]> git.ipfire.org Git - thirdparty/git.git/blob - merge-recursive.c
Merge branch 'jj/stash-reset-only-toplevel'
[thirdparty/git.git] / merge-recursive.c
1 /*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6 #include "cache.h"
7 #include "merge-recursive.h"
8
9 #include "advice.h"
10 #include "alloc.h"
11 #include "attr.h"
12 #include "blob.h"
13 #include "builtin.h"
14 #include "cache-tree.h"
15 #include "commit.h"
16 #include "commit-reach.h"
17 #include "config.h"
18 #include "diff.h"
19 #include "diffcore.h"
20 #include "dir.h"
21 #include "ll-merge.h"
22 #include "lockfile.h"
23 #include "object-store.h"
24 #include "repository.h"
25 #include "revision.h"
26 #include "string-list.h"
27 #include "submodule.h"
28 #include "tag.h"
29 #include "tree-walk.h"
30 #include "unpack-trees.h"
31 #include "xdiff-interface.h"
32
33 struct merge_options_internal {
34 int call_depth;
35 int needed_rename_limit;
36 struct hashmap current_file_dir_set;
37 struct string_list df_conflict_file_set;
38 struct unpack_trees_options unpack_opts;
39 struct index_state orig_index;
40 };
41
42 struct path_hashmap_entry {
43 struct hashmap_entry e;
44 char path[FLEX_ARRAY];
45 };
46
47 static int path_hashmap_cmp(const void *cmp_data,
48 const struct hashmap_entry *eptr,
49 const struct hashmap_entry *entry_or_key,
50 const void *keydata)
51 {
52 const struct path_hashmap_entry *a, *b;
53 const char *key = keydata;
54
55 a = container_of(eptr, const struct path_hashmap_entry, e);
56 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
57
58 if (ignore_case)
59 return strcasecmp(a->path, key ? key : b->path);
60 else
61 return strcmp(a->path, key ? key : b->path);
62 }
63
64 static unsigned int path_hash(const char *path)
65 {
66 return ignore_case ? strihash(path) : strhash(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 == NULL)
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 *unused_cmp_data,
100 const struct hashmap_entry *eptr,
101 const struct hashmap_entry *entry_or_key,
102 const void *unused_keydata)
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(&entry->possible_new_dirs, 0);
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 *unused_cmp_data,
145 const struct hashmap_entry *eptr,
146 const struct hashmap_entry *entry_or_key,
147 const void *unused_keydata)
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 static int err(struct merge_options *opt, const char *err, ...)
171 {
172 va_list params;
173
174 if (opt->buffer_output < 2)
175 flush_output(opt);
176 else {
177 strbuf_complete(&opt->obuf, '\n');
178 strbuf_addstr(&opt->obuf, "error: ");
179 }
180 va_start(params, err);
181 strbuf_vaddf(&opt->obuf, err, params);
182 va_end(params);
183 if (opt->buffer_output > 1)
184 strbuf_addch(&opt->obuf, '\n');
185 else {
186 error("%s", opt->obuf.buf);
187 strbuf_reset(&opt->obuf);
188 }
189
190 return -1;
191 }
192
193 static struct tree *shift_tree_object(struct repository *repo,
194 struct tree *one, struct tree *two,
195 const char *subtree_shift)
196 {
197 struct object_id shifted;
198
199 if (!*subtree_shift) {
200 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
201 } else {
202 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
203 subtree_shift);
204 }
205 if (oideq(&two->object.oid, &shifted))
206 return two;
207 return lookup_tree(repo, &shifted);
208 }
209
210 static inline void set_commit_tree(struct commit *c, struct tree *t)
211 {
212 c->maybe_tree = t;
213 }
214
215 static struct commit *make_virtual_commit(struct repository *repo,
216 struct tree *tree,
217 const char *comment)
218 {
219 struct commit *commit = alloc_commit_node(repo);
220
221 set_merge_remote_desc(commit, comment, (struct object *)commit);
222 set_commit_tree(commit, tree);
223 commit->object.parsed = 1;
224 return commit;
225 }
226
227 /*
228 * Since we use get_tree_entry(), which does not put the read object into
229 * the object pool, we cannot rely on a == b.
230 */
231 static int oid_eq(const struct object_id *a, const struct object_id *b)
232 {
233 if (!a && !b)
234 return 2;
235 return a && b && oideq(a, b);
236 }
237
238 enum rename_type {
239 RENAME_NORMAL = 0,
240 RENAME_VIA_DIR,
241 RENAME_ADD,
242 RENAME_DELETE,
243 RENAME_ONE_FILE_TO_ONE,
244 RENAME_ONE_FILE_TO_TWO,
245 RENAME_TWO_FILES_TO_ONE
246 };
247
248 /*
249 * Since we want to write the index eventually, we cannot reuse the index
250 * for these (temporary) data.
251 */
252 struct stage_data {
253 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
254 struct rename_conflict_info *rename_conflict_info;
255 unsigned processed:1;
256 };
257
258 struct rename {
259 unsigned processed:1;
260 struct diff_filepair *pair;
261 const char *branch; /* branch that the rename occurred on */
262 /*
263 * If directory rename detection affected this rename, what was its
264 * original type ('A' or 'R') and it's original destination before
265 * the directory rename (otherwise, '\0' and NULL for these two vars).
266 */
267 char dir_rename_original_type;
268 char *dir_rename_original_dest;
269 /*
270 * Purpose of src_entry and dst_entry:
271 *
272 * If 'before' is renamed to 'after' then src_entry will contain
273 * the versions of 'before' from the merge_base, HEAD, and MERGE in
274 * stages 1, 2, and 3; dst_entry will contain the respective
275 * versions of 'after' in corresponding locations. Thus, we have a
276 * total of six modes and oids, though some will be null. (Stage 0
277 * is ignored; we're interested in handling conflicts.)
278 *
279 * Since we don't turn on break-rewrites by default, neither
280 * src_entry nor dst_entry can have all three of their stages have
281 * non-null oids, meaning at most four of the six will be non-null.
282 * Also, since this is a rename, both src_entry and dst_entry will
283 * have at least one non-null oid, meaning at least two will be
284 * non-null. Of the six oids, a typical rename will have three be
285 * non-null. Only two implies a rename/delete, and four implies a
286 * rename/add.
287 */
288 struct stage_data *src_entry;
289 struct stage_data *dst_entry;
290 };
291
292 struct rename_conflict_info {
293 enum rename_type rename_type;
294 struct rename *ren1;
295 struct rename *ren2;
296 };
297
298 static inline void setup_rename_conflict_info(enum rename_type rename_type,
299 struct merge_options *opt,
300 struct rename *ren1,
301 struct rename *ren2)
302 {
303 struct rename_conflict_info *ci;
304
305 /*
306 * When we have two renames involved, it's easiest to get the
307 * correct things into stage 2 and 3, and to make sure that the
308 * content merge puts HEAD before the other branch if we just
309 * ensure that branch1 == opt->branch1. So, simply flip arguments
310 * around if we don't have that.
311 */
312 if (ren2 && ren1->branch != opt->branch1) {
313 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
314 return;
315 }
316
317 ci = xcalloc(1, sizeof(struct rename_conflict_info));
318 ci->rename_type = rename_type;
319 ci->ren1 = ren1;
320 ci->ren2 = ren2;
321
322 ci->ren1->dst_entry->processed = 0;
323 ci->ren1->dst_entry->rename_conflict_info = ci;
324 if (ren2) {
325 ci->ren2->dst_entry->rename_conflict_info = ci;
326 }
327 }
328
329 static int show(struct merge_options *opt, int v)
330 {
331 return (!opt->priv->call_depth && opt->verbosity >= v) ||
332 opt->verbosity >= 5;
333 }
334
335 __attribute__((format (printf, 3, 4)))
336 static void output(struct merge_options *opt, int v, const char *fmt, ...)
337 {
338 va_list ap;
339
340 if (!show(opt, v))
341 return;
342
343 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
344
345 va_start(ap, fmt);
346 strbuf_vaddf(&opt->obuf, fmt, ap);
347 va_end(ap);
348
349 strbuf_addch(&opt->obuf, '\n');
350 if (!opt->buffer_output)
351 flush_output(opt);
352 }
353
354 static void output_commit_title(struct merge_options *opt, struct commit *commit)
355 {
356 struct merge_remote_desc *desc;
357
358 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
359 desc = merge_remote_util(commit);
360 if (desc)
361 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
362 else {
363 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
364 DEFAULT_ABBREV);
365 strbuf_addch(&opt->obuf, ' ');
366 if (parse_commit(commit) != 0)
367 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
368 else {
369 const char *title;
370 const char *msg = get_commit_buffer(commit, NULL);
371 int len = find_commit_subject(msg, &title);
372 if (len)
373 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
374 unuse_commit_buffer(commit, msg);
375 }
376 }
377 flush_output(opt);
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 parse_tree(tree);
414 init_tree_desc(desc, tree->buffer, tree->size);
415 }
416
417 static int unpack_trees_start(struct merge_options *opt,
418 struct tree *common,
419 struct tree *head,
420 struct tree *merge)
421 {
422 int rc;
423 struct tree_desc t[3];
424 struct index_state tmp_index = { NULL };
425
426 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
427 if (opt->priv->call_depth)
428 opt->priv->unpack_opts.index_only = 1;
429 else
430 opt->priv->unpack_opts.update = 1;
431 opt->priv->unpack_opts.merge = 1;
432 opt->priv->unpack_opts.head_idx = 2;
433 opt->priv->unpack_opts.fn = threeway_merge;
434 opt->priv->unpack_opts.src_index = opt->repo->index;
435 opt->priv->unpack_opts.dst_index = &tmp_index;
436 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
437 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
438
439 init_tree_desc_from_tree(t+0, common);
440 init_tree_desc_from_tree(t+1, head);
441 init_tree_desc_from_tree(t+2, merge);
442
443 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
444 cache_tree_free(&opt->repo->index->cache_tree);
445
446 /*
447 * Update opt->repo->index to match the new results, AFTER saving a
448 * copy in opt->priv->orig_index. Update src_index to point to the
449 * saved copy. (verify_uptodate() checks src_index, and the original
450 * index is the one that had the necessary modification timestamps.)
451 */
452 opt->priv->orig_index = *opt->repo->index;
453 *opt->repo->index = tmp_index;
454 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
455
456 return rc;
457 }
458
459 static void unpack_trees_finish(struct merge_options *opt)
460 {
461 discard_index(&opt->priv->orig_index);
462 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
463 }
464
465 static int save_files_dirs(const struct object_id *oid,
466 struct strbuf *base, const char *path,
467 unsigned int mode, int stage, void *context)
468 {
469 struct path_hashmap_entry *entry;
470 int baselen = base->len;
471 struct merge_options *opt = context;
472
473 strbuf_addstr(base, path);
474
475 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
476 hashmap_entry_init(&entry->e, path_hash(entry->path));
477 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
478
479 strbuf_setlen(base, baselen);
480 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
481 }
482
483 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
484 {
485 struct pathspec match_all;
486 memset(&match_all, 0, sizeof(match_all));
487 read_tree_recursive(opt->repo, tree, "", 0, 0,
488 &match_all, save_files_dirs, opt);
489 }
490
491 static int get_tree_entry_if_blob(struct repository *r,
492 const struct object_id *tree,
493 const char *path,
494 struct diff_filespec *dfs)
495 {
496 int ret;
497
498 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
499 if (S_ISDIR(dfs->mode)) {
500 oidcpy(&dfs->oid, &null_oid);
501 dfs->mode = 0;
502 }
503 return ret;
504 }
505
506 /*
507 * Returns an index_entry instance which doesn't have to correspond to
508 * a real cache entry in Git's index.
509 */
510 static struct stage_data *insert_stage_data(struct repository *r,
511 const char *path,
512 struct tree *o, struct tree *a, struct tree *b,
513 struct string_list *entries)
514 {
515 struct string_list_item *item;
516 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
517 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
518 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
519 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
520 item = string_list_insert(entries, path);
521 item->util = e;
522 return e;
523 }
524
525 /*
526 * Create a dictionary mapping file names to stage_data objects. The
527 * dictionary contains one entry for every path with a non-zero stage entry.
528 */
529 static struct string_list *get_unmerged(struct index_state *istate)
530 {
531 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
532 int i;
533
534 unmerged->strdup_strings = 1;
535
536 for (i = 0; i < istate->cache_nr; i++) {
537 struct string_list_item *item;
538 struct stage_data *e;
539 const struct cache_entry *ce = istate->cache[i];
540 if (!ce_stage(ce))
541 continue;
542
543 item = string_list_lookup(unmerged, ce->name);
544 if (!item) {
545 item = string_list_insert(unmerged, ce->name);
546 item->util = xcalloc(1, sizeof(struct stage_data));
547 }
548 e = item->util;
549 e->stages[ce_stage(ce)].mode = ce->ce_mode;
550 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
551 }
552
553 return unmerged;
554 }
555
556 static int string_list_df_name_compare(const char *one, const char *two)
557 {
558 int onelen = strlen(one);
559 int twolen = strlen(two);
560 /*
561 * Here we only care that entries for D/F conflicts are
562 * adjacent, in particular with the file of the D/F conflict
563 * appearing before files below the corresponding directory.
564 * The order of the rest of the list is irrelevant for us.
565 *
566 * To achieve this, we sort with df_name_compare and provide
567 * the mode S_IFDIR so that D/F conflicts will sort correctly.
568 * We use the mode S_IFDIR for everything else for simplicity,
569 * since in other cases any changes in their order due to
570 * sorting cause no problems for us.
571 */
572 int cmp = df_name_compare(one, onelen, S_IFDIR,
573 two, twolen, S_IFDIR);
574 /*
575 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
576 * that 'foo' comes before 'foo/bar'.
577 */
578 if (cmp)
579 return cmp;
580 return onelen - twolen;
581 }
582
583 static void record_df_conflict_files(struct merge_options *opt,
584 struct string_list *entries)
585 {
586 /* If there is a D/F conflict and the file for such a conflict
587 * currently exists in the working tree, we want to allow it to be
588 * removed to make room for the corresponding directory if needed.
589 * The files underneath the directories of such D/F conflicts will
590 * be processed before the corresponding file involved in the D/F
591 * conflict. If the D/F directory ends up being removed by the
592 * merge, then we won't have to touch the D/F file. If the D/F
593 * directory needs to be written to the working copy, then the D/F
594 * file will simply be removed (in make_room_for_path()) to make
595 * room for the necessary paths. Note that if both the directory
596 * and the file need to be present, then the D/F file will be
597 * reinstated with a new unique name at the time it is processed.
598 */
599 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
600 const char *last_file = NULL;
601 int last_len = 0;
602 int i;
603
604 /*
605 * If we're merging merge-bases, we don't want to bother with
606 * any working directory changes.
607 */
608 if (opt->priv->call_depth)
609 return;
610
611 /* Ensure D/F conflicts are adjacent in the entries list. */
612 for (i = 0; i < entries->nr; i++) {
613 struct string_list_item *next = &entries->items[i];
614 string_list_append(&df_sorted_entries, next->string)->util =
615 next->util;
616 }
617 df_sorted_entries.cmp = string_list_df_name_compare;
618 string_list_sort(&df_sorted_entries);
619
620 string_list_clear(&opt->priv->df_conflict_file_set, 1);
621 for (i = 0; i < df_sorted_entries.nr; i++) {
622 const char *path = df_sorted_entries.items[i].string;
623 int len = strlen(path);
624 struct stage_data *e = df_sorted_entries.items[i].util;
625
626 /*
627 * Check if last_file & path correspond to a D/F conflict;
628 * i.e. whether path is last_file+'/'+<something>.
629 * If so, record that it's okay to remove last_file to make
630 * room for path and friends if needed.
