git

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

7
#define USE_THE_REPOSITORY_VARIABLE
8

9
#include "git-compat-util.h"
10
#include "merge-recursive.h"
11

12
#include "alloc.h"
13
#include "cache-tree.h"
14
#include "commit.h"
15
#include "commit-reach.h"
16
#include "config.h"
17
#include "diff.h"
18
#include "diffcore.h"
19
#include "dir.h"
20
#include "environment.h"
21
#include "gettext.h"
22
#include "hex.h"
23
#include "merge-ll.h"
24
#include "lockfile.h"
25
#include "match-trees.h"
26
#include "name-hash.h"
27
#include "object-file.h"
28
#include "object-name.h"
29
#include "object-store-ll.h"
30
#include "path.h"
31
#include "repository.h"
32
#include "revision.h"
33
#include "sparse-index.h"
34
#include "string-list.h"
35
#include "symlinks.h"
36
#include "tag.h"
37
#include "tree-walk.h"
38
#include "unpack-trees.h"
39
#include "xdiff-interface.h"
40

41
struct merge_options_internal {
42
	int call_depth;
43
	int needed_rename_limit;
44
	struct hashmap current_file_dir_set;
45
	struct string_list df_conflict_file_set;
46
	struct unpack_trees_options unpack_opts;
47
	struct index_state orig_index;
48
};
49

50
struct path_hashmap_entry {
51
	struct hashmap_entry e;
52
	char path[FLEX_ARRAY];
53
};
54

55
static int path_hashmap_cmp(const void *cmp_data UNUSED,
56
			    const struct hashmap_entry *eptr,
57
			    const struct hashmap_entry *entry_or_key,
58
			    const void *keydata)
59
{
60
	const struct path_hashmap_entry *a, *b;
61
	const char *key = keydata;
62

63
	a = container_of(eptr, const struct path_hashmap_entry, e);
64
	b = container_of(entry_or_key, const struct path_hashmap_entry, e);
65

66
	return fspathcmp(a->path, key ? key : b->path);
67
}
68

69
/*
70
 * For dir_rename_entry, directory names are stored as a full path from the
71
 * toplevel of the repository and do not include a trailing '/'.  Also:
72
 *
73
 *   dir:                original name of directory being renamed
74
 *   non_unique_new_dir: if true, could not determine new_dir
75
 *   new_dir:            final name of directory being renamed
76
 *   possible_new_dirs:  temporary used to help determine new_dir; see comments
77
 *                       in get_directory_renames() for details
78
 */
79
struct dir_rename_entry {
80
	struct hashmap_entry ent;
81
	char *dir;
82
	unsigned non_unique_new_dir:1;
83
	struct strbuf new_dir;
84
	struct string_list possible_new_dirs;
85
};
86

87
static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
88
						      char *dir)
89
{
90
	struct dir_rename_entry key;
91

92
	if (!dir)
93
		return NULL;
94
	hashmap_entry_init(&key.ent, strhash(dir));
95
	key.dir = dir;
96
	return hashmap_get_entry(hashmap, &key, ent, NULL);
97
}
98

99
static int dir_rename_cmp(const void *cmp_data UNUSED,
100
			  const struct hashmap_entry *eptr,
101
			  const struct hashmap_entry *entry_or_key,
102
			  const void *keydata UNUSED)
103
{
104
	const struct dir_rename_entry *e1, *e2;
105

106
	e1 = container_of(eptr, const struct dir_rename_entry, ent);
107
	e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
108

109
	return strcmp(e1->dir, e2->dir);
110
}
111

112
static void dir_rename_init(struct hashmap *map)
113
{
114
	hashmap_init(map, dir_rename_cmp, NULL, 0);
115
}
116

117
static void dir_rename_entry_init(struct dir_rename_entry *entry,
118
				  char *directory)
119
{
120
	hashmap_entry_init(&entry->ent, strhash(directory));
121
	entry->dir = directory;
122
	entry->non_unique_new_dir = 0;
123
	strbuf_init(&entry->new_dir, 0);
124
	string_list_init_nodup(&entry->possible_new_dirs);
125
}
126

127
struct collision_entry {
128
	struct hashmap_entry ent;
129
	char *target_file;
130
	struct string_list source_files;
131
	unsigned reported_already:1;
132
};
133

134
static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
135
						    char *target_file)
136
{
137
	struct collision_entry key;
138

139
	hashmap_entry_init(&key.ent, strhash(target_file));
140
	key.target_file = target_file;
141
	return hashmap_get_entry(hashmap, &key, ent, NULL);
142
}
143

144
static int collision_cmp(const void *cmp_data UNUSED,
145
			 const struct hashmap_entry *eptr,
146
			 const struct hashmap_entry *entry_or_key,
147
			 const void *keydata UNUSED)
148
{
149
	const struct collision_entry *e1, *e2;
150

151
	e1 = container_of(eptr, const struct collision_entry, ent);
152
	e2 = container_of(entry_or_key, const struct collision_entry, ent);
153

154
	return strcmp(e1->target_file, e2->target_file);
155
}
156

157
static void collision_init(struct hashmap *map)
158
{
159
	hashmap_init(map, collision_cmp, NULL, 0);
160
}
161

162
static void flush_output(struct merge_options *opt)
163
{
164
	if (opt->buffer_output < 2 && opt->obuf.len) {
165
		fputs(opt->obuf.buf, stdout);
166
		strbuf_reset(&opt->obuf);
167
	}
168
}
169

170
__attribute__((format (printf, 2, 3)))
171
static int err(struct merge_options *opt, const char *err, ...)
172
{
173
	va_list params;
174

175
	if (opt->buffer_output < 2)
176
		flush_output(opt);
177
	else {
178
		strbuf_complete(&opt->obuf, '\n');
179
		strbuf_addstr(&opt->obuf, "error: ");
180
	}
181
	va_start(params, err);
182
	strbuf_vaddf(&opt->obuf, err, params);
183
	va_end(params);
184
	if (opt->buffer_output > 1)
185
		strbuf_addch(&opt->obuf, '\n');
186
	else {
187
		error("%s", opt->obuf.buf);
188
		strbuf_reset(&opt->obuf);
189
	}
190

191
	return -1;
192
}
193

194
static struct tree *shift_tree_object(struct repository *repo,
195
				      struct tree *one, struct tree *two,
196
				      const char *subtree_shift)
197
{
198
	struct object_id shifted;
199

200
	if (!*subtree_shift) {
201
		shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
202
	} else {
203
		shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
204
			      subtree_shift);
205
	}
206
	if (oideq(&two->object.oid, &shifted))
207
		return two;
208
	return lookup_tree(repo, &shifted);
209
}
210

211
static inline void set_commit_tree(struct commit *c, struct tree *t)
212
{
213
	c->maybe_tree = t;
214
}
215

216
static struct commit *make_virtual_commit(struct repository *repo,
217
					  struct tree *tree,
218
					  const char *comment)
219
{
220
	struct commit *commit = alloc_commit_node(repo);
221

222
	set_merge_remote_desc(commit, comment, (struct object *)commit);
223
	set_commit_tree(commit, tree);
224
	commit->object.parsed = 1;
225
	return commit;
226
}
227

228
enum rename_type {
229
	RENAME_NORMAL = 0,
230
	RENAME_VIA_DIR,
231
	RENAME_ADD,
232
	RENAME_DELETE,
233
	RENAME_ONE_FILE_TO_ONE,
234
	RENAME_ONE_FILE_TO_TWO,
235
	RENAME_TWO_FILES_TO_ONE
236
};
237

238
/*
239
 * Since we want to write the index eventually, we cannot reuse the index
240
 * for these (temporary) data.
241
 */
242
struct stage_data {
243
	struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
244
	struct rename_conflict_info *rename_conflict_info;
245
	unsigned processed:1,
246
		 rename_conflict_info_owned:1;
247
};
248

249
struct rename {
250
	unsigned processed:1;
251
	struct diff_filepair *pair;
252
	const char *branch; /* branch that the rename occurred on */
253
	/*
254
	 * If directory rename detection affected this rename, what was its
255
	 * original type ('A' or 'R') and it's original destination before
256
	 * the directory rename (otherwise, '\0' and NULL for these two vars).
257
	 */
258
	char dir_rename_original_type;
259
	char *dir_rename_original_dest;
260
	/*
261
	 * Purpose of src_entry and dst_entry:
262
	 *
263
	 * If 'before' is renamed to 'after' then src_entry will contain
264
	 * the versions of 'before' from the merge_base, HEAD, and MERGE in
265
	 * stages 1, 2, and 3; dst_entry will contain the respective
266
	 * versions of 'after' in corresponding locations.  Thus, we have a
267
	 * total of six modes and oids, though some will be null.  (Stage 0
268
	 * is ignored; we're interested in handling conflicts.)
269
	 *
270
	 * Since we don't turn on break-rewrites by default, neither
271
	 * src_entry nor dst_entry can have all three of their stages have
272
	 * non-null oids, meaning at most four of the six will be non-null.
273
	 * Also, since this is a rename, both src_entry and dst_entry will
274
	 * have at least one non-null oid, meaning at least two will be
275
	 * non-null.  Of the six oids, a typical rename will have three be
276
	 * non-null.  Only two implies a rename/delete, and four implies a
277
	 * rename/add.
278
	 */
279
	struct stage_data *src_entry;
280
	struct stage_data *dst_entry;
281
};
282

283
struct rename_conflict_info {
284
	enum rename_type rename_type;
285
	struct rename *ren1;
286
	struct rename *ren2;
287
};
288

289
static inline void setup_rename_conflict_info(enum rename_type rename_type,
290
					      struct merge_options *opt,
291
					      struct rename *ren1,
292
					      struct rename *ren2)
293
{
294
	struct rename_conflict_info *ci;
295

296
	/*
297
	 * When we have two renames involved, it's easiest to get the
298
	 * correct things into stage 2 and 3, and to make sure that the
299
	 * content merge puts HEAD before the other branch if we just
300
	 * ensure that branch1 == opt->branch1.  So, simply flip arguments
301
	 * around if we don't have that.
302
	 */
303
	if (ren2 && ren1->branch != opt->branch1) {
304
		setup_rename_conflict_info(rename_type, opt, ren2, ren1);
305
		return;
306
	}
307

308
	CALLOC_ARRAY(ci, 1);
309
	ci->rename_type = rename_type;
310
	ci->ren1 = ren1;
311
	ci->ren2 = ren2;
312

313
	ci->ren1->dst_entry->processed = 0;
314
	ci->ren1->dst_entry->rename_conflict_info = ci;
315
	ci->ren1->dst_entry->rename_conflict_info_owned = 1;
316
	if (ren2) {
317
		ci->ren2->dst_entry->rename_conflict_info = ci;
318
	}
319
}
320

321
static int show(struct merge_options *opt, int v)
322
{
323
	return (!opt->priv->call_depth && opt->verbosity >= v) ||
324
		opt->verbosity >= 5;
325
}
326

327
__attribute__((format (printf, 3, 4)))
328
static void output(struct merge_options *opt, int v, const char *fmt, ...)
329
{
330
	va_list ap;
331

332
	if (!show(opt, v))
333
		return;
334

335
	strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
336

337
	va_start(ap, fmt);
338
	strbuf_vaddf(&opt->obuf, fmt, ap);
339
	va_end(ap);
340

341
	strbuf_addch(&opt->obuf, '\n');
342
	if (!opt->buffer_output)
343
		flush_output(opt);
344
}
345

346
static void repo_output_commit_title(struct merge_options *opt,
347
				     struct repository *repo,
348
				     struct commit *commit)
349
{
350
	struct merge_remote_desc *desc;
351

352
	strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
353
	desc = merge_remote_util(commit);
354
	if (desc)
355
		strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
356
	else {
357
		strbuf_repo_add_unique_abbrev(&opt->obuf, repo,
358
					      &commit->object.oid,
359
					      DEFAULT_ABBREV);
360
		strbuf_addch(&opt->obuf, ' ');
361
		if (repo_parse_commit(repo, commit) != 0)
362
			strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
363
		else {
364
			const char *title;
365
			const char *msg = repo_get_commit_buffer(repo, commit, NULL);
366
			int len = find_commit_subject(msg, &title);
367
			if (len)
368
				strbuf_addf(&opt->obuf, "%.*s\n", len, title);
369
			repo_unuse_commit_buffer(repo, commit, msg);
370
		}
371
	}
372
	flush_output(opt);
373
}
374

375
static void output_commit_title(struct merge_options *opt, struct commit *commit)
376
{
377
	repo_output_commit_title(opt, the_repository, commit);
378
}
379

380
static int add_cacheinfo(struct merge_options *opt,
381
			 const struct diff_filespec *blob,
382
			 const char *path, int stage, int refresh, int options)
383
{
384
	struct index_state *istate = opt->repo->index;
385
	struct cache_entry *ce;
386
	int ret;
387

388
	ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
389
	if (!ce)
390
		return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
391

392
	ret = add_index_entry(istate, ce, options);
393
	if (refresh) {
394
		struct cache_entry *nce;
395

396
		nce = refresh_cache_entry(istate, ce,
397
					  CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
398
		if (!nce)
399
			return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
400
		if (nce != ce)
401
			ret = add_index_entry(istate, nce, options);
402
	}
403
	return ret;
404
}
405

406
static inline int merge_detect_rename(struct merge_options *opt)
407
{
408
	return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
409
}
410

411
static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
412
{
413
	if (parse_tree(tree) < 0)
414
		exit(128);
415
	init_tree_desc(desc, &tree->object.oid, tree->buffer, tree->size);
416
}
417

418
static int unpack_trees_start(struct merge_options *opt,
419
			      struct tree *common,
420
			      struct tree *head,
421
			      struct tree *merge)
422
{
423
	int rc;
424
	struct tree_desc t[3];
425
	struct index_state tmp_index = INDEX_STATE_INIT(opt->repo);
426

427
	memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
428
	if (opt->priv->call_depth)
429
		opt->priv->unpack_opts.index_only = 1;
430
	else {
431
		opt->priv->unpack_opts.update = 1;
432
		/* FIXME: should only do this if !overwrite_ignore */
433
		opt->priv->unpack_opts.preserve_ignored = 0;
434
	}
435
	opt->priv->unpack_opts.merge = 1;
436
	opt->priv->unpack_opts.head_idx = 2;
437
	opt->priv->unpack_opts.fn = threeway_merge;
438
	opt->priv->unpack_opts.src_index = opt->repo->index;
439
	opt->priv->unpack_opts.dst_index = &tmp_index;
440
	opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
441
	setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
442

443
	init_tree_desc_from_tree(t+0, common);
444
	init_tree_desc_from_tree(t+1, head);
445
	init_tree_desc_from_tree(t+2, merge);
446

447
	rc = unpack_trees(3, t, &opt->priv->unpack_opts);
448
	cache_tree_free(&opt->repo->index->cache_tree);
449

450
	/*
451
	 * Update opt->repo->index to match the new results, AFTER saving a
452
	 * copy in opt->priv->orig_index.  Update src_index to point to the
453
	 * saved copy.  (verify_uptodate() checks src_index, and the original
454
	 * index is the one that had the necessary modification timestamps.)
455
	 */
456
	opt->priv->orig_index = *opt->repo->index;
457
	*opt->repo->index = tmp_index;
458
	opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
459

460
	return rc;
461
}
462

463
static void unpack_trees_finish(struct merge_options *opt)
464
{
465
	discard_index(&opt->priv->orig_index);
466
	clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
467
}
468

469
static int save_files_dirs(const struct object_id *oid UNUSED,
470
			   struct strbuf *base, const char *path,
471
			   unsigned int mode, void *context)
472
{
473
	struct path_hashmap_entry *entry;
474
	int baselen = base->len;
475
	struct merge_options *opt = context;
476

477
	strbuf_addstr(base, path);
478

479
	FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
480
	hashmap_entry_init(&entry->e, fspathhash(entry->path));
481
	hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
482

483
	strbuf_setlen(base, baselen);
484
	return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
485
}
486

487
static void get_files_dirs(struct merge_options *opt, struct tree *tree)
488
{
489
	struct pathspec match_all;
490
	memset(&match_all, 0, sizeof(match_all));
491
	read_tree(opt->repo, tree,
492
		  &match_all, save_files_dirs, opt);
493
}
494

495
static int get_tree_entry_if_blob(struct repository *r,
496
				  const struct object_id *tree,
497
				  const char *path,
498
				  struct diff_filespec *dfs)
499
{
500
	int ret;
501

502
	ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
503
	if (S_ISDIR(dfs->mode)) {
504
		oidcpy(&dfs->oid, null_oid());
505
		dfs->mode = 0;
506
	}
507
	return ret;
508
}
509

510
/*
511
 * Returns an index_entry instance which doesn't have to correspond to
512
 * a real cache entry in Git's index.
513
 */
514
static struct stage_data *insert_stage_data(struct repository *r,
515
		const char *path,
516
		struct tree *o, struct tree *a, struct tree *b,
517
		struct string_list *entries)
518
{
519
	struct string_list_item *item;
520
	struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
521
	get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
522
	get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
523
	get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
524
	item = string_list_insert(entries, path);
525
	item->util = e;
526
	return e;
527
}
528

529
/*
530
 * Create a dictionary mapping file names to stage_data objects. The
531
 * dictionary contains one entry for every path with a non-zero stage entry.
532
 */
533
static struct string_list *get_unmerged(struct index_state *istate)
534
{
535
	struct string_list *unmerged = xmalloc(sizeof(struct string_list));
536
	int i;
537

538
	string_list_init_dup(unmerged);
539

540
	/* TODO: audit for interaction with sparse-index. */
541
	ensure_full_index(istate);
542
	for (i = 0; i < istate->cache_nr; i++) {
543
		struct string_list_item *item;
544
		struct stage_data *e;
545
		const struct cache_entry *ce = istate->cache[i];
546
		if (!ce_stage(ce))
547
			continue;
548

549
		item = string_list_lookup(unmerged, ce->name);
550
		if (!item) {
551
			item = string_list_insert(unmerged, ce->name);
552
			item->util = xcalloc(1, sizeof(struct stage_data));
553
		}
554
		e = item->util;
555
		e->stages[ce_stage(ce)].mode = ce->ce_mode;
556
		oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
557
	}
558

559
	return unmerged;
560
}
561

562
static int string_list_df_name_compare(const char *one, const char *two)
563
{
564
	int onelen = strlen(one);
565
	int twolen = strlen(two);
566
	/*
567
	 * Here we only care that entries for D/F conflicts are
568
	 * adjacent, in particular with the file of the D/F conflict
569
	 * appearing before files below the corresponding directory.
570
	 * The order of the rest of the list is irrelevant for us.
571
	 *
572
	 * To achieve this, we sort with df_name_compare and provide
573
	 * the mode S_IFDIR so that D/F conflicts will sort correctly.
574
	 * We use the mode S_IFDIR for everything else for simplicity,
575
	 * since in other cases any changes in their order due to
576
	 * sorting cause no problems for us.
577
	 */
578
	int cmp = df_name_compare(one, onelen, S_IFDIR,
579
				  two, twolen, S_IFDIR);
580
	/*
581
	 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
582
	 * that 'foo' comes before 'foo/bar'.
583
	 */
584
	if (cmp)
585
		return cmp;
586
	return onelen - twolen;
587
}
588

