]> git.ipfire.org Git - thirdparty/git.git/blob - merge-recursive.c
merge-recursive: fix the diff3 common ancestor label for virtual commits
[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 void *entry,
49 const void *entry_or_key,
50 const void *keydata)
51 {
52 const struct path_hashmap_entry *a = entry;
53 const struct path_hashmap_entry *b = entry_or_key;
54 const char *key = keydata;
55
56 if (ignore_case)
57 return strcasecmp(a->path, key ? key : b->path);
58 else
59 return strcmp(a->path, key ? key : b->path);
60 }
61
62 static unsigned int path_hash(const char *path)
63 {
64 return ignore_case ? strihash(path) : strhash(path);
65 }
66
67 /*
68 * For dir_rename_entry, directory names are stored as a full path from the
69 * toplevel of the repository and do not include a trailing '/'. Also:
70 *
71 * dir: original name of directory being renamed
72 * non_unique_new_dir: if true, could not determine new_dir
73 * new_dir: final name of directory being renamed
74 * possible_new_dirs: temporary used to help determine new_dir; see comments
75 * in get_directory_renames() for details
76 */
77 struct dir_rename_entry {
78 struct hashmap_entry ent; /* must be the first member! */
79 char *dir;
80 unsigned non_unique_new_dir:1;
81 struct strbuf new_dir;
82 struct string_list possible_new_dirs;
83 };
84
85 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
86 char *dir)
87 {
88 struct dir_rename_entry key;
89
90 if (dir == NULL)
91 return NULL;
92 hashmap_entry_init(&key, strhash(dir));
93 key.dir = dir;
94 return hashmap_get(hashmap, &key, NULL);
95 }
96
97 static int dir_rename_cmp(const void *unused_cmp_data,
98 const void *entry,
99 const void *entry_or_key,
100 const void *unused_keydata)
101 {
102 const struct dir_rename_entry *e1 = entry;
103 const struct dir_rename_entry *e2 = entry_or_key;
104
105 return strcmp(e1->dir, e2->dir);
106 }
107
108 static void dir_rename_init(struct hashmap *map)
109 {
110 hashmap_init(map, dir_rename_cmp, NULL, 0);
111 }
112
113 static void dir_rename_entry_init(struct dir_rename_entry *entry,
114 char *directory)
115 {
116 hashmap_entry_init(entry, strhash(directory));
117 entry->dir = directory;
118 entry->non_unique_new_dir = 0;
119 strbuf_init(&entry->new_dir, 0);
120 string_list_init(&entry->possible_new_dirs, 0);
121 }
122
123 struct collision_entry {
124 struct hashmap_entry ent; /* must be the first member! */
125 char *target_file;
126 struct string_list source_files;
127 unsigned reported_already:1;
128 };
129
130 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
131 char *target_file)
132 {
133 struct collision_entry key;
134
135 hashmap_entry_init(&key, strhash(target_file));
136 key.target_file = target_file;
137 return hashmap_get(hashmap, &key, NULL);
138 }
139
140 static int collision_cmp(void *unused_cmp_data,
141 const struct collision_entry *e1,
142 const struct collision_entry *e2,
143 const void *unused_keydata)
144 {
145 return strcmp(e1->target_file, e2->target_file);
146 }
147
148 static void collision_init(struct hashmap *map)
149 {
150 hashmap_init(map, (hashmap_cmp_fn) collision_cmp, NULL, 0);
151 }
152
153 static void flush_output(struct merge_options *opt)
154 {
155 if (opt->buffer_output < 2 && opt->obuf.len) {
156 fputs(opt->obuf.buf, stdout);
157 strbuf_reset(&opt->obuf);
158 }
159 }
160
161 static int err(struct merge_options *opt, const char *err, ...)
162 {
163 va_list params;
164
165 if (opt->buffer_output < 2)
166 flush_output(opt);
167 else {
168 strbuf_complete(&opt->obuf, '\n');
169 strbuf_addstr(&opt->obuf, "error: ");
170 }
171 va_start(params, err);
172 strbuf_vaddf(&opt->obuf, err, params);
173 va_end(params);
174 if (opt->buffer_output > 1)
175 strbuf_addch(&opt->obuf, '\n');
176 else {
177 error("%s", opt->obuf.buf);
178 strbuf_reset(&opt->obuf);
179 }
180
181 return -1;
182 }
183
184 static struct tree *shift_tree_object(struct repository *repo,
185 struct tree *one, struct tree *two,
186 const char *subtree_shift)
187 {
188 struct object_id shifted;
189
190 if (!*subtree_shift) {
191 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
192 } else {
193 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
194 subtree_shift);
195 }
196 if (oideq(&two->object.oid, &shifted))
197 return two;
198 return lookup_tree(repo, &shifted);
199 }
200
201 static inline void set_commit_tree(struct commit *c, struct tree *t)
202 {
203 c->maybe_tree = t;
204 }
205
206 static struct commit *make_virtual_commit(struct repository *repo,
207 struct tree *tree,
208 const char *comment)
209 {
210 struct commit *commit = alloc_commit_node(repo);
211
212 set_merge_remote_desc(commit, comment, (struct object *)commit);
213 set_commit_tree(commit, tree);
214 commit->object.parsed = 1;
215 return commit;
216 }
217
218 /*
219 * Since we use get_tree_entry(), which does not put the read object into
220 * the object pool, we cannot rely on a == b.
221 */
222 static int oid_eq(const struct object_id *a, const struct object_id *b)
223 {
224 if (!a && !b)
225 return 2;
226 return a && b && oideq(a, b);
227 }
228
229 enum rename_type {
230 RENAME_NORMAL = 0,
231 RENAME_VIA_DIR,
232 RENAME_ADD,
233 RENAME_DELETE,
234 RENAME_ONE_FILE_TO_ONE,
235 RENAME_ONE_FILE_TO_TWO,
236 RENAME_TWO_FILES_TO_ONE
237 };
238
239 /*
240 * Since we want to write the index eventually, we cannot reuse the index
241 * for these (temporary) data.
242 */
243 struct stage_data {
244 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
245 struct rename_conflict_info *rename_conflict_info;
246 unsigned processed:1;
247 };
248
249 struct rename {
250 unsigned processed:1;
251 struct diff_filepair *pair;
252 const char *branch; /* branch that the rename occurred on */
253 /*
254 * If directory rename detection affected this rename, what was its
255 * original type ('A' or 'R') and it's original destination before
256 * the directory rename (otherwise, '\0' and NULL for these two vars).
257 */
258 char dir_rename_original_type;
259 char *dir_rename_original_dest;
260 /*
261 * Purpose of src_entry and dst_entry:
262 *
263 * If 'before' is renamed to 'after' then src_entry will contain
264 * the versions of 'before' from the merge_base, HEAD, and MERGE in
265 * stages 1, 2, and 3; dst_entry will contain the respective
266 * versions of 'after' in corresponding locations. Thus, we have a
267 * total of six modes and oids, though some will be null. (Stage 0
268 * is ignored; we're interested in handling conflicts.)
269 *
270 * Since we don't turn on break-rewrites by default, neither
271 * src_entry nor dst_entry can have all three of their stages have
272 * non-null oids, meaning at most four of the six will be non-null.
273 * Also, since this is a rename, both src_entry and dst_entry will
274 * have at least one non-null oid, meaning at least two will be
275 * non-null. Of the six oids, a typical rename will have three be
276 * non-null. Only two implies a rename/delete, and four implies a
277 * rename/add.
278 */
279 struct stage_data *src_entry;
280 struct stage_data *dst_entry;
281 };
282
283 struct rename_conflict_info {
284 enum rename_type rename_type;
285 struct rename *ren1;
286 struct rename *ren2;
287 };
288
289 static inline void setup_rename_conflict_info(enum rename_type rename_type,
290 struct merge_options *opt,
291 struct rename *ren1,
292 struct rename *ren2)
293 {
294 struct rename_conflict_info *ci;
295
296 /*
297 * When we have two renames involved, it's easiest to get the
298 * correct things into stage 2 and 3, and to make sure that the
299 * content merge puts HEAD before the other branch if we just
300 * ensure that branch1 == opt->branch1. So, simply flip arguments
301 * around if we don't have that.
302 */
303 if (ren2 && ren1->branch != opt->branch1) {
304 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
305 return;
306 }
307
308 ci = xcalloc(1, sizeof(struct rename_conflict_info));
309 ci->rename_type = rename_type;
310 ci->ren1 = ren1;
311 ci->ren2 = ren2;
312
313 ci->ren1->dst_entry->processed = 0;
314 ci->ren1->dst_entry->rename_conflict_info = ci;
315 if (ren2) {
316 ci->ren2->dst_entry->rename_conflict_info = ci;
317 }
318 }
319
320 static int show(struct merge_options *opt, int v)
321 {
322 return (!opt->priv->call_depth && opt->verbosity >= v) ||
323 opt->verbosity >= 5;
324 }
325
326 __attribute__((format (printf, 3, 4)))
327 static void output(struct merge_options *opt, int v, const char *fmt, ...)
328 {
329 va_list ap;
330
331 if (!show(opt, v))
332 return;
333
334 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
335
336 va_start(ap, fmt);
337 strbuf_vaddf(&opt->obuf, fmt, ap);
338 va_end(ap);
339
340 strbuf_addch(&opt->obuf, '\n');
341 if (!opt->buffer_output)
342 flush_output(opt);
343 }
344
345 static void output_commit_title(struct merge_options *opt, struct commit *commit)
346 {
347 struct merge_remote_desc *desc;
348
349 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
350 desc = merge_remote_util(commit);
351 if (desc)
352 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
353 else {
354 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
355 DEFAULT_ABBREV);
356 strbuf_addch(&opt->obuf, ' ');
357 if (parse_commit(commit) != 0)
358 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
359 else {
360 const char *title;
361 const char *msg = get_commit_buffer(commit, NULL);
362 int len = find_commit_subject(msg, &title);
363 if (len)
364 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
365 unuse_commit_buffer(commit, msg);
366 }
367 }
368 flush_output(opt);
369 }
370
371 static int add_cacheinfo(struct merge_options *opt,
372 const struct diff_filespec *blob,
373 const char *path, int stage, int refresh, int options)
374 {
375 struct index_state *istate = opt->repo->index;
376 struct cache_entry *ce;
377 int ret;
378
379 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
380 if (!ce)
381 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
382
383 ret = add_index_entry(istate, ce, options);
384 if (refresh) {
385 struct cache_entry *nce;
386
387 nce = refresh_cache_entry(istate, ce,
388 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
389 if (!nce)
390 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
391 if (nce != ce)
392 ret = add_index_entry(istate, nce, options);
393 }
394 return ret;
395 }
396
397 static inline int merge_detect_rename(struct merge_options *opt)
398 {
399 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
400 }
401
402 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
403 {
404 parse_tree(tree);
405 init_tree_desc(desc, tree->buffer, tree->size);
406 }
407
408 static int unpack_trees_start(struct merge_options *opt,
409 struct tree *common,
410 struct tree *head,
411 struct tree *merge)
412 {
413 int rc;
414 struct tree_desc t[3];
415 struct index_state tmp_index = { NULL };
416
417 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
418 if (opt->priv->call_depth)
419 opt->priv->unpack_opts.index_only = 1;
420 else
421 opt->priv->unpack_opts.update = 1;
422 opt->priv->unpack_opts.merge = 1;
423 opt->priv->unpack_opts.head_idx = 2;
424 opt->priv->unpack_opts.fn = threeway_merge;
425 opt->priv->unpack_opts.src_index = opt->repo->index;
426 opt->priv->unpack_opts.dst_index = &tmp_index;
427 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
428 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
429
430 init_tree_desc_from_tree(t+0, common);
431 init_tree_desc_from_tree(t+1, head);
432 init_tree_desc_from_tree(t+2, merge);
433
434 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
435 cache_tree_free(&opt->repo->index->cache_tree);
436
437 /*
438 * Update opt->repo->index to match the new results, AFTER saving a
439 * copy in opt->priv->orig_index. Update src_index to point to the
440 * saved copy. (verify_uptodate() checks src_index, and the original
441 * index is the one that had the necessary modification timestamps.)