631 */
632 if (last_file &&
633 len > last_len &&
634 memcmp(path, last_file, last_len) == 0 &&
635 path[last_len] == '/') {
636 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
637 }
638
639 /*
640 * Determine whether path could exist as a file in the
641 * working directory as a possible D/F conflict. This
642 * will only occur when it exists in stage 2 as a
643 * file.
644 */
645 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
646 last_file = path;
647 last_len = len;
648 } else {
649 last_file = NULL;
650 }
651 }
652 string_list_clear(&df_sorted_entries, 0);
653 }
654
655 static int update_stages(struct merge_options *opt, const char *path,
656 const struct diff_filespec *o,
657 const struct diff_filespec *a,
658 const struct diff_filespec *b)
659 {
660
661 /*
662 * NOTE: It is usually a bad idea to call update_stages on a path
663 * before calling update_file on that same path, since it can
664 * sometimes lead to spurious "refusing to lose untracked file..."
665 * messages from update_file (via make_room_for path via
666 * would_lose_untracked). Instead, reverse the order of the calls
667 * (executing update_file first and then update_stages).
668 */
669 int clear = 1;
670 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
671 if (clear)
672 if (remove_file_from_index(opt->repo->index, path))
673 return -1;
674 if (o)
675 if (add_cacheinfo(opt, o, path, 1, 0, options))
676 return -1;
677 if (a)
678 if (add_cacheinfo(opt, a, path, 2, 0, options))
679 return -1;
680 if (b)
681 if (add_cacheinfo(opt, b, path, 3, 0, options))
682 return -1;
683 return 0;
684 }
685
686 static void update_entry(struct stage_data *entry,
687 struct diff_filespec *o,
688 struct diff_filespec *a,
689 struct diff_filespec *b)
690 {
691 entry->processed = 0;
692 entry->stages[1].mode = o->mode;
693 entry->stages[2].mode = a->mode;
694 entry->stages[3].mode = b->mode;
695 oidcpy(&entry->stages[1].oid, &o->oid);
696 oidcpy(&entry->stages[2].oid, &a->oid);
697 oidcpy(&entry->stages[3].oid, &b->oid);
698 }
699
700 static int remove_file(struct merge_options *opt, int clean,
701 const char *path, int no_wd)
702 {
703 int update_cache = opt->priv->call_depth || clean;
704 int update_working_directory = !opt->priv->call_depth && !no_wd;
705
706 if (update_cache) {
707 if (remove_file_from_index(opt->repo->index, path))
708 return -1;
709 }
710 if (update_working_directory) {
711 if (ignore_case) {
712 struct cache_entry *ce;
713 ce = index_file_exists(opt->repo->index, path, strlen(path),
714 ignore_case);
715 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
716 return 0;
717 }
718 if (remove_path(path))
719 return -1;
720 }
721 return 0;
722 }
723
724 /* add a string to a strbuf, but converting "/" to "_" */
725 static void add_flattened_path(struct strbuf *out, const char *s)
726 {
727 size_t i = out->len;
728 strbuf_addstr(out, s);
729 for (; i < out->len; i++)
730 if (out->buf[i] == '/')
731 out->buf[i] = '_';
732 }
733
734 static char *unique_path(struct merge_options *opt,
735 const char *path,
736 const char *branch)
737 {
738 struct path_hashmap_entry *entry;
739 struct strbuf newpath = STRBUF_INIT;
740 int suffix = 0;
741 size_t base_len;
742
743 strbuf_addf(&newpath, "%s~", path);
744 add_flattened_path(&newpath, branch);
745
746 base_len = newpath.len;
747 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
748 path_hash(newpath.buf), newpath.buf) ||
749 (!opt->priv->call_depth && file_exists(newpath.buf))) {
750 strbuf_setlen(&newpath, base_len);
751 strbuf_addf(&newpath, "_%d", suffix++);
752 }
753
754 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
755 hashmap_entry_init(&entry->e, path_hash(entry->path));
756 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
757 return strbuf_detach(&newpath, NULL);
758 }
759
760 /**
761 * Check whether a directory in the index is in the way of an incoming
762 * file. Return 1 if so. If check_working_copy is non-zero, also
763 * check the working directory. If empty_ok is non-zero, also return
764 * 0 in the case where the working-tree dir exists but is empty.
765 */
766 static int dir_in_way(struct index_state *istate, const char *path,
767 int check_working_copy, int empty_ok)
768 {
769 int pos;
770 struct strbuf dirpath = STRBUF_INIT;
771 struct stat st;
772
773 strbuf_addstr(&dirpath, path);
774 strbuf_addch(&dirpath, '/');
775
776 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
777
778 if (pos < 0)
779 pos = -1 - pos;
780 if (pos < istate->cache_nr &&
781 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
782 strbuf_release(&dirpath);
783 return 1;
784 }
785
786 strbuf_release(&dirpath);
787 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
788 !(empty_ok && is_empty_dir(path)) &&
789 !has_symlink_leading_path(path, strlen(path));
790 }
791
792 /*
793 * Returns whether path was tracked in the index before the merge started,
794 * and its oid and mode match the specified values
795 */
796 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
797 const struct diff_filespec *blob)
798 {
799 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
800 struct cache_entry *ce;
801
802 if (0 > pos)
803 /* we were not tracking this path before the merge */
804 return 0;
805
806 /* See if the file we were tracking before matches */
807 ce = opt->priv->orig_index.cache[pos];
808 return (oid_eq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
809 }
810
811 /*
812 * Returns whether path was tracked in the index before the merge started
813 */
814 static int was_tracked(struct merge_options *opt, const char *path)
815 {
816 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
817
818 if (0 <= pos)
819 /* we were tracking this path before the merge */
820 return 1;
821
822 return 0;
823 }
824
825 static int would_lose_untracked(struct merge_options *opt, const char *path)
826 {
827 struct index_state *istate = opt->repo->index;
828
829 /*
830 * This may look like it can be simplified to:
831 * return !was_tracked(opt, path) && file_exists(path)
832 * but it can't. This function needs to know whether path was in
833 * the working tree due to EITHER having been tracked in the index
834 * before the merge OR having been put into the working copy and
835 * index by unpack_trees(). Due to that either-or requirement, we
836 * check the current index instead of the original one.
837 *
838 * Note that we do not need to worry about merge-recursive itself
839 * updating the index after unpack_trees() and before calling this
840 * function, because we strictly require all code paths in
841 * merge-recursive to update the working tree first and the index
842 * second. Doing otherwise would break
843 * update_file()/would_lose_untracked(); see every comment in this
844 * file which mentions "update_stages".
845 */
846 int pos = index_name_pos(istate, path, strlen(path));
847
848 if (pos < 0)
849 pos = -1 - pos;
850 while (pos < istate->cache_nr &&
851 !strcmp(path, istate->cache[pos]->name)) {
852 /*
853 * If stage #0, it is definitely tracked.
854 * If it has stage #2 then it was tracked
855 * before this merge started. All other
856 * cases the path was not tracked.
857 */
858 switch (ce_stage(istate->cache[pos])) {
859 case 0:
860 case 2:
861 return 0;
862 }
863 pos++;
864 }
865 return file_exists(path);
866 }
867
868 static int was_dirty(struct merge_options *opt, const char *path)
869 {
870 struct cache_entry *ce;
871 int dirty = 1;
872
873 if (opt->priv->call_depth || !was_tracked(opt, path))
874 return !dirty;
875
876 ce = index_file_exists(opt->priv->unpack_opts.src_index,
877 path, strlen(path), ignore_case);
878 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
879 return dirty;
880 }
881
882 static int make_room_for_path(struct merge_options *opt, const char *path)
883 {
884 int status, i;
885 const char *msg = _("failed to create path '%s'%s");
886
887 /* Unlink any D/F conflict files that are in the way */
888 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
889 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
890 size_t pathlen = strlen(path);
891 size_t df_pathlen = strlen(df_path);
892 if (df_pathlen < pathlen &&
893 path[df_pathlen] == '/' &&
894 strncmp(path, df_path, df_pathlen) == 0) {
895 output(opt, 3,
896 _("Removing %s to make room for subdirectory\n"),
897 df_path);
898 unlink(df_path);
899 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
900 i, 0);
901 break;
902 }
903 }
904
905 /* Make sure leading directories are created */
906 status = safe_create_leading_directories_const(path);
907 if (status) {
908 if (status == SCLD_EXISTS)
909 /* something else exists */
910 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
911 return err(opt, msg, path, "");
912 }
913
914 /*
915 * Do not unlink a file in the work tree if we are not
916 * tracking it.
917 */
918 if (would_lose_untracked(opt, path))
919 return err(opt, _("refusing to lose untracked file at '%s'"),
920 path);
921
922 /* Successful unlink is good.. */
923 if (!unlink(path))
924 return 0;
925 /* .. and so is no existing file */
926 if (errno == ENOENT)
927 return 0;
928 /* .. but not some other error (who really cares what?) */
929 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
930 }
931
932 static int update_file_flags(struct merge_options *opt,
933 const struct diff_filespec *contents,
934 const char *path,
935 int update_cache,
936 int update_wd)
937 {
938 int ret = 0;
939
940 if (opt->priv->call_depth)
941 update_wd = 0;
942
943 if (update_wd) {
944 enum object_type type;
945 void *buf;
946 unsigned long size;
947
948 if (S_ISGITLINK(contents->mode)) {
949 /*
950 * We may later decide to recursively descend into
951 * the submodule directory and update its index
952 * and/or work tree, but we do not do that now.
953 */
954 update_wd = 0;
955 goto update_index;
956 }
957
958 buf = read_object_file(&contents->oid, &type, &size);
959 if (!buf) {
960 ret = err(opt, _("cannot read object %s '%s'"),
961 oid_to_hex(&contents->oid), path);
962 goto free_buf;
963 }
964 if (type != OBJ_BLOB) {
965 ret = err(opt, _("blob expected for %s '%s'"),
966 oid_to_hex(&contents->oid), path);
967 goto free_buf;
968 }
969 if (S_ISREG(contents->mode)) {
970 struct strbuf strbuf = STRBUF_INIT;
971 if (convert_to_working_tree(opt->repo->index,
972 path, buf, size, &strbuf)) {
973 free(buf);
974 size = strbuf.len;
975 buf = strbuf_detach(&strbuf, NULL);
976 }
977 }
978
979 if (make_room_for_path(opt, path) < 0) {
980 update_wd = 0;
981 goto free_buf;
982 }
983 if (S_ISREG(contents->mode) ||
984 (!has_symlinks && S_ISLNK(contents->mode))) {
985 int fd;
986 int mode = (contents->mode & 0100 ? 0777 : 0666);
987
988 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
989 if (fd < 0) {
990 ret = err(opt, _("failed to open '%s': %s"),
991 path, strerror(errno));
992 goto free_buf;
993 }
994 write_in_full(fd, buf, size);
995 close(fd);
996 } else if (S_ISLNK(contents->mode)) {
997 char *lnk = xmemdupz(buf, size);
998 safe_create_leading_directories_const(path);
999 unlink(path);
1000 if (symlink(lnk, path))
1001 ret = err(opt, _("failed to symlink '%s': %s"),
1002 path, strerror(errno));
1003 free(lnk);
1004 } else
1005 ret = err(opt,
1006 _("do not know what to do with %06o %s '%s'"),
1007 contents->mode, oid_to_hex(&contents->oid), path);
1008 free_buf:
1009 free(buf);
1010 }
1011 update_index:
1012 if (!ret && update_cache)
1013 if (add_cacheinfo(opt, contents, path, 0, update_wd,
1014 ADD_CACHE_OK_TO_ADD))
1015 return -1;
1016 return ret;
1017 }
1018
1019 static int update_file(struct merge_options *opt,
1020 int clean,
1021 const struct diff_filespec *contents,
1022 const char *path)
1023 {
1024 return update_file_flags(opt, contents, path,
1025 opt->priv->call_depth || clean, !opt->priv->call_depth);
1026 }
1027
1028 /* Low level file merging, update and removal */
1029
1030 struct merge_file_info {
1031 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1032 unsigned clean:1,
1033 merge:1;
1034 };
1035
1036 static int merge_3way(struct merge_options *opt,
1037 mmbuffer_t *result_buf,
1038 const struct diff_filespec *o,
1039 const struct diff_filespec *a,
1040 const struct diff_filespec *b,
1041 const char *branch1,
1042 const char *branch2,
1043 const int extra_marker_size)
1044 {
1045 mmfile_t orig, src1, src2;
1046 struct ll_merge_options ll_opts = {0};
1047 char *base, *name1, *name2;
1048 int merge_status;
1049
1050 ll_opts.renormalize = opt->renormalize;
1051 ll_opts.extra_marker_size = extra_marker_size;
1052 ll_opts.xdl_opts = opt->xdl_opts;
1053
1054 if (opt->priv->call_depth) {
1055 ll_opts.virtual_ancestor = 1;
1056 ll_opts.variant = 0;
1057 } else {
1058 switch (opt->recursive_variant) {
1059 case MERGE_VARIANT_OURS:
1060 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1061 break;
1062 case MERGE_VARIANT_THEIRS:
1063 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1064 break;
1065 default:
1066 ll_opts.variant = 0;
1067 break;
1068 }
1069 }
1070
1071 assert(a->path && b->path && o->path && opt->ancestor);
1072 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1073 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1074 name1 = mkpathdup("%s:%s", branch1, a->path);
1075 name2 = mkpathdup("%s:%s", branch2, b->path);
1076 } else {
1077 base = mkpathdup("%s", opt->ancestor);
1078 name1 = mkpathdup("%s", branch1);
1079 name2 = mkpathdup("%s", branch2);
1080 }
1081
1082 read_mmblob(&orig, &o->oid);
1083 read_mmblob(&src1, &a->oid);
1084 read_mmblob(&src2, &b->oid);
1085
1086 merge_status = ll_merge(result_buf, a->path, &orig, base,
1087 &src1, name1, &src2, name2,
1088 opt->repo->index, &ll_opts);
1089
1090 free(base);
1091 free(name1);
1092 free(name2);
1093 free(orig.ptr);
1094 free(src1.ptr);
1095 free(src2.ptr);
1096 return merge_status;
1097 }
1098
1099 static int find_first_merges(struct repository *repo,
1100 struct object_array *result, const char *path,
1101 struct commit *a, struct commit *b)
1102 {
1103 int i, j;
1104 struct object_array merges = OBJECT_ARRAY_INIT;
1105 struct commit *commit;
1106 int contains_another;
1107
1108 char merged_revision[GIT_MAX_HEXSZ + 2];
1109 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1110 "--all", merged_revision, NULL };
1111 struct rev_info revs;
1112 struct setup_revision_opt rev_opts;
1113
1114 memset(result, 0, sizeof(struct object_array));
1115 memset(&rev_opts, 0, sizeof(rev_opts));
1116
1117 /* get all revisions that merge commit a */
1118 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1119 oid_to_hex(&a->object.oid));
1120 repo_init_revisions(repo, &revs, NULL);
1121 rev_opts.submodule = path;
1122 /* FIXME: can't handle linked worktrees in submodules yet */
1123 revs.single_worktree = path != NULL;
1124 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1125
1126 /* save all revisions from the above list that contain b */
1127 if (prepare_revision_walk(&revs))
1128 die("revision walk setup failed");
1129 while ((commit = get_revision(&revs)) != NULL) {
1130 struct object *o = &(commit->object);
1131 if (in_merge_bases(b, commit))
1132 add_object_array(o, NULL, &merges);
1133 }
1134 reset_revision_walk();
1135
1136 /* Now we've got all merges that contain a and b. Prune all
1137 * merges that contain another found merge and save them in
1138 * result.