589
static void record_df_conflict_files(struct merge_options *opt,
590
				     struct string_list *entries)
591
{
592
	/* If there is a D/F conflict and the file for such a conflict
593
	 * currently exists in the working tree, we want to allow it to be
594
	 * removed to make room for the corresponding directory if needed.
595
	 * The files underneath the directories of such D/F conflicts will
596
	 * be processed before the corresponding file involved in the D/F
597
	 * conflict.  If the D/F directory ends up being removed by the
598
	 * merge, then we won't have to touch the D/F file.  If the D/F
599
	 * directory needs to be written to the working copy, then the D/F
600
	 * file will simply be removed (in make_room_for_path()) to make
601
	 * room for the necessary paths.  Note that if both the directory
602
	 * and the file need to be present, then the D/F file will be
603
	 * reinstated with a new unique name at the time it is processed.
604
	 */
605
	struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
606
	const char *last_file = NULL;
607
	int last_len = 0;
608
	int i;
609

610
	/*
611
	 * If we're merging merge-bases, we don't want to bother with
612
	 * any working directory changes.
613
	 */
614
	if (opt->priv->call_depth)
615
		return;
616

617
	/* Ensure D/F conflicts are adjacent in the entries list. */
618
	for (i = 0; i < entries->nr; i++) {
619
		struct string_list_item *next = &entries->items[i];
620
		string_list_append(&df_sorted_entries, next->string)->util =
621
				   next->util;
622
	}
623
	df_sorted_entries.cmp = string_list_df_name_compare;
624
	string_list_sort(&df_sorted_entries);
625

626
	string_list_clear(&opt->priv->df_conflict_file_set, 1);
627
	for (i = 0; i < df_sorted_entries.nr; i++) {
628
		const char *path = df_sorted_entries.items[i].string;
629
		int len = strlen(path);
630
		struct stage_data *e = df_sorted_entries.items[i].util;
631

632
		/*
633
		 * Check if last_file & path correspond to a D/F conflict;
634
		 * i.e. whether path is last_file+'/'+<something>.
635
		 * If so, record that it's okay to remove last_file to make
636
		 * room for path and friends if needed.
637
		 */
638
		if (last_file &&
639
		    len > last_len &&
640
		    memcmp(path, last_file, last_len) == 0 &&
641
		    path[last_len] == '/') {
642
			string_list_insert(&opt->priv->df_conflict_file_set, last_file);
643
		}
644

645
		/*
646
		 * Determine whether path could exist as a file in the
647
		 * working directory as a possible D/F conflict.  This
648
		 * will only occur when it exists in stage 2 as a
649
		 * file.
650
		 */
651
		if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
652
			last_file = path;
653
			last_len = len;
654
		} else {
655
			last_file = NULL;
656
		}
657
	}
658
	string_list_clear(&df_sorted_entries, 0);
659
}
660

661
static int update_stages(struct merge_options *opt, const char *path,
662
			 const struct diff_filespec *o,
663
			 const struct diff_filespec *a,
664
			 const struct diff_filespec *b)
665
{
666

667
	/*
668
	 * NOTE: It is usually a bad idea to call update_stages on a path
669
	 * before calling update_file on that same path, since it can
670
	 * sometimes lead to spurious "refusing to lose untracked file..."
671
	 * messages from update_file (via make_room_for path via
672
	 * would_lose_untracked).  Instead, reverse the order of the calls
673
	 * (executing update_file first and then update_stages).
674
	 */
675
	int clear = 1;
676
	int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
677
	if (clear)
678
		if (remove_file_from_index(opt->repo->index, path))
679
			return -1;
680
	if (o)
681
		if (add_cacheinfo(opt, o, path, 1, 0, options))
682
			return -1;
683
	if (a)
684
		if (add_cacheinfo(opt, a, path, 2, 0, options))
685
			return -1;
686
	if (b)
687
		if (add_cacheinfo(opt, b, path, 3, 0, options))
688
			return -1;
689
	return 0;
690
}
691

692
static void update_entry(struct stage_data *entry,
693
			 struct diff_filespec *o,
694
			 struct diff_filespec *a,
695
			 struct diff_filespec *b)
696
{
697
	entry->processed = 0;
698
	entry->stages[1].mode = o->mode;
699
	entry->stages[2].mode = a->mode;
700
	entry->stages[3].mode = b->mode;
701
	oidcpy(&entry->stages[1].oid, &o->oid);
702
	oidcpy(&entry->stages[2].oid, &a->oid);
703
	oidcpy(&entry->stages[3].oid, &b->oid);
704
}
705

706
static int remove_file(struct merge_options *opt, int clean,
707
		       const char *path, int no_wd)
708
{
709
	int update_cache = opt->priv->call_depth || clean;
710
	int update_working_directory = !opt->priv->call_depth && !no_wd;
711

712
	if (update_cache) {
713
		if (remove_file_from_index(opt->repo->index, path))
714
			return -1;
715
	}
716
	if (update_working_directory) {
717
		if (ignore_case) {
718
			struct cache_entry *ce;
719
			ce = index_file_exists(opt->repo->index, path, strlen(path),
720
					       ignore_case);
721
			if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
722
				return 0;
723
		}
724
		if (remove_path(path))
725
			return -1;
726
	}
727
	return 0;
728
}
729

730
/* add a string to a strbuf, but converting "/" to "_" */
731
static void add_flattened_path(struct strbuf *out, const char *s)
732
{
733
	size_t i = out->len;
734
	strbuf_addstr(out, s);
735
	for (; i < out->len; i++)
736
		if (out->buf[i] == '/')
737
			out->buf[i] = '_';
738
}
739

740
static char *unique_path(struct merge_options *opt,
741
			 const char *path,
742
			 const char *branch)
743
{
744
	struct path_hashmap_entry *entry;
745
	struct strbuf newpath = STRBUF_INIT;
746
	int suffix = 0;
747
	size_t base_len;
748

749
	strbuf_addf(&newpath, "%s~", path);
750
	add_flattened_path(&newpath, branch);
751

752
	base_len = newpath.len;
753
	while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
754
				     fspathhash(newpath.buf), newpath.buf) ||
755
	       (!opt->priv->call_depth && file_exists(newpath.buf))) {
756
		strbuf_setlen(&newpath, base_len);
757
		strbuf_addf(&newpath, "_%d", suffix++);
758
	}
759

760
	FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
761
	hashmap_entry_init(&entry->e, fspathhash(entry->path));
762
	hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
763
	return strbuf_detach(&newpath, NULL);
764
}
765

766
/**
767
 * Check whether a directory in the index is in the way of an incoming
768
 * file.  Return 1 if so.  If check_working_copy is non-zero, also
769
 * check the working directory.  If empty_ok is non-zero, also return
770
 * 0 in the case where the working-tree dir exists but is empty.
771
 */
772
static int dir_in_way(struct index_state *istate, const char *path,
773
		      int check_working_copy, int empty_ok)
774
{
775
	int pos;
776
	struct strbuf dirpath = STRBUF_INIT;
777
	struct stat st;
778

779
	strbuf_addstr(&dirpath, path);
780
	strbuf_addch(&dirpath, '/');
781

782
	pos = index_name_pos(istate, dirpath.buf, dirpath.len);
783

784
	if (pos < 0)
785
		pos = -1 - pos;
786
	if (pos < istate->cache_nr &&
787
	    !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
788
		strbuf_release(&dirpath);
789
		return 1;
790
	}
791

792
	strbuf_release(&dirpath);
793
	return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
794
		!(empty_ok && is_empty_dir(path)) &&
795
		!has_symlink_leading_path(path, strlen(path));
796
}
797

798
/*
799
 * Returns whether path was tracked in the index before the merge started,
800
 * and its oid and mode match the specified values
801
 */
802
static int was_tracked_and_matches(struct merge_options *opt, const char *path,
803
				   const struct diff_filespec *blob)
804
{
805
	int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
806
	struct cache_entry *ce;
807

808
	if (0 > pos)
809
		/* we were not tracking this path before the merge */
810
		return 0;
811

812
	/* See if the file we were tracking before matches */
813
	ce = opt->priv->orig_index.cache[pos];
814
	return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
815
}
816

817
/*
818
 * Returns whether path was tracked in the index before the merge started
819
 */
820
static int was_tracked(struct merge_options *opt, const char *path)
821
{
822
	int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
823

824
	if (0 <= pos)
825
		/* we were tracking this path before the merge */
826
		return 1;
827

828
	return 0;
829
}
830

831
static int would_lose_untracked(struct merge_options *opt, const char *path)
832
{
833
	struct index_state *istate = opt->repo->index;
834

835
	/*
836
	 * This may look like it can be simplified to:
837
	 *   return !was_tracked(opt, path) && file_exists(path)
838
	 * but it can't.  This function needs to know whether path was in
839
	 * the working tree due to EITHER having been tracked in the index
840
	 * before the merge OR having been put into the working copy and
841
	 * index by unpack_trees().  Due to that either-or requirement, we
842
	 * check the current index instead of the original one.
843
	 *
844
	 * Note that we do not need to worry about merge-recursive itself
845
	 * updating the index after unpack_trees() and before calling this
846
	 * function, because we strictly require all code paths in
847
	 * merge-recursive to update the working tree first and the index
848
	 * second.  Doing otherwise would break
849
	 * update_file()/would_lose_untracked(); see every comment in this
850
	 * file which mentions "update_stages".
851
	 */
852
	int pos = index_name_pos(istate, path, strlen(path));
853

854
	if (pos < 0)
855
		pos = -1 - pos;
856
	while (pos < istate->cache_nr &&
857
	       !strcmp(path, istate->cache[pos]->name)) {
858
		/*
859
		 * If stage #0, it is definitely tracked.
860
		 * If it has stage #2 then it was tracked
861
		 * before this merge started.  All other
862
		 * cases the path was not tracked.
863
		 */
864
		switch (ce_stage(istate->cache[pos])) {
865
		case 0:
866
		case 2:
867
			return 0;
868
		}
869
		pos++;
870
	}
871
	return file_exists(path);
872
}
873

874
static int was_dirty(struct merge_options *opt, const char *path)
875
{
876
	struct cache_entry *ce;
877
	int dirty = 1;
878

879
	if (opt->priv->call_depth || !was_tracked(opt, path))
880
		return !dirty;
881

882
	ce = index_file_exists(opt->priv->unpack_opts.src_index,
883
			       path, strlen(path), ignore_case);
884
	dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
885
	return dirty;
886
}
887

888
static int make_room_for_path(struct merge_options *opt, const char *path)
889
{
890
	int status, i;
891
	const char *msg = _("failed to create path '%s'%s");
892

893
	/* Unlink any D/F conflict files that are in the way */
894
	for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
895
		const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
896
		size_t pathlen = strlen(path);
897
		size_t df_pathlen = strlen(df_path);
898
		if (df_pathlen < pathlen &&
899
		    path[df_pathlen] == '/' &&
900
		    strncmp(path, df_path, df_pathlen) == 0) {
901
			output(opt, 3,
902
			       _("Removing %s to make room for subdirectory\n"),
903
			       df_path);
904
			unlink(df_path);
905
			unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
906
							 i, 0);
907
			break;
908
		}
909
	}
910

911
	/* Make sure leading directories are created */
912
	status = safe_create_leading_directories_const(path);
913
	if (status) {
914
		if (status == SCLD_EXISTS)
915
			/* something else exists */
916
			return err(opt, msg, path, _(": perhaps a D/F conflict?"));
917
		return err(opt, msg, path, "");
918
	}
919

920
	/*
921
	 * Do not unlink a file in the work tree if we are not
922
	 * tracking it.
923
	 */
924
	if (would_lose_untracked(opt, path))
925
		return err(opt, _("refusing to lose untracked file at '%s'"),
926
			   path);
927

928
	/* Successful unlink is good.. */
929
	if (!unlink(path))
930
		return 0;
931
	/* .. and so is no existing file */
932
	if (errno == ENOENT)
933
		return 0;
934
	/* .. but not some other error (who really cares what?) */
935
	return err(opt, msg, path, _(": perhaps a D/F conflict?"));
936
}
937

938
static int update_file_flags(struct merge_options *opt,
939
			     const struct diff_filespec *contents,
940
			     const char *path,
941
			     int update_cache,
942
			     int update_wd)
943
{
944
	int ret = 0;
945

946
	if (opt->priv->call_depth)
947
		update_wd = 0;
948

949
	if (update_wd) {
950
		enum object_type type;
951
		void *buf;
952
		unsigned long size;
953

954
		if (S_ISGITLINK(contents->mode)) {
955
			/*
956
			 * We may later decide to recursively descend into
957
			 * the submodule directory and update its index
958
			 * and/or work tree, but we do not do that now.
959
			 */
960
			update_wd = 0;
961
			goto update_index;
962
		}
963

964
		buf = repo_read_object_file(the_repository, &contents->oid,
965
					    &type, &size);
966
		if (!buf) {
967
			ret = err(opt, _("cannot read object %s '%s'"),
968
				  oid_to_hex(&contents->oid), path);
969
			goto free_buf;
970
		}
971
		if (type != OBJ_BLOB) {
972
			ret = err(opt, _("blob expected for %s '%s'"),
973
				  oid_to_hex(&contents->oid), path);
974
			goto free_buf;
975
		}
976
		if (S_ISREG(contents->mode)) {
977
			struct strbuf strbuf = STRBUF_INIT;
978
			if (convert_to_working_tree(opt->repo->index,
979
						    path, buf, size, &strbuf, NULL)) {
980
				free(buf);
981
				size = strbuf.len;
982
				buf = strbuf_detach(&strbuf, NULL);
983
			}
984
		}
985

986
		if (make_room_for_path(opt, path) < 0) {
987
			update_wd = 0;
988
			goto free_buf;
989
		}
990
		if (S_ISREG(contents->mode) ||
991
		    (!has_symlinks && S_ISLNK(contents->mode))) {
992
			int fd;
993
			int mode = (contents->mode & 0100 ? 0777 : 0666);
994

995
			fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
996
			if (fd < 0) {
997
				ret = err(opt, _("failed to open '%s': %s"),
998
					  path, strerror(errno));
999
				goto free_buf;
1000
			}
1001
			write_in_full(fd, buf, size);
1002
			close(fd);
1003
		} else if (S_ISLNK(contents->mode)) {
1004
			char *lnk = xmemdupz(buf, size);
1005
			safe_create_leading_directories_const(path);
1006
			unlink(path);
1007
			if (symlink(lnk, path))
1008
				ret = err(opt, _("failed to symlink '%s': %s"),
1009
					  path, strerror(errno));
1010
			free(lnk);
1011
		} else
1012
			ret = err(opt,
1013
				  _("do not know what to do with %06o %s '%s'"),
1014
				  contents->mode, oid_to_hex(&contents->oid), path);
1015
	free_buf:
1016
		free(buf);
1017
	}
1018
update_index:
1019
	if (!ret && update_cache) {
1020
		int refresh = (!opt->priv->call_depth &&
1021
			       contents->mode != S_IFGITLINK);
1022
		if (add_cacheinfo(opt, contents, path, 0, refresh,
1023
				  ADD_CACHE_OK_TO_ADD))
1024
			return -1;
1025
	}
1026
	return ret;
1027
}
1028

1029
static int update_file(struct merge_options *opt,
1030
		       int clean,
1031
		       const struct diff_filespec *contents,
1032
		       const char *path)
1033
{
1034
	return update_file_flags(opt, contents, path,
1035
				 opt->priv->call_depth || clean, !opt->priv->call_depth);
1036
}
1037

1038
/* Low level file merging, update and removal */
1039

1040
struct merge_file_info {
1041
	struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1042
	unsigned clean:1,
1043
		 merge:1;
1044
};
1045

1046
static int merge_3way(struct merge_options *opt,
1047
		      mmbuffer_t *result_buf,
1048
		      const struct diff_filespec *o,
1049
		      const struct diff_filespec *a,
1050
		      const struct diff_filespec *b,
1051
		      const char *branch1,
1052
		      const char *branch2,
1053
		      const int extra_marker_size)
1054
{
1055
	mmfile_t orig, src1, src2;
1056
	struct ll_merge_options ll_opts = LL_MERGE_OPTIONS_INIT;
1057
	char *base, *name1, *name2;
1058
	enum ll_merge_result merge_status;
1059

1060
	ll_opts.renormalize = opt->renormalize;
1061
	ll_opts.extra_marker_size = extra_marker_size;
1062
	ll_opts.xdl_opts = opt->xdl_opts;
1063
	ll_opts.conflict_style = opt->conflict_style;
1064

1065
	if (opt->priv->call_depth) {
1066
		ll_opts.virtual_ancestor = 1;
1067
		ll_opts.variant = 0;
1068
	} else {
1069
		switch (opt->recursive_variant) {
1070
		case MERGE_VARIANT_OURS:
1071
			ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1072
			break;
1073
		case MERGE_VARIANT_THEIRS:
1074
			ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1075
			break;
1076
		default:
1077
			ll_opts.variant = 0;
1078
			break;
1079
		}
1080
	}
1081

1082
	assert(a->path && b->path && o->path && opt->ancestor);
1083
	if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1084
		base  = mkpathdup("%s:%s", opt->ancestor, o->path);
1085
		name1 = mkpathdup("%s:%s", branch1, a->path);
1086
		name2 = mkpathdup("%s:%s", branch2, b->path);
1087
	} else {
1088
		base  = mkpathdup("%s", opt->ancestor);
1089
		name1 = mkpathdup("%s", branch1);
1090
		name2 = mkpathdup("%s", branch2);
1091
	}
1092

1093
	read_mmblob(&orig, &o->oid);
1094
	read_mmblob(&src1, &a->oid);
1095
	read_mmblob(&src2, &b->oid);
1096

1097
	/*
1098
	 * FIXME: Using a->path for normalization rules in ll_merge could be
1099
	 * wrong if we renamed from a->path to b->path.  We should use the
1100
	 * target path for where the file will be written.
1101
	 */
1102
	merge_status = ll_merge(result_buf, a->path, &orig, base,
1103
				&src1, name1, &src2, name2,
1104
				opt->repo->index, &ll_opts);
1105
	if (merge_status == LL_MERGE_BINARY_CONFLICT)
1106
		warning("Cannot merge binary files: %s (%s vs. %s)",
1107
			a->path, name1, name2);
1108

