dir, &new_dir);
1956 if (!old_dir)
1957 /* Directory didn't change at all; ignore this one. */
1958 continue;
1960 entry = dir_rename_find_entry(dir_renames, old_dir);
1961 if (!entry) {
1962 entry = xmalloc(sizeof(*entry));
1963 dir_rename_entry_init(entry, old_dir);
1964 hashmap_put(dir_renames, entry);
1965 } else {
1966 free(old_dir);
1968 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1969 if (!item) {
1970 item = string_list_insert(&entry->possible_new_dirs,
1971 new_dir);
1972 item->util = xcalloc(1, sizeof(int));
1973 } else {
1974 free(new_dir);
1976 count = item->util;
1977 *count += 1;
1981 * For each directory with files moved out of it, we find out which
1982 * target directory received the most files so we can declare it to
1983 * be the "winning" target location for the directory rename. This
1984 * winner gets recorded in new_dir. If there is no winner
1985 * (multiple target directories received the same number of files),
1986 * we set non_unique_new_dir. Once we've determined the winner (or
1987 * that there is no winner), we no longer need possible_new_dirs.
1989 hashmap_iter_init(dir_renames, &iter);
1990 while ((entry = hashmap_iter_next(&iter))) {
1991 int max = 0;
1992 int bad_max = 0;
1993 char *best = NULL;
1995 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1996 int *count = entry->possible_new_dirs.items[i].util;
1998 if (*count == max)
1999 bad_max = max;
2000 else if (*count > max) {
2001 max = *count;
2002 best = entry->possible_new_dirs.items[i].string;
2005 if (bad_max == max)
2006 entry->non_unique_new_dir = 1;
2007 else {
2008 assert(entry->new_dir.len == 0);
2009 strbuf_addstr(&entry->new_dir, best);
2012 * The relevant directory sub-portion of the original full
2013 * filepaths were xstrndup'ed before inserting into
2014 * possible_new_dirs, and instead of manually iterating the
2015 * list and free'ing each, just lie and tell
2016 * possible_new_dirs that it did the strdup'ing so that it
2017 * will free them for us.
2019 entry->possible_new_dirs.strdup_strings = 1;
2020 string_list_clear(&entry->possible_new_dirs, 1);
2023 return dir_renames;
2026 static struct dir_rename_entry *check_dir_renamed(const char *path,
2027 struct hashmap *dir_renames)
2029 char temp[PATH_MAX];
2030 char *end;
2031 struct dir_rename_entry *entry;
2033 strcpy(temp, path);
2034 while ((end = strrchr(temp, '/'))) {
2035 *end = '\0';
2036 entry = dir_rename_find_entry(dir_renames, temp);
2037 if (entry)
2038 return entry;
2040 return NULL;
2043 static void compute_collisions(struct hashmap *collisions,
2044 struct hashmap *dir_renames,
2045 struct diff_queue_struct *pairs)
2047 int i;
2050 * Multiple files can be mapped to the same path due to directory
2051 * renames done by the other side of history. Since that other
2052 * side of history could have merged multiple directories into one,
2053 * if our side of history added the same file basename to each of
2054 * those directories, then all N of them would get implicitly
2055 * renamed by the directory rename detection into the same path,
2056 * and we'd get an add/add/.../add conflict, and all those adds
2057 * from *this* side of history. This is not representable in the
2058 * index, and users aren't going to easily be able to make sense of
2059 * it. So we need to provide a good warning about what's
2060 * happening, and fall back to no-directory-rename detection
2061 * behavior for those paths.
2063 * See testcases 9e and all of section 5 from t6043 for examples.
2065 collision_init(collisions);
2067 for (i = 0; i < pairs->nr; ++i) {
2068 struct dir_rename_entry *dir_rename_ent;
2069 struct collision_entry *collision_ent;
2070 char *new_path;
2071 struct diff_filepair *pair = pairs->queue[i];
2073 if (pair->status != 'A' && pair->status != 'R')
2074 continue;
2075 dir_rename_ent = check_dir_renamed(pair->two->path,
2076 dir_renames);
2077 if (!dir_rename_ent)
2078 continue;
2080 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2081 if (!new_path)
2083 * dir_rename_ent->non_unique_new_path is true, which
2084 * means there is no directory rename for us to use,
2085 * which means it won't cause us any additional
2086 * collisions.
2088 continue;
2089 collision_ent = collision_find_entry(collisions, new_path);
2090 if (!collision_ent) {
2091 collision_ent = xcalloc(1,
2092 sizeof(struct collision_entry));
2093 hashmap_entry_init(collision_ent, strhash(new_path));
2094 hashmap_put(collisions, collision_ent);
2095 collision_ent->target_file = new_path;
2096 } else {
2097 free(new_path);
2099 string_list_insert(&collision_ent->source_files,
2100 pair->two->path);
2104 static char *check_for_directory_rename(struct merge_options *o,
2105 const char *path,
2106 struct tree *tree,
2107 struct hashmap *dir_renames,
2108 struct hashmap *dir_rename_exclusions,
2109 struct hashmap *collisions,
2110 int *clean_merge)
2112 char *new_path = NULL;
2113 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2114 struct dir_rename_entry *oentry = NULL;
2116 if (!entry)
2117 return new_path;
2120 * This next part is a little weird. We do not want to do an
2121 * implicit rename into a directory we renamed on our side, because
2122 * that will result in a spurious rename/rename(1to2) conflict. An
2123 * example:
2124 * Base commit: dumbdir/afile, otherdir/bfile
2125 * Side 1: smrtdir/afile, otherdir/bfile
2126 * Side 2: dumbdir/afile, dumbdir/bfile
2127 * Here, while working on Side 1, we could notice that otherdir was
2128 * renamed/merged to dumbdir, and change the diff_filepair for
2129 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2130 * 2 will notice the rename from dumbdir to smrtdir, and do the
2131 * transitive rename to move it from dumbdir/bfile to
2132 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2133 * smrtdir, a rename/rename(1to2) conflict. We really just want
2134 * the file to end up in smrtdir. And the way to achieve that is
2135 * to not let Side1 do the rename to dumbdir, since we know that is
2136 * the source of one of our directory renames.