1139 */
1140 for (i = 0; i < merges.nr; i++) {
1141 struct commit *m1 = (struct commit *) merges.objects[i].item;
1142
1143 contains_another = 0;
1144 for (j = 0; j < merges.nr; j++) {
1145 struct commit *m2 = (struct commit *) merges.objects[j].item;
1146 if (i != j && in_merge_bases(m2, m1)) {
1147 contains_another = 1;
1148 break;
1149 }
1150 }
1151
1152 if (!contains_another)
1153 add_object_array(merges.objects[i].item, NULL, result);
1154 }
1155
1156 object_array_clear(&merges);
1157 return result->nr;
1158 }
1159
1160 static void print_commit(struct commit *commit)
1161 {
1162 struct strbuf sb = STRBUF_INIT;
1163 struct pretty_print_context ctx = {0};
1164 ctx.date_mode.type = DATE_NORMAL;
1165 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1166 fprintf(stderr, "%s\n", sb.buf);
1167 strbuf_release(&sb);
1168 }
1169
1170 static int is_valid(const struct diff_filespec *dfs)
1171 {
1172 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1173 }
1174
1175 static int merge_submodule(struct merge_options *opt,
1176 struct object_id *result, const char *path,
1177 const struct object_id *base, const struct object_id *a,
1178 const struct object_id *b)
1179 {
1180 struct commit *commit_base, *commit_a, *commit_b;
1181 int parent_count;
1182 struct object_array merges;
1183
1184 int i;
1185 int search = !opt->priv->call_depth;
1186
1187 /* store a in result in case we fail */
1188 oidcpy(result, a);
1189
1190 /* we can not handle deletion conflicts */
1191 if (is_null_oid(base))
1192 return 0;
1193 if (is_null_oid(a))
1194 return 0;
1195 if (is_null_oid(b))
1196 return 0;
1197
1198 if (add_submodule_odb(path)) {
1199 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1200 return 0;
1201 }
1202
1203 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1204 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1205 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1206 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1207 return 0;
1208 }
1209
1210 /* check whether both changes are forward */
1211 if (!in_merge_bases(commit_base, commit_a) ||
1212 !in_merge_bases(commit_base, commit_b)) {
1213 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1214 return 0;
1215 }
1216
1217 /* Case #1: a is contained in b or vice versa */
1218 if (in_merge_bases(commit_a, commit_b)) {
1219 oidcpy(result, b);
1220 if (show(opt, 3)) {
1221 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1222 output_commit_title(opt, commit_b);
1223 } else if (show(opt, 2))
1224 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1225 else
1226 ; /* no output */
1227
1228 return 1;
1229 }
1230 if (in_merge_bases(commit_b, commit_a)) {
1231 oidcpy(result, a);
1232 if (show(opt, 3)) {
1233 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1234 output_commit_title(opt, commit_a);
1235 } else if (show(opt, 2))
1236 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1237 else
1238 ; /* no output */
1239
1240 return 1;
1241 }
1242
1243 /*
1244 * Case #2: There are one or more merges that contain a and b in
1245 * the submodule. If there is only one, then present it as a
1246 * suggestion to the user, but leave it marked unmerged so the
1247 * user needs to confirm the resolution.
1248 */
1249
1250 /* Skip the search if makes no sense to the calling context. */
1251 if (!search)
1252 return 0;
1253
1254 /* find commit which merges them */
1255 parent_count = find_first_merges(opt->repo, &merges, path,
1256 commit_a, commit_b);
1257 switch (parent_count) {
1258 case 0:
1259 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1260 break;
1261
1262 case 1:
1263 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1264 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1265 print_commit((struct commit *) merges.objects[0].item);
1266 output(opt, 2, _(
1267 "If this is correct simply add it to the index "
1268 "for example\n"
1269 "by using:\n\n"
1270 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1271 "which will accept this suggestion.\n"),
1272 oid_to_hex(&merges.objects[0].item->oid), path);
1273 break;
1274
1275 default:
1276 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1277 for (i = 0; i < merges.nr; i++)
1278 print_commit((struct commit *) merges.objects[i].item);
1279 }
1280
1281 object_array_clear(&merges);
1282 return 0;
1283 }
1284
1285 static int merge_mode_and_contents(struct merge_options *opt,
1286 const struct diff_filespec *o,
1287 const struct diff_filespec *a,
1288 const struct diff_filespec *b,
1289 const char *filename,
1290 const char *branch1,
1291 const char *branch2,
1292 const int extra_marker_size,
1293 struct merge_file_info *result)
1294 {
1295 if (opt->branch1 != branch1) {
1296 /*
1297 * It's weird getting a reverse merge with HEAD on the bottom
1298 * side of the conflict markers and the other branch on the
1299 * top. Fix that.
1300 */
1301 return merge_mode_and_contents(opt, o, b, a,
1302 filename,
1303 branch2, branch1,
1304 extra_marker_size, result);
1305 }
1306
1307 result->merge = 0;
1308 result->clean = 1;
1309
1310 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1311 result->clean = 0;
1312 if (S_ISREG(a->mode)) {
1313 result->blob.mode = a->mode;
1314 oidcpy(&result->blob.oid, &a->oid);
1315 } else {
1316 result->blob.mode = b->mode;
1317 oidcpy(&result->blob.oid, &b->oid);
1318 }
1319 } else {
1320 if (!oid_eq(&a->oid, &o->oid) && !oid_eq(&b->oid, &o->oid))
1321 result->merge = 1;
1322
1323 /*
1324 * Merge modes
1325 */
1326 if (a->mode == b->mode || a->mode == o->mode)
1327 result->blob.mode = b->mode;
1328 else {
1329 result->blob.mode = a->mode;
1330 if (b->mode != o->mode) {
1331 result->clean = 0;
1332 result->merge = 1;
1333 }
1334 }
1335
1336 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &o->oid))
1337 oidcpy(&result->blob.oid, &b->oid);
1338 else if (oid_eq(&b->oid, &o->oid))
1339 oidcpy(&result->blob.oid, &a->oid);
1340 else if (S_ISREG(a->mode)) {
1341 mmbuffer_t result_buf;
1342 int ret = 0, merge_status;
1343
1344 merge_status = merge_3way(opt, &result_buf, o, a, b,
1345 branch1, branch2,
1346 extra_marker_size);
1347
1348 if ((merge_status < 0) || !result_buf.ptr)
1349 ret = err(opt, _("Failed to execute internal merge"));
1350
1351 if (!ret &&
1352 write_object_file(result_buf.ptr, result_buf.size,
1353 blob_type, &result->blob.oid))
1354 ret = err(opt, _("Unable to add %s to database"),
1355 a->path);
1356
1357 free(result_buf.ptr);
1358 if (ret)
1359 return ret;
1360 result->clean = (merge_status == 0);
1361 } else if (S_ISGITLINK(a->mode)) {
1362 result->clean = merge_submodule(opt, &result->blob.oid,
1363 o->path,
1364 &o->oid,
1365 &a->oid,
1366 &b->oid);
1367 } else if (S_ISLNK(a->mode)) {
1368 switch (opt->recursive_variant) {
1369 case MERGE_VARIANT_NORMAL:
1370 oidcpy(&result->blob.oid, &a->oid);
1371 if (!oid_eq(&a->oid, &b->oid))
1372 result->clean = 0;
1373 break;
1374 case MERGE_VARIANT_OURS:
1375 oidcpy(&result->blob.oid, &a->oid);
1376 break;
1377 case MERGE_VARIANT_THEIRS:
1378 oidcpy(&result->blob.oid, &b->oid);
1379 break;
1380 }
1381 } else
1382 BUG("unsupported object type in the tree");
1383 }
1384
1385 if (result->merge)
1386 output(opt, 2, _("Auto-merging %s"), filename);
1387
1388 return 0;
1389 }
1390
1391 static int handle_rename_via_dir(struct merge_options *opt,
1392 struct rename_conflict_info *ci)
1393 {
1394 /*
1395 * Handle file adds that need to be renamed due to directory rename
1396 * detection. This differs from handle_rename_normal, because
1397 * there is no content merge to do; just move the file into the
1398 * desired final location.
1399 */
1400 const struct rename *ren = ci->ren1;
1401 const struct diff_filespec *dest = ren->pair->two;
1402 char *file_path = dest->path;
1403 int mark_conflicted = (opt->detect_directory_renames ==
1404 MERGE_DIRECTORY_RENAMES_CONFLICT);
1405 assert(ren->dir_rename_original_dest);
1406
1407 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1408 mark_conflicted = 1;
1409 file_path = unique_path(opt, dest->path, ren->branch);
1410 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1411 "writing to %s instead."),
1412 dest->path, file_path);
1413 }
1414
1415 if (mark_conflicted) {
1416 /*
1417 * Write the file in worktree at file_path. In the index,
1418 * only record the file at dest->path in the appropriate
1419 * higher stage.
1420 */
1421 if (update_file(opt, 0, dest, file_path))
1422 return -1;
1423 if (file_path != dest->path)
1424 free(file_path);
1425 if (update_stages(opt, dest->path, NULL,
1426 ren->branch == opt->branch1 ? dest : NULL,
1427 ren->branch == opt->branch1 ? NULL : dest))
1428 return -1;
1429 return 0; /* not clean, but conflicted */
1430 } else {
1431 /* Update dest->path both in index and in worktree */
1432 if (update_file(opt, 1, dest, dest->path))
1433 return -1;
1434 return 1; /* clean */
1435 }
1436 }
1437
1438 static int handle_change_delete(struct merge_options *opt,
1439 const char *path, const char *old_path,
1440 const struct diff_filespec *o,
1441 const struct diff_filespec *changed,
1442 const char *change_branch,
1443 const char *delete_branch,
1444 const char *change, const char *change_past)
1445 {
1446 char *alt_path = NULL;
1447 const char *update_path = path;
1448 int ret = 0;
1449
1450 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1451 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1452 update_path = alt_path = unique_path(opt, path, change_branch);
1453 }
1454
1455 if (opt->priv->call_depth) {
1456 /*
1457 * We cannot arbitrarily accept either a_sha or b_sha as
1458 * correct; since there is no true "middle point" between
1459 * them, simply reuse the base version for virtual merge base.
1460 */
1461 ret = remove_file_from_index(opt->repo->index, path);
1462 if (!ret)
1463 ret = update_file(opt, 0, o, update_path);
1464 } else {
1465 /*
1466 * Despite the four nearly duplicate messages and argument
1467 * lists below and the ugliness of the nested if-statements,
1468 * having complete messages makes the job easier for
1469 * translators.
1470 *
1471 * The slight variance among the cases is due to the fact
1472 * that:
1473 * 1) directory/file conflicts (in effect if
1474 * !alt_path) could cause us to need to write the
1475 * file to a different path.
1476 * 2) renames (in effect if !old_path) could mean that
1477 * there are two names for the path that the user
1478 * may know the file by.
1479 */
1480 if (!alt_path) {
1481 if (!old_path) {
1482 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1483 "and %s in %s. Version %s of %s left in tree."),
1484 change, path, delete_branch, change_past,
1485 change_branch, change_branch, path);
1486 } else {
1487 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1488 "and %s to %s in %s. Version %s of %s left in tree."),
1489 change, old_path, delete_branch, change_past, path,
1490 change_branch, change_branch, path);
1491 }
1492 } else {
1493 if (!old_path) {
1494 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1495 "and %s in %s. Version %s of %s left in tree at %s."),
1496 change, path, delete_branch, change_past,
1497 change_branch, change_branch, path, alt_path);
1498 } else {
1499 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1500 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1501 change, old_path, delete_branch, change_past, path,
1502 change_branch, change_branch, path, alt_path);
1503 }
1504 }
1505 /*
1506 * No need to call update_file() on path when change_branch ==
1507 * opt->branch1 && !alt_path, since that would needlessly touch
1508 * path. We could call update_file_flags() with update_cache=0
1509 * and update_wd=0, but that's a no-op.
1510 */
1511 if (change_branch != opt->branch1 || alt_path)
1512 ret = update_file(opt, 0, changed, update_path);
1513 }
1514 free(alt_path);
1515
1516 return ret;
1517 }
1518
1519 static int handle_rename_delete(struct merge_options *opt,
1520 struct rename_conflict_info *ci)
1521 {
1522 const struct rename *ren = ci->ren1;
1523 const struct diff_filespec *orig = ren->pair->one;
1524 const struct diff_filespec *dest = ren->pair->two;
1525 const char *rename_branch = ren->branch;
1526 const char *delete_branch = (opt->branch1 == ren->branch ?
1527 opt->branch2 : opt->branch1);
1528
1529 if (handle_change_delete(opt,
1530 opt->priv->call_depth ? orig->path : dest->path,
1531 opt->priv->call_depth ? NULL : orig->path,
1532 orig, dest,
1533 rename_branch, delete_branch,
1534 _("rename"), _("renamed")))
1535 return -1;
1536
1537 if (opt->priv->call_depth)
1538 return remove_file_from_index(opt->repo->index, dest->path);
1539 else
1540 return update_stages(opt, dest->path, NULL,
1541 rename_branch == opt->branch1 ? dest : NULL,
1542 rename_branch == opt->branch1 ? NULL : dest);
1543 }
1544
1545 static int handle_file_collision(struct merge_options *opt,
1546 const char *collide_path,
1547 const char *prev_path1,
1548 const char *prev_path2,
1549 const char *branch1, const char *branch2,
1550 struct diff_filespec *a,
1551 struct diff_filespec *b)
1552 {
1553 struct merge_file_info mfi;
1554 struct diff_filespec null;
1555 char *alt_path = NULL;
1556 const char *update_path = collide_path;
1557
1558 /*
1559 * It's easiest to get the correct things into stage 2 and 3, and
1560 * to make sure that the content merge puts HEAD before the other
1561 * branch if we just ensure that branch1 == opt->branch1. So, simply
1562 * flip arguments around if we don't have that.
1563 */
1564 if (branch1 != opt->branch1) {
1565 return handle_file_collision(opt, collide_path,
1566 prev_path2, prev_path1,
1567 branch2, branch1,
1568 b, a);
1569 }
1570
1571 /*
1572 * In the recursive case, we just opt to undo renames
1573 */
1574 if (opt->priv->call_depth && (prev_path1 || prev_path2)) {
1575 /* Put first file (a->oid, a->mode) in its original spot */
1576 if (prev_path1) {
1577 if (update_file(opt, 1, a, prev_path1))
1578 return -1;
1579 } else {
1580 if (update_file(opt, 1, a, collide_path))
1581 return -1;
1582 }
1583
1584 /* Put second file (b->oid, b->mode) in its original spot */
1585 if (prev_path2) {
1586 if (update_file(opt, 1, b, prev_path2))
1587 return -1;
1588 } else {
1589 if (update_file(opt, 1, b, collide_path))
1590 return -1;
1591 }
1592
1593 /* Don't leave something at collision path if unrenaming both */
1594 if (prev_path1 && prev_path2)
1595 remove_file(opt, 1, collide_path, 0);
1596
1597 return 0;
1598 }
1599
1600 /* Remove rename sources if rename/add or rename/rename(2to1) */
1601 if (prev_path1)
1602 remove_file(opt, 1, prev_path1,
1603 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1604 if (prev_path2)
1605 remove_file(opt, 1, prev_path2,
1606 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1607
1608 /*
1609 * Remove the collision path, if it wouldn't cause dirty contents
1610 * or an untracked file to get lost. We'll either overwrite with
1611 * merged contents, or just write out to differently named files.
1612 */
1613 if (was_dirty(opt, collide_path)) {
1614 output(opt, 1, _("Refusing to lose dirty file at %s"),
1615 collide_path);
1616 update_path = alt_path = unique_path(opt, collide_path, "merged");
1617 } else if (would_lose_untracked(opt, collide_path)) {
1618 /*
1619 * Only way we get here is if both renames were from
1620 * a directory rename AND user had an untracked file
1621 * at the location where both files end up after the
1622 * two directory renames. See testcase 10d of t6043.