1109
	free(base);
1110
	free(name1);
1111
	free(name2);
1112
	free(orig.ptr);
1113
	free(src1.ptr);
1114
	free(src2.ptr);
1115
	return merge_status;
1116
}
1117

1118
static int find_first_merges(struct repository *repo,
1119
			     struct object_array *result, const char *path,
1120
			     struct commit *a, struct commit *b)
1121
{
1122
	int i, j;
1123
	struct object_array merges = OBJECT_ARRAY_INIT;
1124
	struct commit *commit;
1125
	int contains_another;
1126

1127
	char merged_revision[GIT_MAX_HEXSZ + 2];
1128
	const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1129
				   "--all", merged_revision, NULL };
1130
	struct rev_info revs;
1131
	struct setup_revision_opt rev_opts;
1132

1133
	memset(result, 0, sizeof(struct object_array));
1134
	memset(&rev_opts, 0, sizeof(rev_opts));
1135

1136
	/* get all revisions that merge commit a */
1137
	xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1138
		  oid_to_hex(&a->object.oid));
1139
	repo_init_revisions(repo, &revs, NULL);
1140
	/* FIXME: can't handle linked worktrees in submodules yet */
1141
	revs.single_worktree = path != NULL;
1142
	setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1143

1144
	/* save all revisions from the above list that contain b */
1145
	if (prepare_revision_walk(&revs))
1146
		die("revision walk setup failed");
1147
	while ((commit = get_revision(&revs)) != NULL) {
1148
		struct object *o = &(commit->object);
1149
		int ret = repo_in_merge_bases(repo, b, commit);
1150
		if (ret < 0) {
1151
			object_array_clear(&merges);
1152
			release_revisions(&revs);
1153
			return ret;
1154
		}
1155
		if (ret)
1156
			add_object_array(o, NULL, &merges);
1157
	}
1158
	reset_revision_walk();
1159

1160
	/* Now we've got all merges that contain a and b. Prune all
1161
	 * merges that contain another found merge and save them in
1162
	 * result.
1163
	 */
1164
	for (i = 0; i < merges.nr; i++) {
1165
		struct commit *m1 = (struct commit *) merges.objects[i].item;
1166

1167
		contains_another = 0;
1168
		for (j = 0; j < merges.nr; j++) {
1169
			struct commit *m2 = (struct commit *) merges.objects[j].item;
1170
			if (i != j) {
1171
				int ret = repo_in_merge_bases(repo, m2, m1);
1172
				if (ret < 0) {
1173
					object_array_clear(&merges);
1174
					release_revisions(&revs);
1175
					return ret;
1176
				}
1177
				if (ret > 0) {
1178
					contains_another = 1;
1179
					break;
1180
				}
1181
			}
1182
		}
1183

1184
		if (!contains_another)
1185
			add_object_array(merges.objects[i].item, NULL, result);
1186
	}
1187

1188
	object_array_clear(&merges);
1189
	release_revisions(&revs);
1190
	return result->nr;
1191
}
1192

1193
static void print_commit(struct repository *repo, struct commit *commit)
1194
{
1195
	struct strbuf sb = STRBUF_INIT;
1196
	struct pretty_print_context ctx = {0};
1197
	ctx.date_mode.type = DATE_NORMAL;
1198
	/* FIXME: Merge this with output_commit_title() */
1199
	assert(!merge_remote_util(commit));
1200
	repo_format_commit_message(repo, commit, " %h: %m %s", &sb, &ctx);
1201
	fprintf(stderr, "%s\n", sb.buf);
1202
	strbuf_release(&sb);
1203
}
1204

1205
static int is_valid(const struct diff_filespec *dfs)
1206
{
1207
	return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1208
}
1209

1210
static int merge_submodule(struct merge_options *opt,
1211
			   struct object_id *result, const char *path,
1212
			   const struct object_id *base, const struct object_id *a,
1213
			   const struct object_id *b)
1214
{
1215
	struct repository subrepo;
1216
	int ret = 0, ret2;
1217
	struct commit *commit_base, *commit_a, *commit_b;
1218
	int parent_count;
1219
	struct object_array merges;
1220

1221
	int i;
1222
	int search = !opt->priv->call_depth;
1223

1224
	/* store a in result in case we fail */
1225
	/* FIXME: This is the WRONG resolution for the recursive case when
1226
	 * we need to be careful to avoid accidentally matching either side.
1227
	 * Should probably use o instead there, much like we do for merging
1228
	 * binaries.
1229
	 */
1230
	oidcpy(result, a);
1231

1232
	/* we can not handle deletion conflicts */
1233
	if (is_null_oid(base))
1234
		return 0;
1235
	if (is_null_oid(a))
1236
		return 0;
1237
	if (is_null_oid(b))
1238
		return 0;
1239

1240
	if (repo_submodule_init(&subrepo, opt->repo, path, null_oid())) {
1241
		output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1242
		return 0;
1243
	}
1244

1245
	if (!(commit_base = lookup_commit_reference(&subrepo, base)) ||
1246
	    !(commit_a = lookup_commit_reference(&subrepo, a)) ||
1247
	    !(commit_b = lookup_commit_reference(&subrepo, b))) {
1248
		output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1249
		goto cleanup;
1250
	}
1251

1252
	/* check whether both changes are forward */
1253
	ret2 = repo_in_merge_bases(&subrepo, commit_base, commit_a);
1254
	if (ret2 < 0) {
1255
		output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1256
		ret = -1;
1257
		goto cleanup;
1258
	}
1259
	if (ret2 > 0)
1260
		ret2 = repo_in_merge_bases(&subrepo, commit_base, commit_b);
1261
	if (ret2 < 0) {
1262
		output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1263
		ret = -1;
1264
		goto cleanup;
1265
	}
1266
	if (!ret2) {
1267
		output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1268
		goto cleanup;
1269
	}
1270

1271
	/* Case #1: a is contained in b or vice versa */
1272
	ret2 = repo_in_merge_bases(&subrepo, commit_a, commit_b);
1273
	if (ret2 < 0) {
1274
		output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1275
		ret = -1;
1276
		goto cleanup;
1277
	}
1278
	if (ret2) {
1279
		oidcpy(result, b);
1280
		if (show(opt, 3)) {
1281
			output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1282
			repo_output_commit_title(opt, &subrepo, commit_b);
1283
		} else if (show(opt, 2))
1284
			output(opt, 2, _("Fast-forwarding submodule %s"), path);
1285
		else
1286
			; /* no output */
1287

1288
		ret = 1;
1289
		goto cleanup;
1290
	}
1291
	ret2 = repo_in_merge_bases(&subrepo, commit_b, commit_a);
1292
	if (ret2 < 0) {
1293
		output(opt, 1, _("Failed to merge submodule %s (repository corrupt)"), path);
1294
		ret = -1;
1295
		goto cleanup;
1296
	}
1297
	if (ret2) {
1298
		oidcpy(result, a);
1299
		if (show(opt, 3)) {
1300
			output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1301
			repo_output_commit_title(opt, &subrepo, commit_a);
1302
		} else if (show(opt, 2))
1303
			output(opt, 2, _("Fast-forwarding submodule %s"), path);
1304
		else
1305
			; /* no output */
1306

1307
		ret = 1;
1308
		goto cleanup;
1309
	}
1310

1311
	/*
1312
	 * Case #2: There are one or more merges that contain a and b in
1313
	 * the submodule. If there is only one, then present it as a
1314
	 * suggestion to the user, but leave it marked unmerged so the
1315
	 * user needs to confirm the resolution.
1316
	 */
1317

1318
	/* Skip the search if makes no sense to the calling context.  */
1319
	if (!search)
1320
		goto cleanup;
1321

1322
	/* find commit which merges them */
1323
	parent_count = find_first_merges(&subrepo, &merges, path,
1324
					 commit_a, commit_b);
1325
	switch (parent_count) {
1326
	case -1:
1327
		output(opt, 1,_("Failed to merge submodule %s (repository corrupt)"), path);
1328
		ret = -1;
1329
		break;
1330
	case 0:
1331
		output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1332
		break;
1333

1334
	case 1:
1335
		output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1336
		output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1337
		print_commit(&subrepo, (struct commit *) merges.objects[0].item);
1338
		output(opt, 2, _(
1339
		       "If this is correct simply add it to the index "
1340
		       "for example\n"
1341
		       "by using:\n\n"
1342
		       "  git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1343
		       "which will accept this suggestion.\n"),
1344
		       oid_to_hex(&merges.objects[0].item->oid), path);
1345
		break;
1346

1347
	default:
1348
		output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1349
		for (i = 0; i < merges.nr; i++)
1350
			print_commit(&subrepo, (struct commit *) merges.objects[i].item);
1351
	}
1352

1353
	object_array_clear(&merges);
1354
cleanup:
1355
	repo_clear(&subrepo);
1356
	return ret;
1357
}
1358

1359
static int merge_mode_and_contents(struct merge_options *opt,
1360
				   const struct diff_filespec *o,
1361
				   const struct diff_filespec *a,
1362
				   const struct diff_filespec *b,
1363
				   const char *filename,
1364
				   const char *branch1,
1365
				   const char *branch2,
1366
				   const int extra_marker_size,
1367
				   struct merge_file_info *result)
1368
{
1369
	if (opt->branch1 != branch1) {
1370
		/*
1371
		 * It's weird getting a reverse merge with HEAD on the bottom
1372
		 * side of the conflict markers and the other branch on the
1373
		 * top.  Fix that.
1374
		 */
1375
		return merge_mode_and_contents(opt, o, b, a,
1376
					       filename,
1377
					       branch2, branch1,
1378
					       extra_marker_size, result);
1379
	}
1380

1381
	result->merge = 0;
1382
	result->clean = 1;
1383

1384
	if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1385
		result->clean = 0;
1386
		/*
1387
		 * FIXME: This is a bad resolution for recursive case; for
1388
		 * the recursive case we want something that is unlikely to
1389
		 * accidentally match either side.  Also, while it makes
1390
		 * sense to prefer regular files over symlinks, it doesn't
1391
		 * make sense to prefer regular files over submodules.
1392
		 */
1393
		if (S_ISREG(a->mode)) {
1394
			result->blob.mode = a->mode;
1395
			oidcpy(&result->blob.oid, &a->oid);
1396
		} else {
1397
			result->blob.mode = b->mode;
1398
			oidcpy(&result->blob.oid, &b->oid);
1399
		}
1400
	} else {
1401
		if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1402
			result->merge = 1;
1403

1404
		/*
1405
		 * Merge modes
1406
		 */
1407
		if (a->mode == b->mode || a->mode == o->mode)
1408
			result->blob.mode = b->mode;
1409
		else {
1410
			result->blob.mode = a->mode;
1411
			if (b->mode != o->mode) {
1412
				result->clean = 0;
1413
				result->merge = 1;
1414
			}
1415
		}
1416

1417
		if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1418
			oidcpy(&result->blob.oid, &b->oid);
1419
		else if (oideq(&b->oid, &o->oid))
1420
			oidcpy(&result->blob.oid, &a->oid);
1421
		else if (S_ISREG(a->mode)) {
1422
			mmbuffer_t result_buf;
1423
			int ret = 0, merge_status;
1424

1425
			merge_status = merge_3way(opt, &result_buf, o, a, b,
1426
						  branch1, branch2,
1427
						  extra_marker_size);
1428

1429
			if ((merge_status < 0) || !result_buf.ptr)
1430
				ret = err(opt, _("failed to execute internal merge"));
1431

1432
			if (!ret &&
1433
			    write_object_file(result_buf.ptr, result_buf.size,
1434
					      OBJ_BLOB, &result->blob.oid))
1435
				ret = err(opt, _("unable to add %s to database"),
1436
					  a->path);
1437

1438
			free(result_buf.ptr);
1439
			if (ret)
1440
				return ret;
1441
			/* FIXME: bug, what if modes didn't match? */
1442
			result->clean = (merge_status == 0);
1443
		} else if (S_ISGITLINK(a->mode)) {
1444
			int clean = merge_submodule(opt, &result->blob.oid,
1445
						    o->path,
1446
						    &o->oid,
1447
						    &a->oid,
1448
						    &b->oid);
1449
			if (clean < 0)
1450
				return -1;
1451
			result->clean = clean;
1452
		} else if (S_ISLNK(a->mode)) {
1453
			switch (opt->recursive_variant) {
1454
			case MERGE_VARIANT_NORMAL:
1455
				oidcpy(&result->blob.oid, &a->oid);
1456
				if (!oideq(&a->oid, &b->oid))
1457
					result->clean = 0;
1458
				break;
1459
			case MERGE_VARIANT_OURS:
1460
				oidcpy(&result->blob.oid, &a->oid);
1461
				break;
1462
			case MERGE_VARIANT_THEIRS:
1463
				oidcpy(&result->blob.oid, &b->oid);
1464
				break;
1465
			}
1466
		} else
1467
			BUG("unsupported object type in the tree");
1468
	}
1469

1470
	if (result->merge)
1471
		output(opt, 2, _("Auto-merging %s"), filename);
1472

1473
	return 0;
1474
}
1475

1476
static int handle_rename_via_dir(struct merge_options *opt,
1477
				 struct rename_conflict_info *ci)
1478
{
1479
	/*
1480
	 * Handle file adds that need to be renamed due to directory rename
1481
	 * detection.  This differs from handle_rename_normal, because
1482
	 * there is no content merge to do; just move the file into the
1483
	 * desired final location.
1484
	 */
1485
	const struct rename *ren = ci->ren1;
1486
	const struct diff_filespec *dest = ren->pair->two;
1487
	char *file_path = dest->path;
1488
	int mark_conflicted = (opt->detect_directory_renames ==
1489
			       MERGE_DIRECTORY_RENAMES_CONFLICT);
1490
	assert(ren->dir_rename_original_dest);
1491

1492
	if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1493
		mark_conflicted = 1;
1494
		file_path = unique_path(opt, dest->path, ren->branch);
1495
		output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1496
				 "writing to %s instead."),
1497
		       dest->path, file_path);
1498
	}
1499

1500
	if (mark_conflicted) {
1501
		/*
1502
		 * Write the file in worktree at file_path.  In the index,
1503
		 * only record the file at dest->path in the appropriate
1504
		 * higher stage.
1505
		 */
1506
		if (update_file(opt, 0, dest, file_path))
1507
			return -1;
1508
		if (file_path != dest->path)
1509
			free(file_path);
1510
		if (update_stages(opt, dest->path, NULL,
1511
				  ren->branch == opt->branch1 ? dest : NULL,
1512
				  ren->branch == opt->branch1 ? NULL : dest))
1513
			return -1;
1514
		return 0; /* not clean, but conflicted */
1515
	} else {
1516
		/* Update dest->path both in index and in worktree */
1517
		if (update_file(opt, 1, dest, dest->path))
1518
			return -1;
1519
		return 1; /* clean */
1520
	}
1521
}
1522

1523
static int handle_change_delete(struct merge_options *opt,
1524
				const char *path, const char *old_path,
1525
				const struct diff_filespec *o,
1526
				const struct diff_filespec *changed,
1527
				const char *change_branch,
1528
				const char *delete_branch,
1529
				const char *change, const char *change_past)
1530
{
1531
	char *alt_path = NULL;
1532
	const char *update_path = path;
1533
	int ret = 0;
1534

1535
	if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1536
	    (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1537
		update_path = alt_path = unique_path(opt, path, change_branch);
1538
	}
1539

1540
	if (opt->priv->call_depth) {
1541
		/*
1542
		 * We cannot arbitrarily accept either a_sha or b_sha as
1543
		 * correct; since there is no true "middle point" between
1544
		 * them, simply reuse the base version for virtual merge base.
1545
		 */
1546
		ret = remove_file_from_index(opt->repo->index, path);
1547
		if (!ret)
1548
			ret = update_file(opt, 0, o, update_path);
1549
	} else {
1550
		/*
1551
		 * Despite the four nearly duplicate messages and argument
1552
		 * lists below and the ugliness of the nested if-statements,
1553
		 * having complete messages makes the job easier for
1554
		 * translators.
1555
		 *
1556
		 * The slight variance among the cases is due to the fact
1557
		 * that:
1558
		 *   1) directory/file conflicts (in effect if
1559
		 *      !alt_path) could cause us to need to write the
1560
		 *      file to a different path.
1561
		 *   2) renames (in effect if !old_path) could mean that
1562
		 *      there are two names for the path that the user
1563
		 *      may know the file by.
1564
		 */
1565
		if (!alt_path) {
1566
			if (!old_path) {
1567
				output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1568
				       "and %s in %s. Version %s of %s left in tree."),
1569
				       change, path, delete_branch, change_past,
1570
				       change_branch, change_branch, path);
1571
			} else {
1572
				output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1573
				       "and %s to %s in %s. Version %s of %s left in tree."),
1574
				       change, old_path, delete_branch, change_past, path,
1575
				       change_branch, change_branch, path);
1576
			}
1577
		} else {
1578
			if (!old_path) {
1579
				output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1580
				       "and %s in %s. Version %s of %s left in tree at %s."),
1581
				       change, path, delete_branch, change_past,
1582
				       change_branch, change_branch, path, alt_path);
1583
			} else {
1584
				output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1585
				       "and %s to %s in %s. Version %s of %s left in tree at %s."),
1586
				       change, old_path, delete_branch, change_past, path,
1587
				       change_branch, change_branch, path, alt_path);
1588
			}
1589
		}
1590
		/*
1591
		 * No need to call update_file() on path when change_branch ==
1592
		 * opt->branch1 && !alt_path, since that would needlessly touch
1593
		 * path.  We could call update_file_flags() with update_cache=0
1594
		 * and update_wd=0, but that's a no-op.
1595
		 */
1596
		if (change_branch != opt->branch1 || alt_path)
1597
			ret = update_file(opt, 0, changed, update_path);
1598
	}
1599
	free(alt_path);
1600

1601
	return ret;
1602
}
1603

1604
static int handle_rename_delete(struct merge_options *opt,
1605
				struct rename_conflict_info *ci)
1606
{
1607
	const struct rename *ren = ci->ren1;
1608
	const struct diff_filespec *orig = ren->pair->one;
1609
	const struct diff_filespec *dest = ren->pair->two;
1610
	const char *rename_branch = ren->branch;
1611
	const char *delete_branch = (opt->branch1 == ren->branch ?
1612
				     opt->branch2 : opt->branch1);
1613

1614
	if (handle_change_delete(opt,
1615
				 opt->priv->call_depth ? orig->path : dest->path,
1616
				 opt->priv->call_depth ? NULL : orig->path,
1617
				 orig, dest,
1618
				 rename_branch, delete_branch,
1619
				 _("rename"), _("renamed")))
1620
		return -1;
1621

1622
	if (opt->priv->call_depth)
1623
		return remove_file_from_index(opt->repo->index, dest->path);
1624
	else
1625
		return update_stages(opt, dest->path, NULL,
1626
				     rename_branch == opt->branch1 ? dest : NULL,
1627
				     rename_branch == opt->branch1 ? NULL : dest);
1628
}
1629