2138 * That's why oentry and dir_rename_exclusions is here.
2140 * As it turns out, this also prevents N-way transient rename
2141 * confusion; See testcases 9c and 9d of t6043.
2143 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2144 if (oentry) {
2145 output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2146 "to %s, because %s itself was renamed."),
2147 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2148 } else {
2149 new_path = handle_path_level_conflicts(o, path, entry,
2150 collisions, tree);
2151 *clean_merge &= (new_path != NULL);
2154 return new_path;
2157 static void apply_directory_rename_modifications(struct merge_options *o,
2158 struct diff_filepair *pair,
2159 char *new_path,
2160 struct rename *re,
2161 struct tree *tree,
2162 struct tree *o_tree,
2163 struct tree *a_tree,
2164 struct tree *b_tree,
2165 struct string_list *entries,
2166 int *clean)
2168 struct string_list_item *item;
2169 int stage = (tree == a_tree ? 2 : 3);
2170 int update_wd;
2173 * In all cases where we can do directory rename detection,
2174 * unpack_trees() will have read pair->two->path into the
2175 * index and the working copy. We need to remove it so that
2176 * we can instead place it at new_path. It is guaranteed to
2177 * not be untracked (unpack_trees() would have errored out
2178 * saying the file would have been overwritten), but it might
2179 * be dirty, though.
2181 update_wd = !was_dirty(o, pair->two->path);
2182 if (!update_wd)
2183 output(o, 1, _("Refusing to lose dirty file at %s"),
2184 pair->two->path);
2185 remove_file(o, 1, pair->two->path, !update_wd);
2187 /* Find or create a new re->dst_entry */
2188 item = string_list_lookup(entries, new_path);
2189 if (item) {
2191 * Since we're renaming on this side of history, and it's
2192 * due to a directory rename on the other side of history
2193 * (which we only allow when the directory in question no
2194 * longer exists on the other side of history), the
2195 * original entry for re->dst_entry is no longer
2196 * necessary...
2198 re->dst_entry->processed = 1;
2201 * ...because we'll be using this new one.
2203 re->dst_entry = item->util;
2204 } else {
2206 * re->dst_entry is for the before-dir-rename path, and we
2207 * need it to hold information for the after-dir-rename
2208 * path. Before creating a new entry, we need to mark the
2209 * old one as unnecessary (...unless it is shared by
2210 * src_entry, i.e. this didn't use to be a rename, in which
2211 * case we can just allow the normal processing to happen
2212 * for it).
2214 if (pair->status == 'R')
2215 re->dst_entry->processed = 1;
2217 re->dst_entry = insert_stage_data(new_path,
2218 o_tree, a_tree, b_tree,
2219 entries);
2220 item = string_list_insert(entries, new_path);
2221 item->util = re->dst_entry;
2225 * Update the stage_data with the information about the path we are
2226 * moving into place. That slot will be empty and available for us
2227 * to write to because of the collision checks in
2228 * handle_path_level_conflicts(). In other words,
2229 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2230 * open for us to write to.
2232 * It may be tempting to actually update the index at this point as
2233 * well, using update_stages_for_stage_data(), but as per the big
2234 * "NOTE" in update_stages(), doing so will modify the current
2235 * in-memory index which will break calls to would_lose_untracked()
2236 * that we need to make. Instead, we need to just make sure that
2237 * the various conflict_rename_*() functions update the index
2238 * explicitly rather than relying on unpack_trees() to have done it.
2240 get_tree_entry(&tree->object.oid,
2241 pair->two->path,
2242 &re->dst_entry->stages[stage].oid,
2243 &re->dst_entry->stages[stage].mode);
2245 /* Update pair status */
2246 if (pair->status == 'A') {
2248 * Recording rename information for this add makes it look
2249 * like a rename/delete conflict. Make sure we can
2250 * correctly handle this as an add that was moved to a new
2251 * directory instead of reporting a rename/delete conflict.
2253 re->add_turned_into_rename = 1;
2256 * We don't actually look at pair->status again, but it seems
2257 * pedagogically correct to adjust it.
2259 pair->status = 'R';
2262 * Finally, record the new location.
2264 pair->two->path = new_path;
2268 * Get information of all renames which occurred in 'pairs', making use of
2269 * any implicit directory renames inferred from the other side of history.
2270 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2271 * to be able to associate the correct cache entries with the rename
2272 * information; tree is always equal to either a_tree or b_tree.