1623 */
1624 output(opt, 1, _("Refusing to lose untracked file at "
1625 "%s, even though it's in the way."),
1626 collide_path);
1627 update_path = alt_path = unique_path(opt, collide_path, "merged");
1628 } else {
1629 /*
1630 * FIXME: It's possible that the two files are identical
1631 * and that the current working copy happens to match, in
1632 * which case we are unnecessarily touching the working
1633 * tree file. It's not a likely enough scenario that I
1634 * want to code up the checks for it and a better fix is
1635 * available if we restructure how unpack_trees() and
1636 * merge-recursive interoperate anyway, so punting for
1637 * now...
1638 */
1639 remove_file(opt, 0, collide_path, 0);
1640 }
1641
1642 /* Store things in diff_filespecs for functions that need it */
1643 null.path = (char *)collide_path;
1644 oidcpy(&null.oid, &null_oid);
1645 null.mode = 0;
1646
1647 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1648 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1649 return -1;
1650 mfi.clean &= !alt_path;
1651 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1652 return -1;
1653 if (!mfi.clean && !opt->priv->call_depth &&
1654 update_stages(opt, collide_path, NULL, a, b))
1655 return -1;
1656 free(alt_path);
1657 /*
1658 * FIXME: If both a & b both started with conflicts (only possible
1659 * if they came from a rename/rename(2to1)), but had IDENTICAL
1660 * contents including those conflicts, then in the next line we claim
1661 * it was clean. If someone cares about this case, we should have the
1662 * caller notify us if we started with conflicts.
1663 */
1664 return mfi.clean;
1665 }
1666
1667 static int handle_rename_add(struct merge_options *opt,
1668 struct rename_conflict_info *ci)
1669 {
1670 /* a was renamed to c, and a separate c was added. */
1671 struct diff_filespec *a = ci->ren1->pair->one;
1672 struct diff_filespec *c = ci->ren1->pair->two;
1673 char *path = c->path;
1674 char *prev_path_desc;
1675 struct merge_file_info mfi;
1676
1677 const char *rename_branch = ci->ren1->branch;
1678 const char *add_branch = (opt->branch1 == rename_branch ?
1679 opt->branch2 : opt->branch1);
1680 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1681
1682 output(opt, 1, _("CONFLICT (rename/add): "
1683 "Rename %s->%s in %s. Added %s in %s"),
1684 a->path, c->path, rename_branch,
1685 c->path, add_branch);
1686
1687 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1688 ci->ren1->src_entry->stages[other_stage].path = a->path;
1689 if (merge_mode_and_contents(opt, a, c,
1690 &ci->ren1->src_entry->stages[other_stage],
1691 prev_path_desc,
1692 opt->branch1, opt->branch2,
1693 1 + opt->priv->call_depth * 2, &mfi))
1694 return -1;
1695 free(prev_path_desc);
1696
1697 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1698 return handle_file_collision(opt,
1699 c->path, a->path, NULL,
1700 rename_branch, add_branch,
1701 &mfi.blob,
1702 &ci->ren1->dst_entry->stages[other_stage]);
1703 }
1704
1705 static char *find_path_for_conflict(struct merge_options *opt,
1706 const char *path,
1707 const char *branch1,
1708 const char *branch2)
1709 {
1710 char *new_path = NULL;
1711 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1712 new_path = unique_path(opt, path, branch1);
1713 output(opt, 1, _("%s is a directory in %s adding "
1714 "as %s instead"),
1715 path, branch2, new_path);
1716 } else if (would_lose_untracked(opt, path)) {
1717 new_path = unique_path(opt, path, branch1);
1718 output(opt, 1, _("Refusing to lose untracked file"
1719 " at %s; adding as %s instead"),
1720 path, new_path);
1721 }
1722
1723 return new_path;
1724 }
1725
1726 static int handle_rename_rename_1to2(struct merge_options *opt,
1727 struct rename_conflict_info *ci)
1728 {
1729 /* One file was renamed in both branches, but to different names. */
1730 struct merge_file_info mfi;
1731 struct diff_filespec *add;
1732 struct diff_filespec *o = ci->ren1->pair->one;
1733 struct diff_filespec *a = ci->ren1->pair->two;
1734 struct diff_filespec *b = ci->ren2->pair->two;
1735 char *path_desc;
1736
1737 output(opt, 1, _("CONFLICT (rename/rename): "
1738 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1739 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1740 o->path, a->path, ci->ren1->branch,
1741 o->path, b->path, ci->ren2->branch,
1742 opt->priv->call_depth ? _(" (left unresolved)") : "");
1743
1744 path_desc = xstrfmt("%s and %s, both renamed from %s",
1745 a->path, b->path, o->path);
1746 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1747 ci->ren1->branch, ci->ren2->branch,
1748 opt->priv->call_depth * 2, &mfi))
1749 return -1;
1750 free(path_desc);
1751
1752 if (opt->priv->call_depth) {
1753 /*
1754 * FIXME: For rename/add-source conflicts (if we could detect
1755 * such), this is wrong. We should instead find a unique
1756 * pathname and then either rename the add-source file to that
1757 * unique path, or use that unique path instead of src here.
1758 */
1759 if (update_file(opt, 0, &mfi.blob, o->path))
1760 return -1;
1761
1762 /*
1763 * Above, we put the merged content at the merge-base's
1764 * path. Now we usually need to delete both a->path and
1765 * b->path. However, the rename on each side of the merge
1766 * could also be involved in a rename/add conflict. In
1767 * such cases, we should keep the added file around,
1768 * resolving the conflict at that path in its favor.
1769 */
1770 add = &ci->ren1->dst_entry->stages[2 ^ 1];
1771 if (is_valid(add)) {
1772 if (update_file(opt, 0, add, a->path))
1773 return -1;
1774 }
1775 else
1776 remove_file_from_index(opt->repo->index, a->path);
1777 add = &ci->ren2->dst_entry->stages[3 ^ 1];
1778 if (is_valid(add)) {
1779 if (update_file(opt, 0, add, b->path))
1780 return -1;
1781 }
1782 else
1783 remove_file_from_index(opt->repo->index, b->path);
1784 } else {
1785 /*
1786 * For each destination path, we need to see if there is a
1787 * rename/add collision. If not, we can write the file out
1788 * to the specified location.
1789 */
1790 add = &ci->ren1->dst_entry->stages[2 ^ 1];
1791 if (is_valid(add)) {
1792 add->path = mfi.blob.path = a->path;
1793 if (handle_file_collision(opt, a->path,
1794 NULL, NULL,
1795 ci->ren1->branch,
1796 ci->ren2->branch,
1797 &mfi.blob, add) < 0)
1798 return -1;
1799 } else {
1800 char *new_path = find_path_for_conflict(opt, a->path,
1801 ci->ren1->branch,
1802 ci->ren2->branch);
1803 if (update_file(opt, 0, &mfi.blob,
1804 new_path ? new_path : a->path))
1805 return -1;
1806 free(new_path);
1807 if (update_stages(opt, a->path, NULL, a, NULL))
1808 return -1;
1809 }
1810
1811 add = &ci->ren2->dst_entry->stages[3 ^ 1];
1812 if (is_valid(add)) {
1813 add->path = mfi.blob.path = b->path;
1814 if (handle_file_collision(opt, b->path,
1815 NULL, NULL,
1816 ci->ren1->branch,
1817 ci->ren2->branch,
1818 add, &mfi.blob) < 0)
1819 return -1;
1820 } else {
1821 char *new_path = find_path_for_conflict(opt, b->path,
1822 ci->ren2->branch,
1823 ci->ren1->branch);
1824 if (update_file(opt, 0, &mfi.blob,
1825 new_path ? new_path : b->path))
1826 return -1;
1827 free(new_path);
1828 if (update_stages(opt, b->path, NULL, NULL, b))
1829 return -1;
1830 }
1831 }
1832
1833 return 0;
1834 }
1835
1836 static int handle_rename_rename_2to1(struct merge_options *opt,
1837 struct rename_conflict_info *ci)
1838 {
1839 /* Two files, a & b, were renamed to the same thing, c. */
1840 struct diff_filespec *a = ci->ren1->pair->one;
1841 struct diff_filespec *b = ci->ren2->pair->one;
1842 struct diff_filespec *c1 = ci->ren1->pair->two;
1843 struct diff_filespec *c2 = ci->ren2->pair->two;
1844 char *path = c1->path; /* == c2->path */
1845 char *path_side_1_desc;
1846 char *path_side_2_desc;
1847 struct merge_file_info mfi_c1;
1848 struct merge_file_info mfi_c2;
1849 int ostage1, ostage2;
1850
1851 output(opt, 1, _("CONFLICT (rename/rename): "
1852 "Rename %s->%s in %s. "
1853 "Rename %s->%s in %s"),
1854 a->path, c1->path, ci->ren1->branch,
1855 b->path, c2->path, ci->ren2->branch);
1856
1857 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1858 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1859 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1860 ostage2 = ostage1 ^ 1;
1861 ci->ren1->src_entry->stages[ostage1].path = a->path;
1862 ci->ren2->src_entry->stages[ostage2].path = b->path;
1863 if (merge_mode_and_contents(opt, a, c1,
1864 &ci->ren1->src_entry->stages[ostage1],
1865 path_side_1_desc,
1866 opt->branch1, opt->branch2,
1867 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1868 merge_mode_and_contents(opt, b,
1869 &ci->ren2->src_entry->stages[ostage2],
1870 c2, path_side_2_desc,
1871 opt->branch1, opt->branch2,
1872 1 + opt->priv->call_depth * 2, &mfi_c2))
1873 return -1;
1874 free(path_side_1_desc);
1875 free(path_side_2_desc);
1876 mfi_c1.blob.path = path;
1877 mfi_c2.blob.path = path;
1878
1879 return handle_file_collision(opt, path, a->path, b->path,
1880 ci->ren1->branch, ci->ren2->branch,
1881 &mfi_c1.blob, &mfi_c2.blob);
1882 }
1883
1884 /*
1885 * Get the diff_filepairs changed between o_tree and tree.
1886 */
1887 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1888 struct tree *o_tree,
1889 struct tree *tree)
1890 {
1891 struct diff_queue_struct *ret;
1892 struct diff_options opts;
1893
1894 repo_diff_setup(opt->repo, &opts);
1895 opts.flags.recursive = 1;
1896 opts.flags.rename_empty = 0;
1897 opts.detect_rename = merge_detect_rename(opt);
1898 /*
1899 * We do not have logic to handle the detection of copies. In
1900 * fact, it may not even make sense to add such logic: would we
1901 * really want a change to a base file to be propagated through
1902 * multiple other files by a merge?
1903 */
1904 if (opts.detect_rename > DIFF_DETECT_RENAME)
1905 opts.detect_rename = DIFF_DETECT_RENAME;
1906 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 1000;
1907 opts.rename_score = opt->rename_score;
1908 opts.show_rename_progress = opt->show_rename_progress;
1909 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1910 diff_setup_done(&opts);
1911 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1912 diffcore_std(&opts);
1913 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1914 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1915
1916 ret = xmalloc(sizeof(*ret));
1917 *ret = diff_queued_diff;
1918
1919 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1920 diff_queued_diff.nr = 0;
1921 diff_queued_diff.queue = NULL;
1922 diff_flush(&opts);
1923 return ret;
1924 }
1925
1926 static int tree_has_path(struct repository *r, struct tree *tree,
1927 const char *path)
1928 {
1929 struct object_id hashy;
1930 unsigned short mode_o;
1931
1932 return !get_tree_entry(r,
1933 &tree->object.oid, path,
1934 &hashy, &mode_o);
1935 }
1936
1937 /*
1938 * Return a new string that replaces the beginning portion (which matches
1939 * entry->dir), with entry->new_dir. In perl-speak:
1940 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1941 * NOTE:
1942 * Caller must ensure that old_path starts with entry->dir + '/'.
1943 */
1944 static char *apply_dir_rename(struct dir_rename_entry *entry,
1945 const char *old_path)
1946 {
1947 struct strbuf new_path = STRBUF_INIT;
1948 int oldlen, newlen;
1949
1950 if (entry->non_unique_new_dir)
1951 return NULL;
1952
1953 oldlen = strlen(entry->dir);
1954 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1955 strbuf_grow(&new_path, newlen);
1956 strbuf_addbuf(&new_path, &entry->new_dir);
1957 strbuf_addstr(&new_path, &old_path[oldlen]);
1958
1959 return strbuf_detach(&new_path, NULL);
1960 }
1961
1962 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1963 char **old_dir, char **new_dir)
1964 {
1965 char *end_of_old, *end_of_new;
1966 int old_len, new_len;
1967
1968 *old_dir = NULL;
1969 *new_dir = NULL;
1970
1971 /*
1972 * For
1973 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1974 * the "e/foo.c" part is the same, we just want to know that
1975 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1976 * so, for this example, this function returns "a/b/c/d" in
1977 * *old_dir and "a/b/some/thing/else" in *new_dir.
1978 *
1979 * Also, if the basename of the file changed, we don't care. We
1980 * want to know which portion of the directory, if any, changed.
1981 */
1982 end_of_old = strrchr(old_path, '/');
1983 end_of_new = strrchr(new_path, '/');
1984
1985 if (end_of_old == NULL || end_of_new == NULL)
1986 return;
1987 while (*--end_of_new == *--end_of_old &&
1988 end_of_old != old_path &&
1989 end_of_new != new_path)
1990 ; /* Do nothing; all in the while loop */
1991 /*
1992 * We've found the first non-matching character in the directory
1993 * paths. That means the current directory we were comparing
1994 * represents the rename. Move end_of_old and end_of_new back
1995 * to the full directory name.
1996 */
1997 if (*end_of_old == '/')
1998 end_of_old++;
1999 if (*end_of_old != '/')
2000 end_of_new++;
2001 end_of_old = strchr(end_of_old, '/');
2002 end_of_new = strchr(end_of_new, '/');
2003
2004 /*
2005 * It may have been the case that old_path and new_path were the same
2006 * directory all along. Don't claim a rename if they're the same.
2007 */
2008 old_len = end_of_old - old_path;
2009 new_len = end_of_new - new_path;
2010
2011 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
2012 *old_dir = xstrndup(old_path, old_len);
2013 *new_dir = xstrndup(new_path, new_len);
2014 }
2015 }
2016
2017 static void remove_hashmap_entries(struct hashmap *dir_renames,
2018 struct string_list *items_to_remove)
2019 {
2020 int i;
2021 struct dir_rename_entry *entry;
2022
2023 for (i = 0; i < items_to_remove->nr; i++) {
2024 entry = items_to_remove->items[i].util;
2025 hashmap_remove(dir_renames, &entry->ent, NULL);
2026 }
2027 string_list_clear(items_to_remove, 0);
2028 }
2029
2030 /*
2031 * See if there is a directory rename for path, and if there are any file
2032 * level conflicts for the renamed location. If there is a rename and
2033 * there are no conflicts, return the new name. Otherwise, return NULL.
2034 */
2035 static char *handle_path_level_conflicts(struct merge_options *opt,
2036 const char *path,
2037 struct dir_rename_entry *entry,
2038 struct hashmap *collisions,
2039 struct tree *tree)
2040 {
2041 char *new_path = NULL;
2042 struct collision_entry *collision_ent;
2043 int clean = 1;
2044 struct strbuf collision_paths = STRBUF_INIT;
2045
2046 /*
2047 * entry has the mapping of old directory name to new directory name
2048 * that we want to apply to path.