1630
static int handle_file_collision(struct merge_options *opt,
1631
				 const char *collide_path,
1632
				 const char *prev_path1,
1633
				 const char *prev_path2,
1634
				 const char *branch1, const char *branch2,
1635
				 struct diff_filespec *a,
1636
				 struct diff_filespec *b)
1637
{
1638
	struct merge_file_info mfi;
1639
	struct diff_filespec null;
1640
	char *alt_path = NULL;
1641
	const char *update_path = collide_path;
1642

1643
	/*
1644
	 * It's easiest to get the correct things into stage 2 and 3, and
1645
	 * to make sure that the content merge puts HEAD before the other
1646
	 * branch if we just ensure that branch1 == opt->branch1.  So, simply
1647
	 * flip arguments around if we don't have that.
1648
	 */
1649
	if (branch1 != opt->branch1) {
1650
		return handle_file_collision(opt, collide_path,
1651
					     prev_path2, prev_path1,
1652
					     branch2, branch1,
1653
					     b, a);
1654
	}
1655

1656
	/* Remove rename sources if rename/add or rename/rename(2to1) */
1657
	if (prev_path1)
1658
		remove_file(opt, 1, prev_path1,
1659
			    opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1660
	if (prev_path2)
1661
		remove_file(opt, 1, prev_path2,
1662
			    opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1663

1664
	/*
1665
	 * Remove the collision path, if it wouldn't cause dirty contents
1666
	 * or an untracked file to get lost.  We'll either overwrite with
1667
	 * merged contents, or just write out to differently named files.
1668
	 */
1669
	if (was_dirty(opt, collide_path)) {
1670
		output(opt, 1, _("Refusing to lose dirty file at %s"),
1671
		       collide_path);
1672
		update_path = alt_path = unique_path(opt, collide_path, "merged");
1673
	} else if (would_lose_untracked(opt, collide_path)) {
1674
		/*
1675
		 * Only way we get here is if both renames were from
1676
		 * a directory rename AND user had an untracked file
1677
		 * at the location where both files end up after the
1678
		 * two directory renames.  See testcase 10d of t6043.
1679
		 */
1680
		output(opt, 1, _("Refusing to lose untracked file at "
1681
			       "%s, even though it's in the way."),
1682
		       collide_path);
1683
		update_path = alt_path = unique_path(opt, collide_path, "merged");
1684
	} else {
1685
		/*
1686
		 * FIXME: It's possible that the two files are identical
1687
		 * and that the current working copy happens to match, in
1688
		 * which case we are unnecessarily touching the working
1689
		 * tree file.  It's not a likely enough scenario that I
1690
		 * want to code up the checks for it and a better fix is
1691
		 * available if we restructure how unpack_trees() and
1692
		 * merge-recursive interoperate anyway, so punting for
1693
		 * now...
1694
		 */
1695
		remove_file(opt, 0, collide_path, 0);
1696
	}
1697

1698
	/* Store things in diff_filespecs for functions that need it */
1699
	null.path = (char *)collide_path;
1700
	oidcpy(&null.oid, null_oid());
1701
	null.mode = 0;
1702

1703
	if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1704
				    branch1, branch2, opt->priv->call_depth * 2, &mfi))
1705
		return -1;
1706
	mfi.clean &= !alt_path;
1707
	if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1708
		return -1;
1709
	if (!mfi.clean && !opt->priv->call_depth &&
1710
	    update_stages(opt, collide_path, NULL, a, b))
1711
		return -1;
1712
	free(alt_path);
1713
	/*
1714
	 * FIXME: If both a & b both started with conflicts (only possible
1715
	 * if they came from a rename/rename(2to1)), but had IDENTICAL
1716
	 * contents including those conflicts, then in the next line we claim
1717
	 * it was clean.  If someone cares about this case, we should have the
1718
	 * caller notify us if we started with conflicts.
1719
	 */
1720
	return mfi.clean;
1721
}
1722

1723
static int handle_rename_add(struct merge_options *opt,
1724
			     struct rename_conflict_info *ci)
1725
{
1726
	/* a was renamed to c, and a separate c was added. */
1727
	struct diff_filespec *a = ci->ren1->pair->one;
1728
	struct diff_filespec *c = ci->ren1->pair->two;
1729
	char *path = c->path;
1730
	char *prev_path_desc;
1731
	struct merge_file_info mfi;
1732

1733
	const char *rename_branch = ci->ren1->branch;
1734
	const char *add_branch = (opt->branch1 == rename_branch ?
1735
				  opt->branch2 : opt->branch1);
1736
	int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1737

1738
	output(opt, 1, _("CONFLICT (rename/add): "
1739
	       "Rename %s->%s in %s.  Added %s in %s"),
1740
	       a->path, c->path, rename_branch,
1741
	       c->path, add_branch);
1742

1743
	prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1744
	ci->ren1->src_entry->stages[other_stage].path = a->path;
1745
	if (merge_mode_and_contents(opt, a, c,
1746
				    &ci->ren1->src_entry->stages[other_stage],
1747
				    prev_path_desc,
1748
				    opt->branch1, opt->branch2,
1749
				    1 + opt->priv->call_depth * 2, &mfi))
1750
		return -1;
1751
	free(prev_path_desc);
1752

1753
	ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1754
	return handle_file_collision(opt,
1755
				     c->path, a->path, NULL,
1756
				     rename_branch, add_branch,
1757
				     &mfi.blob,
1758
				     &ci->ren1->dst_entry->stages[other_stage]);
1759
}
1760

1761
static char *find_path_for_conflict(struct merge_options *opt,
1762
				    const char *path,
1763
				    const char *branch1,
1764
				    const char *branch2)
1765
{
1766
	char *new_path = NULL;
1767
	if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1768
		new_path = unique_path(opt, path, branch1);
1769
		output(opt, 1, _("%s is a directory in %s adding "
1770
			       "as %s instead"),
1771
		       path, branch2, new_path);
1772
	} else if (would_lose_untracked(opt, path)) {
1773
		new_path = unique_path(opt, path, branch1);
1774
		output(opt, 1, _("Refusing to lose untracked file"
1775
			       " at %s; adding as %s instead"),
1776
		       path, new_path);
1777
	}
1778

1779
	return new_path;
1780
}
1781

1782
/*
1783
 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1784
 */
1785
static inline int flip_stage(int stage)
1786
{
1787
	return (2 + 3) - stage;
1788
}
1789

1790
static int handle_rename_rename_1to2(struct merge_options *opt,
1791
				     struct rename_conflict_info *ci)
1792
{
1793
	/* One file was renamed in both branches, but to different names. */
1794
	struct merge_file_info mfi;
1795
	struct diff_filespec *add;
1796
	struct diff_filespec *o = ci->ren1->pair->one;
1797
	struct diff_filespec *a = ci->ren1->pair->two;
1798
	struct diff_filespec *b = ci->ren2->pair->two;
1799
	char *path_desc;
1800

1801
	output(opt, 1, _("CONFLICT (rename/rename): "
1802
	       "Rename \"%s\"->\"%s\" in branch \"%s\" "
1803
	       "rename \"%s\"->\"%s\" in \"%s\"%s"),
1804
	       o->path, a->path, ci->ren1->branch,
1805
	       o->path, b->path, ci->ren2->branch,
1806
	       opt->priv->call_depth ? _(" (left unresolved)") : "");
1807

1808
	path_desc = xstrfmt("%s and %s, both renamed from %s",
1809
			    a->path, b->path, o->path);
1810
	if (merge_mode_and_contents(opt, o, a, b, path_desc,
1811
				    ci->ren1->branch, ci->ren2->branch,
1812
				    opt->priv->call_depth * 2, &mfi))
1813
		return -1;
1814
	free(path_desc);
1815

1816
	if (opt->priv->call_depth)
1817
		remove_file_from_index(opt->repo->index, o->path);
1818

1819
	/*
1820
	 * For each destination path, we need to see if there is a
1821
	 * rename/add collision.  If not, we can write the file out
1822
	 * to the specified location.
1823
	 */
1824
	add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1825
	if (is_valid(add)) {
1826
		add->path = mfi.blob.path = a->path;
1827
		if (handle_file_collision(opt, a->path,
1828
					  NULL, NULL,
1829
					  ci->ren1->branch,
1830
					  ci->ren2->branch,
1831
					  &mfi.blob, add) < 0)
1832
			return -1;
1833
	} else {
1834
		char *new_path = find_path_for_conflict(opt, a->path,
1835
							ci->ren1->branch,
1836
							ci->ren2->branch);
1837
		if (update_file(opt, 0, &mfi.blob,
1838
				new_path ? new_path : a->path))
1839
			return -1;
1840
		free(new_path);
1841
		if (!opt->priv->call_depth &&
1842
		    update_stages(opt, a->path, NULL, a, NULL))
1843
			return -1;
1844
	}
1845

1846
	if (!mfi.clean && mfi.blob.mode == a->mode &&
1847
	    oideq(&mfi.blob.oid, &a->oid)) {
1848
		/*
1849
		 * Getting here means we were attempting to merge a binary
1850
		 * blob.  Since we can't merge binaries, the merge algorithm
1851
		 * just takes one side.  But we don't want to copy the
1852
		 * contents of one side to both paths; we'd rather use the
1853
		 * original content at the given path for each path.
1854
		 */
1855
		oidcpy(&mfi.blob.oid, &b->oid);
1856
		mfi.blob.mode = b->mode;
1857
	}
1858
	add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1859
	if (is_valid(add)) {
1860
		add->path = mfi.blob.path = b->path;
1861
		if (handle_file_collision(opt, b->path,
1862
					  NULL, NULL,
1863
					  ci->ren1->branch,
1864
					  ci->ren2->branch,
1865
					  add, &mfi.blob) < 0)
1866
			return -1;
1867
	} else {
1868
		char *new_path = find_path_for_conflict(opt, b->path,
1869
							ci->ren2->branch,
1870
							ci->ren1->branch);
1871
		if (update_file(opt, 0, &mfi.blob,
1872
				new_path ? new_path : b->path))
1873
			return -1;
1874
		free(new_path);
1875
		if (!opt->priv->call_depth &&
1876
		    update_stages(opt, b->path, NULL, NULL, b))
1877
			return -1;
1878
	}
1879

1880
	return 0;
1881
}
1882

1883
static int handle_rename_rename_2to1(struct merge_options *opt,
1884
				     struct rename_conflict_info *ci)
1885
{
1886
	/* Two files, a & b, were renamed to the same thing, c. */
1887
	struct diff_filespec *a = ci->ren1->pair->one;
1888
	struct diff_filespec *b = ci->ren2->pair->one;
1889
	struct diff_filespec *c1 = ci->ren1->pair->two;
1890
	struct diff_filespec *c2 = ci->ren2->pair->two;
1891
	char *path = c1->path; /* == c2->path */
1892
	char *path_side_1_desc;
1893
	char *path_side_2_desc;
1894
	struct merge_file_info mfi_c1;
1895
	struct merge_file_info mfi_c2;
1896
	int ostage1, ostage2;
1897

1898
	output(opt, 1, _("CONFLICT (rename/rename): "
1899
	       "Rename %s->%s in %s. "
1900
	       "Rename %s->%s in %s"),
1901
	       a->path, c1->path, ci->ren1->branch,
1902
	       b->path, c2->path, ci->ren2->branch);
1903

1904
	path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1905
	path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1906
	ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1907
	ostage2 = flip_stage(ostage1);
1908
	ci->ren1->src_entry->stages[ostage1].path = a->path;
1909
	ci->ren2->src_entry->stages[ostage2].path = b->path;
1910
	if (merge_mode_and_contents(opt, a, c1,
1911
				    &ci->ren1->src_entry->stages[ostage1],
1912
				    path_side_1_desc,
1913
				    opt->branch1, opt->branch2,
1914
				    1 + opt->priv->call_depth * 2, &mfi_c1) ||
1915
	    merge_mode_and_contents(opt, b,
1916
				    &ci->ren2->src_entry->stages[ostage2],
1917
				    c2, path_side_2_desc,
1918
				    opt->branch1, opt->branch2,
1919
				    1 + opt->priv->call_depth * 2, &mfi_c2))
1920
		return -1;
1921
	free(path_side_1_desc);
1922
	free(path_side_2_desc);
1923
	mfi_c1.blob.path = path;
1924
	mfi_c2.blob.path = path;
1925

1926
	return handle_file_collision(opt, path, a->path, b->path,
1927
				     ci->ren1->branch, ci->ren2->branch,
1928
				     &mfi_c1.blob, &mfi_c2.blob);
1929
}
1930

1931
/*
1932
 * Get the diff_filepairs changed between o_tree and tree.
1933
 */
1934
static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1935
					       struct tree *o_tree,
1936
					       struct tree *tree)
1937
{
1938
	struct diff_queue_struct *ret;
1939
	struct diff_options opts;
1940

1941
	repo_diff_setup(opt->repo, &opts);
1942
	opts.flags.recursive = 1;
1943
	opts.flags.rename_empty = 0;
1944
	opts.detect_rename = merge_detect_rename(opt);
1945
	/*
1946
	 * We do not have logic to handle the detection of copies.  In
1947
	 * fact, it may not even make sense to add such logic: would we
1948
	 * really want a change to a base file to be propagated through
1949
	 * multiple other files by a merge?
1950
	 */
1951
	if (opts.detect_rename > DIFF_DETECT_RENAME)
1952
		opts.detect_rename = DIFF_DETECT_RENAME;
1953
	opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 7000;
1954
	opts.rename_score = opt->rename_score;
1955
	opts.show_rename_progress = opt->show_rename_progress;
1956
	opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1957
	diff_setup_done(&opts);
1958
	diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1959
	diffcore_std(&opts);
1960
	if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1961
		opt->priv->needed_rename_limit = opts.needed_rename_limit;
1962

1963
	ret = xmalloc(sizeof(*ret));
1964
	*ret = diff_queued_diff;
1965

1966
	opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1967
	diff_queued_diff.nr = 0;
1968
	diff_queued_diff.queue = NULL;
1969
	diff_flush(&opts);
1970
	return ret;
1971
}
1972

1973
static int tree_has_path(struct repository *r, struct tree *tree,
1974
			 const char *path)
1975
{
1976
	struct object_id hashy;
1977
	unsigned short mode_o;
1978

1979
	return !get_tree_entry(r,
1980
			       &tree->object.oid, path,
1981
			       &hashy, &mode_o);
1982
}
1983

1984
/*
1985
 * Return a new string that replaces the beginning portion (which matches
1986
 * entry->dir), with entry->new_dir.  In perl-speak:
1987
 *   new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1988
 * NOTE:
1989
 *   Caller must ensure that old_path starts with entry->dir + '/'.
1990
 */
1991
static char *apply_dir_rename(struct dir_rename_entry *entry,
1992
			      const char *old_path)
1993
{
1994
	struct strbuf new_path = STRBUF_INIT;
1995
	int oldlen, newlen;
1996

1997
	if (entry->non_unique_new_dir)
1998
		return NULL;
1999

2000
	oldlen = strlen(entry->dir);
2001
	if (entry->new_dir.len == 0)
2002
		/*
2003
		 * If someone renamed/merged a subdirectory into the root
2004
		 * directory (e.g. 'some/subdir' -> ''), then we want to
2005
		 * avoid returning
2006
		 *     '' + '/filename'
2007
		 * as the rename; we need to make old_path + oldlen advance
2008
		 * past the '/' character.
2009
		 */
2010
		oldlen++;
2011
	newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
2012
	strbuf_grow(&new_path, newlen);
2013
	strbuf_addbuf(&new_path, &entry->new_dir);
2014
	strbuf_addstr(&new_path, &old_path[oldlen]);
2015

2016
	return strbuf_detach(&new_path, NULL);
2017
}
2018

2019
static void get_renamed_dir_portion(const char *old_path, const char *new_path,
2020
				    char **old_dir, char **new_dir)
2021
{
2022
	char *end_of_old, *end_of_new;
2023

2024
	/* Default return values: NULL, meaning no rename */
2025
	*old_dir = NULL;
2026
	*new_dir = NULL;
2027

2028
	/*
2029
	 * For
2030
	 *    "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
2031
	 * the "e/foo.c" part is the same, we just want to know that
2032
	 *    "a/b/c/d" was renamed to "a/b/some/thing/else"
2033
	 * so, for this example, this function returns "a/b/c/d" in
2034
	 * *old_dir and "a/b/some/thing/else" in *new_dir.
2035
	 */
2036

2037
	/*
2038
	 * If the basename of the file changed, we don't care.  We want
2039
	 * to know which portion of the directory, if any, changed.
2040
	 */
2041
	end_of_old = strrchr(old_path, '/');
2042
	end_of_new = strrchr(new_path, '/');
2043

2044
	/*
2045
	 * If end_of_old is NULL, old_path wasn't in a directory, so there
2046
	 * could not be a directory rename (our rule elsewhere that a
2047
	 * directory which still exists is not considered to have been
2048
	 * renamed means the root directory can never be renamed -- because
2049
	 * the root directory always exists).
2050
	 */
2051
	if (!end_of_old)
2052
		return; /* Note: *old_dir and *new_dir are still NULL */
2053

2054
	/*
2055
	 * If new_path contains no directory (end_of_new is NULL), then we
2056
	 * have a rename of old_path's directory to the root directory.
2057
	 */
2058
	if (!end_of_new) {
2059
		*old_dir = xstrndup(old_path, end_of_old - old_path);
2060
		*new_dir = xstrdup("");
2061
		return;
2062
	}
2063

2064
	/* Find the first non-matching character traversing backwards */
2065
	while (*--end_of_new == *--end_of_old &&
2066
	       end_of_old != old_path &&
2067
	       end_of_new != new_path)
2068
		; /* Do nothing; all in the while loop */
2069

2070
	/*
2071
	 * If both got back to the beginning of their strings, then the
2072
	 * directory didn't change at all, only the basename did.
2073
	 */
2074
	if (end_of_old == old_path && end_of_new == new_path &&
2075
	    *end_of_old == *end_of_new)
2076
		return; /* Note: *old_dir and *new_dir are still NULL */
2077

2078
	/*
2079
	 * If end_of_new got back to the beginning of its string, and
2080
	 * end_of_old got back to the beginning of some subdirectory, then
2081
	 * we have a rename/merge of a subdirectory into the root, which
2082
	 * needs slightly special handling.
2083
	 *
2084
	 * Note: There is no need to consider the opposite case, with a
2085
	 * rename/merge of the root directory into some subdirectory
2086
	 * because as noted above the root directory always exists so it
2087
	 * cannot be considered to be renamed.
2088
	 */
2089
	if (end_of_new == new_path &&
2090
	    end_of_old != old_path && end_of_old[-1] == '/') {
2091
		*old_dir = xstrndup(old_path, --end_of_old - old_path);
2092
		*new_dir = xstrdup("");
2093
		return;
2094
	}
2095