2274 static struct string_list *get_renames(struct merge_options *o,
2275 struct diff_queue_struct *pairs,
2276 struct hashmap *dir_renames,
2277 struct hashmap *dir_rename_exclusions,
2278 struct tree *tree,
2279 struct tree *o_tree,
2280 struct tree *a_tree,
2281 struct tree *b_tree,
2282 struct string_list *entries,
2283 int *clean_merge)
2285 int i;
2286 struct hashmap collisions;
2287 struct hashmap_iter iter;
2288 struct collision_entry *e;
2289 struct string_list *renames;
2291 compute_collisions(&collisions, dir_renames, pairs);
2292 renames = xcalloc(1, sizeof(struct string_list));
2294 for (i = 0; i < pairs->nr; ++i) {
2295 struct string_list_item *item;
2296 struct rename *re;
2297 struct diff_filepair *pair = pairs->queue[i];
2298 char *new_path; /* non-NULL only with directory renames */
2300 if (pair->status != 'A' && pair->status != 'R') {
2301 diff_free_filepair(pair);
2302 continue;
2304 new_path = check_for_directory_rename(o, pair->two->path, tree,
2305 dir_renames,
2306 dir_rename_exclusions,
2307 &collisions,
2308 clean_merge);
2309 if (pair->status != 'R' && !new_path) {
2310 diff_free_filepair(pair);
2311 continue;
2314 re = xmalloc(sizeof(*re));
2315 re->processed = 0;
2316 re->add_turned_into_rename = 0;
2317 re->pair = pair;
2318 item = string_list_lookup(entries, re->pair->one->path);
2319 if (!item)
2320 re->src_entry = insert_stage_data(re->pair->one->path,
2321 o_tree, a_tree, b_tree, entries);
2322 else
2323 re->src_entry = item->util;
2325 item = string_list_lookup(entries, re->pair->two->path);
2326 if (!item)
2327 re->dst_entry = insert_stage_data(re->pair->two->path,
2328 o_tree, a_tree, b_tree, entries);
2329 else
2330 re->dst_entry = item->util;
2331 item = string_list_insert(renames, pair->one->path);
2332 item->util = re;
2333 if (new_path)
2334 apply_directory_rename_modifications(o, pair, new_path,
2335 re, tree, o_tree,
2336 a_tree, b_tree,
2337 entries,
2338 clean_merge);
2341 hashmap_iter_init(&collisions, &iter);
2342 while ((e = hashmap_iter_next(&iter))) {
2343 free(e->target_file);
2344 string_list_clear(&e->source_files, 0);
2346 hashmap_free(&collisions, 1);
2347 return renames;
2350 static int process_renames(struct merge_options *o,
2351 struct string_list *a_renames,
2352 struct string_list *b_renames)
2354 int clean_merge = 1, i, j;
2355 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2356 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2357 const struct rename *sre;
2359 for (i = 0; i < a_renames->nr; i++) {
2360 sre = a_renames->items[i].util;
2361 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2362 = (void *)sre;
2364 for (i = 0; i < b_renames->nr; i++) {
2365 sre = b_renames->items[i].util;
2366 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2367 = (void *)sre;
2370 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2371 struct string_list *renames1, *renames2Dst;
2372 struct rename *ren1 = NULL, *ren2 = NULL;
2373 const char *branch1, *branch2;
2374 const char *ren1_src, *ren1_dst;
2375 struct string_list_item *lookup;
2377 if (i >= a_renames->nr) {
2378 ren2 = b_renames->items[j++].util;
2379 } else if (j >= b_renames->nr) {
2380 ren1 = a_renames->items[i++].util;
2381 } else {
2382 int compare = strcmp(a_renames->items[i].string,
2383 b_renames->items[j].string);
2384 if (compare <= 0)
2385 ren1 = a_renames->items[i++].util;
2386 if (compare >= 0)
2387 ren2 = b_renames->items[j++].util;
2390 /* TODO: refactor, so that 1/2 are not needed */
2391 if (ren1) {
2392 renames1 = a_renames;
2393 renames2Dst = &b_by_dst;
2394 branch1 = o->branch1;
2395 branch2 = o->branch2;
2396 } else {
2397 renames1 = b_renames;
2398 renames2Dst = &a_by_dst;
2399 branch1 = o->branch2;
2400 branch2 = o->branch1;
2401 SWAP(ren2, ren1);
2404 if (ren1->processed)
2405 continue;
2406 ren1->processed = 1;
2407 ren1->dst_entry->processed = 1;
2408 /* BUG: We should only mark src_entry as processed if we
2409 * are not dealing with a rename + add-source case.
2411 ren1->src_entry->processed = 1;
2413 ren1_src = ren1->pair->one->path;
2414 ren1_dst = ren1->pair->two->path;
2416 if (ren2) {
2417 /* One file renamed on both sides */
2418 const char *ren2_src = ren2->pair->one->path;
2419 const char *ren2_dst = ren2->pair->two->path;
2420 enum rename_type rename_type;
2421 if (strcmp(ren1_src, ren2_src) != 0)
2422 BUG("ren1_src != ren2_src");
2423 ren2->dst_entry->processed = 1;
2424 ren2->processed = 1;
2425 if (strcmp(ren1_dst, ren2_dst) != 0) {
2426 rename_type = RENAME_ONE_FILE_TO_TWO;
2427 clean_merge = 0;
2428 } else {
2429 rename_type = RENAME_ONE_FILE_TO_ONE;
2430 /* BUG: We should only remove ren1_src in
2431 * the base stage (think of rename +
2432 * add-source cases).
2434 remove_file(o, 1, ren1_src, 1);
2435 update_entry(ren1->dst_entry,
2436 ren1->pair->one,
2437 ren1->pair->two,
2438 ren2->pair->two);
2440 setup_rename_conflict_info(rename_type,
2441 ren1->pair,
2442 ren2->pair,
2443 branch1,
2444 branch2,
2445 ren1->dst_entry,
2446 ren2->dst_entry,
2448 NULL,
2449 NULL);
2450 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2451 /* Two different files renamed to the same thing */
2452 char *ren2_dst;
2453 ren2 = lookup->util;
2454 ren2_dst = ren2->pair->two->path;
2455 if (strcmp(ren1_dst, ren2_dst) != 0)
2456 BUG("ren1_dst != ren2_dst");
2458 clean_merge = 0;
2459 ren2->processed = 1;
2461 * BUG: We should only mark src_entry as processed
2462 * if we are not dealing with a rename + add-source
2463 * case.
2465 ren2->src_entry->processed = 1;
2467 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2468 ren1->pair,
2469 ren2->pair,
2470 branch1,
2471 branch2,
2472 ren1->dst_entry,
2473 ren2->dst_entry,
2475 ren1->src_entry,
2476 ren2->src_entry);
2478 } else {
2479 /* Renamed in 1, maybe changed in 2 */
2480 /* we only use sha1 and mode of these */
2481 struct diff_filespec src_other, dst_other;
2482 int try_merge;
2485 * unpack_trees loads entries from common-commit
2486 * into stage 1, from head-commit into stage 2, and
2487 * from merge-commit into stage 3. We keep track
2488 * of which side corresponds to the rename.