2049 */
2050 new_path = apply_dir_rename(entry, path);
2051
2052 if (!new_path) {
2053 /* This should only happen when entry->non_unique_new_dir set */
2054 if (!entry->non_unique_new_dir)
2055 BUG("entry->non_unqiue_dir not set and !new_path");
2056 output(opt, 1, _("CONFLICT (directory rename split): "
2057 "Unclear where to place %s because directory "
2058 "%s was renamed to multiple other directories, "
2059 "with no destination getting a majority of the "
2060 "files."),
2061 path, entry->dir);
2062 clean = 0;
2063 return NULL;
2064 }
2065
2066 /*
2067 * The caller needs to have ensured that it has pre-populated
2068 * collisions with all paths that map to new_path. Do a quick check
2069 * to ensure that's the case.
2070 */
2071 collision_ent = collision_find_entry(collisions, new_path);
2072 if (collision_ent == NULL)
2073 BUG("collision_ent is NULL");
2074
2075 /*
2076 * Check for one-sided add/add/.../add conflicts, i.e.
2077 * where implicit renames from the other side doing
2078 * directory rename(s) can affect this side of history
2079 * to put multiple paths into the same location. Warn
2080 * and bail on directory renames for such paths.
2081 */
2082 if (collision_ent->reported_already) {
2083 clean = 0;
2084 } else if (tree_has_path(opt->repo, tree, new_path)) {
2085 collision_ent->reported_already = 1;
2086 strbuf_add_separated_string_list(&collision_paths, ", ",
2087 &collision_ent->source_files);
2088 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2089 "file/dir at %s in the way of implicit "
2090 "directory rename(s) putting the following "
2091 "path(s) there: %s."),
2092 new_path, collision_paths.buf);
2093 clean = 0;
2094 } else if (collision_ent->source_files.nr > 1) {
2095 collision_ent->reported_already = 1;
2096 strbuf_add_separated_string_list(&collision_paths, ", ",
2097 &collision_ent->source_files);
2098 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2099 "more than one path to %s; implicit directory "
2100 "renames tried to put these paths there: %s"),
2101 new_path, collision_paths.buf);
2102 clean = 0;
2103 }
2104
2105 /* Free memory we no longer need */
2106 strbuf_release(&collision_paths);
2107 if (!clean && new_path) {
2108 free(new_path);
2109 return NULL;
2110 }
2111
2112 return new_path;
2113 }
2114
2115 /*
2116 * There are a couple things we want to do at the directory level:
2117 * 1. Check for both sides renaming to the same thing, in order to avoid
2118 * implicit renaming of files that should be left in place. (See
2119 * testcase 6b in t6043 for details.)
2120 * 2. Prune directory renames if there are still files left in the
2121 * the original directory. These represent a partial directory rename,
2122 * i.e. a rename where only some of the files within the directory
2123 * were renamed elsewhere. (Technically, this could be done earlier
2124 * in get_directory_renames(), except that would prevent us from
2125 * doing the previous check and thus failing testcase 6b.)
2126 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2127 * In the future, we could potentially record this info as well and
2128 * omit reporting rename/rename(1to2) conflicts for each path within
2129 * the affected directories, thus cleaning up the merge output.
2130 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2131 * directory level, because merging directories is fine. If it
2132 * causes conflicts for files within those merged directories, then
2133 * that should be detected at the individual path level.
2134 */
2135 static void handle_directory_level_conflicts(struct merge_options *opt,
2136 struct hashmap *dir_re_head,
2137 struct tree *head,
2138 struct hashmap *dir_re_merge,
2139 struct tree *merge)
2140 {
2141 struct hashmap_iter iter;
2142 struct dir_rename_entry *head_ent;
2143 struct dir_rename_entry *merge_ent;
2144
2145 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2146 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2147
2148 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2149 ent /* member name */) {
2150 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2151 if (merge_ent &&
2152 !head_ent->non_unique_new_dir &&
2153 !merge_ent->non_unique_new_dir &&
2154 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2155 /* 1. Renamed identically; remove it from both sides */
2156 string_list_append(&remove_from_head,
2157 head_ent->dir)->util = head_ent;
2158 strbuf_release(&head_ent->new_dir);
2159 string_list_append(&remove_from_merge,
2160 merge_ent->dir)->util = merge_ent;
2161 strbuf_release(&merge_ent->new_dir);
2162 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2163 /* 2. This wasn't a directory rename after all */
2164 string_list_append(&remove_from_head,
2165 head_ent->dir)->util = head_ent;
2166 strbuf_release(&head_ent->new_dir);
2167 }
2168 }
2169
2170 remove_hashmap_entries(dir_re_head, &remove_from_head);
2171 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2172
2173 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2174 ent /* member name */) {
2175 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2176 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2177 /* 2. This wasn't a directory rename after all */
2178 string_list_append(&remove_from_merge,
2179 merge_ent->dir)->util = merge_ent;
2180 } else if (head_ent &&
2181 !head_ent->non_unique_new_dir &&
2182 !merge_ent->non_unique_new_dir) {
2183 /* 3. rename/rename(1to2) */
2184 /*
2185 * We can assume it's not rename/rename(1to1) because
2186 * that was case (1), already checked above. So we
2187 * know that head_ent->new_dir and merge_ent->new_dir
2188 * are different strings.
2189 */
2190 output(opt, 1, _("CONFLICT (rename/rename): "
2191 "Rename directory %s->%s in %s. "
2192 "Rename directory %s->%s in %s"),
2193 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2194 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2195 string_list_append(&remove_from_head,
2196 head_ent->dir)->util = head_ent;
2197 strbuf_release(&head_ent->new_dir);
2198 string_list_append(&remove_from_merge,
2199 merge_ent->dir)->util = merge_ent;
2200 strbuf_release(&merge_ent->new_dir);
2201 }
2202 }
2203
2204 remove_hashmap_entries(dir_re_head, &remove_from_head);
2205 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2206 }
2207
2208 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2209 {
2210 struct hashmap *dir_renames;
2211 struct hashmap_iter iter;
2212 struct dir_rename_entry *entry;
2213 int i;
2214
2215 /*
2216 * Typically, we think of a directory rename as all files from a
2217 * certain directory being moved to a target directory. However,
2218 * what if someone first moved two files from the original
2219 * directory in one commit, and then renamed the directory
2220 * somewhere else in a later commit? At merge time, we just know
2221 * that files from the original directory went to two different
2222 * places, and that the bulk of them ended up in the same place.
2223 * We want each directory rename to represent where the bulk of the
2224 * files from that directory end up; this function exists to find
2225 * where the bulk of the files went.
2226 *
2227 * The first loop below simply iterates through the list of file
2228 * renames, finding out how often each directory rename pair
2229 * possibility occurs.
2230 */
2231 dir_renames = xmalloc(sizeof(*dir_renames));
2232 dir_rename_init(dir_renames);
2233 for (i = 0; i < pairs->nr; ++i) {
2234 struct string_list_item *item;
2235 int *count;
2236 struct diff_filepair *pair = pairs->queue[i];
2237 char *old_dir, *new_dir;
2238
2239 /* File not part of directory rename if it wasn't renamed */
2240 if (pair->status != 'R')
2241 continue;
2242
2243 get_renamed_dir_portion(pair->one->path, pair->two->path,
2244 &old_dir, &new_dir);
2245 if (!old_dir)
2246 /* Directory didn't change at all; ignore this one. */
2247 continue;
2248
2249 entry = dir_rename_find_entry(dir_renames, old_dir);
2250 if (!entry) {
2251 entry = xmalloc(sizeof(*entry));
2252 dir_rename_entry_init(entry, old_dir);
2253 hashmap_put(dir_renames, &entry->ent);
2254 } else {
2255 free(old_dir);
2256 }
2257 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2258 if (!item) {
2259 item = string_list_insert(&entry->possible_new_dirs,
2260 new_dir);
2261 item->util = xcalloc(1, sizeof(int));
2262 } else {
2263 free(new_dir);
2264 }
2265 count = item->util;
2266 *count += 1;
2267 }
2268
2269 /*
2270 * For each directory with files moved out of it, we find out which
2271 * target directory received the most files so we can declare it to
2272 * be the "winning" target location for the directory rename. This
2273 * winner gets recorded in new_dir. If there is no winner
2274 * (multiple target directories received the same number of files),
2275 * we set non_unique_new_dir. Once we've determined the winner (or
2276 * that there is no winner), we no longer need possible_new_dirs.
2277 */
2278 hashmap_for_each_entry(dir_renames, &iter, entry,
2279 ent /* member name */) {
2280 int max = 0;
2281 int bad_max = 0;
2282 char *best = NULL;
2283
2284 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2285 int *count = entry->possible_new_dirs.items[i].util;
2286
2287 if (*count == max)
2288 bad_max = max;
2289 else if (*count > max) {
2290 max = *count;
2291 best = entry->possible_new_dirs.items[i].string;
2292 }
2293 }
2294 if (bad_max == max)
2295 entry->non_unique_new_dir = 1;
2296 else {
2297 assert(entry->new_dir.len == 0);
2298 strbuf_addstr(&entry->new_dir, best);
2299 }
2300 /*
2301 * The relevant directory sub-portion of the original full
2302 * filepaths were xstrndup'ed before inserting into
2303 * possible_new_dirs, and instead of manually iterating the
2304 * list and free'ing each, just lie and tell
2305 * possible_new_dirs that it did the strdup'ing so that it
2306 * will free them for us.
2307 */
2308 entry->possible_new_dirs.strdup_strings = 1;
2309 string_list_clear(&entry->possible_new_dirs, 1);
2310 }
2311
2312 return dir_renames;
2313 }
2314
2315 static struct dir_rename_entry *check_dir_renamed(const char *path,
2316 struct hashmap *dir_renames)
2317 {
2318 char *temp = xstrdup(path);
2319 char *end;
2320 struct dir_rename_entry *entry = NULL;
2321
2322 while ((end = strrchr(temp, '/'))) {
2323 *end = '\0';
2324 entry = dir_rename_find_entry(dir_renames, temp);
2325 if (entry)
2326 break;
2327 }
2328 free(temp);
2329 return entry;
2330 }
2331
2332 static void compute_collisions(struct hashmap *collisions,
2333 struct hashmap *dir_renames,
2334 struct diff_queue_struct *pairs)
2335 {
2336 int i;
2337
2338 /*
2339 * Multiple files can be mapped to the same path due to directory
2340 * renames done by the other side of history. Since that other
2341 * side of history could have merged multiple directories into one,
2342 * if our side of history added the same file basename to each of
2343 * those directories, then all N of them would get implicitly
2344 * renamed by the directory rename detection into the same path,
2345 * and we'd get an add/add/.../add conflict, and all those adds
2346 * from *this* side of history. This is not representable in the
2347 * index, and users aren't going to easily be able to make sense of
2348 * it. So we need to provide a good warning about what's
2349 * happening, and fall back to no-directory-rename detection
2350 * behavior for those paths.
2351 *
2352 * See testcases 9e and all of section 5 from t6043 for examples.
2353 */
2354 collision_init(collisions);
2355
2356 for (i = 0; i < pairs->nr; ++i) {
2357 struct dir_rename_entry *dir_rename_ent;
2358 struct collision_entry *collision_ent;
2359 char *new_path;
2360 struct diff_filepair *pair = pairs->queue[i];
2361
2362 if (pair->status != 'A' && pair->status != 'R')
2363 continue;
2364 dir_rename_ent = check_dir_renamed(pair->two->path,
2365 dir_renames);
2366 if (!dir_rename_ent)
2367 continue;
2368
2369 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2370 if (!new_path)
2371 /*
2372 * dir_rename_ent->non_unique_new_path is true, which
2373 * means there is no directory rename for us to use,
2374 * which means it won't cause us any additional
2375 * collisions.
2376 */
2377 continue;
2378 collision_ent = collision_find_entry(collisions, new_path);
2379 if (!collision_ent) {
2380 collision_ent = xcalloc(1,
2381 sizeof(struct collision_entry));
2382 hashmap_entry_init(&collision_ent->ent,
2383 strhash(new_path));
2384 hashmap_put(collisions, &collision_ent->ent);
2385 collision_ent->target_file = new_path;
2386 } else {
2387 free(new_path);
2388 }
2389 string_list_insert(&collision_ent->source_files,
2390 pair->two->path);
2391 }
2392 }
2393
2394 static char *check_for_directory_rename(struct merge_options *opt,
2395 const char *path,
2396 struct tree *tree,
2397 struct hashmap *dir_renames,
2398 struct hashmap *dir_rename_exclusions,
2399 struct hashmap *collisions,
2400 int *clean_merge)
2401 {
2402 char *new_path = NULL;
2403 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2404 struct dir_rename_entry *oentry = NULL;
2405
2406 if (!entry)
2407 return new_path;
2408
2409 /*
2410 * This next part is a little weird. We do not want to do an
2411 * implicit rename into a directory we renamed on our side, because
2412 * that will result in a spurious rename/rename(1to2) conflict. An
2413 * example:
2414 * Base commit: dumbdir/afile, otherdir/bfile
2415 * Side 1: smrtdir/afile, otherdir/bfile
2416 * Side 2: dumbdir/afile, dumbdir/bfile
2417 * Here, while working on Side 1, we could notice that otherdir was
2418 * renamed/merged to dumbdir, and change the diff_filepair for
2419 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2420 * 2 will notice the rename from dumbdir to smrtdir, and do the
2421 * transitive rename to move it from dumbdir/bfile to
2422 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2423 * smrtdir, a rename/rename(1to2) conflict. We really just want
2424 * the file to end up in smrtdir. And the way to achieve that is
2425 * to not let Side1 do the rename to dumbdir, since we know that is
2426 * the source of one of our directory renames.
2427 *
2428 * That's why oentry and dir_rename_exclusions is here.
2429 *
2430 * As it turns out, this also prevents N-way transient rename
2431 * confusion; See testcases 9c and 9d of t6043.
2432 */
2433 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2434 if (oentry) {
2435 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2436 "to %s, because %s itself was renamed."),
2437 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2438 } else {
2439 new_path = handle_path_level_conflicts(opt, path, entry,
2440 collisions, tree);
2441 *clean_merge &= (new_path != NULL);
2442 }
2443
2444 return new_path;
2445 }
2446
2447 static void apply_directory_rename_modifications(struct merge_options *opt,
2448 struct diff_filepair *pair,
2449 char *new_path,
2450 struct rename *re,
2451 struct tree *tree,
2452 struct tree *o_tree,
2453 struct tree *a_tree,
2454 struct tree *b_tree,
2455 struct string_list *entries)
2456 {
2457 struct string_list_item *item;
2458 int stage = (tree == a_tree ? 2 : 3);
2459 int update_wd;
2460
2461 /*
2462 * In all cases where we can do directory rename detection,
2463 * unpack_trees() will have read pair->two->path into the
2464 * index and the working copy. We need to remove it so that
2465 * we can instead place it at new_path. It is guaranteed to
2466 * not be untracked (unpack_trees() would have errored out
2467 * saying the file would have been overwritten), but it might
2468 * be dirty, though.
2469 */
2470 update_wd = !was_dirty(opt, pair->two->path);
2471 if (!update_wd)
2472 output(opt, 1, _("Refusing to lose dirty file at %s"),
2473 pair->two->path);
2474 remove_file(opt, 1, pair->two->path, !update_wd);
2475
2476 /* Find or create a new re->dst_entry */
2477 item = string_list_lookup(entries, new_path);
2478 if (item) {
2479 /*
2480 * Since we're renaming on this side of history, and it's
2481 * due to a directory rename on the other side of history
2482 * (which we only allow when the directory in question no
2483 * longer exists on the other side of history), the
2484 * original entry for re->dst_entry is no longer
2485 * necessary...