2096
	/*
2097
	 * We've found the first non-matching character in the directory
2098
	 * paths.  That means the current characters we were looking at
2099
	 * were part of the first non-matching subdir name going back from
2100
	 * the end of the strings.  Get the whole name by advancing both
2101
	 * end_of_old and end_of_new to the NEXT '/' character.  That will
2102
	 * represent the entire directory rename.
2103
	 *
2104
	 * The reason for the increment is cases like
2105
	 *    a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2106
	 * After dropping the basename and going back to the first
2107
	 * non-matching character, we're now comparing:
2108
	 *    a/b/s          and         a/b/
2109
	 * and we want to be comparing:
2110
	 *    a/b/star/      and         a/b/tar/
2111
	 * but without the pre-increment, the one on the right would stay
2112
	 * a/b/.
2113
	 */
2114
	end_of_old = strchr(++end_of_old, '/');
2115
	end_of_new = strchr(++end_of_new, '/');
2116

2117
	/* Copy the old and new directories into *old_dir and *new_dir. */
2118
	*old_dir = xstrndup(old_path, end_of_old - old_path);
2119
	*new_dir = xstrndup(new_path, end_of_new - new_path);
2120
}
2121

2122
static void remove_hashmap_entries(struct hashmap *dir_renames,
2123
				   struct string_list *items_to_remove)
2124
{
2125
	int i;
2126
	struct dir_rename_entry *entry;
2127

2128
	for (i = 0; i < items_to_remove->nr; i++) {
2129
		entry = items_to_remove->items[i].util;
2130
		hashmap_remove(dir_renames, &entry->ent, NULL);
2131
	}
2132
	string_list_clear(items_to_remove, 0);
2133
}
2134

2135
/*
2136
 * See if there is a directory rename for path, and if there are any file
2137
 * level conflicts for the renamed location.  If there is a rename and
2138
 * there are no conflicts, return the new name.  Otherwise, return NULL.
2139
 */
2140
static char *handle_path_level_conflicts(struct merge_options *opt,
2141
					 const char *path,
2142
					 struct dir_rename_entry *entry,
2143
					 struct hashmap *collisions,
2144
					 struct tree *tree)
2145
{
2146
	char *new_path = NULL;
2147
	struct collision_entry *collision_ent;
2148
	int clean = 1;
2149
	struct strbuf collision_paths = STRBUF_INIT;
2150

2151
	/*
2152
	 * entry has the mapping of old directory name to new directory name
2153
	 * that we want to apply to path.
2154
	 */
2155
	new_path = apply_dir_rename(entry, path);
2156

2157
	if (!new_path) {
2158
		/* This should only happen when entry->non_unique_new_dir set */
2159
		if (!entry->non_unique_new_dir)
2160
			BUG("entry->non_unique_new_dir not set and !new_path");
2161
		output(opt, 1, _("CONFLICT (directory rename split): "
2162
			       "Unclear where to place %s because directory "
2163
			       "%s was renamed to multiple other directories, "
2164
			       "with no destination getting a majority of the "
2165
			       "files."),
2166
		       path, entry->dir);
2167
		clean = 0;
2168
		return NULL;
2169
	}
2170

2171
	/*
2172
	 * The caller needs to have ensured that it has pre-populated
2173
	 * collisions with all paths that map to new_path.  Do a quick check
2174
	 * to ensure that's the case.
2175
	 */
2176
	collision_ent = collision_find_entry(collisions, new_path);
2177
	if (!collision_ent)
2178
		BUG("collision_ent is NULL");
2179

2180
	/*
2181
	 * Check for one-sided add/add/.../add conflicts, i.e.
2182
	 * where implicit renames from the other side doing
2183
	 * directory rename(s) can affect this side of history
2184
	 * to put multiple paths into the same location.  Warn
2185
	 * and bail on directory renames for such paths.
2186
	 */
2187
	if (collision_ent->reported_already) {
2188
		clean = 0;
2189
	} else if (tree_has_path(opt->repo, tree, new_path)) {
2190
		collision_ent->reported_already = 1;
2191
		strbuf_add_separated_string_list(&collision_paths, ", ",
2192
						 &collision_ent->source_files);
2193
		output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2194
			       "file/dir at %s in the way of implicit "
2195
			       "directory rename(s) putting the following "
2196
			       "path(s) there: %s."),
2197
		       new_path, collision_paths.buf);
2198
		clean = 0;
2199
	} else if (collision_ent->source_files.nr > 1) {
2200
		collision_ent->reported_already = 1;
2201
		strbuf_add_separated_string_list(&collision_paths, ", ",
2202
						 &collision_ent->source_files);
2203
		output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2204
			       "more than one path to %s; implicit directory "
2205
			       "renames tried to put these paths there: %s"),
2206
		       new_path, collision_paths.buf);
2207
		clean = 0;
2208
	}
2209

2210
	/* Free memory we no longer need */
2211
	strbuf_release(&collision_paths);
2212
	if (!clean && new_path) {
2213
		free(new_path);
2214
		return NULL;
2215
	}
2216

2217
	return new_path;
2218
}
2219

2220
/*
2221
 * There are a couple things we want to do at the directory level:
2222
 *   1. Check for both sides renaming to the same thing, in order to avoid
2223
 *      implicit renaming of files that should be left in place.  (See
2224
 *      testcase 6b in t6043 for details.)
2225
 *   2. Prune directory renames if there are still files left in the
2226
 *      original directory.  These represent a partial directory rename,
2227
 *      i.e. a rename where only some of the files within the directory
2228
 *      were renamed elsewhere.  (Technically, this could be done earlier
2229
 *      in get_directory_renames(), except that would prevent us from
2230
 *      doing the previous check and thus failing testcase 6b.)
2231
 *   3. Check for rename/rename(1to2) conflicts (at the directory level).
2232
 *      In the future, we could potentially record this info as well and
2233
 *      omit reporting rename/rename(1to2) conflicts for each path within
2234
 *      the affected directories, thus cleaning up the merge output.
2235
 *   NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2236
 *         directory level, because merging directories is fine.  If it
2237
 *         causes conflicts for files within those merged directories, then
2238
 *         that should be detected at the individual path level.
2239
 */
2240
static void handle_directory_level_conflicts(struct merge_options *opt,
2241
					     struct hashmap *dir_re_head,
2242
					     struct tree *head,
2243
					     struct hashmap *dir_re_merge,
2244
					     struct tree *merge)
2245
{
2246
	struct hashmap_iter iter;
2247
	struct dir_rename_entry *head_ent;
2248
	struct dir_rename_entry *merge_ent;
2249

2250
	struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2251
	struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2252

2253
	hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2254
				ent /* member name */) {
2255
		merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2256
		if (merge_ent &&
2257
		    !head_ent->non_unique_new_dir &&
2258
		    !merge_ent->non_unique_new_dir &&
2259
		    !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2260
			/* 1. Renamed identically; remove it from both sides */
2261
			string_list_append(&remove_from_head,
2262
					   head_ent->dir)->util = head_ent;
2263
			strbuf_release(&head_ent->new_dir);
2264
			string_list_append(&remove_from_merge,
2265
					   merge_ent->dir)->util = merge_ent;
2266
			strbuf_release(&merge_ent->new_dir);
2267
		} else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2268
			/* 2. This wasn't a directory rename after all */
2269
			string_list_append(&remove_from_head,
2270
					   head_ent->dir)->util = head_ent;
2271
			strbuf_release(&head_ent->new_dir);
2272
		}
2273
	}
2274

2275
	remove_hashmap_entries(dir_re_head, &remove_from_head);
2276
	remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2277

2278
	hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2279
				ent /* member name */) {
2280
		head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2281
		if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2282
			/* 2. This wasn't a directory rename after all */
2283
			string_list_append(&remove_from_merge,
2284
					   merge_ent->dir)->util = merge_ent;
2285
		} else if (head_ent &&
2286
			   !head_ent->non_unique_new_dir &&
2287
			   !merge_ent->non_unique_new_dir) {
2288
			/* 3. rename/rename(1to2) */
2289
			/*
2290
			 * We can assume it's not rename/rename(1to1) because
2291
			 * that was case (1), already checked above.  So we
2292
			 * know that head_ent->new_dir and merge_ent->new_dir
2293
			 * are different strings.
2294
			 */
2295
			output(opt, 1, _("CONFLICT (rename/rename): "
2296
				       "Rename directory %s->%s in %s. "
2297
				       "Rename directory %s->%s in %s"),
2298
			       head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2299
			       head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2300
			string_list_append(&remove_from_head,
2301
					   head_ent->dir)->util = head_ent;
2302
			strbuf_release(&head_ent->new_dir);
2303
			string_list_append(&remove_from_merge,
2304
					   merge_ent->dir)->util = merge_ent;
2305
			strbuf_release(&merge_ent->new_dir);
2306
		}
2307
	}
2308

2309
	remove_hashmap_entries(dir_re_head, &remove_from_head);
2310
	remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2311
}
2312

2313
static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2314
{
2315
	struct hashmap *dir_renames;
2316
	struct hashmap_iter iter;
2317
	struct dir_rename_entry *entry;
2318
	int i;
2319

2320
	/*
2321
	 * Typically, we think of a directory rename as all files from a
2322
	 * certain directory being moved to a target directory.  However,
2323
	 * what if someone first moved two files from the original
2324
	 * directory in one commit, and then renamed the directory
2325
	 * somewhere else in a later commit?  At merge time, we just know
2326
	 * that files from the original directory went to two different
2327
	 * places, and that the bulk of them ended up in the same place.
2328
	 * We want each directory rename to represent where the bulk of the
2329
	 * files from that directory end up; this function exists to find
2330
	 * where the bulk of the files went.
2331
	 *
2332
	 * The first loop below simply iterates through the list of file
2333
	 * renames, finding out how often each directory rename pair
2334
	 * possibility occurs.
2335
	 */
2336
	dir_renames = xmalloc(sizeof(*dir_renames));
2337
	dir_rename_init(dir_renames);
2338
	for (i = 0; i < pairs->nr; ++i) {
2339
		struct string_list_item *item;
2340
		int *count;
2341
		struct diff_filepair *pair = pairs->queue[i];
2342
		char *old_dir, *new_dir;
2343

2344
		/* File not part of directory rename if it wasn't renamed */
2345
		if (pair->status != 'R')
2346
			continue;
2347

2348
		get_renamed_dir_portion(pair->one->path, pair->two->path,
2349
					&old_dir,        &new_dir);
2350
		if (!old_dir)
2351
			/* Directory didn't change at all; ignore this one. */
2352
			continue;
2353

2354
		entry = dir_rename_find_entry(dir_renames, old_dir);
2355
		if (!entry) {
2356
			entry = xmalloc(sizeof(*entry));
2357
			dir_rename_entry_init(entry, old_dir);
2358
			hashmap_put(dir_renames, &entry->ent);
2359
		} else {
2360
			free(old_dir);
2361
		}
2362
		item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2363
		if (!item) {
2364
			item = string_list_insert(&entry->possible_new_dirs,
2365
						  new_dir);
2366
			item->util = xcalloc(1, sizeof(int));
2367
		} else {
2368
			free(new_dir);
2369
		}
2370
		count = item->util;
2371
		*count += 1;
2372
	}
2373

2374
	/*
2375
	 * For each directory with files moved out of it, we find out which
2376
	 * target directory received the most files so we can declare it to
2377
	 * be the "winning" target location for the directory rename.  This
2378
	 * winner gets recorded in new_dir.  If there is no winner
2379
	 * (multiple target directories received the same number of files),
2380
	 * we set non_unique_new_dir.  Once we've determined the winner (or
2381
	 * that there is no winner), we no longer need possible_new_dirs.
2382
	 */
2383
	hashmap_for_each_entry(dir_renames, &iter, entry,
2384
				ent /* member name */) {
2385
		int max = 0;
2386
		int bad_max = 0;
2387
		char *best = NULL;
2388

2389
		for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2390
			int *count = entry->possible_new_dirs.items[i].util;
2391

2392
			if (*count == max)
2393
				bad_max = max;
2394
			else if (*count > max) {
2395
				max = *count;
2396
				best = entry->possible_new_dirs.items[i].string;
2397
			}
2398
		}
2399
		if (bad_max == max)
2400
			entry->non_unique_new_dir = 1;
2401
		else {
2402
			assert(entry->new_dir.len == 0);
2403
			strbuf_addstr(&entry->new_dir, best);
2404
		}
2405
		/*
2406
		 * The relevant directory sub-portion of the original full
2407
		 * filepaths were xstrndup'ed before inserting into
2408
		 * possible_new_dirs, and instead of manually iterating the
2409
		 * list and free'ing each, just lie and tell
2410
		 * possible_new_dirs that it did the strdup'ing so that it
2411
		 * will free them for us.
2412
		 */
2413
		entry->possible_new_dirs.strdup_strings = 1;
2414
		string_list_clear(&entry->possible_new_dirs, 1);
2415
	}
2416

2417
	return dir_renames;
2418
}
2419

2420
static struct dir_rename_entry *check_dir_renamed(const char *path,
2421
						  struct hashmap *dir_renames)
2422
{
2423
	char *temp = xstrdup(path);
2424
	char *end;
2425
	struct dir_rename_entry *entry = NULL;
2426

2427
	while ((end = strrchr(temp, '/'))) {
2428
		*end = '\0';
2429
		entry = dir_rename_find_entry(dir_renames, temp);
2430
		if (entry)
2431
			break;
2432
	}
2433
	free(temp);
2434
	return entry;
2435
}
2436

2437
static void compute_collisions(struct hashmap *collisions,
2438
			       struct hashmap *dir_renames,
2439
			       struct diff_queue_struct *pairs)
2440
{
2441
	int i;
2442

2443
	/*
2444
	 * Multiple files can be mapped to the same path due to directory
2445
	 * renames done by the other side of history.  Since that other
2446
	 * side of history could have merged multiple directories into one,
2447
	 * if our side of history added the same file basename to each of
2448
	 * those directories, then all N of them would get implicitly
2449
	 * renamed by the directory rename detection into the same path,
2450
	 * and we'd get an add/add/.../add conflict, and all those adds
2451
	 * from *this* side of history.  This is not representable in the
2452
	 * index, and users aren't going to easily be able to make sense of
2453
	 * it.  So we need to provide a good warning about what's
2454
	 * happening, and fall back to no-directory-rename detection
2455
	 * behavior for those paths.
2456
	 *
2457
	 * See testcases 9e and all of section 5 from t6043 for examples.
2458
	 */
2459
	collision_init(collisions);
2460

2461
	for (i = 0; i < pairs->nr; ++i) {
2462
		struct dir_rename_entry *dir_rename_ent;
2463
		struct collision_entry *collision_ent;
2464
		char *new_path;
2465
		struct diff_filepair *pair = pairs->queue[i];
2466

2467
		if (pair->status != 'A' && pair->status != 'R')
2468
			continue;
2469
		dir_rename_ent = check_dir_renamed(pair->two->path,
2470
						   dir_renames);
2471
		if (!dir_rename_ent)
2472
			continue;
2473

2474
		new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2475
		if (!new_path)
2476
			/*
2477
			 * dir_rename_ent->non_unique_new_path is true, which
2478
			 * means there is no directory rename for us to use,
2479
			 * which means it won't cause us any additional
2480
			 * collisions.
2481
			 */
2482
			continue;
2483
		collision_ent = collision_find_entry(collisions, new_path);
2484
		if (!collision_ent) {
2485
			CALLOC_ARRAY(collision_ent, 1);
2486
			hashmap_entry_init(&collision_ent->ent,
2487
						strhash(new_path));
2488
			hashmap_put(collisions, &collision_ent->ent);
2489
			collision_ent->target_file = new_path;
2490
		} else {
2491
			free(new_path);
2492
		}
2493
		string_list_insert(&collision_ent->source_files,
2494
				   pair->two->path);
2495
	}
2496
}
2497

2498
static char *check_for_directory_rename(struct merge_options *opt,
2499
					const char *path,
2500
					struct tree *tree,
2501
					struct hashmap *dir_renames,
2502
					struct hashmap *dir_rename_exclusions,
2503
					struct hashmap *collisions,
2504
					int *clean_merge)
2505
{
2506
	char *new_path = NULL;
2507
	struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2508
	struct dir_rename_entry *oentry = NULL;
2509

2510
	if (!entry)
2511
		return new_path;
2512

2513
	/*
2514
	 * This next part is a little weird.  We do not want to do an
2515
	 * implicit rename into a directory we renamed on our side, because
2516
	 * that will result in a spurious rename/rename(1to2) conflict.  An
2517
	 * example:
2518
	 *   Base commit: dumbdir/afile, otherdir/bfile
2519
	 *   Side 1:      smrtdir/afile, otherdir/bfile
2520
	 *   Side 2:      dumbdir/afile, dumbdir/bfile
2521
	 * Here, while working on Side 1, we could notice that otherdir was
2522
	 * renamed/merged to dumbdir, and change the diff_filepair for
2523
	 * otherdir/bfile into a rename into dumbdir/bfile.  However, Side
2524
	 * 2 will notice the rename from dumbdir to smrtdir, and do the
2525
	 * transitive rename to move it from dumbdir/bfile to
2526
	 * smrtdir/bfile.  That gives us bfile in dumbdir vs being in
2527
	 * smrtdir, a rename/rename(1to2) conflict.  We really just want
2528
	 * the file to end up in smrtdir.  And the way to achieve that is
2529
	 * to not let Side1 do the rename to dumbdir, since we know that is
2530
	 * the source of one of our directory renames.
2531
	 *
2532
	 * That's why oentry and dir_rename_exclusions is here.
2533
	 *
2534
	 * As it turns out, this also prevents N-way transient rename
2535
	 * confusion; See testcases 9c and 9d of t6043.
2536
	 */
2537
	oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2538
	if (oentry) {
2539
		output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2540
			       "to %s, because %s itself was renamed."),
2541
		       entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2542
	} else {
2543
		new_path = handle_path_level_conflicts(opt, path, entry,
2544
						       collisions, tree);
2545
		*clean_merge &= (new_path != NULL);
2546
	}
2547

2548
	return new_path;
2549
}
2550

2551
static void apply_directory_rename_modifications(struct merge_options *opt,
2552
						 struct diff_filepair *pair,
2553
						 char *new_path,
2554
						 struct rename *re,
2555
						 struct tree *tree,
2556
						 struct tree *o_tree,
2557
						 struct tree *a_tree,
2558
						 struct tree *b_tree,
2559
						 struct string_list *entries)
2560
{
2561
	struct string_list_item *item;
2562
	int stage = (tree == a_tree ? 2 : 3);
2563
	int update_wd;
2564