442 */
443 opt->priv->orig_index = *opt->repo->index;
444 *opt->repo->index = tmp_index;
445 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
446
447 return rc;
448 }
449
450 static void unpack_trees_finish(struct merge_options *opt)
451 {
452 discard_index(&opt->priv->orig_index);
453 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
454 }
455
456 static int save_files_dirs(const struct object_id *oid,
457 struct strbuf *base, const char *path,
458 unsigned int mode, int stage, void *context)
459 {
460 struct path_hashmap_entry *entry;
461 int baselen = base->len;
462 struct merge_options *opt = context;
463
464 strbuf_addstr(base, path);
465
466 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
467 hashmap_entry_init(entry, path_hash(entry->path));
468 hashmap_add(&opt->priv->current_file_dir_set, entry);
469
470 strbuf_setlen(base, baselen);
471 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
472 }
473
474 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
475 {
476 struct pathspec match_all;
477 memset(&match_all, 0, sizeof(match_all));
478 read_tree_recursive(opt->repo, tree, "", 0, 0,
479 &match_all, save_files_dirs, opt);
480 }
481
482 static int get_tree_entry_if_blob(struct repository *r,
483 const struct object_id *tree,
484 const char *path,
485 struct diff_filespec *dfs)
486 {
487 int ret;
488
489 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
490 if (S_ISDIR(dfs->mode)) {
491 oidcpy(&dfs->oid, &null_oid);
492 dfs->mode = 0;
493 }
494 return ret;
495 }
496
497 /*
498 * Returns an index_entry instance which doesn't have to correspond to
499 * a real cache entry in Git's index.
500 */
501 static struct stage_data *insert_stage_data(struct repository *r,
502 const char *path,
503 struct tree *o, struct tree *a, struct tree *b,
504 struct string_list *entries)
505 {
506 struct string_list_item *item;
507 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
508 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
509 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
510 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
511 item = string_list_insert(entries, path);
512 item->util = e;
513 return e;
514 }
515
516 /*
517 * Create a dictionary mapping file names to stage_data objects. The
518 * dictionary contains one entry for every path with a non-zero stage entry.
519 */
520 static struct string_list *get_unmerged(struct index_state *istate)
521 {
522 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
523 int i;
524
525 unmerged->strdup_strings = 1;
526
527 for (i = 0; i < istate->cache_nr; i++) {
528 struct string_list_item *item;
529 struct stage_data *e;
530 const struct cache_entry *ce = istate->cache[i];
531 if (!ce_stage(ce))
532 continue;
533
534 item = string_list_lookup(unmerged, ce->name);
535 if (!item) {
536 item = string_list_insert(unmerged, ce->name);
537 item->util = xcalloc(1, sizeof(struct stage_data));
538 }
539 e = item->util;
540 e->stages[ce_stage(ce)].mode = ce->ce_mode;
541 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
542 }
543
544 return unmerged;
545 }
546
547 static int string_list_df_name_compare(const char *one, const char *two)
548 {
549 int onelen = strlen(one);
550 int twolen = strlen(two);
551 /*
552 * Here we only care that entries for D/F conflicts are
553 * adjacent, in particular with the file of the D/F conflict
554 * appearing before files below the corresponding directory.
555 * The order of the rest of the list is irrelevant for us.
556 *
557 * To achieve this, we sort with df_name_compare and provide
558 * the mode S_IFDIR so that D/F conflicts will sort correctly.
559 * We use the mode S_IFDIR for everything else for simplicity,
560 * since in other cases any changes in their order due to
561 * sorting cause no problems for us.
562 */
563 int cmp = df_name_compare(one, onelen, S_IFDIR,
564 two, twolen, S_IFDIR);
565 /*
566 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
567 * that 'foo' comes before 'foo/bar'.
568 */
569 if (cmp)
570 return cmp;
571 return onelen - twolen;
572 }
573
574 static void record_df_conflict_files(struct merge_options *opt,
575 struct string_list *entries)
576 {
577 /* If there is a D/F conflict and the file for such a conflict
578 * currently exists in the working tree, we want to allow it to be
579 * removed to make room for the corresponding directory if needed.
580 * The files underneath the directories of such D/F conflicts will
581 * be processed before the corresponding file involved in the D/F
582 * conflict. If the D/F directory ends up being removed by the
583 * merge, then we won't have to touch the D/F file. If the D/F
584 * directory needs to be written to the working copy, then the D/F
585 * file will simply be removed (in make_room_for_path()) to make
586 * room for the necessary paths. Note that if both the directory
587 * and the file need to be present, then the D/F file will be
588 * reinstated with a new unique name at the time it is processed.
589 */
590 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
591 const char *last_file = NULL;
592 int last_len = 0;
593 int i;
594
595 /*
596 * If we're merging merge-bases, we don't want to bother with
597 * any working directory changes.
598 */
599 if (opt->priv->call_depth)
600 return;
601
602 /* Ensure D/F conflicts are adjacent in the entries list. */
603 for (i = 0; i < entries->nr; i++) {
604 struct string_list_item *next = &entries->items[i];
605 string_list_append(&df_sorted_entries, next->string)->util =
606 next->util;
607 }
608 df_sorted_entries.cmp = string_list_df_name_compare;
609 string_list_sort(&df_sorted_entries);
610
611 string_list_clear(&opt->priv->df_conflict_file_set, 1);
612 for (i = 0; i < df_sorted_entries.nr; i++) {
613 const char *path = df_sorted_entries.items[i].string;
614 int len = strlen(path);
615 struct stage_data *e = df_sorted_entries.items[i].util;
616
617 /*
618 * Check if last_file & path correspond to a D/F conflict;
619 * i.e. whether path is last_file+'/'+<something>.
620 * If so, record that it's okay to remove last_file to make
621 * room for path and friends if needed.
622 */
623 if (last_file &&
624 len > last_len &&
625 memcmp(path, last_file, last_len) == 0 &&
626 path[last_len] == '/') {
627 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
628 }
629
630 /*
631 * Determine whether path could exist as a file in the
632 * working directory as a possible D/F conflict. This
633 * will only occur when it exists in stage 2 as a
634 * file.
635 */
636 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
637 last_file = path;
638 last_len = len;
639 } else {
640 last_file = NULL;
641 }
642 }
643 string_list_clear(&df_sorted_entries, 0);
644 }
645
646 static int update_stages(struct merge_options *opt, const char *path,
647 const struct diff_filespec *o,
648 const struct diff_filespec *a,
649 const struct diff_filespec *b)
650 {
651
652 /*
653 * NOTE: It is usually a bad idea to call update_stages on a path
654 * before calling update_file on that same path, since it can
655 * sometimes lead to spurious "refusing to lose untracked file..."
656 * messages from update_file (via make_room_for path via
657 * would_lose_untracked). Instead, reverse the order of the calls
658 * (executing update_file first and then update_stages).
659 */
660 int clear = 1;
661 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
662 if (clear)
663 if (remove_file_from_index(opt->repo->index, path))
664 return -1;
665 if (o)
666 if (add_cacheinfo(opt, o, path, 1, 0, options))
667 return -1;
668 if (a)
669 if (add_cacheinfo(opt, a, path, 2, 0, options))
670 return -1;
671 if (b)
672 if (add_cacheinfo(opt, b, path, 3, 0, options))
673 return -1;
674 return 0;
675 }
676
677 static void update_entry(struct stage_data *entry,
678 struct diff_filespec *o,
679 struct diff_filespec *a,
680 struct diff_filespec *b)
681 {
682 entry->processed = 0;
683 entry->stages[1].mode = o->mode;
684 entry->stages[2].mode = a->mode;
685 entry->stages[3].mode = b->mode;
686 oidcpy(&entry->stages[1].oid, &o->oid);
687 oidcpy(&entry->stages[2].oid, &a->oid);
688 oidcpy(&entry->stages[3].oid, &b->oid);
689 }
690
691 static int remove_file(struct merge_options *opt, int clean,
692 const char *path, int no_wd)
693 {
694 int update_cache = opt->priv->call_depth || clean;
695 int update_working_directory = !opt->priv->call_depth && !no_wd;
696
697 if (update_cache) {
698 if (remove_file_from_index(opt->repo->index, path))
699 return -1;
700 }
701 if (update_working_directory) {
702 if (ignore_case) {
703 struct cache_entry *ce;
704 ce = index_file_exists(opt->repo->index, path, strlen(path),
705 ignore_case);
706 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
707 return 0;
708 }
709 if (remove_path(path))
710 return -1;
711 }
712 return 0;
713 }
714
715 /* add a string to a strbuf, but converting "/" to "_" */
716 static void add_flattened_path(struct strbuf *out, const char *s)
717 {
718 size_t i = out->len;
719 strbuf_addstr(out, s);
720 for (; i < out->len; i++)
721 if (out->buf[i] == '/')
722 out->buf[i] = '_';
723 }
724
725 static char *unique_path(struct merge_options *opt,
726 const char *path,
727 const char *branch)
728 {
729 struct path_hashmap_entry *entry;
730 struct strbuf newpath = STRBUF_INIT;
731 int suffix = 0;
732 size_t base_len;
733
734 strbuf_addf(&newpath, "%s~", path);
735 add_flattened_path(&newpath, branch);
736
737 base_len = newpath.len;
738 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
739 path_hash(newpath.buf), newpath.buf) ||
740 (!opt->priv->call_depth && file_exists(newpath.buf))) {
741 strbuf_setlen(&newpath, base_len);
742 strbuf_addf(&newpath, "_%d", suffix++);
743 }
744
745 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
746 hashmap_entry_init(entry, path_hash(entry->path));
747 hashmap_add(&opt->priv->current_file_dir_set, entry);
748 return strbuf_detach(&newpath, NULL);
749 }
750
751 /**
752 * Check whether a directory in the index is in the way of an incoming
753 * file. Return 1 if so. If check_working_copy is non-zero, also
754 * check the working directory. If empty_ok is non-zero, also return
755 * 0 in the case where the working-tree dir exists but is empty.
756 */
757 static int dir_in_way(struct index_state *istate, const char *path,
758 int check_working_copy, int empty_ok)
759 {
760 int pos;
761 struct strbuf dirpath = STRBUF_INIT;
762 struct stat st;
763
764 strbuf_addstr(&dirpath, path);
765 strbuf_addch(&dirpath, '/');
766
767 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
768
769 if (pos < 0)
770 pos = -1 - pos;
771 if (pos < istate->cache_nr &&
772 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
773 strbuf_release(&dirpath);
774 return 1;
775 }
776
777 strbuf_release(&dirpath);
778 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
779 !(empty_ok && is_empty_dir(path));
780 }
781
782 /*
783 * Returns whether path was tracked in the index before the merge started,
784 * and its oid and mode match the specified values
785 */
786 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
787 const struct diff_filespec *blob)
788 {
789 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
790 struct cache_entry *ce;
791
792 if (0 > pos)
793 /* we were not tracking this path before the merge */
794 return 0;
795
796 /* See if the file we were tracking before matches */
797 ce = opt->priv->orig_index.cache[pos];
798 return (oid_eq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
799 }
800
801 /*
802 * Returns whether path was tracked in the index before the merge started
803 */
804 static int was_tracked(struct merge_options *opt, const char *path)
805 {
806 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
807
808 if (0 <= pos)
809 /* we were tracking this path before the merge */
810 return 1;
811
812 return 0;
813 }
814
815 static int would_lose_untracked(struct merge_options *opt, const char *path)
816 {
817 struct index_state *istate = opt->repo->index;
818
819 /*
820 * This may look like it can be simplified to:
821 * return !was_tracked(opt, path) && file_exists(path)
822 * but it can't. This function needs to know whether path was in
823 * the working tree due to EITHER having been tracked in the index
824 * before the merge OR having been put into the working copy and
825 * index by unpack_trees(). Due to that either-or requirement, we
826 * check the current index instead of the original one.
827 *
828 * Note that we do not need to worry about merge-recursive itself
829 * updating the index after unpack_trees() and before calling this
830 * function, because we strictly require all code paths in
831 * merge-recursive to update the working tree first and the index
832 * second. Doing otherwise would break
833 * update_file()/would_lose_untracked(); see every comment in this
834 * file which mentions "update_stages".
835 */
836 int pos = index_name_pos(istate, path, strlen(path));
837
838 if (pos < 0)
839 pos = -1 - pos;
840 while (pos < istate->cache_nr &&
841 !strcmp(path, istate->cache[pos]->name)) {
842 /*
843 * If stage #0, it is definitely tracked.
844 * If it has stage #2 then it was tracked
845 * before this merge started. All other
846 * cases the path was not tracked.