2490 int renamed_stage = a_renames == renames1 ? 2 : 3;
2491 int other_stage = a_renames == renames1 ? 3 : 2;
2493 /* BUG: We should only remove ren1_src in the base
2494 * stage and in other_stage (think of rename +
2495 * add-source case).
2497 remove_file(o, 1, ren1_src,
2498 renamed_stage == 2 || !was_tracked(o, ren1_src));
2500 oidcpy(&src_other.oid,
2501 &ren1->src_entry->stages[other_stage].oid);
2502 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2503 oidcpy(&dst_other.oid,
2504 &ren1->dst_entry->stages[other_stage].oid);
2505 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2506 try_merge = 0;
2508 if (oid_eq(&src_other.oid, &null_oid) &&
2509 ren1->add_turned_into_rename) {
2510 setup_rename_conflict_info(RENAME_DIR,
2511 ren1->pair,
2512 NULL,
2513 branch1,
2514 branch2,
2515 ren1->dst_entry,
2516 NULL,
2518 NULL,
2519 NULL);
2520 } else if (oid_eq(&src_other.oid, &null_oid)) {
2521 setup_rename_conflict_info(RENAME_DELETE,
2522 ren1->pair,
2523 NULL,
2524 branch1,
2525 branch2,
2526 ren1->dst_entry,
2527 NULL,
2529 NULL,
2530 NULL);
2531 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2532 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2534 * Added file on the other side identical to
2535 * the file being renamed: clean merge.
2536 * Also, there is no need to overwrite the
2537 * file already in the working copy, so call
2538 * update_file_flags() instead of
2539 * update_file().
2541 if (update_file_flags(o,
2542 &ren1->pair->two->oid,
2543 ren1->pair->two->mode,
2544 ren1_dst,
2545 1, /* update_cache */
2546 0 /* update_wd */))
2547 clean_merge = -1;
2548 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2549 clean_merge = 0;
2550 try_merge = 1;
2551 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2552 "%s added in %s"),
2553 ren1_src, ren1_dst, branch1,
2554 ren1_dst, branch2);
2555 if (o->call_depth) {
2556 struct merge_file_info mfi;
2557 if (merge_file_one(o, ren1_dst, &null_oid, 0,
2558 &ren1->pair->two->oid,
2559 ren1->pair->two->mode,
2560 &dst_other.oid,
2561 dst_other.mode,
2562 branch1, branch2, &mfi)) {
2563 clean_merge = -1;
2564 goto cleanup_and_return;
2566 output(o, 1, _("Adding merged %s"), ren1_dst);
2567 if (update_file(o, 0, &mfi.oid,
2568 mfi.mode, ren1_dst))
2569 clean_merge = -1;
2570 try_merge = 0;
2571 } else {
2572 char *new_path = unique_path(o, ren1_dst, branch2);
2573 output(o, 1, _("Adding as %s instead"), new_path);
2574 if (update_file(o, 0, &dst_other.oid,
2575 dst_other.mode, new_path))
2576 clean_merge = -1;
2577 free(new_path);
2579 } else
2580 try_merge = 1;
2582 if (clean_merge < 0)
2583 goto cleanup_and_return;
2584 if (try_merge) {
2585 struct diff_filespec *one, *a, *b;
2586 src_other.path = (char *)ren1_src;
2588 one = ren1->pair->one;
2589 if (a_renames == renames1) {
2590 a = ren1->pair->two;
2591 b = &src_other;
2592 } else {
2593 b = ren1->pair->two;
2594 a = &src_other;
2596 update_entry(ren1->dst_entry, one, a, b);
2597 setup_rename_conflict_info(RENAME_NORMAL,
2598 ren1->pair,
2599 NULL,
2600 branch1,
2601 NULL,
2602 ren1->dst_entry,
2603 NULL,
2605 NULL,
2606 NULL);
2610 cleanup_and_return:
2611 string_list_clear(&a_by_dst, 0);
2612 string_list_clear(&b_by_dst, 0);
2614 return clean_merge;
2617 struct rename_info {
2618 struct string_list *head_renames;
2619 struct string_list *merge_renames;
2622 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2623 struct hashmap *dir_renames)
2625 struct hashmap_iter iter;
2626 struct dir_rename_entry *e;
2628 hashmap_iter_init(dir_renames, &iter);
2629 while ((e = hashmap_iter_next(&iter))) {
2630 free(e->dir);
2631 strbuf_release(&e->new_dir);
2632 /* possible_new_dirs already cleared in get_directory_renames */
2634 hashmap_free(dir_renames, 1);
2635 free(dir_renames);
2637 free(pairs->queue);
2638 free(pairs);
2641 static int handle_renames(struct merge_options *o,
2642 struct tree *common,
2643 struct tree *head,
2644 struct tree *merge,
2645 struct string_list *entries,
2646 struct rename_info *ri)
2648 struct diff_queue_struct *head_pairs, *merge_pairs;
2649 struct hashmap *dir_re_head, *dir_re_merge;
2650 int clean = 1;
2652 ri->head_renames = NULL;
2653 ri->merge_renames = NULL;
2655 if (!merge_detect_rename(o))
2656 return 1;
2658 head_pairs = get_diffpairs(o, common, head);
2659 merge_pairs = get_diffpairs(o, common, merge);
2661 dir_re_head = get_directory_renames(head_pairs, head);
2662 dir_re_merge = get_directory_renames(merge_pairs, merge);
2664 handle_directory_level_conflicts(o,
2665 dir_re_head, head,
2666 dir_re_merge, merge);
2668 ri->head_renames = get_renames(o, head_pairs,
2669 dir_re_merge, dir_re_head, head,
2670 common, head, merge, entries,
2671 &clean);
2672 if (clean < 0)
2673 goto cleanup;
2674 ri->merge_renames = get_renames(o, merge_pairs,
2675 dir_re_head, dir_re_merge, merge,
2676 common, head, merge, entries,
2677 &clean);
2678 if (clean < 0)
2679 goto cleanup;
2680 clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2682 cleanup:
2684 * Some cleanup is deferred until cleanup_renames() because the
2685 * data structures are still needed and referenced in
2686 * process_entry(). But there are a few things we can free now.