2565
	/*
2566
	 * In all cases where we can do directory rename detection,
2567
	 * unpack_trees() will have read pair->two->path into the
2568
	 * index and the working copy.  We need to remove it so that
2569
	 * we can instead place it at new_path.  It is guaranteed to
2570
	 * not be untracked (unpack_trees() would have errored out
2571
	 * saying the file would have been overwritten), but it might
2572
	 * be dirty, though.
2573
	 */
2574
	update_wd = !was_dirty(opt, pair->two->path);
2575
	if (!update_wd)
2576
		output(opt, 1, _("Refusing to lose dirty file at %s"),
2577
		       pair->two->path);
2578
	remove_file(opt, 1, pair->two->path, !update_wd);
2579

2580
	/* Find or create a new re->dst_entry */
2581
	item = string_list_lookup(entries, new_path);
2582
	if (item) {
2583
		/*
2584
		 * Since we're renaming on this side of history, and it's
2585
		 * due to a directory rename on the other side of history
2586
		 * (which we only allow when the directory in question no
2587
		 * longer exists on the other side of history), the
2588
		 * original entry for re->dst_entry is no longer
2589
		 * necessary...
2590
		 */
2591
		re->dst_entry->processed = 1;
2592

2593
		/*
2594
		 * ...because we'll be using this new one.
2595
		 */
2596
		re->dst_entry = item->util;
2597
	} else {
2598
		/*
2599
		 * re->dst_entry is for the before-dir-rename path, and we
2600
		 * need it to hold information for the after-dir-rename
2601
		 * path.  Before creating a new entry, we need to mark the
2602
		 * old one as unnecessary (...unless it is shared by
2603
		 * src_entry, i.e. this didn't use to be a rename, in which
2604
		 * case we can just allow the normal processing to happen
2605
		 * for it).
2606
		 */
2607
		if (pair->status == 'R')
2608
			re->dst_entry->processed = 1;
2609

2610
		re->dst_entry = insert_stage_data(opt->repo, new_path,
2611
						  o_tree, a_tree, b_tree,
2612
						  entries);
2613
		item = string_list_insert(entries, new_path);
2614
		item->util = re->dst_entry;
2615
	}
2616

2617
	/*
2618
	 * Update the stage_data with the information about the path we are
2619
	 * moving into place.  That slot will be empty and available for us
2620
	 * to write to because of the collision checks in
2621
	 * handle_path_level_conflicts().  In other words,
2622
	 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2623
	 * open for us to write to.
2624
	 *
2625
	 * It may be tempting to actually update the index at this point as
2626
	 * well, using update_stages_for_stage_data(), but as per the big
2627
	 * "NOTE" in update_stages(), doing so will modify the current
2628
	 * in-memory index which will break calls to would_lose_untracked()
2629
	 * that we need to make.  Instead, we need to just make sure that
2630
	 * the various handle_rename_*() functions update the index
2631
	 * explicitly rather than relying on unpack_trees() to have done it.
2632
	 */
2633
	get_tree_entry(opt->repo,
2634
		       &tree->object.oid,
2635
		       pair->two->path,
2636
		       &re->dst_entry->stages[stage].oid,
2637
		       &re->dst_entry->stages[stage].mode);
2638

2639
	/*
2640
	 * Record the original change status (or 'type' of change).  If it
2641
	 * was originally an add ('A'), this lets us differentiate later
2642
	 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2643
	 * otherwise look the same).  If it was originally a rename ('R'),
2644
	 * this lets us remember and report accurately about the transitive
2645
	 * renaming that occurred via the directory rename detection.  Also,
2646
	 * record the original destination name.
2647
	 */
2648
	re->dir_rename_original_type = pair->status;
2649
	re->dir_rename_original_dest = pair->two->path;
2650

2651
	/*
2652
	 * We don't actually look at pair->status again, but it seems
2653
	 * pedagogically correct to adjust it.
2654
	 */
2655
	pair->status = 'R';
2656

2657
	/*
2658
	 * Finally, record the new location.
2659
	 */
2660
	pair->two->path = new_path;
2661
}
2662

2663
/*
2664
 * Get information of all renames which occurred in 'pairs', making use of
2665
 * any implicit directory renames inferred from the other side of history.
2666
 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2667
 * to be able to associate the correct cache entries with the rename
2668
 * information; tree is always equal to either a_tree or b_tree.
2669
 */
2670
static struct string_list *get_renames(struct merge_options *opt,
2671
				       const char *branch,
2672
				       struct diff_queue_struct *pairs,
2673
				       struct hashmap *dir_renames,
2674
				       struct hashmap *dir_rename_exclusions,
2675
				       struct tree *tree,
2676
				       struct tree *o_tree,
2677
				       struct tree *a_tree,
2678
				       struct tree *b_tree,
2679
				       struct string_list *entries,
2680
				       int *clean_merge)
2681
{
2682
	int i;
2683
	struct hashmap collisions;
2684
	struct hashmap_iter iter;
2685
	struct collision_entry *e;
2686
	struct string_list *renames;
2687

2688
	compute_collisions(&collisions, dir_renames, pairs);
2689
	CALLOC_ARRAY(renames, 1);
2690

2691
	for (i = 0; i < pairs->nr; ++i) {
2692
		struct string_list_item *item;
2693
		struct rename *re;
2694
		struct diff_filepair *pair = pairs->queue[i];
2695
		char *new_path; /* non-NULL only with directory renames */
2696

2697
		if (pair->status != 'A' && pair->status != 'R') {
2698
			diff_free_filepair(pair);
2699
			continue;
2700
		}
2701
		new_path = check_for_directory_rename(opt, pair->two->path, tree,
2702
						      dir_renames,
2703
						      dir_rename_exclusions,
2704
						      &collisions,
2705
						      clean_merge);
2706
		if (pair->status != 'R' && !new_path) {
2707
			diff_free_filepair(pair);
2708
			continue;
2709
		}
2710

2711
		re = xmalloc(sizeof(*re));
2712
		re->processed = 0;
2713
		re->pair = pair;
2714
		re->branch = branch;
2715
		re->dir_rename_original_type = '\0';
2716
		re->dir_rename_original_dest = NULL;
2717
		item = string_list_lookup(entries, re->pair->one->path);
2718
		if (!item)
2719
			re->src_entry = insert_stage_data(opt->repo,
2720
					re->pair->one->path,
2721
					o_tree, a_tree, b_tree, entries);
2722
		else
2723
			re->src_entry = item->util;
2724

2725
		item = string_list_lookup(entries, re->pair->two->path);
2726
		if (!item)
2727
			re->dst_entry = insert_stage_data(opt->repo,
2728
					re->pair->two->path,
2729
					o_tree, a_tree, b_tree, entries);
2730
		else
2731
			re->dst_entry = item->util;
2732
		item = string_list_insert(renames, pair->one->path);
2733
		item->util = re;
2734
		if (new_path)
2735
			apply_directory_rename_modifications(opt, pair, new_path,
2736
							     re, tree, o_tree,
2737
							     a_tree, b_tree,
2738
							     entries);
2739
	}
2740

2741
	hashmap_for_each_entry(&collisions, &iter, e,
2742
				ent /* member name */) {
2743
		free(e->target_file);
2744
		string_list_clear(&e->source_files, 0);
2745
	}
2746
	hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2747
	return renames;
2748
}
2749

2750
static int process_renames(struct merge_options *opt,
2751
			   struct string_list *a_renames,
2752
			   struct string_list *b_renames)
2753
{
2754
	int clean_merge = 1, i, j;
2755
	struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2756
	struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2757
	const struct rename *sre;
2758

2759
	/*
2760
	 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2761
	 * string_list one-by-one, but O(n log n) to build it unsorted and
2762
	 * then sort it.  Note that as we build the list, we do not need to
2763
	 * check if the existing destination path is already in the list,
2764
	 * because the structure of diffcore_rename guarantees we won't
2765
	 * have duplicates.
2766
	 */
2767
	for (i = 0; i < a_renames->nr; i++) {
2768
		sre = a_renames->items[i].util;
2769
		string_list_insert(&a_by_dst, sre->pair->two->path)->util
2770
			= (void *)sre;
2771
	}
2772
	for (i = 0; i < b_renames->nr; i++) {
2773
		sre = b_renames->items[i].util;
2774
		string_list_insert(&b_by_dst, sre->pair->two->path)->util
2775
			= (void *)sre;
2776
	}
2777

2778
	for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2779
		struct string_list *renames1, *renames2Dst;
2780
		struct rename *ren1 = NULL, *ren2 = NULL;
2781
		const char *ren1_src, *ren1_dst;
2782
		struct string_list_item *lookup;
2783

2784
		if (i >= a_renames->nr) {
2785
			ren2 = b_renames->items[j++].util;
2786
		} else if (j >= b_renames->nr) {
2787
			ren1 = a_renames->items[i++].util;
2788
		} else {
2789
			int compare = strcmp(a_renames->items[i].string,
2790
					     b_renames->items[j].string);
2791
			if (compare <= 0)
2792
				ren1 = a_renames->items[i++].util;
2793
			if (compare >= 0)
2794
				ren2 = b_renames->items[j++].util;
2795
		}
2796

2797
		/* TODO: refactor, so that 1/2 are not needed */
2798
		if (ren1) {
2799
			renames1 = a_renames;
2800
			renames2Dst = &b_by_dst;
2801
		} else {
2802
			renames1 = b_renames;
2803
			renames2Dst = &a_by_dst;
2804
			SWAP(ren2, ren1);
2805
		}
2806

2807
		if (ren1->processed)
2808
			continue;
2809
		ren1->processed = 1;
2810
		ren1->dst_entry->processed = 1;
2811
		/* BUG: We should only mark src_entry as processed if we
2812
		 * are not dealing with a rename + add-source case.
2813
		 */
2814
		ren1->src_entry->processed = 1;
2815

2816
		ren1_src = ren1->pair->one->path;
2817
		ren1_dst = ren1->pair->two->path;
2818

2819
		if (ren2) {
2820
			/* One file renamed on both sides */
2821
			const char *ren2_src = ren2->pair->one->path;
2822
			const char *ren2_dst = ren2->pair->two->path;
2823
			enum rename_type rename_type;
2824
			if (strcmp(ren1_src, ren2_src) != 0)
2825
				BUG("ren1_src != ren2_src");
2826
			ren2->dst_entry->processed = 1;
2827
			ren2->processed = 1;
2828
			if (strcmp(ren1_dst, ren2_dst) != 0) {
2829
				rename_type = RENAME_ONE_FILE_TO_TWO;
2830
				clean_merge = 0;
2831
			} else {
2832
				rename_type = RENAME_ONE_FILE_TO_ONE;
2833
				/* BUG: We should only remove ren1_src in
2834
				 * the base stage (think of rename +
2835
				 * add-source cases).
2836
				 */
2837
				remove_file(opt, 1, ren1_src, 1);
2838
				update_entry(ren1->dst_entry,
2839
					     ren1->pair->one,
2840
					     ren1->pair->two,
2841
					     ren2->pair->two);
2842
			}
2843
			setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2844
		} else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2845
			/* Two different files renamed to the same thing */
2846
			char *ren2_dst;
2847
			ren2 = lookup->util;
2848
			ren2_dst = ren2->pair->two->path;
2849
			if (strcmp(ren1_dst, ren2_dst) != 0)
2850
				BUG("ren1_dst != ren2_dst");
2851

2852
			clean_merge = 0;
2853
			ren2->processed = 1;
2854
			/*
2855
			 * BUG: We should only mark src_entry as processed
2856
			 * if we are not dealing with a rename + add-source
2857
			 * case.
2858
			 */
2859
			ren2->src_entry->processed = 1;
2860

2861
			setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2862
						   opt, ren1, ren2);
2863
		} else {
2864
			/* Renamed in 1, maybe changed in 2 */
2865
			/* we only use sha1 and mode of these */
2866
			struct diff_filespec src_other, dst_other;
2867
			int try_merge;
2868

2869
			/*
2870
			 * unpack_trees loads entries from common-commit
2871
			 * into stage 1, from head-commit into stage 2, and
2872
			 * from merge-commit into stage 3.  We keep track
2873
			 * of which side corresponds to the rename.
2874
			 */
2875
			int renamed_stage = a_renames == renames1 ? 2 : 3;
2876
			int other_stage =   a_renames == renames1 ? 3 : 2;
2877

2878
			/*
2879
			 * Directory renames have a funny corner case...
2880
			 */
2881
			int renamed_to_self = !strcmp(ren1_src, ren1_dst);
2882

2883
			/* BUG: We should only remove ren1_src in the base
2884
			 * stage and in other_stage (think of rename +
2885
			 * add-source case).
2886
			 */
2887
			if (!renamed_to_self)
2888
				remove_file(opt, 1, ren1_src,
2889
					    renamed_stage == 2 ||
2890
					    !was_tracked(opt, ren1_src));
2891

2892
			oidcpy(&src_other.oid,
2893
			       &ren1->src_entry->stages[other_stage].oid);
2894
			src_other.mode = ren1->src_entry->stages[other_stage].mode;
2895
			oidcpy(&dst_other.oid,
2896
			       &ren1->dst_entry->stages[other_stage].oid);
2897
			dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2898
			try_merge = 0;
2899

2900
			if (oideq(&src_other.oid, null_oid()) &&
2901
			    ren1->dir_rename_original_type == 'A') {
2902
				setup_rename_conflict_info(RENAME_VIA_DIR,
2903
							   opt, ren1, NULL);
2904
			} else if (renamed_to_self) {
2905
				setup_rename_conflict_info(RENAME_NORMAL,
2906
							   opt, ren1, NULL);
2907
			} else if (oideq(&src_other.oid, null_oid())) {
2908
				setup_rename_conflict_info(RENAME_DELETE,
2909
							   opt, ren1, NULL);
2910
			} else if ((dst_other.mode == ren1->pair->two->mode) &&
2911
				   oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2912
				/*
2913
				 * Added file on the other side identical to
2914
				 * the file being renamed: clean merge.
2915
				 * Also, there is no need to overwrite the
2916
				 * file already in the working copy, so call
2917
				 * update_file_flags() instead of
2918
				 * update_file().
2919
				 */
2920
				if (update_file_flags(opt,
2921
						      ren1->pair->two,
2922
						      ren1_dst,
2923
						      1, /* update_cache */
2924
						      0  /* update_wd    */))
2925
					clean_merge = -1;
2926
			} else if (!oideq(&dst_other.oid, null_oid())) {
2927
				/*
2928
				 * Probably not a clean merge, but it's
2929
				 * premature to set clean_merge to 0 here,
2930
				 * because if the rename merges cleanly and
2931
				 * the merge exactly matches the newly added
2932
				 * file, then the merge will be clean.
2933
				 */
2934
				setup_rename_conflict_info(RENAME_ADD,
2935
							   opt, ren1, NULL);
2936
			} else
2937
				try_merge = 1;
2938

2939
			if (clean_merge < 0)
2940
				goto cleanup_and_return;
2941
			if (try_merge) {
2942
				struct diff_filespec *o, *a, *b;
2943
				src_other.path = (char *)ren1_src;
2944

2945
				o = ren1->pair->one;
2946
				if (a_renames == renames1) {
2947
					a = ren1->pair->two;
2948
					b = &src_other;
2949
				} else {
2950
					b = ren1->pair->two;
2951
					a = &src_other;
2952
				}
2953
				update_entry(ren1->dst_entry, o, a, b);
2954
				setup_rename_conflict_info(RENAME_NORMAL,
2955
							   opt, ren1, NULL);
2956
			}
2957
		}
2958
	}
2959
cleanup_and_return:
2960
	string_list_clear(&a_by_dst, 0);
2961
	string_list_clear(&b_by_dst, 0);
2962

2963
	return clean_merge;
2964
}
2965

2966
struct rename_info {
2967
	struct string_list *head_renames;
2968
	struct string_list *merge_renames;
2969
};
2970

2971
static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2972
				   struct hashmap *dir_renames)
2973
{
2974
	struct hashmap_iter iter;
2975
	struct dir_rename_entry *e;
2976

2977
	hashmap_for_each_entry(dir_renames, &iter, e,
2978
				ent /* member name */) {
2979
		free(e->dir);
2980
		strbuf_release(&e->new_dir);
2981
		/* possible_new_dirs already cleared in get_directory_renames */
2982
	}
2983
	hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2984
	free(dir_renames);
2985

2986
	free(pairs->queue);
2987
	free(pairs);
2988
}
2989

2990
static int detect_and_process_renames(struct merge_options *opt,
2991
				      struct tree *common,
2992
				      struct tree *head,
2993
				      struct tree *merge,
2994
				      struct string_list *entries,
2995
				      struct rename_info *ri)
2996
{
2997
	struct diff_queue_struct *head_pairs, *merge_pairs;
2998
	struct hashmap *dir_re_head, *dir_re_merge;
2999
	int clean = 1;
3000

3001
	ri->head_renames = NULL;
3002
	ri->merge_renames = NULL;
3003

3004
	if (!merge_detect_rename(opt))
3005
		return 1;
3006

3007
	head_pairs = get_diffpairs(opt, common, head);
3008
	merge_pairs = get_diffpairs(opt, common, merge);
3009

3010
	if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
3011
	    (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3012
	     !opt->priv->call_depth)) {
3013
		dir_re_head = get_directory_renames(head_pairs);
3014
		dir_re_merge = get_directory_renames(merge_pairs);
3015

3016
		handle_directory_level_conflicts(opt,
3017
						 dir_re_head, head,
3018
						 dir_re_merge, merge);
3019
	} else {
3020
		dir_re_head  = xmalloc(sizeof(*dir_re_head));
3021
		dir_re_merge = xmalloc(sizeof(*dir_re_merge));
3022
		dir_rename_init(dir_re_head);
3023
		dir_rename_init(dir_re_merge);
3024
	}
3025

3026
	ri->head_renames  = get_renames(opt, opt->branch1, head_pairs,
3027
					dir_re_merge, dir_re_head, head,
3028
					common, head, merge, entries,
3029
					&clean);
3030
	if (clean < 0)
3031
		goto cleanup;
3032
	ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
3033
					dir_re_head, dir_re_merge, merge,
3034
					common, head, merge, entries,
3035
					&clean);
3036
	if (clean < 0)
3037
		goto cleanup;
3038
	clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
3039

3040
cleanup:
3041
	/*
3042
	 * Some cleanup is deferred until cleanup_renames() because the
3043
	 * data structures are still needed and referenced in
3044
	 * process_entry().  But there are a few things we can free now.
3045
	 */
3046
	initial_cleanup_rename(head_pairs, dir_re_head);
3047
	initial_cleanup_rename(merge_pairs, dir_re_merge);
3048

3049
	return clean;
3050
}
3051

3052
static void final_cleanup_rename(struct string_list *rename)
3053
{
3054
	const struct rename *re;
3055
	int i;
3056