847 */
848 switch (ce_stage(istate->cache[pos])) {
849 case 0:
850 case 2:
851 return 0;
852 }
853 pos++;
854 }
855 return file_exists(path);
856 }
857
858 static int was_dirty(struct merge_options *opt, const char *path)
859 {
860 struct cache_entry *ce;
861 int dirty = 1;
862
863 if (opt->priv->call_depth || !was_tracked(opt, path))
864 return !dirty;
865
866 ce = index_file_exists(opt->priv->unpack_opts.src_index,
867 path, strlen(path), ignore_case);
868 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
869 return dirty;
870 }
871
872 static int make_room_for_path(struct merge_options *opt, const char *path)
873 {
874 int status, i;
875 const char *msg = _("failed to create path '%s'%s");
876
877 /* Unlink any D/F conflict files that are in the way */
878 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
879 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
880 size_t pathlen = strlen(path);
881 size_t df_pathlen = strlen(df_path);
882 if (df_pathlen < pathlen &&
883 path[df_pathlen] == '/' &&
884 strncmp(path, df_path, df_pathlen) == 0) {
885 output(opt, 3,
886 _("Removing %s to make room for subdirectory\n"),
887 df_path);
888 unlink(df_path);
889 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
890 i, 0);
891 break;
892 }
893 }
894
895 /* Make sure leading directories are created */
896 status = safe_create_leading_directories_const(path);
897 if (status) {
898 if (status == SCLD_EXISTS)
899 /* something else exists */
900 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
901 return err(opt, msg, path, "");
902 }
903
904 /*
905 * Do not unlink a file in the work tree if we are not
906 * tracking it.
907 */
908 if (would_lose_untracked(opt, path))
909 return err(opt, _("refusing to lose untracked file at '%s'"),
910 path);
911
912 /* Successful unlink is good.. */
913 if (!unlink(path))
914 return 0;
915 /* .. and so is no existing file */
916 if (errno == ENOENT)
917 return 0;
918 /* .. but not some other error (who really cares what?) */
919 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
920 }
921
922 static int update_file_flags(struct merge_options *opt,
923 const struct diff_filespec *contents,
924 const char *path,
925 int update_cache,
926 int update_wd)
927 {
928 int ret = 0;
929
930 if (opt->priv->call_depth)
931 update_wd = 0;
932
933 if (update_wd) {
934 enum object_type type;
935 void *buf;
936 unsigned long size;
937
938 if (S_ISGITLINK(contents->mode)) {
939 /*
940 * We may later decide to recursively descend into
941 * the submodule directory and update its index
942 * and/or work tree, but we do not do that now.
943 */
944 update_wd = 0;
945 goto update_index;
946 }
947
948 buf = read_object_file(&contents->oid, &type, &size);
949 if (!buf) {
950 ret = err(opt, _("cannot read object %s '%s'"),
951 oid_to_hex(&contents->oid), path);
952 goto free_buf;
953 }
954 if (type != OBJ_BLOB) {
955 ret = err(opt, _("blob expected for %s '%s'"),
956 oid_to_hex(&contents->oid), path);
957 goto free_buf;
958 }
959 if (S_ISREG(contents->mode)) {
960 struct strbuf strbuf = STRBUF_INIT;
961 if (convert_to_working_tree(opt->repo->index,
962 path, buf, size, &strbuf)) {
963 free(buf);
964 size = strbuf.len;
965 buf = strbuf_detach(&strbuf, NULL);
966 }
967 }
968
969 if (make_room_for_path(opt, path) < 0) {
970 update_wd = 0;
971 goto free_buf;
972 }
973 if (S_ISREG(contents->mode) ||
974 (!has_symlinks && S_ISLNK(contents->mode))) {
975 int fd;
976 int mode = (contents->mode & 0100 ? 0777 : 0666);
977
978 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
979 if (fd < 0) {
980 ret = err(opt, _("failed to open '%s': %s"),
981 path, strerror(errno));
982 goto free_buf;
983 }
984 write_in_full(fd, buf, size);
985 close(fd);
986 } else if (S_ISLNK(contents->mode)) {
987 char *lnk = xmemdupz(buf, size);
988 safe_create_leading_directories_const(path);
989 unlink(path);
990 if (symlink(lnk, path))
991 ret = err(opt, _("failed to symlink '%s': %s"),
992 path, strerror(errno));
993 free(lnk);
994 } else
995 ret = err(opt,
996 _("do not know what to do with %06o %s '%s'"),
997 contents->mode, oid_to_hex(&contents->oid), path);
998 free_buf:
999 free(buf);
1000 }
1001 update_index:
1002 if (!ret && update_cache)
1003 if (add_cacheinfo(opt, contents, path, 0, update_wd,
1004 ADD_CACHE_OK_TO_ADD))
1005 return -1;
1006 return ret;
1007 }
1008
1009 static int update_file(struct merge_options *opt,
1010 int clean,
1011 const struct diff_filespec *contents,
1012 const char *path)
1013 {
1014 return update_file_flags(opt, contents, path,
1015 opt->priv->call_depth || clean, !opt->priv->call_depth);
1016 }
1017
1018 /* Low level file merging, update and removal */
1019
1020 struct merge_file_info {
1021 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1022 unsigned clean:1,
1023 merge:1;
1024 };
1025
1026 static int merge_3way(struct merge_options *opt,
1027 mmbuffer_t *result_buf,
1028 const struct diff_filespec *o,
1029 const struct diff_filespec *a,
1030 const struct diff_filespec *b,
1031 const char *branch1,
1032 const char *branch2,
1033 const int extra_marker_size)
1034 {
1035 mmfile_t orig, src1, src2;
1036 struct ll_merge_options ll_opts = {0};
1037 char *base, *name1, *name2;
1038 int merge_status;
1039
1040 ll_opts.renormalize = opt->renormalize;
1041 ll_opts.extra_marker_size = extra_marker_size;
1042 ll_opts.xdl_opts = opt->xdl_opts;
1043
1044 if (opt->priv->call_depth) {
1045 ll_opts.virtual_ancestor = 1;
1046 ll_opts.variant = 0;
1047 } else {
1048 switch (opt->recursive_variant) {
1049 case MERGE_VARIANT_OURS:
1050 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1051 break;
1052 case MERGE_VARIANT_THEIRS:
1053 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1054 break;
1055 default:
1056 ll_opts.variant = 0;
1057 break;
1058 }
1059 }
1060
1061 assert(a->path && b->path && o->path && opt->ancestor);
1062 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1063 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1064 name1 = mkpathdup("%s:%s", branch1, a->path);
1065 name2 = mkpathdup("%s:%s", branch2, b->path);
1066 } else {
1067 base = mkpathdup("%s", opt->ancestor);
1068 name1 = mkpathdup("%s", branch1);
1069 name2 = mkpathdup("%s", branch2);
1070 }
1071
1072 read_mmblob(&orig, &o->oid);
1073 read_mmblob(&src1, &a->oid);
1074 read_mmblob(&src2, &b->oid);
1075
1076 merge_status = ll_merge(result_buf, a->path, &orig, base,
1077 &src1, name1, &src2, name2,
1078 opt->repo->index, &ll_opts);
1079
1080 free(base);
1081 free(name1);
1082 free(name2);
1083 free(orig.ptr);
1084 free(src1.ptr);
1085 free(src2.ptr);
1086 return merge_status;
1087 }
1088
1089 static int find_first_merges(struct repository *repo,
1090 struct object_array *result, const char *path,
1091 struct commit *a, struct commit *b)
1092 {
1093 int i, j;
1094 struct object_array merges = OBJECT_ARRAY_INIT;
1095 struct commit *commit;
1096 int contains_another;
1097
1098 char merged_revision[GIT_MAX_HEXSZ + 2];
1099 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1100 "--all", merged_revision, NULL };
1101 struct rev_info revs;
1102 struct setup_revision_opt rev_opts;
1103
1104 memset(result, 0, sizeof(struct object_array));
1105 memset(&rev_opts, 0, sizeof(rev_opts));
1106
1107 /* get all revisions that merge commit a */
1108 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1109 oid_to_hex(&a->object.oid));
1110 repo_init_revisions(repo, &revs, NULL);
1111 rev_opts.submodule = path;
1112 /* FIXME: can't handle linked worktrees in submodules yet */
1113 revs.single_worktree = path != NULL;
1114 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1115
1116 /* save all revisions from the above list that contain b */
1117 if (prepare_revision_walk(&revs))
1118 die("revision walk setup failed");
1119 while ((commit = get_revision(&revs)) != NULL) {
1120 struct object *o = &(commit->object);
1121 if (in_merge_bases(b, commit))
1122 add_object_array(o, NULL, &merges);
1123 }
1124 reset_revision_walk();
1125
1126 /* Now we've got all merges that contain a and b. Prune all
1127 * merges that contain another found merge and save them in
1128 * result.
1129 */
1130 for (i = 0; i < merges.nr; i++) {
1131 struct commit *m1 = (struct commit *) merges.objects[i].item;
1132
1133 contains_another = 0;
1134 for (j = 0; j < merges.nr; j++) {
1135 struct commit *m2 = (struct commit *) merges.objects[j].item;
1136 if (i != j && in_merge_bases(m2, m1)) {
1137 contains_another = 1;
1138 break;
1139 }
1140 }
1141
1142 if (!contains_another)
1143 add_object_array(merges.objects[i].item, NULL, result);
1144 }
1145
1146 object_array_clear(&merges);
1147 return result->nr;
1148 }
1149
1150 static void print_commit(struct commit *commit)
1151 {
1152 struct strbuf sb = STRBUF_INIT;
1153 struct pretty_print_context ctx = {0};
1154 ctx.date_mode.type = DATE_NORMAL;
1155 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1156 fprintf(stderr, "%s\n", sb.buf);
1157 strbuf_release(&sb);
1158 }
1159
1160 static int is_valid(const struct diff_filespec *dfs)
1161 {
1162 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1163 }
1164
1165 static int merge_submodule(struct merge_options *opt,
1166 struct object_id *result, const char *path,
1167 const struct object_id *base, const struct object_id *a,
1168 const struct object_id *b)
1169 {
1170 struct commit *commit_base, *commit_a, *commit_b;
1171 int parent_count;
1172 struct object_array merges;
1173
1174 int i;
1175 int search = !opt->priv->call_depth;
1176
1177 /* store a in result in case we fail */
1178 oidcpy(result, a);
1179
1180 /* we can not handle deletion conflicts */
1181 if (is_null_oid(base))
1182 return 0;
1183 if (is_null_oid(a))
1184 return 0;
1185 if (is_null_oid(b))
1186 return 0;
1187
1188 if (add_submodule_odb(path)) {
1189 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1190 return 0;
1191 }
1192
1193 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1194 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1195 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1196 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1197 return 0;
1198 }
1199
1200 /* check whether both changes are forward */
1201 if (!in_merge_bases(commit_base, commit_a) ||
1202 !in_merge_bases(commit_base, commit_b)) {
1203 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1204 return 0;
1205 }
1206
1207 /* Case #1: a is contained in b or vice versa */
1208 if (in_merge_bases(commit_a, commit_b)) {
1209 oidcpy(result, b);
1210 if (show(opt, 3)) {
1211 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1212 output_commit_title(opt, commit_b);
1213 } else if (show(opt, 2))
1214 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1215 else
1216 ; /* no output */
1217
1218 return 1;
1219 }
1220 if (in_merge_bases(commit_b, commit_a)) {
1221 oidcpy(result, a);
1222 if (show(opt, 3)) {
1223 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1224 output_commit_title(opt, commit_a);
1225 } else if (show(opt, 2))
1226 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1227 else
1228 ; /* no output */
1229
1230 return 1;
1231 }
1232
1233 /*
1234 * Case #2: There are one or more merges that contain a and b in
1235 * the submodule. If there is only one, then present it as a
1236 * suggestion to the user, but leave it marked unmerged so the
1237 * user needs to confirm the resolution.
1238 */
1239
1240 /* Skip the search if makes no sense to the calling context. */
1241 if (!search)
1242 return 0;
1243
1244 /* find commit which merges them */
1245 parent_count = find_first_merges(opt->repo, &merges, path,
1246 commit_a, commit_b);
1247 switch (parent_count) {
1248 case 0:
1249 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1250 break;
1251
1252 case 1:
1253 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1254 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1255 print_commit((struct commit *) merges.objects[0].item);
1256 output(opt, 2, _(
1257 "If this is correct simply add it to the index "
1258 "for example\n"
1259 "by using:\n\n"
1260 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1261 "which will accept this suggestion.\n"),
1262 oid_to_hex(&merges.objects[0].item->oid), path);
1263 break;
1264
1265 default:
1266 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1267 for (i = 0; i < merges.nr; i++)
1268 print_commit((struct commit *) merges.objects[i].item);
1269 }
1270
1271 object_array_clear(&merges);
1272 return 0;
1273 }
1274
1275 static int merge_mode_and_contents(struct merge_options *opt,
1276 const struct diff_filespec *o,
1277 const struct diff_filespec *a,
1278 const struct diff_filespec *b,
1279 const char *filename,
1280 const char *branch1,
1281 const char *branch2,
1282 const int extra_marker_size,
1283 struct merge_file_info *result)
1284 {
1285 if (opt->branch1 != branch1) {
1286 /*
1287 * It's weird getting a reverse merge with HEAD on the bottom
1288 * side of the conflict markers and the other branch on the
1289 * top. Fix that.