2486 */
2487 re->dst_entry->processed = 1;
2488
2489 /*
2490 * ...because we'll be using this new one.
2491 */
2492 re->dst_entry = item->util;
2493 } else {
2494 /*
2495 * re->dst_entry is for the before-dir-rename path, and we
2496 * need it to hold information for the after-dir-rename
2497 * path. Before creating a new entry, we need to mark the
2498 * old one as unnecessary (...unless it is shared by
2499 * src_entry, i.e. this didn't use to be a rename, in which
2500 * case we can just allow the normal processing to happen
2501 * for it).
2502 */
2503 if (pair->status == 'R')
2504 re->dst_entry->processed = 1;
2505
2506 re->dst_entry = insert_stage_data(opt->repo, new_path,
2507 o_tree, a_tree, b_tree,
2508 entries);
2509 item = string_list_insert(entries, new_path);
2510 item->util = re->dst_entry;
2511 }
2512
2513 /*
2514 * Update the stage_data with the information about the path we are
2515 * moving into place. That slot will be empty and available for us
2516 * to write to because of the collision checks in
2517 * handle_path_level_conflicts(). In other words,
2518 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2519 * open for us to write to.
2520 *
2521 * It may be tempting to actually update the index at this point as
2522 * well, using update_stages_for_stage_data(), but as per the big
2523 * "NOTE" in update_stages(), doing so will modify the current
2524 * in-memory index which will break calls to would_lose_untracked()
2525 * that we need to make. Instead, we need to just make sure that
2526 * the various handle_rename_*() functions update the index
2527 * explicitly rather than relying on unpack_trees() to have done it.
2528 */
2529 get_tree_entry(opt->repo,
2530 &tree->object.oid,
2531 pair->two->path,
2532 &re->dst_entry->stages[stage].oid,
2533 &re->dst_entry->stages[stage].mode);
2534
2535 /*
2536 * Record the original change status (or 'type' of change). If it
2537 * was originally an add ('A'), this lets us differentiate later
2538 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2539 * otherwise look the same). If it was originally a rename ('R'),
2540 * this lets us remember and report accurately about the transitive
2541 * renaming that occurred via the directory rename detection. Also,
2542 * record the original destination name.
2543 */
2544 re->dir_rename_original_type = pair->status;
2545 re->dir_rename_original_dest = pair->two->path;
2546
2547 /*
2548 * We don't actually look at pair->status again, but it seems
2549 * pedagogically correct to adjust it.
2550 */
2551 pair->status = 'R';
2552
2553 /*
2554 * Finally, record the new location.
2555 */
2556 pair->two->path = new_path;
2557 }
2558
2559 /*
2560 * Get information of all renames which occurred in 'pairs', making use of
2561 * any implicit directory renames inferred from the other side of history.
2562 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2563 * to be able to associate the correct cache entries with the rename
2564 * information; tree is always equal to either a_tree or b_tree.
2565 */
2566 static struct string_list *get_renames(struct merge_options *opt,
2567 const char *branch,
2568 struct diff_queue_struct *pairs,
2569 struct hashmap *dir_renames,
2570 struct hashmap *dir_rename_exclusions,
2571 struct tree *tree,
2572 struct tree *o_tree,
2573 struct tree *a_tree,
2574 struct tree *b_tree,
2575 struct string_list *entries,
2576 int *clean_merge)
2577 {
2578 int i;
2579 struct hashmap collisions;
2580 struct hashmap_iter iter;
2581 struct collision_entry *e;
2582 struct string_list *renames;
2583
2584 compute_collisions(&collisions, dir_renames, pairs);
2585 renames = xcalloc(1, sizeof(struct string_list));
2586
2587 for (i = 0; i < pairs->nr; ++i) {
2588 struct string_list_item *item;
2589 struct rename *re;
2590 struct diff_filepair *pair = pairs->queue[i];
2591 char *new_path; /* non-NULL only with directory renames */
2592
2593 if (pair->status != 'A' && pair->status != 'R') {
2594 diff_free_filepair(pair);
2595 continue;
2596 }
2597 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2598 dir_renames,
2599 dir_rename_exclusions,
2600 &collisions,
2601 clean_merge);
2602 if (pair->status != 'R' && !new_path) {
2603 diff_free_filepair(pair);
2604 continue;
2605 }
2606
2607 re = xmalloc(sizeof(*re));
2608 re->processed = 0;
2609 re->pair = pair;
2610 re->branch = branch;
2611 re->dir_rename_original_type = '\0';
2612 re->dir_rename_original_dest = NULL;
2613 item = string_list_lookup(entries, re->pair->one->path);
2614 if (!item)
2615 re->src_entry = insert_stage_data(opt->repo,
2616 re->pair->one->path,
2617 o_tree, a_tree, b_tree, entries);
2618 else
2619 re->src_entry = item->util;
2620
2621 item = string_list_lookup(entries, re->pair->two->path);
2622 if (!item)
2623 re->dst_entry = insert_stage_data(opt->repo,
2624 re->pair->two->path,
2625 o_tree, a_tree, b_tree, entries);
2626 else
2627 re->dst_entry = item->util;
2628 item = string_list_insert(renames, pair->one->path);
2629 item->util = re;
2630 if (new_path)
2631 apply_directory_rename_modifications(opt, pair, new_path,
2632 re, tree, o_tree,
2633 a_tree, b_tree,
2634 entries);
2635 }
2636
2637 hashmap_for_each_entry(&collisions, &iter, e,
2638 ent /* member name */) {
2639 free(e->target_file);
2640 string_list_clear(&e->source_files, 0);
2641 }
2642 hashmap_free_entries(&collisions, struct collision_entry, ent);
2643 return renames;
2644 }
2645
2646 static int process_renames(struct merge_options *opt,
2647 struct string_list *a_renames,
2648 struct string_list *b_renames)
2649 {
2650 int clean_merge = 1, i, j;
2651 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2652 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2653 const struct rename *sre;
2654
2655 for (i = 0; i < a_renames->nr; i++) {
2656 sre = a_renames->items[i].util;
2657 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2658 = (void *)sre;
2659 }
2660 for (i = 0; i < b_renames->nr; i++) {
2661 sre = b_renames->items[i].util;
2662 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2663 = (void *)sre;
2664 }
2665
2666 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2667 struct string_list *renames1, *renames2Dst;
2668 struct rename *ren1 = NULL, *ren2 = NULL;
2669 const char *ren1_src, *ren1_dst;
2670 struct string_list_item *lookup;
2671
2672 if (i >= a_renames->nr) {
2673 ren2 = b_renames->items[j++].util;
2674 } else if (j >= b_renames->nr) {
2675 ren1 = a_renames->items[i++].util;
2676 } else {
2677 int compare = strcmp(a_renames->items[i].string,
2678 b_renames->items[j].string);
2679 if (compare <= 0)
2680 ren1 = a_renames->items[i++].util;
2681 if (compare >= 0)
2682 ren2 = b_renames->items[j++].util;
2683 }
2684
2685 /* TODO: refactor, so that 1/2 are not needed */
2686 if (ren1) {
2687 renames1 = a_renames;
2688 renames2Dst = &b_by_dst;
2689 } else {
2690 renames1 = b_renames;
2691 renames2Dst = &a_by_dst;
2692 SWAP(ren2, ren1);
2693 }
2694
2695 if (ren1->processed)
2696 continue;
2697 ren1->processed = 1;
2698 ren1->dst_entry->processed = 1;
2699 /* BUG: We should only mark src_entry as processed if we
2700 * are not dealing with a rename + add-source case.
2701 */
2702 ren1->src_entry->processed = 1;
2703
2704 ren1_src = ren1->pair->one->path;
2705 ren1_dst = ren1->pair->two->path;
2706
2707 if (ren2) {
2708 /* One file renamed on both sides */
2709 const char *ren2_src = ren2->pair->one->path;
2710 const char *ren2_dst = ren2->pair->two->path;
2711 enum rename_type rename_type;
2712 if (strcmp(ren1_src, ren2_src) != 0)
2713 BUG("ren1_src != ren2_src");
2714 ren2->dst_entry->processed = 1;
2715 ren2->processed = 1;
2716 if (strcmp(ren1_dst, ren2_dst) != 0) {
2717 rename_type = RENAME_ONE_FILE_TO_TWO;
2718 clean_merge = 0;
2719 } else {
2720 rename_type = RENAME_ONE_FILE_TO_ONE;
2721 /* BUG: We should only remove ren1_src in
2722 * the base stage (think of rename +
2723 * add-source cases).
2724 */
2725 remove_file(opt, 1, ren1_src, 1);
2726 update_entry(ren1->dst_entry,
2727 ren1->pair->one,
2728 ren1->pair->two,
2729 ren2->pair->two);
2730 }
2731 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2732 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2733 /* Two different files renamed to the same thing */
2734 char *ren2_dst;
2735 ren2 = lookup->util;
2736 ren2_dst = ren2->pair->two->path;
2737 if (strcmp(ren1_dst, ren2_dst) != 0)
2738 BUG("ren1_dst != ren2_dst");
2739
2740 clean_merge = 0;
2741 ren2->processed = 1;
2742 /*
2743 * BUG: We should only mark src_entry as processed
2744 * if we are not dealing with a rename + add-source
2745 * case.
2746 */
2747 ren2->src_entry->processed = 1;
2748
2749 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2750 opt, ren1, ren2);
2751 } else {
2752 /* Renamed in 1, maybe changed in 2 */
2753 /* we only use sha1 and mode of these */
2754 struct diff_filespec src_other, dst_other;
2755 int try_merge;
2756
2757 /*
2758 * unpack_trees loads entries from common-commit
2759 * into stage 1, from head-commit into stage 2, and
2760 * from merge-commit into stage 3. We keep track
2761 * of which side corresponds to the rename.
2762 */
2763 int renamed_stage = a_renames == renames1 ? 2 : 3;
2764 int other_stage = a_renames == renames1 ? 3 : 2;
2765
2766 /* BUG: We should only remove ren1_src in the base
2767 * stage and in other_stage (think of rename +
2768 * add-source case).
2769 */
2770 remove_file(opt, 1, ren1_src,
2771 renamed_stage == 2 || !was_tracked(opt, ren1_src));
2772
2773 oidcpy(&src_other.oid,
2774 &ren1->src_entry->stages[other_stage].oid);
2775 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2776 oidcpy(&dst_other.oid,
2777 &ren1->dst_entry->stages[other_stage].oid);
2778 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2779 try_merge = 0;
2780
2781 if (oid_eq(&src_other.oid, &null_oid) &&
2782 ren1->dir_rename_original_type == 'A') {
2783 setup_rename_conflict_info(RENAME_VIA_DIR,
2784 opt, ren1, NULL);
2785 } else if (oid_eq(&src_other.oid, &null_oid)) {
2786 setup_rename_conflict_info(RENAME_DELETE,
2787 opt, ren1, NULL);
2788 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2789 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2790 /*
2791 * Added file on the other side identical to
2792 * the file being renamed: clean merge.
2793 * Also, there is no need to overwrite the
2794 * file already in the working copy, so call
2795 * update_file_flags() instead of
2796 * update_file().
2797 */
2798 if (update_file_flags(opt,
2799 ren1->pair->two,
2800 ren1_dst,
2801 1, /* update_cache */
2802 0 /* update_wd */))
2803 clean_merge = -1;
2804 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2805 /*
2806 * Probably not a clean merge, but it's
2807 * premature to set clean_merge to 0 here,
2808 * because if the rename merges cleanly and
2809 * the merge exactly matches the newly added
2810 * file, then the merge will be clean.
2811 */
2812 setup_rename_conflict_info(RENAME_ADD,
2813 opt, ren1, NULL);
2814 } else
2815 try_merge = 1;
2816
2817 if (clean_merge < 0)
2818 goto cleanup_and_return;
2819 if (try_merge) {
2820 struct diff_filespec *o, *a, *b;
2821 src_other.path = (char *)ren1_src;
2822
2823 o = ren1->pair->one;
2824 if (a_renames == renames1) {
2825 a = ren1->pair->two;
2826 b = &src_other;
2827 } else {
2828 b = ren1->pair->two;
2829 a = &src_other;
2830 }
2831 update_entry(ren1->dst_entry, o, a, b);
2832 setup_rename_conflict_info(RENAME_NORMAL,
2833 opt, ren1, NULL);
2834 }
2835 }
2836 }
2837 cleanup_and_return:
2838 string_list_clear(&a_by_dst, 0);
2839 string_list_clear(&b_by_dst, 0);
2840
2841 return clean_merge;
2842 }
2843
2844 struct rename_info {
2845 struct string_list *head_renames;
2846 struct string_list *merge_renames;
2847 };
2848
2849 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2850 struct hashmap *dir_renames)
2851 {
2852 struct hashmap_iter iter;
2853 struct dir_rename_entry *e;
2854
2855 hashmap_for_each_entry(dir_renames, &iter, e,
2856 ent /* member name */) {
2857 free(e->dir);
2858 strbuf_release(&e->new_dir);
2859 /* possible_new_dirs already cleared in get_directory_renames */
2860 }
2861 hashmap_free_entries(dir_renames, struct dir_rename_entry, ent);
2862 free(dir_renames);
2863
2864 free(pairs->queue);
2865 free(pairs);
2866 }
2867
2868 static int detect_and_process_renames(struct merge_options *opt,
2869 struct tree *common,
2870 struct tree *head,
2871 struct tree *merge,
2872 struct string_list *entries,
2873 struct rename_info *ri)
2874 {
2875 struct diff_queue_struct *head_pairs, *merge_pairs;
2876 struct hashmap *dir_re_head, *dir_re_merge;
2877 int clean = 1;
2878
2879 ri->head_renames = NULL;
2880 ri->merge_renames = NULL;
2881
2882 if (!merge_detect_rename(opt))
2883 return 1;
2884
2885 head_pairs = get_diffpairs(opt, common, head);
2886 merge_pairs = get_diffpairs(opt, common, merge);
2887
2888 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2889 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2890 !opt->priv->call_depth)) {
2891 dir_re_head = get_directory_renames(head_pairs);
2892 dir_re_merge = get_directory_renames(merge_pairs);
2893
2894 handle_directory_level_conflicts(opt,
2895 dir_re_head, head,
2896 dir_re_merge, merge);
2897 } else {
2898 dir_re_head = xmalloc(sizeof(*dir_re_head));
2899 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2900 dir_rename_init(dir_re_head);
2901 dir_rename_init(dir_re_merge);
2902 }
2903
2904 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2905 dir_re_merge, dir_re_head, head,
2906 common, head, merge, entries,
2907 &clean);
2908 if (clean < 0)
2909 goto cleanup;
2910 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2911 dir_re_head, dir_re_merge, merge,
2912 common, head, merge, entries,
2913 &clean);
2914 if (clean < 0)
2915 goto cleanup;
2916 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2917
2918 cleanup:
2919 /*
2920 * Some cleanup is deferred until cleanup_renames() because the
2921 * data structures are still needed and referenced in
2922 * process_entry(). But there are a few things we can free now.