3057
	if (!rename)
3058
		return;
3059

3060
	for (i = 0; i < rename->nr; i++) {
3061
		re = rename->items[i].util;
3062
		diff_free_filepair(re->pair);
3063
		if (re->src_entry->rename_conflict_info_owned)
3064
			FREE_AND_NULL(re->src_entry->rename_conflict_info);
3065
		if (re->dst_entry->rename_conflict_info_owned)
3066
			FREE_AND_NULL(re->dst_entry->rename_conflict_info);
3067
	}
3068
	string_list_clear(rename, 1);
3069
	free(rename);
3070
}
3071

3072
static void final_cleanup_renames(struct rename_info *re_info)
3073
{
3074
	final_cleanup_rename(re_info->head_renames);
3075
	final_cleanup_rename(re_info->merge_renames);
3076
}
3077

3078
static int read_oid_strbuf(struct merge_options *opt,
3079
			   const struct object_id *oid,
3080
			   struct strbuf *dst)
3081
{
3082
	void *buf;
3083
	enum object_type type;
3084
	unsigned long size;
3085
	buf = repo_read_object_file(the_repository, oid, &type, &size);
3086
	if (!buf)
3087
		return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3088
	if (type != OBJ_BLOB) {
3089
		free(buf);
3090
		return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3091
	}
3092
	strbuf_attach(dst, buf, size, size + 1);
3093
	return 0;
3094
}
3095

3096
static int blob_unchanged(struct merge_options *opt,
3097
			  const struct diff_filespec *o,
3098
			  const struct diff_filespec *a,
3099
			  int renormalize, const char *path)
3100
{
3101
	struct strbuf obuf = STRBUF_INIT;
3102
	struct strbuf abuf = STRBUF_INIT;
3103
	int ret = 0; /* assume changed for safety */
3104
	struct index_state *idx = opt->repo->index;
3105

3106
	if (a->mode != o->mode)
3107
		return 0;
3108
	if (oideq(&o->oid, &a->oid))
3109
		return 1;
3110
	if (!renormalize)
3111
		return 0;
3112

3113
	if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3114
	    read_oid_strbuf(opt, &a->oid, &abuf))
3115
		goto error_return;
3116
	/*
3117
	 * Note: binary | is used so that both renormalizations are
3118
	 * performed.  Comparison can be skipped if both files are
3119
	 * unchanged since their sha1s have already been compared.
3120
	 */
3121
	if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3122
	    renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3123
		ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3124

3125
error_return:
3126
	strbuf_release(&obuf);
3127
	strbuf_release(&abuf);
3128
	return ret;
3129
}
3130

3131
static int handle_modify_delete(struct merge_options *opt,
3132
				const char *path,
3133
				const struct diff_filespec *o,
3134
				const struct diff_filespec *a,
3135
				const struct diff_filespec *b)
3136
{
3137
	const char *modify_branch, *delete_branch;
3138
	const struct diff_filespec *changed;
3139

3140
	if (is_valid(a)) {
3141
		modify_branch = opt->branch1;
3142
		delete_branch = opt->branch2;
3143
		changed = a;
3144
	} else {
3145
		modify_branch = opt->branch2;
3146
		delete_branch = opt->branch1;
3147
		changed = b;
3148
	}
3149

3150
	return handle_change_delete(opt,
3151
				    path, NULL,
3152
				    o, changed,
3153
				    modify_branch, delete_branch,
3154
				    _("modify"), _("modified"));
3155
}
3156

3157
static int handle_content_merge(struct merge_file_info *mfi,
3158
				struct merge_options *opt,
3159
				const char *path,
3160
				int is_dirty,
3161
				const struct diff_filespec *o,
3162
				const struct diff_filespec *a,
3163
				const struct diff_filespec *b,
3164
				struct rename_conflict_info *ci)
3165
{
3166
	const char *reason = _("content");
3167
	unsigned df_conflict_remains = 0;
3168

3169
	if (!is_valid(o))
3170
		reason = _("add/add");
3171

3172
	assert(o->path && a->path && b->path);
3173
	if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3174
			     S_ISGITLINK(ci->ren1->pair->two->mode)))
3175
		df_conflict_remains = 1;
3176

3177
	if (merge_mode_and_contents(opt, o, a, b, path,
3178
				    opt->branch1, opt->branch2,
3179
				    opt->priv->call_depth * 2, mfi))
3180
		return -1;
3181

3182
	/*
3183
	 * We can skip updating the working tree file iff:
3184
	 *   a) The merge is clean
3185
	 *   b) The merge matches what was in HEAD (content, mode, pathname)
3186
	 *   c) The target path is usable (i.e. not involved in D/F conflict)
3187
	 */
3188
	if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3189
	    !df_conflict_remains) {
3190
		int pos;
3191
		struct cache_entry *ce;
3192

3193
		output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3194
		if (add_cacheinfo(opt, &mfi->blob, path,
3195
				  0, (!opt->priv->call_depth && !is_dirty), 0))
3196
			return -1;
3197
		/*
3198
		 * However, add_cacheinfo() will delete the old cache entry
3199
		 * and add a new one.  We need to copy over any skip_worktree
3200
		 * flag to avoid making the file appear as if it were
3201
		 * deleted by the user.
3202
		 */
3203
		pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3204
		ce = opt->priv->orig_index.cache[pos];
3205
		if (ce_skip_worktree(ce)) {
3206
			pos = index_name_pos(opt->repo->index, path, strlen(path));
3207
			ce = opt->repo->index->cache[pos];
3208
			ce->ce_flags |= CE_SKIP_WORKTREE;
3209
		}
3210
		return mfi->clean;
3211
	}
3212

3213
	if (!mfi->clean) {
3214
		if (S_ISGITLINK(mfi->blob.mode))
3215
			reason = _("submodule");
3216
		output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3217
				reason, path);
3218
		if (ci && !df_conflict_remains)
3219
			if (update_stages(opt, path, o, a, b))
3220
				return -1;
3221
	}
3222

3223
	if (df_conflict_remains || is_dirty) {
3224
		char *new_path;
3225
		if (opt->priv->call_depth) {
3226
			remove_file_from_index(opt->repo->index, path);
3227
		} else {
3228
			if (!mfi->clean) {
3229
				if (update_stages(opt, path, o, a, b))
3230
					return -1;
3231
			} else {
3232
				int file_from_stage2 = was_tracked(opt, path);
3233

3234
				if (update_stages(opt, path, NULL,
3235
						  file_from_stage2 ? &mfi->blob : NULL,
3236
						  file_from_stage2 ? NULL : &mfi->blob))
3237
					return -1;
3238
			}
3239

3240
		}
3241
		new_path = unique_path(opt, path, ci->ren1->branch);
3242
		if (is_dirty) {
3243
			output(opt, 1, _("Refusing to lose dirty file at %s"),
3244
			       path);
3245
		}
3246
		output(opt, 1, _("Adding as %s instead"), new_path);
3247
		if (update_file(opt, 0, &mfi->blob, new_path)) {
3248
			free(new_path);
3249
			return -1;
3250
		}
3251
		free(new_path);
3252
		mfi->clean = 0;
3253
	} else if (update_file(opt, mfi->clean, &mfi->blob, path))
3254
		return -1;
3255
	return !is_dirty && mfi->clean;
3256
}
3257

3258
static int handle_rename_normal(struct merge_options *opt,
3259
				const char *path,
3260
				const struct diff_filespec *o,
3261
				const struct diff_filespec *a,
3262
				const struct diff_filespec *b,
3263
				struct rename_conflict_info *ci)
3264
{
3265
	struct rename *ren = ci->ren1;
3266
	struct merge_file_info mfi;
3267
	int clean;
3268

3269
	/* Merge the content and write it out */
3270
	clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3271
				     o, a, b, ci);
3272

3273
	if (clean &&
3274
	    opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3275
	    ren->dir_rename_original_dest) {
3276
		if (update_stages(opt, path,
3277
				  &mfi.blob, &mfi.blob, &mfi.blob))
3278
			return -1;
3279
		clean = 0; /* not clean, but conflicted */
3280
	}
3281
	return clean;
3282
}
3283

3284
static void dir_rename_warning(const char *msg,
3285
			       int is_add,
3286
			       int clean,
3287
			       struct merge_options *opt,
3288
			       struct rename *ren)
3289
{
3290
	const char *other_branch;
3291
	other_branch = (ren->branch == opt->branch1 ?
3292
			opt->branch2 : opt->branch1);
3293
	if (is_add) {
3294
		output(opt, clean ? 2 : 1, msg,
3295
		       ren->pair->one->path, ren->branch,
3296
		       other_branch, ren->pair->two->path);
3297
		return;
3298
	}
3299
	output(opt, clean ? 2 : 1, msg,
3300
	       ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3301
	       other_branch, ren->pair->two->path);
3302
}
3303
static int warn_about_dir_renamed_entries(struct merge_options *opt,
3304
					  struct rename *ren)
3305
{
3306
	const char *msg;
3307
	int clean = 1, is_add;
3308

3309
	if (!ren)
3310
		return clean;
3311

3312
	/* Return early if ren was not affected/created by a directory rename */
3313
	if (!ren->dir_rename_original_dest)
3314
		return clean;
3315

3316
	/* Sanity checks */
3317
	assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3318
	assert(ren->dir_rename_original_type == 'A' ||
3319
	       ren->dir_rename_original_type == 'R');
3320

3321
	/* Check whether to treat directory renames as a conflict */
3322
	clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3323

3324
	is_add = (ren->dir_rename_original_type == 'A');
3325
	if (ren->dir_rename_original_type == 'A' && clean) {
3326
		msg = _("Path updated: %s added in %s inside a "
3327
			"directory that was renamed in %s; moving it to %s.");
3328
	} else if (ren->dir_rename_original_type == 'A' && !clean) {
3329
		msg = _("CONFLICT (file location): %s added in %s "
3330
			"inside a directory that was renamed in %s, "
3331
			"suggesting it should perhaps be moved to %s.");
3332
	} else if (ren->dir_rename_original_type == 'R' && clean) {
3333
		msg = _("Path updated: %s renamed to %s in %s, inside a "
3334
			"directory that was renamed in %s; moving it to %s.");
3335
	} else if (ren->dir_rename_original_type == 'R' && !clean) {
3336
		msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3337
			"inside a directory that was renamed in %s, "
3338
			"suggesting it should perhaps be moved to %s.");
3339
	} else {
3340
		BUG("Impossible dir_rename_original_type/clean combination");
3341
	}
3342
	dir_rename_warning(msg, is_add, clean, opt, ren);
3343

3344
	return clean;
3345
}
3346

3347
/* Per entry merge function */
3348
static int process_entry(struct merge_options *opt,
3349
			 const char *path, struct stage_data *entry)
3350
{
3351
	int clean_merge = 1;
3352
	int normalize = opt->renormalize;
3353

3354
	struct diff_filespec *o = &entry->stages[1];
3355
	struct diff_filespec *a = &entry->stages[2];
3356
	struct diff_filespec *b = &entry->stages[3];
3357
	int o_valid = is_valid(o);
3358
	int a_valid = is_valid(a);
3359
	int b_valid = is_valid(b);
3360
	o->path = a->path = b->path = (char*)path;
3361

3362
	entry->processed = 1;
3363
	if (entry->rename_conflict_info) {
3364
		struct rename_conflict_info *ci = entry->rename_conflict_info;
3365
		struct diff_filespec *temp;
3366
		int path_clean;
3367

3368
		path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3369
		path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3370

3371
		/*
3372
		 * For cases with a single rename, {o,a,b}->path have all been
3373
		 * set to the rename target path; we need to set two of these
3374
		 * back to the rename source.
3375
		 * For rename/rename conflicts, we'll manually fix paths below.
3376
		 */
3377
		temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3378
		o->path = temp->path = ci->ren1->pair->one->path;
3379
		if (ci->ren2) {
3380
			assert(opt->branch1 == ci->ren1->branch);
3381
		}
3382

3383
		switch (ci->rename_type) {
3384
		case RENAME_NORMAL:
3385
		case RENAME_ONE_FILE_TO_ONE:
3386
			clean_merge = handle_rename_normal(opt, path, o, a, b,
3387
							   ci);
3388
			break;
3389
		case RENAME_VIA_DIR:
3390
			clean_merge = handle_rename_via_dir(opt, ci);
3391
			break;
3392
		case RENAME_ADD:
3393
			/*
3394
			 * Probably unclean merge, but if the renamed file
3395
			 * merges cleanly and the result can then be
3396
			 * two-way merged cleanly with the added file, I
3397
			 * guess it's a clean merge?
3398
			 */
3399
			clean_merge = handle_rename_add(opt, ci);
3400
			break;
3401
		case RENAME_DELETE:
3402
			clean_merge = 0;
3403
			if (handle_rename_delete(opt, ci))
3404
				clean_merge = -1;
3405
			break;
3406
		case RENAME_ONE_FILE_TO_TWO:
3407
			/*
3408
			 * Manually fix up paths; note:
3409
			 * ren[12]->pair->one->path are equal.
3410
			 */
3411
			o->path = ci->ren1->pair->one->path;
3412
			a->path = ci->ren1->pair->two->path;
3413
			b->path = ci->ren2->pair->two->path;
3414

3415
			clean_merge = 0;
3416
			if (handle_rename_rename_1to2(opt, ci))
3417
				clean_merge = -1;
3418
			break;
3419
		case RENAME_TWO_FILES_TO_ONE:
3420
			/*
3421
			 * Manually fix up paths; note,
3422
			 * ren[12]->pair->two->path are actually equal.
3423
			 */
3424
			o->path = NULL;
3425
			a->path = ci->ren1->pair->two->path;
3426
			b->path = ci->ren2->pair->two->path;
3427

3428
			/*
3429
			 * Probably unclean merge, but if the two renamed
3430
			 * files merge cleanly and the two resulting files
3431
			 * can then be two-way merged cleanly, I guess it's
3432
			 * a clean merge?
3433
			 */
3434
			clean_merge = handle_rename_rename_2to1(opt, ci);
3435
			break;
3436
		default:
3437
			entry->processed = 0;
3438
			break;
3439
		}
3440
		if (path_clean < clean_merge)
3441
			clean_merge = path_clean;
3442
	} else if (o_valid && (!a_valid || !b_valid)) {
3443
		/* Case A: Deleted in one */
3444
		if ((!a_valid && !b_valid) ||
3445
		    (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3446
		    (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3447
			/* Deleted in both or deleted in one and
3448
			 * unchanged in the other */
3449
			if (a_valid)
3450
				output(opt, 2, _("Removing %s"), path);
3451
			/* do not touch working file if it did not exist */
3452
			remove_file(opt, 1, path, !a_valid);
3453
		} else {
3454
			/* Modify/delete; deleted side may have put a directory in the way */
3455
			clean_merge = 0;
3456
			if (handle_modify_delete(opt, path, o, a, b))
3457
				clean_merge = -1;
3458
		}
3459
	} else if ((!o_valid && a_valid && !b_valid) ||
3460
		   (!o_valid && !a_valid && b_valid)) {
3461
		/* Case B: Added in one. */
3462
		/* [nothing|directory] -> ([nothing|directory], file) */
3463

3464
		const char *add_branch;
3465
		const char *other_branch;
3466
		const char *conf;
3467
		const struct diff_filespec *contents;
3468

3469
		if (a_valid) {
3470
			add_branch = opt->branch1;
3471
			other_branch = opt->branch2;
3472
			contents = a;
3473
			conf = _("file/directory");
3474
		} else {
3475
			add_branch = opt->branch2;
3476
			other_branch = opt->branch1;
3477
			contents = b;
3478
			conf = _("directory/file");
3479
		}
3480
		if (dir_in_way(opt->repo->index, path,
3481
			       !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3482
			       0)) {
3483
			char *new_path = unique_path(opt, path, add_branch);
3484
			clean_merge = 0;
3485
			output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3486
			       "Adding %s as %s"),
3487
			       conf, path, other_branch, path, new_path);
3488
			if (update_file(opt, 0, contents, new_path))
3489
				clean_merge = -1;
3490
			else if (opt->priv->call_depth)
3491
				remove_file_from_index(opt->repo->index, path);
3492
			free(new_path);
3493
		} else {
3494
			output(opt, 2, _("Adding %s"), path);
3495
			/* do not overwrite file if already present */
3496
			if (update_file_flags(opt, contents, path, 1, !a_valid))
3497
				clean_merge = -1;
3498
		}
3499
	} else if (a_valid && b_valid) {
3500
		if (!o_valid) {
3501
			/* Case C: Added in both (check for same permissions) */
3502
			output(opt, 1,
3503
			       _("CONFLICT (add/add): Merge conflict in %s"),
3504
			       path);
3505
			clean_merge = handle_file_collision(opt,
3506
							    path, NULL, NULL,
3507
							    opt->branch1,
3508
							    opt->branch2,
3509
							    a, b);
3510
		} else {
3511
			/* case D: Modified in both, but differently. */
3512
			struct merge_file_info mfi;
3513
			int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3514
			clean_merge = handle_content_merge(&mfi, opt, path,
3515
							   is_dirty,
3516
							   o, a, b, NULL);
3517
		}
3518
	} else if (!o_valid && !a_valid && !b_valid) {
3519
		/*
3520
		 * this entry was deleted altogether. a_mode == 0 means
3521
		 * we had that path and want to actively remove it.
3522
		 */
3523
		remove_file(opt, 1, path, !a->mode);
3524
	} else
3525
		BUG("fatal merge failure, shouldn't happen.");
3526

3527
	return clean_merge;
3528
}
3529

3530
static int merge_trees_internal(struct merge_options *opt,
3531
				struct tree *head,
3532
				struct tree *merge,
3533
				struct tree *merge_base,
3534
				struct tree **result)
3535
{
3536
	struct index_state *istate = opt->repo->index;
3537
	int code, clean;
3538

3539
	if (opt->subtree_shift) {
3540
		merge = shift_tree_object(opt->repo, head, merge,
3541
					  opt->subtree_shift);
3542
		merge_base = shift_tree_object(opt->repo, head, merge_base,
3543
					       opt->subtree_shift);
3544
	}
3545

3546
	if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3547
		output(opt, 0, _("Already up to date."));
3548
		*result = head;
3549
		return 1;
3550
	}
3551

3552
	code = unpack_trees_start(opt, merge_base, head, merge);
3553

3554
	if (code != 0) {
3555
		if (show(opt, 4) || opt->priv->call_depth)
3556
			err(opt, _("merging of trees %s and %s failed"),
3557
			    oid_to_hex(&head->object.oid),
3558
			    oid_to_hex(&merge->object.oid));
3559
		unpack_trees_finish(opt);
3560
		return -1;
3561
	}
3562