1290 */
1291 return merge_mode_and_contents(opt, o, b, a,
1292 filename,
1293 branch2, branch1,
1294 extra_marker_size, result);
1295 }
1296
1297 result->merge = 0;
1298 result->clean = 1;
1299
1300 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1301 result->clean = 0;
1302 if (S_ISREG(a->mode)) {
1303 result->blob.mode = a->mode;
1304 oidcpy(&result->blob.oid, &a->oid);
1305 } else {
1306 result->blob.mode = b->mode;
1307 oidcpy(&result->blob.oid, &b->oid);
1308 }
1309 } else {
1310 if (!oid_eq(&a->oid, &o->oid) && !oid_eq(&b->oid, &o->oid))
1311 result->merge = 1;
1312
1313 /*
1314 * Merge modes
1315 */
1316 if (a->mode == b->mode || a->mode == o->mode)
1317 result->blob.mode = b->mode;
1318 else {
1319 result->blob.mode = a->mode;
1320 if (b->mode != o->mode) {
1321 result->clean = 0;
1322 result->merge = 1;
1323 }
1324 }
1325
1326 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &o->oid))
1327 oidcpy(&result->blob.oid, &b->oid);
1328 else if (oid_eq(&b->oid, &o->oid))
1329 oidcpy(&result->blob.oid, &a->oid);
1330 else if (S_ISREG(a->mode)) {
1331 mmbuffer_t result_buf;
1332 int ret = 0, merge_status;
1333
1334 merge_status = merge_3way(opt, &result_buf, o, a, b,
1335 branch1, branch2,
1336 extra_marker_size);
1337
1338 if ((merge_status < 0) || !result_buf.ptr)
1339 ret = err(opt, _("Failed to execute internal merge"));
1340
1341 if (!ret &&
1342 write_object_file(result_buf.ptr, result_buf.size,
1343 blob_type, &result->blob.oid))
1344 ret = err(opt, _("Unable to add %s to database"),
1345 a->path);
1346
1347 free(result_buf.ptr);
1348 if (ret)
1349 return ret;
1350 result->clean = (merge_status == 0);
1351 } else if (S_ISGITLINK(a->mode)) {
1352 result->clean = merge_submodule(opt, &result->blob.oid,
1353 o->path,
1354 &o->oid,
1355 &a->oid,
1356 &b->oid);
1357 } else if (S_ISLNK(a->mode)) {
1358 switch (opt->recursive_variant) {
1359 case MERGE_VARIANT_NORMAL:
1360 oidcpy(&result->blob.oid, &a->oid);
1361 if (!oid_eq(&a->oid, &b->oid))
1362 result->clean = 0;
1363 break;
1364 case MERGE_VARIANT_OURS:
1365 oidcpy(&result->blob.oid, &a->oid);
1366 break;
1367 case MERGE_VARIANT_THEIRS:
1368 oidcpy(&result->blob.oid, &b->oid);
1369 break;
1370 }
1371 } else
1372 BUG("unsupported object type in the tree");
1373 }
1374
1375 if (result->merge)
1376 output(opt, 2, _("Auto-merging %s"), filename);
1377
1378 return 0;
1379 }
1380
1381 static int handle_rename_via_dir(struct merge_options *opt,
1382 struct rename_conflict_info *ci)
1383 {
1384 /*
1385 * Handle file adds that need to be renamed due to directory rename
1386 * detection. This differs from handle_rename_normal, because
1387 * there is no content merge to do; just move the file into the
1388 * desired final location.
1389 */
1390 const struct rename *ren = ci->ren1;
1391 const struct diff_filespec *dest = ren->pair->two;
1392 char *file_path = dest->path;
1393 int mark_conflicted = (opt->detect_directory_renames ==
1394 MERGE_DIRECTORY_RENAMES_CONFLICT);
1395 assert(ren->dir_rename_original_dest);
1396
1397 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1398 mark_conflicted = 1;
1399 file_path = unique_path(opt, dest->path, ren->branch);
1400 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1401 "writing to %s instead."),
1402 dest->path, file_path);
1403 }
1404
1405 if (mark_conflicted) {
1406 /*
1407 * Write the file in worktree at file_path. In the index,
1408 * only record the file at dest->path in the appropriate
1409 * higher stage.
1410 */
1411 if (update_file(opt, 0, dest, file_path))
1412 return -1;
1413 if (file_path != dest->path)
1414 free(file_path);
1415 if (update_stages(opt, dest->path, NULL,
1416 ren->branch == opt->branch1 ? dest : NULL,
1417 ren->branch == opt->branch1 ? NULL : dest))
1418 return -1;
1419 return 0; /* not clean, but conflicted */
1420 } else {
1421 /* Update dest->path both in index and in worktree */
1422 if (update_file(opt, 1, dest, dest->path))
1423 return -1;
1424 return 1; /* clean */
1425 }
1426 }
1427
1428 static int handle_change_delete(struct merge_options *opt,
1429 const char *path, const char *old_path,
1430 const struct diff_filespec *o,
1431 const struct diff_filespec *changed,
1432 const char *change_branch,
1433 const char *delete_branch,
1434 const char *change, const char *change_past)
1435 {
1436 char *alt_path = NULL;
1437 const char *update_path = path;
1438 int ret = 0;
1439
1440 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1441 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1442 update_path = alt_path = unique_path(opt, path, change_branch);
1443 }
1444
1445 if (opt->priv->call_depth) {
1446 /*
1447 * We cannot arbitrarily accept either a_sha or b_sha as
1448 * correct; since there is no true "middle point" between
1449 * them, simply reuse the base version for virtual merge base.
1450 */
1451 ret = remove_file_from_index(opt->repo->index, path);
1452 if (!ret)
1453 ret = update_file(opt, 0, o, update_path);
1454 } else {
1455 /*
1456 * Despite the four nearly duplicate messages and argument
1457 * lists below and the ugliness of the nested if-statements,
1458 * having complete messages makes the job easier for
1459 * translators.
1460 *
1461 * The slight variance among the cases is due to the fact
1462 * that:
1463 * 1) directory/file conflicts (in effect if
1464 * !alt_path) could cause us to need to write the
1465 * file to a different path.
1466 * 2) renames (in effect if !old_path) could mean that
1467 * there are two names for the path that the user
1468 * may know the file by.
1469 */
1470 if (!alt_path) {
1471 if (!old_path) {
1472 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1473 "and %s in %s. Version %s of %s left in tree."),
1474 change, path, delete_branch, change_past,
1475 change_branch, change_branch, path);
1476 } else {
1477 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1478 "and %s to %s in %s. Version %s of %s left in tree."),
1479 change, old_path, delete_branch, change_past, path,
1480 change_branch, change_branch, path);
1481 }
1482 } else {
1483 if (!old_path) {
1484 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1485 "and %s in %s. Version %s of %s left in tree at %s."),
1486 change, path, delete_branch, change_past,
1487 change_branch, change_branch, path, alt_path);
1488 } else {
1489 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1490 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1491 change, old_path, delete_branch, change_past, path,
1492 change_branch, change_branch, path, alt_path);
1493 }
1494 }
1495 /*
1496 * No need to call update_file() on path when change_branch ==
1497 * opt->branch1 && !alt_path, since that would needlessly touch
1498 * path. We could call update_file_flags() with update_cache=0
1499 * and update_wd=0, but that's a no-op.
1500 */
1501 if (change_branch != opt->branch1 || alt_path)
1502 ret = update_file(opt, 0, changed, update_path);
1503 }
1504 free(alt_path);
1505
1506 return ret;
1507 }
1508
1509 static int handle_rename_delete(struct merge_options *opt,
1510 struct rename_conflict_info *ci)
1511 {
1512 const struct rename *ren = ci->ren1;
1513 const struct diff_filespec *orig = ren->pair->one;
1514 const struct diff_filespec *dest = ren->pair->two;
1515 const char *rename_branch = ren->branch;
1516 const char *delete_branch = (opt->branch1 == ren->branch ?
1517 opt->branch2 : opt->branch1);
1518
1519 if (handle_change_delete(opt,
1520 opt->priv->call_depth ? orig->path : dest->path,
1521 opt->priv->call_depth ? NULL : orig->path,
1522 orig, dest,
1523 rename_branch, delete_branch,
1524 _("rename"), _("renamed")))
1525 return -1;
1526
1527 if (opt->priv->call_depth)
1528 return remove_file_from_index(opt->repo->index, dest->path);
1529 else
1530 return update_stages(opt, dest->path, NULL,
1531 rename_branch == opt->branch1 ? dest : NULL,
1532 rename_branch == opt->branch1 ? NULL : dest);
1533 }
1534
1535 static int handle_file_collision(struct merge_options *opt,
1536 const char *collide_path,
1537 const char *prev_path1,
1538 const char *prev_path2,
1539 const char *branch1, const char *branch2,
1540 struct diff_filespec *a,
1541 struct diff_filespec *b)
1542 {
1543 struct merge_file_info mfi;
1544 struct diff_filespec null;
1545 char *alt_path = NULL;
1546 const char *update_path = collide_path;
1547
1548 /*
1549 * It's easiest to get the correct things into stage 2 and 3, and
1550 * to make sure that the content merge puts HEAD before the other
1551 * branch if we just ensure that branch1 == opt->branch1. So, simply
1552 * flip arguments around if we don't have that.
1553 */
1554 if (branch1 != opt->branch1) {
1555 return handle_file_collision(opt, collide_path,
1556 prev_path2, prev_path1,
1557 branch2, branch1,
1558 b, a);
1559 }
1560
1561 /*
1562 * In the recursive case, we just opt to undo renames
1563 */
1564 if (opt->priv->call_depth && (prev_path1 || prev_path2)) {
1565 /* Put first file (a->oid, a->mode) in its original spot */
1566 if (prev_path1) {
1567 if (update_file(opt, 1, a, prev_path1))
1568 return -1;
1569 } else {
1570 if (update_file(opt, 1, a, collide_path))
1571 return -1;
1572 }
1573
1574 /* Put second file (b->oid, b->mode) in its original spot */
1575 if (prev_path2) {
1576 if (update_file(opt, 1, b, prev_path2))
1577 return -1;
1578 } else {
1579 if (update_file(opt, 1, b, collide_path))
1580 return -1;
1581 }
1582
1583 /* Don't leave something at collision path if unrenaming both */
1584 if (prev_path1 && prev_path2)
1585 remove_file(opt, 1, collide_path, 0);
1586
1587 return 0;
1588 }
1589
1590 /* Remove rename sources if rename/add or rename/rename(2to1) */
1591 if (prev_path1)
1592 remove_file(opt, 1, prev_path1,
1593 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1594 if (prev_path2)
1595 remove_file(opt, 1, prev_path2,
1596 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1597
1598 /*
1599 * Remove the collision path, if it wouldn't cause dirty contents
1600 * or an untracked file to get lost. We'll either overwrite with
1601 * merged contents, or just write out to differently named files.
1602 */
1603 if (was_dirty(opt, collide_path)) {
1604 output(opt, 1, _("Refusing to lose dirty file at %s"),
1605 collide_path);
1606 update_path = alt_path = unique_path(opt, collide_path, "merged");
1607 } else if (would_lose_untracked(opt, collide_path)) {
1608 /*
1609 * Only way we get here is if both renames were from
1610 * a directory rename AND user had an untracked file
1611 * at the location where both files end up after the
1612 * two directory renames. See testcase 10d of t6043.
1613 */
1614 output(opt, 1, _("Refusing to lose untracked file at "
1615 "%s, even though it's in the way."),
1616 collide_path);
1617 update_path = alt_path = unique_path(opt, collide_path, "merged");
1618 } else {
1619 /*
1620 * FIXME: It's possible that the two files are identical
1621 * and that the current working copy happens to match, in
1622 * which case we are unnecessarily touching the working
1623 * tree file. It's not a likely enough scenario that I
1624 * want to code up the checks for it and a better fix is
1625 * available if we restructure how unpack_trees() and
1626 * merge-recursive interoperate anyway, so punting for
1627 * now...
1628 */
1629 remove_file(opt, 0, collide_path, 0);
1630 }
1631
1632 /* Store things in diff_filespecs for functions that need it */
1633 null.path = (char *)collide_path;
1634 oidcpy(&null.oid, &null_oid);
1635 null.mode = 0;
1636
1637 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1638 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1639 return -1;
1640 mfi.clean &= !alt_path;
1641 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1642 return -1;
1643 if (!mfi.clean && !opt->priv->call_depth &&
1644 update_stages(opt, collide_path, NULL, a, b))
1645 return -1;
1646 free(alt_path);
1647 /*
1648 * FIXME: If both a & b both started with conflicts (only possible
1649 * if they came from a rename/rename(2to1)), but had IDENTICAL
1650 * contents including those conflicts, then in the next line we claim
1651 * it was clean. If someone cares about this case, we should have the
1652 * caller notify us if we started with conflicts.