2923 */
2924 initial_cleanup_rename(head_pairs, dir_re_head);
2925 initial_cleanup_rename(merge_pairs, dir_re_merge);
2926
2927 return clean;
2928 }
2929
2930 static void final_cleanup_rename(struct string_list *rename)
2931 {
2932 const struct rename *re;
2933 int i;
2934
2935 if (rename == NULL)
2936 return;
2937
2938 for (i = 0; i < rename->nr; i++) {
2939 re = rename->items[i].util;
2940 diff_free_filepair(re->pair);
2941 }
2942 string_list_clear(rename, 1);
2943 free(rename);
2944 }
2945
2946 static void final_cleanup_renames(struct rename_info *re_info)
2947 {
2948 final_cleanup_rename(re_info->head_renames);
2949 final_cleanup_rename(re_info->merge_renames);
2950 }
2951
2952 static int read_oid_strbuf(struct merge_options *opt,
2953 const struct object_id *oid,
2954 struct strbuf *dst)
2955 {
2956 void *buf;
2957 enum object_type type;
2958 unsigned long size;
2959 buf = read_object_file(oid, &type, &size);
2960 if (!buf)
2961 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
2962 if (type != OBJ_BLOB) {
2963 free(buf);
2964 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
2965 }
2966 strbuf_attach(dst, buf, size, size + 1);
2967 return 0;
2968 }
2969
2970 static int blob_unchanged(struct merge_options *opt,
2971 const struct diff_filespec *o,
2972 const struct diff_filespec *a,
2973 int renormalize, const char *path)
2974 {
2975 struct strbuf obuf = STRBUF_INIT;
2976 struct strbuf abuf = STRBUF_INIT;
2977 int ret = 0; /* assume changed for safety */
2978 const struct index_state *idx = opt->repo->index;
2979
2980 if (a->mode != o->mode)
2981 return 0;
2982 if (oid_eq(&o->oid, &a->oid))
2983 return 1;
2984 if (!renormalize)
2985 return 0;
2986
2987 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
2988 read_oid_strbuf(opt, &a->oid, &abuf))
2989 goto error_return;
2990 /*
2991 * Note: binary | is used so that both renormalizations are
2992 * performed. Comparison can be skipped if both files are
2993 * unchanged since their sha1s have already been compared.
2994 */
2995 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
2996 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
2997 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
2998
2999 error_return:
3000 strbuf_release(&obuf);
3001 strbuf_release(&abuf);
3002 return ret;
3003 }
3004
3005 static int handle_modify_delete(struct merge_options *opt,
3006 const char *path,
3007 const struct diff_filespec *o,
3008 const struct diff_filespec *a,
3009 const struct diff_filespec *b)
3010 {
3011 const char *modify_branch, *delete_branch;
3012 const struct diff_filespec *changed;
3013
3014 if (is_valid(a)) {
3015 modify_branch = opt->branch1;
3016 delete_branch = opt->branch2;
3017 changed = a;
3018 } else {
3019 modify_branch = opt->branch2;
3020 delete_branch = opt->branch1;
3021 changed = b;
3022 }
3023
3024 return handle_change_delete(opt,
3025 path, NULL,
3026 o, changed,
3027 modify_branch, delete_branch,
3028 _("modify"), _("modified"));
3029 }
3030
3031 static int handle_content_merge(struct merge_file_info *mfi,
3032 struct merge_options *opt,
3033 const char *path,
3034 int is_dirty,
3035 const struct diff_filespec *o,
3036 const struct diff_filespec *a,
3037 const struct diff_filespec *b,
3038 struct rename_conflict_info *ci)
3039 {
3040 const char *reason = _("content");
3041 unsigned df_conflict_remains = 0;
3042
3043 if (!is_valid(o))
3044 reason = _("add/add");
3045
3046 assert(o->path && a->path && b->path);
3047 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3048 S_ISGITLINK(ci->ren1->pair->two->mode)))
3049 df_conflict_remains = 1;
3050
3051 if (merge_mode_and_contents(opt, o, a, b, path,
3052 opt->branch1, opt->branch2,
3053 opt->priv->call_depth * 2, mfi))
3054 return -1;
3055
3056 /*
3057 * We can skip updating the working tree file iff:
3058 * a) The merge is clean
3059 * b) The merge matches what was in HEAD (content, mode, pathname)
3060 * c) The target path is usable (i.e. not involved in D/F conflict)
3061 */
3062 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3063 !df_conflict_remains) {
3064 int pos;
3065 struct cache_entry *ce;
3066
3067 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3068 if (add_cacheinfo(opt, &mfi->blob, path,
3069 0, (!opt->priv->call_depth && !is_dirty), 0))
3070 return -1;
3071 /*
3072 * However, add_cacheinfo() will delete the old cache entry
3073 * and add a new one. We need to copy over any skip_worktree
3074 * flag to avoid making the file appear as if it were
3075 * deleted by the user.
3076 */
3077 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3078 ce = opt->priv->orig_index.cache[pos];
3079 if (ce_skip_worktree(ce)) {
3080 pos = index_name_pos(opt->repo->index, path, strlen(path));
3081 ce = opt->repo->index->cache[pos];
3082 ce->ce_flags |= CE_SKIP_WORKTREE;
3083 }
3084 return mfi->clean;
3085 }
3086
3087 if (!mfi->clean) {
3088 if (S_ISGITLINK(mfi->blob.mode))
3089 reason = _("submodule");
3090 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3091 reason, path);
3092 if (ci && !df_conflict_remains)
3093 if (update_stages(opt, path, o, a, b))
3094 return -1;
3095 }
3096
3097 if (df_conflict_remains || is_dirty) {
3098 char *new_path;
3099 if (opt->priv->call_depth) {
3100 remove_file_from_index(opt->repo->index, path);
3101 } else {
3102 if (!mfi->clean) {
3103 if (update_stages(opt, path, o, a, b))
3104 return -1;
3105 } else {
3106 int file_from_stage2 = was_tracked(opt, path);
3107
3108 if (update_stages(opt, path, NULL,
3109 file_from_stage2 ? &mfi->blob : NULL,
3110 file_from_stage2 ? NULL : &mfi->blob))
3111 return -1;
3112 }
3113
3114 }
3115 new_path = unique_path(opt, path, ci->ren1->branch);
3116 if (is_dirty) {
3117 output(opt, 1, _("Refusing to lose dirty file at %s"),
3118 path);
3119 }
3120 output(opt, 1, _("Adding as %s instead"), new_path);
3121 if (update_file(opt, 0, &mfi->blob, new_path)) {
3122 free(new_path);
3123 return -1;
3124 }
3125 free(new_path);
3126 mfi->clean = 0;
3127 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3128 return -1;
3129 return !is_dirty && mfi->clean;
3130 }
3131
3132 static int handle_rename_normal(struct merge_options *opt,
3133 const char *path,
3134 const struct diff_filespec *o,
3135 const struct diff_filespec *a,
3136 const struct diff_filespec *b,
3137 struct rename_conflict_info *ci)
3138 {
3139 struct rename *ren = ci->ren1;
3140 struct merge_file_info mfi;
3141 int clean;
3142 int side = (ren->branch == opt->branch1 ? 2 : 3);
3143
3144 /* Merge the content and write it out */
3145 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3146 o, a, b, ci);
3147
3148 if (clean &&
3149 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3150 ren->dir_rename_original_dest) {
3151 if (update_stages(opt, path,
3152 NULL,
3153 side == 2 ? &mfi.blob : NULL,
3154 side == 2 ? NULL : &mfi.blob))
3155 return -1;
3156 clean = 0; /* not clean, but conflicted */
3157 }
3158 return clean;
3159 }
3160
3161 static void dir_rename_warning(const char *msg,
3162 int is_add,
3163 int clean,
3164 struct merge_options *opt,
3165 struct rename *ren)
3166 {
3167 const char *other_branch;
3168 other_branch = (ren->branch == opt->branch1 ?
3169 opt->branch2 : opt->branch1);
3170 if (is_add) {
3171 output(opt, clean ? 2 : 1, msg,
3172 ren->pair->one->path, ren->branch,
3173 other_branch, ren->pair->two->path);
3174 return;
3175 }
3176 output(opt, clean ? 2 : 1, msg,
3177 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3178 other_branch, ren->pair->two->path);
3179 }
3180 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3181 struct rename *ren)
3182 {
3183 const char *msg;
3184 int clean = 1, is_add;
3185
3186 if (!ren)
3187 return clean;
3188
3189 /* Return early if ren was not affected/created by a directory rename */
3190 if (!ren->dir_rename_original_dest)
3191 return clean;
3192
3193 /* Sanity checks */
3194 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3195 assert(ren->dir_rename_original_type == 'A' ||
3196 ren->dir_rename_original_type == 'R');
3197
3198 /* Check whether to treat directory renames as a conflict */
3199 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3200
3201 is_add = (ren->dir_rename_original_type == 'A');
3202 if (ren->dir_rename_original_type == 'A' && clean) {
3203 msg = _("Path updated: %s added in %s inside a "
3204 "directory that was renamed in %s; moving it to %s.");
3205 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3206 msg = _("CONFLICT (file location): %s added in %s "
3207 "inside a directory that was renamed in %s, "
3208 "suggesting it should perhaps be moved to %s.");
3209 } else if (ren->dir_rename_original_type == 'R' && clean) {
3210 msg = _("Path updated: %s renamed to %s in %s, inside a "
3211 "directory that was renamed in %s; moving it to %s.");
3212 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3213 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3214 "inside a directory that was renamed in %s, "
3215 "suggesting it should perhaps be moved to %s.");
3216 } else {
3217 BUG("Impossible dir_rename_original_type/clean combination");
3218 }
3219 dir_rename_warning(msg, is_add, clean, opt, ren);
3220
3221 return clean;
3222 }
3223
3224 /* Per entry merge function */
3225 static int process_entry(struct merge_options *opt,
3226 const char *path, struct stage_data *entry)
3227 {
3228 int clean_merge = 1;
3229 int normalize = opt->renormalize;
3230
3231 struct diff_filespec *o = &entry->stages[1];
3232 struct diff_filespec *a = &entry->stages[2];
3233 struct diff_filespec *b = &entry->stages[3];
3234 int o_valid = is_valid(o);
3235 int a_valid = is_valid(a);
3236 int b_valid = is_valid(b);
3237 o->path = a->path = b->path = (char*)path;
3238
3239 entry->processed = 1;
3240 if (entry->rename_conflict_info) {
3241 struct rename_conflict_info *ci = entry->rename_conflict_info;
3242 struct diff_filespec *temp;
3243 int path_clean;
3244
3245 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3246 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3247
3248 /*
3249 * For cases with a single rename, {o,a,b}->path have all been
3250 * set to the rename target path; we need to set two of these
3251 * back to the rename source.
3252 * For rename/rename conflicts, we'll manually fix paths below.
3253 */
3254 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3255 o->path = temp->path = ci->ren1->pair->one->path;
3256 if (ci->ren2) {
3257 assert(opt->branch1 == ci->ren1->branch);
3258 }
3259
3260 switch (ci->rename_type) {
3261 case RENAME_NORMAL:
3262 case RENAME_ONE_FILE_TO_ONE:
3263 clean_merge = handle_rename_normal(opt, path, o, a, b,
3264 ci);
3265 break;
3266 case RENAME_VIA_DIR:
3267 clean_merge = handle_rename_via_dir(opt, ci);
3268 break;
3269 case RENAME_ADD:
3270 /*
3271 * Probably unclean merge, but if the renamed file
3272 * merges cleanly and the result can then be
3273 * two-way merged cleanly with the added file, I
3274 * guess it's a clean merge?
3275 */
3276 clean_merge = handle_rename_add(opt, ci);
3277 break;
3278 case RENAME_DELETE:
3279 clean_merge = 0;
3280 if (handle_rename_delete(opt, ci))
3281 clean_merge = -1;
3282 break;
3283 case RENAME_ONE_FILE_TO_TWO:
3284 /*
3285 * Manually fix up paths; note:
3286 * ren[12]->pair->one->path are equal.
3287 */
3288 o->path = ci->ren1->pair->one->path;
3289 a->path = ci->ren1->pair->two->path;
3290 b->path = ci->ren2->pair->two->path;
3291
3292 clean_merge = 0;
3293 if (handle_rename_rename_1to2(opt, ci))
3294 clean_merge = -1;
3295 break;
3296 case RENAME_TWO_FILES_TO_ONE:
3297 /*
3298 * Manually fix up paths; note,
3299 * ren[12]->pair->two->path are actually equal.
3300 */
3301 o->path = NULL;
3302 a->path = ci->ren1->pair->two->path;
3303 b->path = ci->ren2->pair->two->path;
3304
3305 /*
3306 * Probably unclean merge, but if the two renamed
3307 * files merge cleanly and the two resulting files
3308 * can then be two-way merged cleanly, I guess it's
3309 * a clean merge?
3310 */
3311 clean_merge = handle_rename_rename_2to1(opt, ci);
3312 break;
3313 default:
3314 entry->processed = 0;
3315 break;
3316 }
3317 if (path_clean < clean_merge)
3318 clean_merge = path_clean;
3319 } else if (o_valid && (!a_valid || !b_valid)) {
3320 /* Case A: Deleted in one */
3321 if ((!a_valid && !b_valid) ||
3322 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3323 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3324 /* Deleted in both or deleted in one and
3325 * unchanged in the other */
3326 if (a_valid)
3327 output(opt, 2, _("Removing %s"), path);
3328 /* do not touch working file if it did not exist */
3329 remove_file(opt, 1, path, !a_valid);
3330 } else {
3331 /* Modify/delete; deleted side may have put a directory in the way */
3332 clean_merge = 0;
3333 if (handle_modify_delete(opt, path, o, a, b))
3334 clean_merge = -1;
3335 }
3336 } else if ((!o_valid && a_valid && !b_valid) ||
3337 (!o_valid && !a_valid && b_valid)) {
3338 /* Case B: Added in one. */
3339 /* [nothing|directory] -> ([nothing|directory], file) */
3340
3341 const char *add_branch;
3342 const char *other_branch;
3343 const char *conf;
3344 const struct diff_filespec *contents;
3345
3346 if (a_valid) {
3347 add_branch = opt->branch1;
3348 other_branch = opt->branch2;
3349 contents = a;
3350 conf = _("file/directory");
3351 } else {
3352 add_branch = opt->branch2;
3353 other_branch = opt->branch1;
3354 contents = b;
3355 conf = _("directory/file");
3356 }
3357 if (dir_in_way(opt->repo->index, path,
3358 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3359 0)) {
3360 char *new_path = unique_path(opt, path, add_branch);
3361 clean_merge = 0;
3362 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3363 "Adding %s as %s"),
3364 conf, path, other_branch, path, new_path);
3365 if (update_file(opt, 0, contents, new_path))
3366 clean_merge = -1;
3367 else if (opt->priv->call_depth)
3368 remove_file_from_index(opt->repo->index, path);
3369 free(new_path);
3370 } else {
3371 output(opt, 2, _("Adding %s"), path);
3372 /* do not overwrite file if already present */
3373 if (update_file_flags(opt, contents, path, 1, !a_valid))
3374 clean_merge = -1;
3375 }
3376 } else if (a_valid && b_valid) {
3377 if (!o_valid) {
3378 /* Case C: Added in both (check for same permissions) */
3379 output(opt, 1,
3380 _("CONFLICT (add/add): Merge conflict in %s"),
3381 path);
3382 clean_merge = handle_file_collision(opt,
3383 path, NULL, NULL,
3384 opt->branch1,
3385 opt->branch2,
3386 a, b);
3387 } else {
3388 /* case D: Modified in both, but differently. */
3389 struct merge_file_info mfi;
3390 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3391 clean_merge = handle_content_merge(&mfi, opt, path,
3392 is_dirty,
3393 o, a, b, NULL);
3394 }
3395 } else if (!o_valid && !a_valid && !b_valid) {
3396 /*
3397 * this entry was deleted altogether. a_mode == 0 means
3398 * we had that path and want to actively remove it.