2688 initial_cleanup_rename(head_pairs, dir_re_head);
2689 initial_cleanup_rename(merge_pairs, dir_re_merge);
2691 return clean;
2694 static void final_cleanup_rename(struct string_list *rename)
2696 const struct rename *re;
2697 int i;
2699 if (rename == NULL)
2700 return;
2702 for (i = 0; i < rename->nr; i++) {
2703 re = rename->items[i].util;
2704 diff_free_filepair(re->pair);
2706 string_list_clear(rename, 1);
2707 free(rename);
2710 static void final_cleanup_renames(struct rename_info *re_info)
2712 final_cleanup_rename(re_info->head_renames);
2713 final_cleanup_rename(re_info->merge_renames);
2716 static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2718 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2721 static int read_oid_strbuf(struct merge_options *o,
2722 const struct object_id *oid, struct strbuf *dst)
2724 void *buf;
2725 enum object_type type;
2726 unsigned long size;
2727 buf = read_object_file(oid, &type, &size);
2728 if (!buf)
2729 return err(o, _("cannot read object %s"), oid_to_hex(oid));
2730 if (type != OBJ_BLOB) {
2731 free(buf);
2732 return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2734 strbuf_attach(dst, buf, size, size + 1);
2735 return 0;
2738 static int blob_unchanged(struct merge_options *opt,
2739 const struct object_id *o_oid,
2740 unsigned o_mode,
2741 const struct object_id *a_oid,
2742 unsigned a_mode,
2743 int renormalize, const char *path)
2745 struct strbuf o = STRBUF_INIT;
2746 struct strbuf a = STRBUF_INIT;
2747 int ret = 0; /* assume changed for safety */
2749 if (a_mode != o_mode)
2750 return 0;
2751 if (oid_eq(o_oid, a_oid))
2752 return 1;
2753 if (!renormalize)
2754 return 0;
2756 assert(o_oid && a_oid);
2757 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2758 goto error_return;
2760 * Note: binary | is used so that both renormalizations are
2761 * performed. Comparison can be skipped if both files are
2762 * unchanged since their sha1s have already been compared.
2764 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2765 renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2766 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2768 error_return:
2769 strbuf_release(&o);
2770 strbuf_release(&a);
2771 return ret;
2774 static int handle_modify_delete(struct merge_options *o,
2775 const char *path,
2776 struct object_id *o_oid, int o_mode,
2777 struct object_id *a_oid, int a_mode,
2778 struct object_id *b_oid, int b_mode)
2780 const char *modify_branch, *delete_branch;
2781 struct object_id *changed_oid;
2782 int changed_mode;
2784 if (a_oid) {
2785 modify_branch = o->branch1;
2786 delete_branch = o->branch2;
2787 changed_oid = a_oid;
2788 changed_mode = a_mode;
2789 } else {
2790 modify_branch = o->branch2;
2791 delete_branch = o->branch1;
2792 changed_oid = b_oid;
2793 changed_mode = b_mode;
2796 return handle_change_delete(o,
2797 path, NULL,
2798 o_oid, o_mode,
2799 changed_oid, changed_mode,
2800 modify_branch, delete_branch,
2801 _("modify"), _("modified"));
2804 static int merge_content(struct merge_options *o,
2805 const char *path,
2806 int is_dirty,
2807 struct object_id *o_oid, int o_mode,
2808 struct object_id *a_oid, int a_mode,
2809 struct object_id *b_oid, int b_mode,
2810 struct rename_conflict_info *rename_conflict_info)
2812 const char *reason = _("content");
2813 const char *path1 = NULL, *path2 = NULL;
2814 struct merge_file_info mfi;
2815 struct diff_filespec one, a, b;
2816 unsigned df_conflict_remains = 0;
2818 if (!o_oid) {
2819 reason = _("add/add");
2820 o_oid = (struct object_id *)&null_oid;
2822 one.path = a.path = b.path = (char *)path;
2823 oidcpy(&one.oid, o_oid);
2824 one.mode = o_mode;
2825 oidcpy(&a.oid, a_oid);
2826 a.mode = a_mode;
2827 oidcpy(&b.oid, b_oid);
2828 b.mode = b_mode;
2830 if (rename_conflict_info) {
2831 struct diff_filepair *pair1 = rename_conflict_info->pair1;
2833 path1 = (o->branch1 == rename_conflict_info->branch1) ?
2834 pair1->two->path : pair1->one->path;
2835 /* If rename_conflict_info->pair2 != NULL, we are in
2836 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
2837 * normal rename.
2839 path2 = (rename_conflict_info->pair2 ||
2840 o->branch2 == rename_conflict_info->branch1) ?