1653 */
1654 return mfi.clean;
1655 }
1656
1657 static int handle_rename_add(struct merge_options *opt,
1658 struct rename_conflict_info *ci)
1659 {
1660 /* a was renamed to c, and a separate c was added. */
1661 struct diff_filespec *a = ci->ren1->pair->one;
1662 struct diff_filespec *c = ci->ren1->pair->two;
1663 char *path = c->path;
1664 char *prev_path_desc;
1665 struct merge_file_info mfi;
1666
1667 const char *rename_branch = ci->ren1->branch;
1668 const char *add_branch = (opt->branch1 == rename_branch ?
1669 opt->branch2 : opt->branch1);
1670 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1671
1672 output(opt, 1, _("CONFLICT (rename/add): "
1673 "Rename %s->%s in %s. Added %s in %s"),
1674 a->path, c->path, rename_branch,
1675 c->path, add_branch);
1676
1677 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1678 ci->ren1->src_entry->stages[other_stage].path = a->path;
1679 if (merge_mode_and_contents(opt, a, c,
1680 &ci->ren1->src_entry->stages[other_stage],
1681 prev_path_desc,
1682 opt->branch1, opt->branch2,
1683 1 + opt->priv->call_depth * 2, &mfi))
1684 return -1;
1685 free(prev_path_desc);
1686
1687 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1688 return handle_file_collision(opt,
1689 c->path, a->path, NULL,
1690 rename_branch, add_branch,
1691 &mfi.blob,
1692 &ci->ren1->dst_entry->stages[other_stage]);
1693 }
1694
1695 static char *find_path_for_conflict(struct merge_options *opt,
1696 const char *path,
1697 const char *branch1,
1698 const char *branch2)
1699 {
1700 char *new_path = NULL;
1701 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1702 new_path = unique_path(opt, path, branch1);
1703 output(opt, 1, _("%s is a directory in %s adding "
1704 "as %s instead"),
1705 path, branch2, new_path);
1706 } else if (would_lose_untracked(opt, path)) {
1707 new_path = unique_path(opt, path, branch1);
1708 output(opt, 1, _("Refusing to lose untracked file"
1709 " at %s; adding as %s instead"),
1710 path, new_path);
1711 }
1712
1713 return new_path;
1714 }
1715
1716 static int handle_rename_rename_1to2(struct merge_options *opt,
1717 struct rename_conflict_info *ci)
1718 {
1719 /* One file was renamed in both branches, but to different names. */
1720 struct merge_file_info mfi;
1721 struct diff_filespec *add;
1722 struct diff_filespec *o = ci->ren1->pair->one;
1723 struct diff_filespec *a = ci->ren1->pair->two;
1724 struct diff_filespec *b = ci->ren2->pair->two;
1725 char *path_desc;
1726
1727 output(opt, 1, _("CONFLICT (rename/rename): "
1728 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1729 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1730 o->path, a->path, ci->ren1->branch,
1731 o->path, b->path, ci->ren2->branch,
1732 opt->priv->call_depth ? _(" (left unresolved)") : "");
1733
1734 path_desc = xstrfmt("%s and %s, both renamed from %s",
1735 a->path, b->path, o->path);
1736 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1737 ci->ren1->branch, ci->ren2->branch,
1738 opt->priv->call_depth * 2, &mfi))
1739 return -1;
1740 free(path_desc);
1741
1742 if (opt->priv->call_depth) {
1743 /*
1744 * FIXME: For rename/add-source conflicts (if we could detect
1745 * such), this is wrong. We should instead find a unique
1746 * pathname and then either rename the add-source file to that
1747 * unique path, or use that unique path instead of src here.
1748 */
1749 if (update_file(opt, 0, &mfi.blob, o->path))
1750 return -1;
1751
1752 /*
1753 * Above, we put the merged content at the merge-base's
1754 * path. Now we usually need to delete both a->path and
1755 * b->path. However, the rename on each side of the merge
1756 * could also be involved in a rename/add conflict. In
1757 * such cases, we should keep the added file around,
1758 * resolving the conflict at that path in its favor.
1759 */
1760 add = &ci->ren1->dst_entry->stages[2 ^ 1];
1761 if (is_valid(add)) {
1762 if (update_file(opt, 0, add, a->path))
1763 return -1;
1764 }
1765 else
1766 remove_file_from_index(opt->repo->index, a->path);
1767 add = &ci->ren2->dst_entry->stages[3 ^ 1];
1768 if (is_valid(add)) {
1769 if (update_file(opt, 0, add, b->path))
1770 return -1;
1771 }
1772 else
1773 remove_file_from_index(opt->repo->index, b->path);
1774 } else {
1775 /*
1776 * For each destination path, we need to see if there is a
1777 * rename/add collision. If not, we can write the file out
1778 * to the specified location.
1779 */
1780 add = &ci->ren1->dst_entry->stages[2 ^ 1];
1781 if (is_valid(add)) {
1782 add->path = mfi.blob.path = a->path;
1783 if (handle_file_collision(opt, a->path,
1784 NULL, NULL,
1785 ci->ren1->branch,
1786 ci->ren2->branch,
1787 &mfi.blob, add) < 0)
1788 return -1;
1789 } else {
1790 char *new_path = find_path_for_conflict(opt, a->path,
1791 ci->ren1->branch,
1792 ci->ren2->branch);
1793 if (update_file(opt, 0, &mfi.blob,
1794 new_path ? new_path : a->path))
1795 return -1;
1796 free(new_path);
1797 if (update_stages(opt, a->path, NULL, a, NULL))
1798 return -1;
1799 }
1800
1801 add = &ci->ren2->dst_entry->stages[3 ^ 1];
1802 if (is_valid(add)) {
1803 add->path = mfi.blob.path = b->path;
1804 if (handle_file_collision(opt, b->path,
1805 NULL, NULL,
1806 ci->ren1->branch,
1807 ci->ren2->branch,
1808 add, &mfi.blob) < 0)
1809 return -1;
1810 } else {
1811 char *new_path = find_path_for_conflict(opt, b->path,
1812 ci->ren2->branch,
1813 ci->ren1->branch);
1814 if (update_file(opt, 0, &mfi.blob,
1815 new_path ? new_path : b->path))
1816 return -1;
1817 free(new_path);
1818 if (update_stages(opt, b->path, NULL, NULL, b))
1819 return -1;
1820 }
1821 }
1822
1823 return 0;
1824 }
1825
1826 static int handle_rename_rename_2to1(struct merge_options *opt,
1827 struct rename_conflict_info *ci)
1828 {
1829 /* Two files, a & b, were renamed to the same thing, c. */
1830 struct diff_filespec *a = ci->ren1->pair->one;
1831 struct diff_filespec *b = ci->ren2->pair->one;
1832 struct diff_filespec *c1 = ci->ren1->pair->two;
1833 struct diff_filespec *c2 = ci->ren2->pair->two;
1834 char *path = c1->path; /* == c2->path */
1835 char *path_side_1_desc;
1836 char *path_side_2_desc;
1837 struct merge_file_info mfi_c1;
1838 struct merge_file_info mfi_c2;
1839 int ostage1, ostage2;
1840
1841 output(opt, 1, _("CONFLICT (rename/rename): "
1842 "Rename %s->%s in %s. "
1843 "Rename %s->%s in %s"),
1844 a->path, c1->path, ci->ren1->branch,
1845 b->path, c2->path, ci->ren2->branch);
1846
1847 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1848 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1849 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1850 ostage2 = ostage1 ^ 1;
1851 ci->ren1->src_entry->stages[ostage1].path = a->path;
1852 ci->ren2->src_entry->stages[ostage2].path = b->path;
1853 if (merge_mode_and_contents(opt, a, c1,
1854 &ci->ren1->src_entry->stages[ostage1],
1855 path_side_1_desc,
1856 opt->branch1, opt->branch2,
1857 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1858 merge_mode_and_contents(opt, b,
1859 &ci->ren2->src_entry->stages[ostage2],
1860 c2, path_side_2_desc,
1861 opt->branch1, opt->branch2,
1862 1 + opt->priv->call_depth * 2, &mfi_c2))
1863 return -1;
1864 free(path_side_1_desc);
1865 free(path_side_2_desc);
1866 mfi_c1.blob.path = path;
1867 mfi_c2.blob.path = path;
1868
1869 return handle_file_collision(opt, path, a->path, b->path,
1870 ci->ren1->branch, ci->ren2->branch,
1871 &mfi_c1.blob, &mfi_c2.blob);
1872 }
1873
1874 /*
1875 * Get the diff_filepairs changed between o_tree and tree.
1876 */
1877 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1878 struct tree *o_tree,
1879 struct tree *tree)
1880 {
1881 struct diff_queue_struct *ret;
1882 struct diff_options opts;
1883
1884 repo_diff_setup(opt->repo, &opts);
1885 opts.flags.recursive = 1;
1886 opts.flags.rename_empty = 0;
1887 opts.detect_rename = merge_detect_rename(opt);
1888 /*
1889 * We do not have logic to handle the detection of copies. In
1890 * fact, it may not even make sense to add such logic: would we
1891 * really want a change to a base file to be propagated through
1892 * multiple other files by a merge?
1893 */
1894 if (opts.detect_rename > DIFF_DETECT_RENAME)
1895 opts.detect_rename = DIFF_DETECT_RENAME;
1896 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 1000;
1897 opts.rename_score = opt->rename_score;
1898 opts.show_rename_progress = opt->show_rename_progress;
1899 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1900 diff_setup_done(&opts);
1901 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1902 diffcore_std(&opts);
1903 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1904 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1905
1906 ret = xmalloc(sizeof(*ret));
1907 *ret = diff_queued_diff;
1908
1909 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1910 diff_queued_diff.nr = 0;
1911 diff_queued_diff.queue = NULL;
1912 diff_flush(&opts);
1913 return ret;
1914 }
1915
1916 static int tree_has_path(struct repository *r, struct tree *tree,
1917 const char *path)
1918 {
1919 struct object_id hashy;
1920 unsigned short mode_o;
1921
1922 return !get_tree_entry(r,
1923 &tree->object.oid, path,
1924 &hashy, &mode_o);
1925 }
1926
1927 /*
1928 * Return a new string that replaces the beginning portion (which matches
1929 * entry->dir), with entry->new_dir. In perl-speak:
1930 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1931 * NOTE:
1932 * Caller must ensure that old_path starts with entry->dir + '/'.
1933 */
1934 static char *apply_dir_rename(struct dir_rename_entry *entry,
1935 const char *old_path)
1936 {
1937 struct strbuf new_path = STRBUF_INIT;
1938 int oldlen, newlen;
1939
1940 if (entry->non_unique_new_dir)
1941 return NULL;
1942
1943 oldlen = strlen(entry->dir);
1944 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1945 strbuf_grow(&new_path, newlen);
1946 strbuf_addbuf(&new_path, &entry->new_dir);
1947 strbuf_addstr(&new_path, &old_path[oldlen]);
1948
1949 return strbuf_detach(&new_path, NULL);
1950 }
1951
1952 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1953 char **old_dir, char **new_dir)
1954 {
1955 char *end_of_old, *end_of_new;
1956 int old_len, new_len;
1957
1958 *old_dir = NULL;
1959 *new_dir = NULL;
1960
1961 /*
1962 * For
1963 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1964 * the "e/foo.c" part is the same, we just want to know that
1965 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1966 * so, for this example, this function returns "a/b/c/d" in
1967 * *old_dir and "a/b/some/thing/else" in *new_dir.
1968 *
1969 * Also, if the basename of the file changed, we don't care. We
1970 * want to know which portion of the directory, if any, changed.
1971 */
1972 end_of_old = strrchr(old_path, '/');
1973 end_of_new = strrchr(new_path, '/');
1974
1975 if (end_of_old == NULL || end_of_new == NULL)
1976 return;
1977 while (*--end_of_new == *--end_of_old &&
1978 end_of_old != old_path &&
1979 end_of_new != new_path)
1980 ; /* Do nothing; all in the while loop */
1981 /*
1982 * We've found the first non-matching character in the directory
1983 * paths. That means the current directory we were comparing
1984 * represents the rename. Move end_of_old and end_of_new back
1985 * to the full directory name.
1986 */
1987 if (*end_of_old == '/')
1988 end_of_old++;
1989 if (*end_of_old != '/')
1990 end_of_new++;
1991 end_of_old = strchr(end_of_old, '/');
1992 end_of_new = strchr(end_of_new, '/');
1993
1994 /*
1995 * It may have been the case that old_path and new_path were the same
1996 * directory all along. Don't claim a rename if they're the same.