3563
	if (unmerged_index(istate)) {
3564
		struct string_list *entries;
3565
		struct rename_info re_info;
3566
		int i;
3567
		/*
3568
		 * Only need the hashmap while processing entries, so
3569
		 * initialize it here and free it when we are done running
3570
		 * through the entries. Keeping it in the merge_options as
3571
		 * opposed to decaring a local hashmap is for convenience
3572
		 * so that we don't have to pass it to around.
3573
		 */
3574
		hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3575
			     NULL, 512);
3576
		get_files_dirs(opt, head);
3577
		get_files_dirs(opt, merge);
3578

3579
		entries = get_unmerged(opt->repo->index);
3580
		clean = detect_and_process_renames(opt, merge_base, head, merge,
3581
						   entries, &re_info);
3582
		record_df_conflict_files(opt, entries);
3583
		if (clean < 0)
3584
			goto cleanup;
3585
		for (i = entries->nr-1; 0 <= i; i--) {
3586
			const char *path = entries->items[i].string;
3587
			struct stage_data *e = entries->items[i].util;
3588
			if (!e->processed) {
3589
				int ret = process_entry(opt, path, e);
3590
				if (!ret)
3591
					clean = 0;
3592
				else if (ret < 0) {
3593
					clean = ret;
3594
					goto cleanup;
3595
				}
3596
			}
3597
		}
3598
		for (i = 0; i < entries->nr; i++) {
3599
			struct stage_data *e = entries->items[i].util;
3600
			if (!e->processed)
3601
				BUG("unprocessed path??? %s",
3602
				    entries->items[i].string);
3603
		}
3604

3605
	cleanup:
3606
		final_cleanup_renames(&re_info);
3607

3608
		string_list_clear(entries, 1);
3609
		free(entries);
3610

3611
		hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3612
					struct path_hashmap_entry, e);
3613

3614
		if (clean < 0) {
3615
			unpack_trees_finish(opt);
3616
			return clean;
3617
		}
3618
	}
3619
	else
3620
		clean = 1;
3621

3622
	unpack_trees_finish(opt);
3623

3624
	if (opt->priv->call_depth &&
3625
	    !(*result = write_in_core_index_as_tree(opt->repo)))
3626
		return -1;
3627

3628
	return clean;
3629
}
3630

3631
/*
3632
 * Merge the commits h1 and h2, returning a flag (int) indicating the
3633
 * cleanness of the merge.  Also, if opt->priv->call_depth, create a
3634
 * virtual commit and write its location to *result.
3635
 */
3636
static int merge_recursive_internal(struct merge_options *opt,
3637
				    struct commit *h1,
3638
				    struct commit *h2,
3639
				    const struct commit_list *_merge_bases,
3640
				    struct commit **result)
3641
{
3642
	struct commit_list *merge_bases = copy_commit_list(_merge_bases);
3643
	struct commit_list *iter;
3644
	struct commit *merged_merge_bases;
3645
	struct tree *result_tree;
3646
	const char *ancestor_name;
3647
	struct strbuf merge_base_abbrev = STRBUF_INIT;
3648
	int ret;
3649

3650
	if (show(opt, 4)) {
3651
		output(opt, 4, _("Merging:"));
3652
		output_commit_title(opt, h1);
3653
		output_commit_title(opt, h2);
3654
	}
3655

3656
	if (!merge_bases) {
3657
		if (repo_get_merge_bases(the_repository, h1, h2,
3658
					 &merge_bases) < 0) {
3659
			ret = -1;
3660
			goto out;
3661
		}
3662
		merge_bases = reverse_commit_list(merge_bases);
3663
	}
3664

3665
	if (show(opt, 5)) {
3666
		unsigned cnt = commit_list_count(merge_bases);
3667

3668
		output(opt, 5, Q_("found %u common ancestor:",
3669
				"found %u common ancestors:", cnt), cnt);
3670
		for (iter = merge_bases; iter; iter = iter->next)
3671
			output_commit_title(opt, iter->item);
3672
	}
3673

3674
	merged_merge_bases = pop_commit(&merge_bases);
3675
	if (!merged_merge_bases) {
3676
		/* if there is no common ancestor, use an empty tree */
3677
		struct tree *tree;
3678

3679
		tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3680
		merged_merge_bases = make_virtual_commit(opt->repo, tree,
3681
							 "ancestor");
3682
		ancestor_name = "empty tree";
3683
	} else if (opt->ancestor && !opt->priv->call_depth) {
3684
		ancestor_name = opt->ancestor;
3685
	} else if (merge_bases) {
3686
		ancestor_name = "merged common ancestors";
3687
	} else {
3688
		strbuf_add_unique_abbrev(&merge_base_abbrev,
3689
					 &merged_merge_bases->object.oid,
3690
					 DEFAULT_ABBREV);
3691
		ancestor_name = merge_base_abbrev.buf;
3692
	}
3693

3694
	for (iter = merge_bases; iter; iter = iter->next) {
3695
		const char *saved_b1, *saved_b2;
3696
		opt->priv->call_depth++;
3697
		/*
3698
		 * When the merge fails, the result contains files
3699
		 * with conflict markers. The cleanness flag is
3700
		 * ignored (unless indicating an error), it was never
3701
		 * actually used, as result of merge_trees has always
3702
		 * overwritten it: the committed "conflicts" were
3703
		 * already resolved.
3704
		 */
3705
		discard_index(opt->repo->index);
3706
		saved_b1 = opt->branch1;
3707
		saved_b2 = opt->branch2;
3708
		opt->branch1 = "Temporary merge branch 1";
3709
		opt->branch2 = "Temporary merge branch 2";
3710
		if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3711
					     NULL, &merged_merge_bases) < 0) {
3712
			ret = -1;
3713
			goto out;
3714
		}
3715
		opt->branch1 = saved_b1;
3716
		opt->branch2 = saved_b2;
3717
		opt->priv->call_depth--;
3718

3719
		if (!merged_merge_bases) {
3720
			ret = err(opt, _("merge returned no commit"));
3721
			goto out;
3722
		}
3723
	}
3724

3725
	/*
3726
	 * FIXME: Since merge_recursive_internal() is only ever called by
3727
	 * places that ensure the index is loaded first
3728
	 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3729
	 * case where the merge base was unique that means when we get here
3730
	 * we immediately discard the index and re-read it, which is a
3731
	 * complete waste of time.  We should only be discarding and
3732
	 * re-reading if we were forced to recurse.
3733
	 */
3734
	discard_index(opt->repo->index);
3735
	if (!opt->priv->call_depth)
3736
		repo_read_index(opt->repo);
3737

3738
	opt->ancestor = ancestor_name;
3739
	ret = merge_trees_internal(opt,
3740
				   repo_get_commit_tree(opt->repo, h1),
3741
				   repo_get_commit_tree(opt->repo, h2),
3742
				   repo_get_commit_tree(opt->repo,
3743
							merged_merge_bases),
3744
				   &result_tree);
3745
	opt->ancestor = NULL;  /* avoid accidental re-use of opt->ancestor */
3746
	if (ret < 0) {
3747
		flush_output(opt);
3748
		goto out;
3749
	}
3750

3751
	if (opt->priv->call_depth) {
3752
		*result = make_virtual_commit(opt->repo, result_tree,
3753
					      "merged tree");
3754
		commit_list_insert(h1, &(*result)->parents);
3755
		commit_list_insert(h2, &(*result)->parents->next);
3756
	}
3757

3758
out:
3759
	strbuf_release(&merge_base_abbrev);
3760
	free_commit_list(merge_bases);
3761
	return ret;
3762
}
3763

3764
static int merge_start(struct merge_options *opt, struct tree *head)
3765
{
3766
	struct strbuf sb = STRBUF_INIT;
3767

3768
	/* Sanity checks on opt */
3769
	assert(opt->repo);
3770

3771
	assert(opt->branch1 && opt->branch2);
3772

3773
	assert(opt->detect_renames >= -1 &&
3774
	       opt->detect_renames <= DIFF_DETECT_COPY);
3775
	assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3776
	       opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3777
	assert(opt->rename_limit >= -1);
3778
	assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3779
	assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3780

3781
	assert(opt->xdl_opts >= 0);
3782
	assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3783
	       opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3784

3785
	assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3786
	assert(opt->buffer_output <= 2);
3787
	assert(opt->obuf.len == 0);
3788

3789
	assert(opt->priv == NULL);
3790

3791
	/* Not supported; option specific to merge-ort */
3792
	assert(!opt->record_conflict_msgs_as_headers);
3793
	assert(!opt->msg_header_prefix);
3794

3795
	/* Sanity check on repo state; index must match head */
3796
	if (repo_index_has_changes(opt->repo, head, &sb)) {
3797
		err(opt, _("Your local changes to the following files would be overwritten by merge:\n  %s"),
3798
		    sb.buf);
3799
		strbuf_release(&sb);
3800
		return -1;
3801
	}
3802

3803
	CALLOC_ARRAY(opt->priv, 1);
3804
	string_list_init_dup(&opt->priv->df_conflict_file_set);
3805
	return 0;
3806
}
3807

3808
static void merge_finalize(struct merge_options *opt)
3809
{
3810
	flush_output(opt);
3811
	if (!opt->priv->call_depth && opt->buffer_output < 2)
3812
		strbuf_release(&opt->obuf);
3813
	if (show(opt, 2))
3814
		diff_warn_rename_limit("merge.renamelimit",
3815
				       opt->priv->needed_rename_limit, 0);
3816
	hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3817
			       struct path_hashmap_entry, e);
3818
	string_list_clear(&opt->priv->df_conflict_file_set, 0);
3819
	FREE_AND_NULL(opt->priv);
3820
}
3821

3822
int merge_trees(struct merge_options *opt,
3823
		struct tree *head,
3824
		struct tree *merge,
3825
		struct tree *merge_base)
3826
{
3827
	int clean;
3828
	struct tree *ignored;
3829

3830
	assert(opt->ancestor != NULL);
3831

3832
	if (merge_start(opt, head))
3833
		return -1;
3834
	clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3835
	merge_finalize(opt);
3836

3837
	return clean;
3838
}
3839

3840
int merge_recursive(struct merge_options *opt,
3841
		    struct commit *h1,
3842
		    struct commit *h2,
3843
		    const struct commit_list *merge_bases,
3844
		    struct commit **result)
3845
{
3846
	int clean;
3847

3848
	assert(opt->ancestor == NULL ||
3849
	       !strcmp(opt->ancestor, "constructed merge base"));
3850

3851
	prepare_repo_settings(opt->repo);
3852
	opt->repo->settings.command_requires_full_index = 1;
3853

3854
	if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3855
		return -1;
3856
	clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3857
	merge_finalize(opt);
3858

3859
	return clean;
3860
}
3861

3862
static struct commit *get_ref(struct repository *repo,
3863
			      const struct object_id *oid,
3864
			      const char *name)
3865
{
3866
	struct object *object;
3867

3868
	object = deref_tag(repo, parse_object(repo, oid),
3869
			   name, strlen(name));
3870
	if (!object)
3871
		return NULL;
3872
	if (object->type == OBJ_TREE)
3873
		return make_virtual_commit(repo, (struct tree*)object, name);
3874
	if (object->type != OBJ_COMMIT)
3875
		return NULL;
3876
	if (repo_parse_commit(repo, (struct commit *)object))
3877
		return NULL;
3878
	return (struct commit *)object;
3879
}
3880

3881
int merge_recursive_generic(struct merge_options *opt,
3882
			    const struct object_id *head,
3883
			    const struct object_id *merge,
3884
			    int num_merge_bases,
3885
			    const struct object_id *merge_bases,
3886
			    struct commit **result)
3887
{
3888
	int clean;
3889
	struct lock_file lock = LOCK_INIT;
3890
	struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3891
	struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3892
	struct commit_list *ca = NULL;
3893

3894
	if (merge_bases) {
3895
		int i;
3896
		for (i = 0; i < num_merge_bases; ++i) {
3897
			struct commit *base;
3898
			if (!(base = get_ref(opt->repo, &merge_bases[i],
3899
					     oid_to_hex(&merge_bases[i]))))
3900
				return err(opt, _("Could not parse object '%s'"),
3901
					   oid_to_hex(&merge_bases[i]));
3902
			commit_list_insert(base, &ca);
3903
		}
3904
		if (num_merge_bases == 1)
3905
			opt->ancestor = "constructed merge base";
3906
	}
3907

3908
	repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3909
	clean = merge_recursive(opt, head_commit, next_commit, ca,
3910
				result);
3911
	free_commit_list(ca);
3912
	if (clean < 0) {
3913
		rollback_lock_file(&lock);
3914
		return clean;
3915
	}
3916

3917
	if (write_locked_index(opt->repo->index, &lock,
3918
			       COMMIT_LOCK | SKIP_IF_UNCHANGED))
3919
		return err(opt, _("Unable to write index."));
3920

3921
	return clean ? 0 : 1;
3922
}
3923

3924
static void merge_recursive_config(struct merge_options *opt, int ui)
3925
{
3926
	char *value = NULL;
3927
	int renormalize = 0;
3928
	git_config_get_int("merge.verbosity", &opt->verbosity);
3929
	git_config_get_int("diff.renamelimit", &opt->rename_limit);
3930
	git_config_get_int("merge.renamelimit", &opt->rename_limit);
3931
	git_config_get_bool("merge.renormalize", &renormalize);
3932
	opt->renormalize = renormalize;
3933
	if (!git_config_get_string("diff.renames", &value)) {
3934
		opt->detect_renames = git_config_rename("diff.renames", value);
3935
		free(value);
3936
	}
3937
	if (!git_config_get_string("merge.renames", &value)) {
3938
		opt->detect_renames = git_config_rename("merge.renames", value);
3939
		free(value);
3940
	}
3941
	if (!git_config_get_string("merge.directoryrenames", &value)) {
3942
		int boolval = git_parse_maybe_bool(value);
3943
		if (0 <= boolval) {
3944
			opt->detect_directory_renames = boolval ?
3945
				MERGE_DIRECTORY_RENAMES_TRUE :
3946
				MERGE_DIRECTORY_RENAMES_NONE;
3947
		} else if (!strcasecmp(value, "conflict")) {
3948
			opt->detect_directory_renames =
3949
				MERGE_DIRECTORY_RENAMES_CONFLICT;
3950
		} /* avoid erroring on values from future versions of git */
3951
		free(value);
3952
	}
3953
	if (ui) {
3954
		if (!git_config_get_string("diff.algorithm", &value)) {
3955
			long diff_algorithm = parse_algorithm_value(value);
3956
			if (diff_algorithm < 0)
3957
				die(_("unknown value for config '%s': %s"), "diff.algorithm", value);
3958
			opt->xdl_opts = (opt->xdl_opts & ~XDF_DIFF_ALGORITHM_MASK) | diff_algorithm;
3959
			free(value);
3960
		}
3961
	}
3962
	git_config(git_xmerge_config, NULL);
3963
}
3964

3965
static void init_merge_options(struct merge_options *opt,
3966
			struct repository *repo, int ui)
3967
{
3968
	const char *merge_verbosity;
3969
	memset(opt, 0, sizeof(struct merge_options));
3970

3971
	opt->repo = repo;
3972

3973
	opt->detect_renames = -1;
3974
	opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3975
	opt->rename_limit = -1;
3976

3977
	opt->verbosity = 2;
3978
	opt->buffer_output = 1;
3979
	strbuf_init(&opt->obuf, 0);
3980

3981
	opt->renormalize = 0;
3982

3983
	opt->conflict_style = -1;
3984

3985
	merge_recursive_config(opt, ui);
3986
	merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3987
	if (merge_verbosity)
3988
		opt->verbosity = strtol(merge_verbosity, NULL, 10);
3989
	if (opt->verbosity >= 5)
3990
		opt->buffer_output = 0;
3991
}
3992

3993
void init_ui_merge_options(struct merge_options *opt,
3994
			struct repository *repo)
3995
{
3996
	init_merge_options(opt, repo, 1);
3997
}
3998

3999
void init_basic_merge_options(struct merge_options *opt,
4000
			struct repository *repo)
4001
{
4002
	init_merge_options(opt, repo, 0);
4003
}
4004

4005
/*
4006
 * For now, members of merge_options do not need deep copying, but
4007
 * it may change in the future, in which case we would need to update
4008
 * this, and also make a matching change to clear_merge_options() to
4009
 * release the resources held by a copied instance.
4010
 */
4011
void copy_merge_options(struct merge_options *dst, struct merge_options *src)
4012
{
4013
	*dst = *src;
4014
}
4015

4016
void clear_merge_options(struct merge_options *opt UNUSED)
4017
{
4018
	; /* no-op as our copy is shallow right now */
4019
}
4020

4021
int parse_merge_opt(struct merge_options *opt, const char *s)
4022
{
4023
	const char *arg;
4024

4025
	if (!s || !*s)
4026
		return -1;
4027
	if (!strcmp(s, "ours"))
4028
		opt->recursive_variant = MERGE_VARIANT_OURS;
4029
	else if (!strcmp(s, "theirs"))
4030
		opt->recursive_variant = MERGE_VARIANT_THEIRS;
4031
	else if (!strcmp(s, "subtree"))
4032
		opt->subtree_shift = "";
4033
	else if (skip_prefix(s, "subtree=", &arg))
4034
		opt->subtree_shift = arg;
4035
	else if (!strcmp(s, "patience"))
4036
		opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
4037
	else if (!strcmp(s, "histogram"))
4038
		opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
4039
	else if (skip_prefix(s, "diff-algorithm=", &arg)) {
4040
		long value = parse_algorithm_value(arg);
4041
		if (value < 0)
4042
			return -1;
4043
		/* clear out previous settings */
4044
		DIFF_XDL_CLR(opt, NEED_MINIMAL);
4045
		opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
4046
		opt->xdl_opts |= value;
4047
	}
4048
	else if (!strcmp(s, "ignore-space-change"))
4049
		DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
4050
	else if (!strcmp(s, "ignore-all-space"))
4051
		DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
4052
	else if (!strcmp(s, "ignore-space-at-eol"))
4053
		DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
4054
	else if (!strcmp(s, "ignore-cr-at-eol"))
4055
		DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
4056
	else if (!strcmp(s, "renormalize"))
4057
		opt->renormalize = 1;
4058
	else if (!strcmp(s, "no-renormalize"))
4059
		opt->renormalize = 0;
4060
	else if (!strcmp(s, "no-renames"))
4061
		opt->detect_renames = 0;
4062
	else if (!strcmp(s, "find-renames")) {
4063
		opt->detect_renames = 1;
4064
		opt->rename_score = 0;
4065
	}
4066
	else if (skip_prefix(s, "find-renames=", &arg) ||
4067
		 skip_prefix(s, "rename-threshold=", &arg)) {
4068
		if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
4069
			return -1;
4070
		opt->detect_renames = 1;
4071
	}
4072
	/*
4073
	 * Please update $__git_merge_strategy_options in
4074
	 * git-completion.bash when you add new options
4075
	 */
4076
	else
4077
		return -1;
4078
	return 0;
4079
}
4080