2841 pair1->two->path : pair1->one->path;
2843 if (dir_in_way(path, !o->call_depth,
2844 S_ISGITLINK(pair1->two->mode)))
2845 df_conflict_remains = 1;
2847 if (merge_file_special_markers(o, &one, &a, &b, path,
2848 o->branch1, path1,
2849 o->branch2, path2, &mfi))
2850 return -1;
2853 * We can skip updating the working tree file iff:
2854 * a) The merge is clean
2855 * b) The merge matches what was in HEAD (content, mode, pathname)
2856 * c) The target path is usable (i.e. not involved in D/F conflict)
2858 if (mfi.clean &&
2859 was_tracked_and_matches(o, path, &mfi.oid, mfi.mode) &&
2860 !df_conflict_remains) {
2861 output(o, 3, _("Skipped %s (merged same as existing)"), path);
2862 if (add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2863 0, (!o->call_depth && !is_dirty), 0))
2864 return -1;
2865 return mfi.clean;
2868 if (!mfi.clean) {
2869 if (S_ISGITLINK(mfi.mode))
2870 reason = _("submodule");
2871 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2872 reason, path);
2873 if (rename_conflict_info && !df_conflict_remains)
2874 if (update_stages(o, path, &one, &a, &b))
2875 return -1;
2878 if (df_conflict_remains || is_dirty) {
2879 char *new_path;
2880 if (o->call_depth) {
2881 remove_file_from_cache(path);
2882 } else {
2883 if (!mfi.clean) {
2884 if (update_stages(o, path, &one, &a, &b))
2885 return -1;
2886 } else {
2887 int file_from_stage2 = was_tracked(o, path);
2888 struct diff_filespec merged;
2889 oidcpy(&merged.oid, &mfi.oid);
2890 merged.mode = mfi.mode;
2892 if (update_stages(o, path, NULL,
2893 file_from_stage2 ? &merged : NULL,
2894 file_from_stage2 ? NULL : &merged))
2895 return -1;
2899 new_path = unique_path(o, path, rename_conflict_info->branch1);
2900 if (is_dirty) {
2901 output(o, 1, _("Refusing to lose dirty file at %s"),
2902 path);
2904 output(o, 1, _("Adding as %s instead"), new_path);
2905 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2906 free(new_path);
2907 return -1;
2909 free(new_path);
2910 mfi.clean = 0;
2911 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2912 return -1;
2913 return !is_dirty && mfi.clean;
2916 static int conflict_rename_normal(struct merge_options *o,
2917 const char *path,
2918 struct object_id *o_oid, unsigned int o_mode,
2919 struct object_id *a_oid, unsigned int a_mode,
2920 struct object_id *b_oid, unsigned int b_mode,
2921 struct rename_conflict_info *ci)
2923 /* Merge the content and write it out */
2924 return merge_content(o, path, was_dirty(o, path),
2925 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2926 ci);
2929 /* Per entry merge function */
2930 static int process_entry(struct merge_options *o,
2931 const char *path, struct stage_data *entry)
2933 int clean_merge = 1;
2934 int normalize = o->renormalize;
2935 unsigned o_mode = entry->stages[1].mode;
2936 unsigned a_mode = entry->stages[2].mode;
2937 unsigned b_mode = entry->stages[3].mode;
2938 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2939 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2940 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2942 entry->processed = 1;
2943 if (entry->rename_conflict_info) {
2944 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2945 switch (conflict_info->rename_type) {
2946 case RENAME_NORMAL:
2947 case RENAME_ONE_FILE_TO_ONE:
2948 clean_merge = conflict_rename_normal(o,
2949 path,
2950 o_oid, o_mode,
2951 a_oid, a_mode,
2952 b_oid, b_mode,
2953 conflict_info);
2954 break;
2955 case RENAME_DIR:
2956 clean_merge = 1;
2957 if (conflict_rename_dir(o,
2958 conflict_info->pair1,
2959 conflict_info->branch1,
2960 conflict_info->branch2))
2961 clean_merge = -1;
2962 break;
2963 case RENAME_DELETE:
2964 clean_merge = 0;
2965 if (conflict_rename_delete(o,
2966 conflict_info->pair1,
2967 conflict_info->branch1,
2968 conflict_info->branch2))
2969 clean_merge = -1;
2970 break;
2971 case RENAME_ONE_FILE_TO_TWO:
2972 clean_merge = 0;
2973 if (conflict_rename_rename_1to2(o, conflict_info))
2974 clean_merge = -1;
2975 break;
2976 case RENAME_TWO_FILES_TO_ONE:
2977 clean_merge = 0;
2978 if (conflict_rename_rename_2to1(o, conflict_info))
2979 clean_merge = -1;
2980 break;
2981 default:
2982 entry->processed = 0;
2983 break;
2985 } else if (o_oid && (!a_oid || !b_oid)) {
2986 /* Case A: Deleted in one */
2987 if ((!a_oid && !b_oid) ||
2988 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2989 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2990 /* Deleted in both or deleted in one and
2991 * unchanged in the other */
2992 if (a_oid)
2993 output(o, 2, _("Removing %s"), path);
2994 /* do not touch working file if it did not exist */
2995 remove_file(o, 1, path, !a_oid);
2996 } else {
2997 /* Modify/delete; deleted side may have put a directory in the way */
2998 clean_merge = 0;
2999 if (handle_modify_delete(o, path, o_oid, o_mode,
3000 a_oid, a_mode, b_oid, b_mode))
3001 clean_merge = -1;
3003 } else if ((!o_oid && a_oid && !b_oid) ||
3004 (!o_oid && !a_oid && b_oid)) {
3005 /* Case B: Added in one. */
3006 /* [nothing|directory] -> ([nothing|directory], file) */
3008 const char *add_branch;
3009 const char *other_branch;
3010 unsigned mode;
3011 const struct object_id *oid;
3012 const char *conf;
3014 if (a_oid) {
3015 add_branch = o->branch1;
3016 other_branch = o->branch2;
3017 mode = a_mode;
3018 oid = a_oid;
3019 conf = _("file/directory");
3020 } else {
3021 add_branch = o->branch2;
3022 other_branch = o->branch1;
3023 mode = b_mode;
3024 oid = b_oid;
3025 conf = _("directory/file");
3027 if (dir_in_way(path,
3028 !o->call_depth && !S_ISGITLINK(a_mode),
3029 0)) {
3030 char *new_path = unique_path(o, path, add_branch);
3031 clean_merge = 0;
3032 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3033 "Adding %s as %s"),
3034 conf, path, other_branch, path, new_path);
3035 if (update_file(o, 0, oid, mode, new_path))
3036 clean_merge = -1;
3037 else if (o->call_depth)
3038 remove_file_from_cache(path);
3039 free(new_path);
3040 } else {
3041 output(o, 2, _("Adding %s"), path);
3042 /* do not overwrite file if already present */
3043 if (update_file_flags(o, oid, mode, path, 1, !a_oid))
3044 clean_merge = -1;
3046 } else if (a_oid && b_oid) {
3047 /* Case C: Added in both (check for same permissions) and */
3048 /* case D: Modified in both, but differently. */
3049 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3050 clean_merge = merge_content(o, path, is_dirty,
3051 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
3052 NULL);
3053 } else if (!o_oid && !a_oid && !b_oid) {
3055 * this entry was deleted altogether. a_mode == 0 means
3056 * we had that path and want to actively remove it.