1997 */
1998 old_len = end_of_old - old_path;
1999 new_len = end_of_new - new_path;
2000
2001 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
2002 *old_dir = xstrndup(old_path, old_len);
2003 *new_dir = xstrndup(new_path, new_len);
2004 }
2005 }
2006
2007 static void remove_hashmap_entries(struct hashmap *dir_renames,
2008 struct string_list *items_to_remove)
2009 {
2010 int i;
2011 struct dir_rename_entry *entry;
2012
2013 for (i = 0; i < items_to_remove->nr; i++) {
2014 entry = items_to_remove->items[i].util;
2015 hashmap_remove(dir_renames, entry, NULL);
2016 }
2017 string_list_clear(items_to_remove, 0);
2018 }
2019
2020 /*
2021 * See if there is a directory rename for path, and if there are any file
2022 * level conflicts for the renamed location. If there is a rename and
2023 * there are no conflicts, return the new name. Otherwise, return NULL.
2024 */
2025 static char *handle_path_level_conflicts(struct merge_options *opt,
2026 const char *path,
2027 struct dir_rename_entry *entry,
2028 struct hashmap *collisions,
2029 struct tree *tree)
2030 {
2031 char *new_path = NULL;
2032 struct collision_entry *collision_ent;
2033 int clean = 1;
2034 struct strbuf collision_paths = STRBUF_INIT;
2035
2036 /*
2037 * entry has the mapping of old directory name to new directory name
2038 * that we want to apply to path.
2039 */
2040 new_path = apply_dir_rename(entry, path);
2041
2042 if (!new_path) {
2043 /* This should only happen when entry->non_unique_new_dir set */
2044 if (!entry->non_unique_new_dir)
2045 BUG("entry->non_unqiue_dir not set and !new_path");
2046 output(opt, 1, _("CONFLICT (directory rename split): "
2047 "Unclear where to place %s because directory "
2048 "%s was renamed to multiple other directories, "
2049 "with no destination getting a majority of the "
2050 "files."),
2051 path, entry->dir);
2052 clean = 0;
2053 return NULL;
2054 }
2055
2056 /*
2057 * The caller needs to have ensured that it has pre-populated
2058 * collisions with all paths that map to new_path. Do a quick check
2059 * to ensure that's the case.
2060 */
2061 collision_ent = collision_find_entry(collisions, new_path);
2062 if (collision_ent == NULL)
2063 BUG("collision_ent is NULL");
2064
2065 /*
2066 * Check for one-sided add/add/.../add conflicts, i.e.
2067 * where implicit renames from the other side doing
2068 * directory rename(s) can affect this side of history
2069 * to put multiple paths into the same location. Warn
2070 * and bail on directory renames for such paths.
2071 */
2072 if (collision_ent->reported_already) {
2073 clean = 0;
2074 } else if (tree_has_path(opt->repo, tree, new_path)) {
2075 collision_ent->reported_already = 1;
2076 strbuf_add_separated_string_list(&collision_paths, ", ",
2077 &collision_ent->source_files);
2078 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2079 "file/dir at %s in the way of implicit "
2080 "directory rename(s) putting the following "
2081 "path(s) there: %s."),
2082 new_path, collision_paths.buf);
2083 clean = 0;
2084 } else if (collision_ent->source_files.nr > 1) {
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): Cannot map "
2089 "more than one path to %s; implicit directory "
2090 "renames tried to put these paths there: %s"),
2091 new_path, collision_paths.buf);
2092 clean = 0;
2093 }
2094
2095 /* Free memory we no longer need */
2096 strbuf_release(&collision_paths);
2097 if (!clean && new_path) {
2098 free(new_path);
2099 return NULL;
2100 }
2101
2102 return new_path;
2103 }
2104
2105 /*
2106 * There are a couple things we want to do at the directory level:
2107 * 1. Check for both sides renaming to the same thing, in order to avoid
2108 * implicit renaming of files that should be left in place. (See
2109 * testcase 6b in t6043 for details.)
2110 * 2. Prune directory renames if there are still files left in the
2111 * the original directory. These represent a partial directory rename,
2112 * i.e. a rename where only some of the files within the directory
2113 * were renamed elsewhere. (Technically, this could be done earlier
2114 * in get_directory_renames(), except that would prevent us from
2115 * doing the previous check and thus failing testcase 6b.)
2116 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2117 * In the future, we could potentially record this info as well and
2118 * omit reporting rename/rename(1to2) conflicts for each path within
2119 * the affected directories, thus cleaning up the merge output.
2120 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2121 * directory level, because merging directories is fine. If it
2122 * causes conflicts for files within those merged directories, then
2123 * that should be detected at the individual path level.
2124 */
2125 static void handle_directory_level_conflicts(struct merge_options *opt,
2126 struct hashmap *dir_re_head,
2127 struct tree *head,
2128 struct hashmap *dir_re_merge,
2129 struct tree *merge)
2130 {
2131 struct hashmap_iter iter;
2132 struct dir_rename_entry *head_ent;
2133 struct dir_rename_entry *merge_ent;
2134
2135 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2136 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2137
2138 hashmap_iter_init(dir_re_head, &iter);
2139 while ((head_ent = hashmap_iter_next(&iter))) {
2140 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2141 if (merge_ent &&
2142 !head_ent->non_unique_new_dir &&
2143 !merge_ent->non_unique_new_dir &&
2144 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2145 /* 1. Renamed identically; remove it from both sides */
2146 string_list_append(&remove_from_head,
2147 head_ent->dir)->util = head_ent;
2148 strbuf_release(&head_ent->new_dir);
2149 string_list_append(&remove_from_merge,
2150 merge_ent->dir)->util = merge_ent;
2151 strbuf_release(&merge_ent->new_dir);
2152 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2153 /* 2. This wasn't a directory rename after all */
2154 string_list_append(&remove_from_head,
2155 head_ent->dir)->util = head_ent;
2156 strbuf_release(&head_ent->new_dir);
2157 }
2158 }
2159
2160 remove_hashmap_entries(dir_re_head, &remove_from_head);
2161 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2162
2163 hashmap_iter_init(dir_re_merge, &iter);
2164 while ((merge_ent = hashmap_iter_next(&iter))) {
2165 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2166 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2167 /* 2. This wasn't a directory rename after all */
2168 string_list_append(&remove_from_merge,
2169 merge_ent->dir)->util = merge_ent;
2170 } else if (head_ent &&
2171 !head_ent->non_unique_new_dir &&
2172 !merge_ent->non_unique_new_dir) {
2173 /* 3. rename/rename(1to2) */
2174 /*
2175 * We can assume it's not rename/rename(1to1) because
2176 * that was case (1), already checked above. So we
2177 * know that head_ent->new_dir and merge_ent->new_dir
2178 * are different strings.
2179 */
2180 output(opt, 1, _("CONFLICT (rename/rename): "
2181 "Rename directory %s->%s in %s. "
2182 "Rename directory %s->%s in %s"),
2183 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2184 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2185 string_list_append(&remove_from_head,
2186 head_ent->dir)->util = head_ent;
2187 strbuf_release(&head_ent->new_dir);
2188 string_list_append(&remove_from_merge,
2189 merge_ent->dir)->util = merge_ent;
2190 strbuf_release(&merge_ent->new_dir);
2191 }
2192 }
2193
2194 remove_hashmap_entries(dir_re_head, &remove_from_head);
2195 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2196 }
2197
2198 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2199 {
2200 struct hashmap *dir_renames;
2201 struct hashmap_iter iter;
2202 struct dir_rename_entry *entry;
2203 int i;
2204
2205 /*
2206 * Typically, we think of a directory rename as all files from a
2207 * certain directory being moved to a target directory. However,
2208 * what if someone first moved two files from the original
2209 * directory in one commit, and then renamed the directory
2210 * somewhere else in a later commit? At merge time, we just know
2211 * that files from the original directory went to two different
2212 * places, and that the bulk of them ended up in the same place.
2213 * We want each directory rename to represent where the bulk of the
2214 * files from that directory end up; this function exists to find
2215 * where the bulk of the files went.
2216 *
2217 * The first loop below simply iterates through the list of file
2218 * renames, finding out how often each directory rename pair
2219 * possibility occurs.
2220 */
2221 dir_renames = xmalloc(sizeof(*dir_renames));
2222 dir_rename_init(dir_renames);
2223 for (i = 0; i < pairs->nr; ++i) {
2224 struct string_list_item *item;
2225 int *count;
2226 struct diff_filepair *pair = pairs->queue[i];
2227 char *old_dir, *new_dir;
2228
2229 /* File not part of directory rename if it wasn't renamed */
2230 if (pair->status != 'R')
2231 continue;
2232
2233 get_renamed_dir_portion(pair->one->path, pair->two->path,
2234 &old_dir, &new_dir);
2235 if (!old_dir)
2236 /* Directory didn't change at all; ignore this one. */
2237 continue;
2238
2239 entry = dir_rename_find_entry(dir_renames, old_dir);
2240 if (!entry) {
2241 entry = xmalloc(sizeof(*entry));
2242 dir_rename_entry_init(entry, old_dir);
2243 hashmap_put(dir_renames, entry);
2244 } else {
2245 free(old_dir);
2246 }
2247 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2248 if (!item) {
2249 item = string_list_insert(&entry->possible_new_dirs,
2250 new_dir);
2251 item->util = xcalloc(1, sizeof(int));
2252 } else {
2253 free(new_dir);
2254 }
2255 count = item->util;
2256 *count += 1;
2257 }
2258
2259 /*
2260 * For each directory with files moved out of it, we find out which
2261 * target directory received the most files so we can declare it to
2262 * be the "winning" target location for the directory rename. This
2263 * winner gets recorded in new_dir. If there is no winner
2264 * (multiple target directories received the same number of files),
2265 * we set non_unique_new_dir. Once we've determined the winner (or
2266 * that there is no winner), we no longer need possible_new_dirs.
2267 */
2268 hashmap_iter_init(dir_renames, &iter);
2269 while ((entry = hashmap_iter_next(&iter))) {
2270 int max = 0;
2271 int bad_max = 0;
2272 char *best = NULL;
2273
2274 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2275 int *count = entry->possible_new_dirs.items[i].util;
2276
2277 if (*count == max)
2278 bad_max = max;
2279 else if (*count > max) {
2280 max = *count;
2281 best = entry->possible_new_dirs.items[i].string;
2282 }
2283 }
2284 if (bad_max == max)
2285 entry->non_unique_new_dir = 1;
2286 else {
2287 assert(entry->new_dir.len == 0);
2288 strbuf_addstr(&entry->new_dir, best);
2289 }
2290 /*
2291 * The relevant directory sub-portion of the original full
2292 * filepaths were xstrndup'ed before inserting into
2293 * possible_new_dirs, and instead of manually iterating the
2294 * list and free'ing each, just lie and tell
2295 * possible_new_dirs that it did the strdup'ing so that it
2296 * will free them for us.
2297 */
2298 entry->possible_new_dirs.strdup_strings = 1;
2299 string_list_clear(&entry->possible_new_dirs, 1);
2300 }
2301
2302 return dir_renames;
2303 }
2304
2305 static struct dir_rename_entry *check_dir_renamed(const char *path,
2306 struct hashmap *dir_renames)
2307 {
2308 char *temp = xstrdup(path);
2309 char *end;
2310 struct dir_rename_entry *entry = NULL;
2311
2312 while ((end = strrchr(temp, '/'))) {
2313 *end = '\0';
2314 entry = dir_rename_find_entry(dir_renames, temp);
2315 if (entry)
2316 break;
2317 }
2318 free(temp);
2319 return entry;
2320 }
2321
2322 static void compute_collisions(struct hashmap *collisions,
2323 struct hashmap *dir_renames,
2324 struct diff_queue_struct *pairs)
2325 {
2326 int i;
2327
2328 /*
2329 * Multiple files can be mapped to the same path due to directory
2330 * renames done by the other side of history. Since that other
2331 * side of history could have merged multiple directories into one,
2332 * if our side of history added the same file basename to each of
2333 * those directories, then all N of them would get implicitly
2334 * renamed by the directory rename detection into the same path,
2335 * and we'd get an add/add/.../add conflict, and all those adds
2336 * from *this* side of history. This is not representable in the
2337 * index, and users aren't going to easily be able to make sense of
2338 * it. So we need to provide a good warning about what's
2339 * happening, and fall back to no-directory-rename detection
2340 * behavior for those paths.
2341 *
2342 * See testcases 9e and all of section 5 from t6043 for examples.
2343 */
2344 collision_init(collisions);
2345
2346 for (i = 0; i < pairs->nr; ++i) {
2347 struct dir_rename_entry *dir_rename_ent;
2348 struct collision_entry *collision_ent;
2349 char *new_path;
2350 struct diff_filepair *pair = pairs->queue[i];
2351
2352 if (pair->status != 'A' && pair->status != 'R')
2353 continue;
2354 dir_rename_ent = check_dir_renamed(pair->two->path,
2355 dir_renames);
2356 if (!dir_rename_ent)
2357 continue;
2358
2359 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2360 if (!new_path)
2361 /*
2362 * dir_rename_ent->non_unique_new_path is true, which
2363 * means there is no directory rename for us to use,
2364 * which means it won't cause us any additional
2365 * collisions.