3399 */
3400 remove_file(opt, 1, path, !a->mode);
3401 } else
3402 BUG("fatal merge failure, shouldn't happen.");
3403
3404 return clean_merge;
3405 }
3406
3407 static int merge_trees_internal(struct merge_options *opt,
3408 struct tree *head,
3409 struct tree *merge,
3410 struct tree *merge_base,
3411 struct tree **result)
3412 {
3413 struct index_state *istate = opt->repo->index;
3414 int code, clean;
3415
3416 if (opt->subtree_shift) {
3417 merge = shift_tree_object(opt->repo, head, merge,
3418 opt->subtree_shift);
3419 merge_base = shift_tree_object(opt->repo, head, merge_base,
3420 opt->subtree_shift);
3421 }
3422
3423 if (oid_eq(&merge_base->object.oid, &merge->object.oid)) {
3424 output(opt, 0, _("Already up to date!"));
3425 *result = head;
3426 return 1;
3427 }
3428
3429 code = unpack_trees_start(opt, merge_base, head, merge);
3430
3431 if (code != 0) {
3432 if (show(opt, 4) || opt->priv->call_depth)
3433 err(opt, _("merging of trees %s and %s failed"),
3434 oid_to_hex(&head->object.oid),
3435 oid_to_hex(&merge->object.oid));
3436 unpack_trees_finish(opt);
3437 return -1;
3438 }
3439
3440 if (unmerged_index(istate)) {
3441 struct string_list *entries;
3442 struct rename_info re_info;
3443 int i;
3444 /*
3445 * Only need the hashmap while processing entries, so
3446 * initialize it here and free it when we are done running
3447 * through the entries. Keeping it in the merge_options as
3448 * opposed to decaring a local hashmap is for convenience
3449 * so that we don't have to pass it to around.
3450 */
3451 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3452 NULL, 512);
3453 get_files_dirs(opt, head);
3454 get_files_dirs(opt, merge);
3455
3456 entries = get_unmerged(opt->repo->index);
3457 clean = detect_and_process_renames(opt, merge_base, head, merge,
3458 entries, &re_info);
3459 record_df_conflict_files(opt, entries);
3460 if (clean < 0)
3461 goto cleanup;
3462 for (i = entries->nr-1; 0 <= i; i--) {
3463 const char *path = entries->items[i].string;
3464 struct stage_data *e = entries->items[i].util;
3465 if (!e->processed) {
3466 int ret = process_entry(opt, path, e);
3467 if (!ret)
3468 clean = 0;
3469 else if (ret < 0) {
3470 clean = ret;
3471 goto cleanup;
3472 }
3473 }
3474 }
3475 for (i = 0; i < entries->nr; i++) {
3476 struct stage_data *e = entries->items[i].util;
3477 if (!e->processed)
3478 BUG("unprocessed path??? %s",
3479 entries->items[i].string);
3480 }
3481
3482 cleanup:
3483 final_cleanup_renames(&re_info);
3484
3485 string_list_clear(entries, 1);
3486 free(entries);
3487
3488 hashmap_free_entries(&opt->priv->current_file_dir_set,
3489 struct path_hashmap_entry, e);
3490
3491 if (clean < 0) {
3492 unpack_trees_finish(opt);
3493 return clean;
3494 }
3495 }
3496 else
3497 clean = 1;
3498
3499 unpack_trees_finish(opt);
3500
3501 if (opt->priv->call_depth &&
3502 !(*result = write_in_core_index_as_tree(opt->repo)))
3503 return -1;
3504
3505 return clean;
3506 }
3507
3508 static struct commit_list *reverse_commit_list(struct commit_list *list)
3509 {
3510 struct commit_list *next = NULL, *current, *backup;
3511 for (current = list; current; current = backup) {
3512 backup = current->next;
3513 current->next = next;
3514 next = current;
3515 }
3516 return next;
3517 }
3518
3519 /*
3520 * Merge the commits h1 and h2, return the resulting virtual
3521 * commit object and a flag indicating the cleanness of the merge.
3522 */
3523 static int merge_recursive_internal(struct merge_options *opt,
3524 struct commit *h1,
3525 struct commit *h2,
3526 struct commit_list *merge_bases,
3527 struct commit **result)
3528 {
3529 struct commit_list *iter;
3530 struct commit *merged_merge_bases;
3531 struct tree *result_tree;
3532 int clean;
3533 const char *ancestor_name;
3534 struct strbuf merge_base_abbrev = STRBUF_INIT;
3535
3536 if (show(opt, 4)) {
3537 output(opt, 4, _("Merging:"));
3538 output_commit_title(opt, h1);
3539 output_commit_title(opt, h2);
3540 }
3541
3542 if (!merge_bases) {
3543 merge_bases = get_merge_bases(h1, h2);
3544 merge_bases = reverse_commit_list(merge_bases);
3545 }
3546
3547 if (show(opt, 5)) {
3548 unsigned cnt = commit_list_count(merge_bases);
3549
3550 output(opt, 5, Q_("found %u common ancestor:",
3551 "found %u common ancestors:", cnt), cnt);
3552 for (iter = merge_bases; iter; iter = iter->next)
3553 output_commit_title(opt, iter->item);
3554 }
3555
3556 merged_merge_bases = pop_commit(&merge_bases);
3557 if (merged_merge_bases == NULL) {
3558 /* if there is no common ancestor, use an empty tree */
3559 struct tree *tree;
3560
3561 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3562 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3563 "ancestor");
3564 ancestor_name = "empty tree";
3565 } else if (opt->ancestor && !opt->priv->call_depth) {
3566 ancestor_name = opt->ancestor;
3567 } else if (merge_bases) {
3568 ancestor_name = "merged common ancestors";
3569 } else {
3570 strbuf_add_unique_abbrev(&merge_base_abbrev,
3571 &merged_merge_bases->object.oid,
3572 DEFAULT_ABBREV);
3573 ancestor_name = merge_base_abbrev.buf;
3574 }
3575
3576 for (iter = merge_bases; iter; iter = iter->next) {
3577 const char *saved_b1, *saved_b2;
3578 opt->priv->call_depth++;
3579 /*
3580 * When the merge fails, the result contains files
3581 * with conflict markers. The cleanness flag is
3582 * ignored (unless indicating an error), it was never
3583 * actually used, as result of merge_trees has always
3584 * overwritten it: the committed "conflicts" were
3585 * already resolved.
3586 */
3587 discard_index(opt->repo->index);
3588 saved_b1 = opt->branch1;
3589 saved_b2 = opt->branch2;
3590 opt->branch1 = "Temporary merge branch 1";
3591 opt->branch2 = "Temporary merge branch 2";
3592 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3593 NULL, &merged_merge_bases) < 0)
3594 return -1;
3595 opt->branch1 = saved_b1;
3596 opt->branch2 = saved_b2;
3597 opt->priv->call_depth--;
3598
3599 if (!merged_merge_bases)
3600 return err(opt, _("merge returned no commit"));
3601 }
3602
3603 discard_index(opt->repo->index);
3604 if (!opt->priv->call_depth)
3605 repo_read_index(opt->repo);
3606
3607 opt->ancestor = ancestor_name;
3608 clean = merge_trees_internal(opt,
3609 repo_get_commit_tree(opt->repo, h1),
3610 repo_get_commit_tree(opt->repo, h2),
3611 repo_get_commit_tree(opt->repo,
3612 merged_merge_bases),
3613 &result_tree);
3614 strbuf_release(&merge_base_abbrev);
3615 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3616 if (clean < 0) {
3617 flush_output(opt);
3618 return clean;
3619 }
3620
3621 if (opt->priv->call_depth) {
3622 *result = make_virtual_commit(opt->repo, result_tree,
3623 "merged tree");
3624 commit_list_insert(h1, &(*result)->parents);
3625 commit_list_insert(h2, &(*result)->parents->next);
3626 }
3627 return clean;
3628 }
3629
3630 static int merge_start(struct merge_options *opt, struct tree *head)
3631 {
3632 struct strbuf sb = STRBUF_INIT;
3633
3634 /* Sanity checks on opt */
3635 assert(opt->repo);
3636
3637 assert(opt->branch1 && opt->branch2);
3638
3639 assert(opt->detect_renames >= -1 &&
3640 opt->detect_renames <= DIFF_DETECT_COPY);
3641 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3642 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3643 assert(opt->rename_limit >= -1);
3644 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3645 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3646
3647 assert(opt->xdl_opts >= 0);
3648 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3649 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3650
3651 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3652 assert(opt->buffer_output <= 2);
3653 assert(opt->obuf.len == 0);
3654
3655 assert(opt->priv == NULL);
3656
3657 /* Sanity check on repo state; index must match head */
3658 if (repo_index_has_changes(opt->repo, head, &sb)) {
3659 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3660 sb.buf);
3661 strbuf_release(&sb);
3662 return -1;
3663 }
3664
3665 opt->priv = xcalloc(1, sizeof(*opt->priv));
3666 string_list_init(&opt->priv->df_conflict_file_set, 1);
3667 return 0;
3668 }
3669
3670 static void merge_finalize(struct merge_options *opt)
3671 {
3672 flush_output(opt);
3673 if (!opt->priv->call_depth && opt->buffer_output < 2)
3674 strbuf_release(&opt->obuf);
3675 if (show(opt, 2))
3676 diff_warn_rename_limit("merge.renamelimit",
3677 opt->priv->needed_rename_limit, 0);
3678 FREE_AND_NULL(opt->priv);
3679 }
3680
3681 int merge_trees(struct merge_options *opt,
3682 struct tree *head,
3683 struct tree *merge,
3684 struct tree *merge_base)
3685 {
3686 int clean;
3687 struct tree *ignored;
3688
3689 assert(opt->ancestor != NULL);
3690
3691 if (merge_start(opt, head))
3692 return -1;
3693 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3694 merge_finalize(opt);
3695
3696 return clean;
3697 }
3698
3699 int merge_recursive(struct merge_options *opt,
3700 struct commit *h1,
3701 struct commit *h2,
3702 struct commit_list *merge_bases,
3703 struct commit **result)
3704 {
3705 int clean;
3706
3707 assert(opt->ancestor == NULL ||
3708 !strcmp(opt->ancestor, "constructed merge base"));
3709
3710 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3711 return -1;
3712 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3713 merge_finalize(opt);
3714
3715 return clean;
3716 }
3717
3718 static struct commit *get_ref(struct repository *repo,
3719 const struct object_id *oid,
3720 const char *name)
3721 {
3722 struct object *object;
3723
3724 object = deref_tag(repo, parse_object(repo, oid),
3725 name, strlen(name));
3726 if (!object)
3727 return NULL;
3728 if (object->type == OBJ_TREE)
3729 return make_virtual_commit(repo, (struct tree*)object, name);
3730 if (object->type != OBJ_COMMIT)
3731 return NULL;
3732 if (parse_commit((struct commit *)object))
3733 return NULL;
3734 return (struct commit *)object;
3735 }
3736
3737 int merge_recursive_generic(struct merge_options *opt,
3738 const struct object_id *head,
3739 const struct object_id *merge,
3740 int num_merge_bases,
3741 const struct object_id **merge_bases,
3742 struct commit **result)
3743 {
3744 int clean;
3745 struct lock_file lock = LOCK_INIT;
3746 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3747 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3748 struct commit_list *ca = NULL;
3749
3750 if (merge_bases) {
3751 int i;
3752 for (i = 0; i < num_merge_bases; ++i) {
3753 struct commit *base;
3754 if (!(base = get_ref(opt->repo, merge_bases[i],
3755 oid_to_hex(merge_bases[i]))))
3756 return err(opt, _("Could not parse object '%s'"),
3757 oid_to_hex(merge_bases[i]));
3758 commit_list_insert(base, &ca);
3759 }
3760 if (num_merge_bases == 1)
3761 opt->ancestor = "constructed merge base";
3762 }
3763
3764 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3765 clean = merge_recursive(opt, head_commit, next_commit, ca,
3766 result);
3767 if (clean < 0) {
3768 rollback_lock_file(&lock);
3769 return clean;
3770 }
3771
3772 if (write_locked_index(opt->repo->index, &lock,
3773 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3774 return err(opt, _("Unable to write index."));
3775
3776 return clean ? 0 : 1;
3777 }
3778
3779 static void merge_recursive_config(struct merge_options *opt)
3780 {
3781 char *value = NULL;
3782 git_config_get_int("merge.verbosity", &opt->verbosity);
3783 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3784 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3785 if (!git_config_get_string("diff.renames", &value)) {
3786 opt->detect_renames = git_config_rename("diff.renames", value);
3787 free(value);
3788 }
3789 if (!git_config_get_string("merge.renames", &value)) {
3790 opt->detect_renames = git_config_rename("merge.renames", value);
3791 free(value);
3792 }
3793 if (!git_config_get_string("merge.directoryrenames", &value)) {
3794 int boolval = git_parse_maybe_bool(value);
3795 if (0 <= boolval) {
3796 opt->detect_directory_renames = boolval ?
3797 MERGE_DIRECTORY_RENAMES_TRUE :
3798 MERGE_DIRECTORY_RENAMES_NONE;
3799 } else if (!strcasecmp(value, "conflict")) {
3800 opt->detect_directory_renames =
3801 MERGE_DIRECTORY_RENAMES_CONFLICT;
3802 } /* avoid erroring on values from future versions of git */
3803 free(value);
3804 }
3805 git_config(git_xmerge_config, NULL);
3806 }
3807
3808 void init_merge_options(struct merge_options *opt,
3809 struct repository *repo)
3810 {
3811 const char *merge_verbosity;
3812 memset(opt, 0, sizeof(struct merge_options));
3813
3814 opt->repo = repo;
3815
3816 opt->detect_renames = -1;
3817 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3818 opt->rename_limit = -1;
3819
3820 opt->verbosity = 2;
3821 opt->buffer_output = 1;
3822 strbuf_init(&opt->obuf, 0);
3823
3824 opt->renormalize = 0;
3825
3826 merge_recursive_config(opt);
3827 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3828 if (merge_verbosity)
3829 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3830 if (opt->verbosity >= 5)
3831 opt->buffer_output = 0;
3832 }
3833
3834 int parse_merge_opt(struct merge_options *opt, const char *s)
3835 {
3836 const char *arg;
3837
3838 if (!s || !*s)
3839 return -1;
3840 if (!strcmp(s, "ours"))
3841 opt->recursive_variant = MERGE_VARIANT_OURS;
3842 else if (!strcmp(s, "theirs"))
3843 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3844 else if (!strcmp(s, "subtree"))
3845 opt->subtree_shift = "";
3846 else if (skip_prefix(s, "subtree=", &arg))
3847 opt->subtree_shift = arg;
3848 else if (!strcmp(s, "patience"))
3849 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3850 else if (!strcmp(s, "histogram"))
3851 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3852 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3853 long value = parse_algorithm_value(arg);
3854 if (value < 0)
3855 return -1;
3856 /* clear out previous settings */
3857 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3858 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3859 opt->xdl_opts |= value;
3860 }
3861 else if (!strcmp(s, "ignore-space-change"))
3862 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3863 else if (!strcmp(s, "ignore-all-space"))
3864 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3865 else if (!strcmp(s, "ignore-space-at-eol"))
3866 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3867 else if (!strcmp(s, "ignore-cr-at-eol"))
3868 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3869 else if (!strcmp(s, "renormalize"))
3870 opt->renormalize = 1;
3871 else if (!strcmp(s, "no-renormalize"))
3872 opt->renormalize = 0;
3873 else if (!strcmp(s, "no-renames"))
3874 opt->detect_renames = 0;
3875 else if (!strcmp(s, "find-renames")) {
3876 opt->detect_renames = 1;
3877 opt->rename_score = 0;
3878 }
3879 else if (skip_prefix(s, "find-renames=", &arg) ||
3880 skip_prefix(s, "rename-threshold=", &arg)) {
3881 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3882 return -1;
3883 opt->detect_renames = 1;
3884 }
3885 /*
3886 * Please update $__git_merge_strategy_options in
3887 * git-completion.bash when you add new options
3888 */
3889 else
3890 return -1;
3891 return 0;
3892 }