3058 remove_file(o, 1, path, !a_mode);
3059 } else
3060 BUG("fatal merge failure, shouldn't happen.");
3062 return clean_merge;
3065 int merge_trees(struct merge_options *o,
3066 struct tree *head,
3067 struct tree *merge,
3068 struct tree *common,
3069 struct tree **result)
3071 int code, clean;
3073 if (o->subtree_shift) {
3074 merge = shift_tree_object(head, merge, o->subtree_shift);
3075 common = shift_tree_object(head, common, o->subtree_shift);
3078 if (oid_eq(&common->object.oid, &merge->object.oid)) {
3079 struct strbuf sb = STRBUF_INIT;
3081 if (!o->call_depth && index_has_changes(&sb)) {
3082 err(o, _("Dirty index: cannot merge (dirty: %s)"),
3083 sb.buf);
3084 return 0;
3086 output(o, 0, _("Already up to date!"));
3087 *result = head;
3088 return 1;
3091 code = git_merge_trees(o, common, head, merge);
3093 if (code != 0) {
3094 if (show(o, 4) || o->call_depth)
3095 err(o, _("merging of trees %s and %s failed"),
3096 oid_to_hex(&head->object.oid),
3097 oid_to_hex(&merge->object.oid));
3098 return -1;
3101 if (unmerged_cache()) {
3102 struct string_list *entries;
3103 struct rename_info re_info;
3104 int i;
3106 * Only need the hashmap while processing entries, so
3107 * initialize it here and free it when we are done running
3108 * through the entries. Keeping it in the merge_options as
3109 * opposed to decaring a local hashmap is for convenience
3110 * so that we don't have to pass it to around.
3112 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3113 get_files_dirs(o, head);
3114 get_files_dirs(o, merge);
3116 entries = get_unmerged();
3117 clean = handle_renames(o, common, head, merge, entries,
3118 &re_info);
3119 record_df_conflict_files(o, entries);
3120 if (clean < 0)
3121 goto cleanup;
3122 for (i = entries->nr-1; 0 <= i; i--) {
3123 const char *path = entries->items[i].string;
3124 struct stage_data *e = entries->items[i].util;
3125 if (!e->processed) {
3126 int ret = process_entry(o, path, e);
3127 if (!ret)
3128 clean = 0;
3129 else if (ret < 0) {
3130 clean = ret;
3131 goto cleanup;
3135 for (i = 0; i < entries->nr; i++) {
3136 struct stage_data *e = entries->items[i].util;
3137 if (!e->processed)
3138 BUG("unprocessed path??? %s",
3139 entries->items[i].string);
3142 cleanup:
3143 final_cleanup_renames(&re_info);
3145 string_list_clear(entries, 1);
3146 free(entries);
3148 hashmap_free(&o->current_file_dir_set, 1);
3150 if (clean < 0)
3151 return clean;
3153 else
3154 clean = 1;
3156 /* Free the extra index left from git_merge_trees() */
3158 * FIXME: Need to also free data allocated by
3159 * setup_unpack_trees_porcelain() tucked away in o->unpack_opts.msgs,
3160 * but the problem is that only half of it refers to dynamically
3161 * allocated data, while the other half points at static strings.
3163 discard_index(&o->orig_index);
3165 if (o->call_depth && !(*result = write_tree_from_memory(o)))
3166 return -1;
3168 return clean;
3171 static struct commit_list *reverse_commit_list(struct commit_list *list)
3173 struct commit_list *next = NULL, *current, *backup;
3174 for (current = list; current; current = backup) {
3175 backup = current->next;
3176 current->next = next;
3177 next = current;
3179 return next;
3183 * Merge the commits h1 and h2, return the resulting virtual
3184 * commit object and a flag indicating the cleanness of the merge.
3186 int merge_recursive(struct merge_options *o,
3187 struct commit *h1,
3188 struct commit *h2,
3189 struct commit_list *ca,
3190 struct commit **result)
3192 struct commit_list *iter;
3193 struct commit *merged_common_ancestors;
3194 struct tree *mrtree;
3195 int clean;
3197 if (show(o, 4)) {
3198 output(o, 4, _("Merging:"));
3199 output_commit_title(o, h1);
3200 output_commit_title(o, h2);
3203 if (!ca) {
3204 ca = get_merge_bases(h1, h2);
3205 ca = reverse_commit_list(ca);
3208 if (show(o, 5)) {
3209 unsigned cnt = commit_list_count(ca);
3211 output(o, 5, Q_("found %u common ancestor:",
3212 "found %u common ancestors:", cnt), cnt);
3213 for (iter = ca; iter; iter = iter->next)
3214 output_commit_title(o, iter->item);
3217 merged_common_ancestors = pop_commit(&ca);
3218 if (merged_common_ancestors == NULL) {
3219 /* if there is no common ancestor, use an empty tree */
3220 struct tree *tree;
3222 tree = lookup_tree(the_hash_algo->empty_tree);
3223 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3226 for (iter = ca; iter; iter = iter->next) {
3227 const char *saved_b1, *saved_b2;
3228 o->call_depth++;
3230 * When the merge fails, the result contains files
3231 * with conflict markers. The cleanness flag is
3232 * ignored (unless indicating an error), it was never
3233 * actually used, as result of merge_trees has always
3234 * overwritten it: the committed "conflicts" were
3235 * already resolved.