2366 */
2367 continue;
2368 collision_ent = collision_find_entry(collisions, new_path);
2369 if (!collision_ent) {
2370 collision_ent = xcalloc(1,
2371 sizeof(struct collision_entry));
2372 hashmap_entry_init(collision_ent, strhash(new_path));
2373 hashmap_put(collisions, collision_ent);
2374 collision_ent->target_file = new_path;
2375 } else {
2376 free(new_path);
2377 }
2378 string_list_insert(&collision_ent->source_files,
2379 pair->two->path);
2380 }
2381 }
2382
2383 static char *check_for_directory_rename(struct merge_options *opt,
2384 const char *path,
2385 struct tree *tree,
2386 struct hashmap *dir_renames,
2387 struct hashmap *dir_rename_exclusions,
2388 struct hashmap *collisions,
2389 int *clean_merge)
2390 {
2391 char *new_path = NULL;
2392 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2393 struct dir_rename_entry *oentry = NULL;
2394
2395 if (!entry)
2396 return new_path;
2397
2398 /*
2399 * This next part is a little weird. We do not want to do an
2400 * implicit rename into a directory we renamed on our side, because
2401 * that will result in a spurious rename/rename(1to2) conflict. An
2402 * example:
2403 * Base commit: dumbdir/afile, otherdir/bfile
2404 * Side 1: smrtdir/afile, otherdir/bfile
2405 * Side 2: dumbdir/afile, dumbdir/bfile
2406 * Here, while working on Side 1, we could notice that otherdir was
2407 * renamed/merged to dumbdir, and change the diff_filepair for
2408 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2409 * 2 will notice the rename from dumbdir to smrtdir, and do the
2410 * transitive rename to move it from dumbdir/bfile to
2411 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2412 * smrtdir, a rename/rename(1to2) conflict. We really just want
2413 * the file to end up in smrtdir. And the way to achieve that is
2414 * to not let Side1 do the rename to dumbdir, since we know that is
2415 * the source of one of our directory renames.
2416 *
2417 * That's why oentry and dir_rename_exclusions is here.
2418 *
2419 * As it turns out, this also prevents N-way transient rename
2420 * confusion; See testcases 9c and 9d of t6043.
2421 */
2422 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2423 if (oentry) {
2424 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2425 "to %s, because %s itself was renamed."),
2426 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2427 } else {
2428 new_path = handle_path_level_conflicts(opt, path, entry,
2429 collisions, tree);
2430 *clean_merge &= (new_path != NULL);
2431 }
2432
2433 return new_path;
2434 }
2435
2436 static void apply_directory_rename_modifications(struct merge_options *opt,
2437 struct diff_filepair *pair,
2438 char *new_path,
2439 struct rename *re,
2440 struct tree *tree,
2441 struct tree *o_tree,
2442 struct tree *a_tree,
2443 struct tree *b_tree,
2444 struct string_list *entries)
2445 {
2446 struct string_list_item *item;
2447 int stage = (tree == a_tree ? 2 : 3);
2448 int update_wd;
2449
2450 /*
2451 * In all cases where we can do directory rename detection,
2452 * unpack_trees() will have read pair->two->path into the
2453 * index and the working copy. We need to remove it so that
2454 * we can instead place it at new_path. It is guaranteed to
2455 * not be untracked (unpack_trees() would have errored out
2456 * saying the file would have been overwritten), but it might
2457 * be dirty, though.
2458 */
2459 update_wd = !was_dirty(opt, pair->two->path);
2460 if (!update_wd)
2461 output(opt, 1, _("Refusing to lose dirty file at %s"),
2462 pair->two->path);
2463 remove_file(opt, 1, pair->two->path, !update_wd);
2464
2465 /* Find or create a new re->dst_entry */
2466 item = string_list_lookup(entries, new_path);
2467 if (item) {
2468 /*
2469 * Since we're renaming on this side of history, and it's
2470 * due to a directory rename on the other side of history
2471 * (which we only allow when the directory in question no
2472 * longer exists on the other side of history), the
2473 * original entry for re->dst_entry is no longer
2474 * necessary...
2475 */
2476 re->dst_entry->processed = 1;
2477
2478 /*
2479 * ...because we'll be using this new one.
2480 */
2481 re->dst_entry = item->util;
2482 } else {
2483 /*
2484 * re->dst_entry is for the before-dir-rename path, and we
2485 * need it to hold information for the after-dir-rename
2486 * path. Before creating a new entry, we need to mark the
2487 * old one as unnecessary (...unless it is shared by
2488 * src_entry, i.e. this didn't use to be a rename, in which
2489 * case we can just allow the normal processing to happen
2490 * for it).
2491 */
2492 if (pair->status == 'R')
2493 re->dst_entry->processed = 1;
2494
2495 re->dst_entry = insert_stage_data(opt->repo, new_path,
2496 o_tree, a_tree, b_tree,
2497 entries);
2498 item = string_list_insert(entries, new_path);
2499 item->util = re->dst_entry;
2500 }
2501
2502 /*
2503 * Update the stage_data with the information about the path we are
2504 * moving into place. That slot will be empty and available for us
2505 * to write to because of the collision checks in
2506 * handle_path_level_conflicts(). In other words,
2507 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2508 * open for us to write to.
2509 *
2510 * It may be tempting to actually update the index at this point as
2511 * well, using update_stages_for_stage_data(), but as per the big
2512 * "NOTE" in update_stages(), doing so will modify the current
2513 * in-memory index which will break calls to would_lose_untracked()
2514 * that we need to make. Instead, we need to just make sure that
2515 * the various handle_rename_*() functions update the index
2516 * explicitly rather than relying on unpack_trees() to have done it.
2517 */
2518 get_tree_entry(opt->repo,
2519 &tree->object.oid,
2520 pair->two->path,
2521 &re->dst_entry->stages[stage].oid,
2522 &re->dst_entry->stages[stage].mode);
2523
2524 /*
2525 * Record the original change status (or 'type' of change). If it
2526 * was originally an add ('A'), this lets us differentiate later
2527 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2528 * otherwise look the same). If it was originally a rename ('R'),
2529 * this lets us remember and report accurately about the transitive
2530 * renaming that occurred via the directory rename detection. Also,
2531 * record the original destination name.
2532 */
2533 re->dir_rename_original_type = pair->status;
2534 re->dir_rename_original_dest = pair->two->path;
2535
2536 /*
2537 * We don't actually look at pair->status again, but it seems
2538 * pedagogically correct to adjust it.
2539 */
2540 pair->status = 'R';
2541
2542 /*
2543 * Finally, record the new location.
2544 */
2545 pair->two->path = new_path;
2546 }
2547
2548 /*
2549 * Get information of all renames which occurred in 'pairs', making use of
2550 * any implicit directory renames inferred from the other side of history.
2551 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2552 * to be able to associate the correct cache entries with the rename
2553 * information; tree is always equal to either a_tree or b_tree.
2554 */
2555 static struct string_list *get_renames(struct merge_options *opt,
2556 const char *branch,
2557 struct diff_queue_struct *pairs,
2558 struct hashmap *dir_renames,
2559 struct hashmap *dir_rename_exclusions,
2560 struct tree *tree,
2561 struct tree *o_tree,
2562 struct tree *a_tree,
2563 struct tree *b_tree,
2564 struct string_list *entries,
2565 int *clean_merge)
2566 {
2567 int i;
2568 struct hashmap collisions;
2569 struct hashmap_iter iter;
2570 struct collision_entry *e;
2571 struct string_list *renames;
2572
2573 compute_collisions(&collisions, dir_renames, pairs);
2574 renames = xcalloc(1, sizeof(struct string_list));
2575
2576 for (i = 0; i < pairs->nr; ++i) {
2577 struct string_list_item *item;
2578 struct rename *re;
2579 struct diff_filepair *pair = pairs->queue[i];
2580 char *new_path; /* non-NULL only with directory renames */
2581
2582 if (pair->status != 'A' && pair->status != 'R') {
2583 diff_free_filepair(pair);
2584 continue;
2585 }
2586 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2587 dir_renames,
2588 dir_rename_exclusions,
2589 &collisions,
2590 clean_merge);
2591 if (pair->status != 'R' && !new_path) {
2592 diff_free_filepair(pair);
2593 continue;
2594 }
2595
2596 re = xmalloc(sizeof(*re));
2597 re->processed = 0;
2598 re->pair = pair;
2599 re->branch = branch;
2600 re->dir_rename_original_type = '\0';
2601 re->dir_rename_original_dest = NULL;
2602 item = string_list_lookup(entries, re->pair->one->path);
2603 if (!item)
2604 re->src_entry = insert_stage_data(opt->repo,
2605 re->pair->one->path,
2606 o_tree, a_tree, b_tree, entries);
2607 else
2608 re->src_entry = item->util;
2609
2610 item = string_list_lookup(entries, re->pair->two->path);
2611 if (!item)
2612 re->dst_entry = insert_stage_data(opt->repo,
2613 re->pair->two->path,
2614 o_tree, a_tree, b_tree, entries);
2615 else
2616 re->dst_entry = item->util;
2617 item = string_list_insert(renames, pair->one->path);
2618 item->util = re;
2619 if (new_path)
2620 apply_directory_rename_modifications(opt, pair, new_path,
2621 re, tree, o_tree,
2622 a_tree, b_tree,
2623 entries);
2624 }
2625
2626 hashmap_iter_init(&collisions, &iter);
2627 while ((e = hashmap_iter_next(&iter))) {
2628 free(e->target_file);
2629 string_list_clear(&e->source_files, 0);
2630 }
2631 hashmap_free(&collisions, 1);
2632 return renames;
2633 }
2634
2635 static int process_renames(struct merge_options *opt,
2636 struct string_list *a_renames,
2637 struct string_list *b_renames)
2638 {
2639 int clean_merge = 1, i, j;
2640 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2641 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2642 const struct rename *sre;
2643
2644 for (i = 0; i < a_renames->nr; i++) {
2645 sre = a_renames->items[i].util;
2646 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2647 = (void *)sre;
2648 }
2649 for (i = 0; i < b_renames->nr; i++) {
2650 sre = b_renames->items[i].util;
2651 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2652 = (void *)sre;
2653 }
2654
2655 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2656 struct string_list *renames1, *renames2Dst;
2657 struct rename *ren1 = NULL, *ren2 = NULL;
2658 const char *ren1_src, *ren1_dst;
2659 struct string_list_item *lookup;
2660
2661 if (i >= a_renames->nr) {
2662 ren2 = b_renames->items[j++].util;
2663 } else if (j >= b_renames->nr) {
2664 ren1 = a_renames->items[i++].util;
2665 } else {
2666 int compare = strcmp(a_renames->items[i].string,
2667 b_renames->items[j].string);
2668 if (compare <= 0)
2669 ren1 = a_renames->items[i++].util;
2670 if (compare >= 0)
2671 ren2 = b_renames->items[j++].util;
2672 }
2673
2674 /* TODO: refactor, so that 1/2 are not needed */
2675 if (ren1) {
2676 renames1 = a_renames;
2677 renames2Dst = &b_by_dst;
2678 } else {
2679 renames1 = b_renames;
2680 renames2Dst = &a_by_dst;
2681 SWAP(ren2, ren1);
2682 }
2683
2684 if (ren1->processed)
2685 continue;
2686 ren1->processed = 1;
2687 ren1->dst_entry->processed = 1;
2688 /* BUG: We should only mark src_entry as processed if we
2689 * are not dealing with a rename + add-source case.
2690 */
2691 ren1->src_entry->processed = 1;
2692
2693 ren1_src = ren1->pair->one->path;
2694 ren1_dst = ren1->pair->two->path;
2695
2696 if (ren2) {
2697 /* One file renamed on both sides */
2698 const char *ren2_src = ren2->pair->one->path;
2699 const char *ren2_dst = ren2->pair->two->path;
2700 enum rename_type rename_type;
2701 if (strcmp(ren1_src, ren2_src) != 0)
2702 BUG("ren1_src != ren2_src");
2703 ren2->dst_entry->processed = 1;
2704 ren2->processed = 1;
2705 if (strcmp(ren1_dst, ren2_dst) != 0) {
2706 rename_type = RENAME_ONE_FILE_TO_TWO;
2707 clean_merge = 0;
2708 } else {
2709 rename_type = RENAME_ONE_FILE_TO_ONE;
2710 /* BUG: We should only remove ren1_src in
2711 * the base stage (think of rename +
2712 * add-source cases).