3237 discard_cache();
3238 saved_b1 = o->branch1;
3239 saved_b2 = o->branch2;
3240 o->branch1 = "Temporary merge branch 1";
3241 o->branch2 = "Temporary merge branch 2";
3242 if (merge_recursive(o, merged_common_ancestors, iter->item,
3243 NULL, &merged_common_ancestors) < 0)
3244 return -1;
3245 o->branch1 = saved_b1;
3246 o->branch2 = saved_b2;
3247 o->call_depth--;
3249 if (!merged_common_ancestors)
3250 return err(o, _("merge returned no commit"));
3253 discard_cache();
3254 if (!o->call_depth)
3255 read_cache();
3257 o->ancestor = "merged common ancestors";
3258 clean = merge_trees(o, get_commit_tree(h1), get_commit_tree(h2),
3259 get_commit_tree(merged_common_ancestors),
3260 &mrtree);
3261 if (clean < 0) {
3262 flush_output(o);
3263 return clean;
3266 if (o->call_depth) {
3267 *result = make_virtual_commit(mrtree, "merged tree");
3268 commit_list_insert(h1, &(*result)->parents);
3269 commit_list_insert(h2, &(*result)->parents->next);
3271 flush_output(o);
3272 if (!o->call_depth && o->buffer_output < 2)
3273 strbuf_release(&o->obuf);
3274 if (show(o, 2))
3275 diff_warn_rename_limit("merge.renamelimit",
3276 o->needed_rename_limit, 0);
3277 return clean;
3280 static struct commit *get_ref(const struct object_id *oid, const char *name)
3282 struct object *object;
3284 object = deref_tag(parse_object(oid), name, strlen(name));
3285 if (!object)
3286 return NULL;
3287 if (object->type == OBJ_TREE)
3288 return make_virtual_commit((struct tree*)object, name);
3289 if (object->type != OBJ_COMMIT)
3290 return NULL;
3291 if (parse_commit((struct commit *)object))
3292 return NULL;
3293 return (struct commit *)object;
3296 int merge_recursive_generic(struct merge_options *o,
3297 const struct object_id *head,
3298 const struct object_id *merge,
3299 int num_base_list,
3300 const struct object_id **base_list,
3301 struct commit **result)
3303 int clean;
3304 struct lock_file lock = LOCK_INIT;
3305 struct commit *head_commit = get_ref(head, o->branch1);
3306 struct commit *next_commit = get_ref(merge, o->branch2);
3307 struct commit_list *ca = NULL;
3309 if (base_list) {
3310 int i;
3311 for (i = 0; i < num_base_list; ++i) {
3312 struct commit *base;
3313 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3314 return err(o, _("Could not parse object '%s'"),
3315 oid_to_hex(base_list[i]));
3316 commit_list_insert(base, &ca);
3320 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3321 clean = merge_recursive(o, head_commit, next_commit, ca,
3322 result);
3323 if (clean < 0) {
3324 rollback_lock_file(&lock);
3325 return clean;
3328 if (write_locked_index(&the_index, &lock,
3329 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3330 return err(o, _("Unable to write index."));
3332 return clean ? 0 : 1;
3335 static void merge_recursive_config(struct merge_options *o)
3337 char *value = NULL;
3338 git_config_get_int("merge.verbosity", &o->verbosity);
3339 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3340 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3341 if (!git_config_get_string("diff.renames", &value)) {
3342 o->diff_detect_rename = git_config_rename("diff.renames", value);
3343 free(value);
3345 if (!git_config_get_string("merge.renames", &value)) {
3346 o->merge_detect_rename = git_config_rename("merge.renames", value);
3347 free(value);
3349 git_config(git_xmerge_config, NULL);
3352 void init_merge_options(struct merge_options *o)
3354 const char *merge_verbosity;
3355 memset(o, 0, sizeof(struct merge_options));
3356 o->verbosity = 2;
3357 o->buffer_output = 1;
3358 o->diff_rename_limit = -1;
3359 o->merge_rename_limit = -1;
3360 o->renormalize = 0;
3361 o->diff_detect_rename = -1;
3362 o->merge_detect_rename = -1;
3363 merge_recursive_config(o);
3364 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3365 if (merge_verbosity)
3366 o->verbosity = strtol(merge_verbosity, NULL, 10);
3367 if (o->verbosity >= 5)
3368 o->buffer_output = 0;
3369 strbuf_init(&o->obuf, 0);
3370 string_list_init(&o->df_conflict_file_set, 1);
3373 int parse_merge_opt(struct merge_options *o, const char *s)
3375 const char *arg;
3377 if (!s || !*s)
3378 return -1;
3379 if (!strcmp(s, "ours"))
3380 o->recursive_variant = MERGE_RECURSIVE_OURS;
3381 else if (!strcmp(s, "theirs"))
3382 o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3383 else if (!strcmp(s, "subtree"))
3384 o->subtree_shift = "";
3385 else if (skip_prefix(s, "subtree=", &arg))
3386 o->subtree_shift = arg;
3387 else if (!strcmp(s, "patience"))
3388 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3389 else if (!strcmp(s, "histogram"))
3390 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3391 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3392 long value = parse_algorithm_value(arg);
3393 if (value < 0)
3394 return -1;
3395 /* clear out previous settings */
3396 DIFF_XDL_CLR(o, NEED_MINIMAL);
3397 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3398 o->xdl_opts |= value;
3400 else if (!strcmp(s, "ignore-space-change"))
3401 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3402 else if (!strcmp(s, "ignore-all-space"))
3403 DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3404 else if (!strcmp(s, "ignore-space-at-eol"))
3405 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3406 else if (!strcmp(s, "ignore-cr-at-eol"))
3407 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3408 else if (!strcmp(s, "renormalize"))
3409 o->renormalize = 1;
3410 else if (!strcmp(s, "no-renormalize"))
3411 o->renormalize = 0;
3412 else if (!strcmp(s, "no-renames"))
3413 o->merge_detect_rename = 0;
3414 else if (!strcmp(s, "find-renames")) {
3415 o->merge_detect_rename = 1;
3416 o->rename_score = 0;
3418 else if (skip_prefix(s, "find-renames=", &arg) ||
3419 skip_prefix(s, "rename-threshold=", &arg)) {
3420 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3421 return -1;
3422 o->merge_detect_rename = 1;
3424 else
3425 return -1;
3426 return 0;