2713 */
2714 remove_file(opt, 1, ren1_src, 1);
2715 update_entry(ren1->dst_entry,
2716 ren1->pair->one,
2717 ren1->pair->two,
2718 ren2->pair->two);
2719 }
2720 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2721 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2722 /* Two different files renamed to the same thing */
2723 char *ren2_dst;
2724 ren2 = lookup->util;
2725 ren2_dst = ren2->pair->two->path;
2726 if (strcmp(ren1_dst, ren2_dst) != 0)
2727 BUG("ren1_dst != ren2_dst");
2728
2729 clean_merge = 0;
2730 ren2->processed = 1;
2731 /*
2732 * BUG: We should only mark src_entry as processed
2733 * if we are not dealing with a rename + add-source
2734 * case.
2735 */
2736 ren2->src_entry->processed = 1;
2737
2738 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2739 opt, ren1, ren2);
2740 } else {
2741 /* Renamed in 1, maybe changed in 2 */
2742 /* we only use sha1 and mode of these */
2743 struct diff_filespec src_other, dst_other;
2744 int try_merge;
2745
2746 /*
2747 * unpack_trees loads entries from common-commit
2748 * into stage 1, from head-commit into stage 2, and
2749 * from merge-commit into stage 3. We keep track
2750 * of which side corresponds to the rename.
2751 */
2752 int renamed_stage = a_renames == renames1 ? 2 : 3;
2753 int other_stage = a_renames == renames1 ? 3 : 2;
2754
2755 /* BUG: We should only remove ren1_src in the base
2756 * stage and in other_stage (think of rename +
2757 * add-source case).
2758 */
2759 remove_file(opt, 1, ren1_src,
2760 renamed_stage == 2 || !was_tracked(opt, ren1_src));
2761
2762 oidcpy(&src_other.oid,
2763 &ren1->src_entry->stages[other_stage].oid);
2764 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2765 oidcpy(&dst_other.oid,
2766 &ren1->dst_entry->stages[other_stage].oid);
2767 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2768 try_merge = 0;
2769
2770 if (oid_eq(&src_other.oid, &null_oid) &&
2771 ren1->dir_rename_original_type == 'A') {
2772 setup_rename_conflict_info(RENAME_VIA_DIR,
2773 opt, ren1, NULL);
2774 } else if (oid_eq(&src_other.oid, &null_oid)) {
2775 setup_rename_conflict_info(RENAME_DELETE,
2776 opt, ren1, NULL);
2777 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2778 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2779 /*
2780 * Added file on the other side identical to
2781 * the file being renamed: clean merge.
2782 * Also, there is no need to overwrite the
2783 * file already in the working copy, so call
2784 * update_file_flags() instead of
2785 * update_file().
2786 */
2787 if (update_file_flags(opt,
2788 ren1->pair->two,
2789 ren1_dst,
2790 1, /* update_cache */
2791 0 /* update_wd */))
2792 clean_merge = -1;
2793 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2794 /*
2795 * Probably not a clean merge, but it's
2796 * premature to set clean_merge to 0 here,
2797 * because if the rename merges cleanly and
2798 * the merge exactly matches the newly added
2799 * file, then the merge will be clean.
2800 */
2801 setup_rename_conflict_info(RENAME_ADD,
2802 opt, ren1, NULL);
2803 } else
2804 try_merge = 1;
2805
2806 if (clean_merge < 0)
2807 goto cleanup_and_return;
2808 if (try_merge) {
2809 struct diff_filespec *o, *a, *b;
2810 src_other.path = (char *)ren1_src;
2811
2812 o = ren1->pair->one;
2813 if (a_renames == renames1) {
2814 a = ren1->pair->two;
2815 b = &src_other;
2816 } else {
2817 b = ren1->pair->two;
2818 a = &src_other;
2819 }
2820 update_entry(ren1->dst_entry, o, a, b);
2821 setup_rename_conflict_info(RENAME_NORMAL,
2822 opt, ren1, NULL);
2823 }
2824 }
2825 }
2826 cleanup_and_return:
2827 string_list_clear(&a_by_dst, 0);
2828 string_list_clear(&b_by_dst, 0);
2829
2830 return clean_merge;
2831 }
2832
2833 struct rename_info {
2834 struct string_list *head_renames;
2835 struct string_list *merge_renames;
2836 };
2837
2838 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2839 struct hashmap *dir_renames)
2840 {
2841 struct hashmap_iter iter;
2842 struct dir_rename_entry *e;
2843
2844 hashmap_iter_init(dir_renames, &iter);
2845 while ((e = hashmap_iter_next(&iter))) {
2846 free(e->dir);
2847 strbuf_release(&e->new_dir);
2848 /* possible_new_dirs already cleared in get_directory_renames */
2849 }
2850 hashmap_free(dir_renames, 1);
2851 free(dir_renames);
2852
2853 free(pairs->queue);
2854 free(pairs);
2855 }
2856
2857 static int detect_and_process_renames(struct merge_options *opt,
2858 struct tree *common,
2859 struct tree *head,
2860 struct tree *merge,
2861 struct string_list *entries,
2862 struct rename_info *ri)
2863 {
2864 struct diff_queue_struct *head_pairs, *merge_pairs;
2865 struct hashmap *dir_re_head, *dir_re_merge;
2866 int clean = 1;
2867
2868 ri->head_renames = NULL;
2869 ri->merge_renames = NULL;
2870
2871 if (!merge_detect_rename(opt))
2872 return 1;
2873
2874 head_pairs = get_diffpairs(opt, common, head);
2875 merge_pairs = get_diffpairs(opt, common, merge);
2876
2877 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2878 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2879 !opt->priv->call_depth)) {
2880 dir_re_head = get_directory_renames(head_pairs);
2881 dir_re_merge = get_directory_renames(merge_pairs);
2882
2883 handle_directory_level_conflicts(opt,
2884 dir_re_head, head,
2885 dir_re_merge, merge);
2886 } else {
2887 dir_re_head = xmalloc(sizeof(*dir_re_head));
2888 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2889 dir_rename_init(dir_re_head);
2890 dir_rename_init(dir_re_merge);
2891 }
2892
2893 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2894 dir_re_merge, dir_re_head, head,
2895 common, head, merge, entries,
2896 &clean);
2897 if (clean < 0)
2898 goto cleanup;
2899 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2900 dir_re_head, dir_re_merge, merge,
2901 common, head, merge, entries,
2902 &clean);
2903 if (clean < 0)
2904 goto cleanup;
2905 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2906
2907 cleanup:
2908 /*
2909 * Some cleanup is deferred until cleanup_renames() because the
2910 * data structures are still needed and referenced in
2911 * process_entry(). But there are a few things we can free now.
2912 */
2913 initial_cleanup_rename(head_pairs, dir_re_head);
2914 initial_cleanup_rename(merge_pairs, dir_re_merge);
2915
2916 return clean;
2917 }
2918
2919 static void final_cleanup_rename(struct string_list *rename)
2920 {
2921 const struct rename *re;
2922 int i;
2923
2924 if (rename == NULL)
2925 return;
2926
2927 for (i = 0; i < rename->nr; i++) {
2928 re = rename->items[i].util;
2929 diff_free_filepair(re->pair);
2930 }
2931 string_list_clear(rename, 1);
2932 free(rename);
2933 }
2934
2935 static void final_cleanup_renames(struct rename_info *re_info)
2936 {
2937 final_cleanup_rename(re_info->head_renames);
2938 final_cleanup_rename(re_info->merge_renames);
2939 }
2940
2941 static int read_oid_strbuf(struct merge_options *opt,
2942 const struct object_id *oid,
2943 struct strbuf *dst)
2944 {
2945 void *buf;
2946 enum object_type type;
2947 unsigned long size;
2948 buf = read_object_file(oid, &type, &size);
2949 if (!buf)
2950 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
2951 if (type != OBJ_BLOB) {
2952 free(buf);
2953 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
2954 }
2955 strbuf_attach(dst, buf, size, size + 1);
2956 return 0;
2957 }
2958
2959 static int blob_unchanged(struct merge_options *opt,
2960 const struct diff_filespec *o,
2961 const struct diff_filespec *a,
2962 int renormalize, const char *path)
2963 {
2964 struct strbuf obuf = STRBUF_INIT;
2965 struct strbuf abuf = STRBUF_INIT;
2966 int ret = 0; /* assume changed for safety */
2967 const struct index_state *idx = opt->repo->index;
2968
2969 if (a->mode != o->mode)
2970 return 0;
2971 if (oid_eq(&o->oid, &a->oid))
2972 return 1;
2973 if (!renormalize)
2974 return 0;
2975
2976 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
2977 read_oid_strbuf(opt, &a->oid, &abuf))
2978 goto error_return;
2979 /*
2980 * Note: binary | is used so that both renormalizations are
2981 * performed. Comparison can be skipped if both files are
2982 * unchanged since their sha1s have already been compared.
2983 */
2984 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
2985 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
2986 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
2987
2988 error_return:
2989 strbuf_release(&obuf);
2990 strbuf_release(&abuf);
2991 return ret;
2992 }
2993
2994 static int handle_modify_delete(struct merge_options *opt,
2995 const char *path,
2996 const struct diff_filespec *o,
2997 const struct diff_filespec *a,
2998 const struct diff_filespec *b)
2999 {
3000 const char *modify_branch, *delete_branch;
3001 const struct diff_filespec *changed;
3002
3003 if (is_valid(a)) {
3004 modify_branch = opt->branch1;
3005 delete_branch = opt->branch2;
3006 changed = a;
3007 } else {
3008 modify_branch = opt->branch2;
3009 delete_branch = opt->branch1;
3010 changed = b;
3011 }
3012
3013 return handle_change_delete(opt,
3014 path, NULL,
3015 o, changed,
3016 modify_branch, delete_branch,
3017 _("modify"), _("modified"));
3018 }
3019
3020 static int handle_content_merge(struct merge_file_info *mfi,
3021 struct merge_options *opt,
3022 const char *path,
3023 int is_dirty,
3024 const struct diff_filespec *o,
3025 const struct diff_filespec *a,
3026 const struct diff_filespec *b,
3027 struct rename_conflict_info *ci)
3028 {
3029 const char *reason = _("content");
3030 unsigned df_conflict_remains = 0;
3031
3032 if (!is_valid(o))
3033 reason = _("add/add");
3034
3035 assert(o->path && a->path && b->path);
3036 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3037 S_ISGITLINK(ci->ren1->pair->two->mode)))
3038 df_conflict_remains = 1;
3039
3040 if (merge_mode_and_contents(opt, o, a, b, path,
3041 opt->branch1, opt->branch2,
3042 opt->priv->call_depth * 2, mfi))
3043 return -1;
3044
3045 /*
3046 * We can skip updating the working tree file iff:
3047 * a) The merge is clean
3048 * b) The merge matches what was in HEAD (content, mode, pathname)
3049 * c) The target path is usable (i.e. not involved in D/F conflict)
3050 */
3051 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3052 !df_conflict_remains) {
3053 int pos;
3054 struct cache_entry *ce;
3055
3056 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3057 if (add_cacheinfo(opt, &mfi->blob, path,
3058 0, (!opt->priv->call_depth && !is_dirty), 0))
3059 return -1;
3060 /*
3061 * However, add_cacheinfo() will delete the old cache entry
3062 * and add a new one. We need to copy over any skip_worktree
3063 * flag to avoid making the file appear as if it were
3064 * deleted by the user.
3065 */
3066 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3067 ce = opt->priv->orig_index.cache[pos];
3068 if (ce_skip_worktree(ce)) {
3069 pos = index_name_pos(opt->repo->index, path, strlen(path));
3070 ce = opt->repo->index->cache[pos];
3071 ce->ce_flags |= CE_SKIP_WORKTREE;
3072 }
3073 return mfi->clean;
3074 }
3075
3076 if (!mfi->clean) {
3077 if (S_ISGITLINK(mfi->blob.mode))
3078 reason = _("submodule");
3079 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3080 reason, path);
3081 if (ci && !df_conflict_remains)
3082 if (update_stages(opt, path, o, a, b))
3083 return -1;
3084 }
3085
3086 if (df_conflict_remains || is_dirty) {
3087 char *new_path;
3088 if (opt->priv->call_depth) {
3089 remove_file_from_index(opt->repo->index, path);
3090 } else {
3091 if (!mfi->clean) {
3092 if (update_stages(opt, path, o, a, b))
3093 return -1;
3094 } else {
3095 int file_from_stage2 = was_tracked(opt, path);
3096
3097 if (update_stages(opt, path, NULL,
3098 file_from_stage2 ? &mfi->blob : NULL,
3099 file_from_stage2 ? NULL : &mfi->blob))
3100 return -1;
3101 }
3102
3103 }
3104 new_path = unique_path(opt, path, ci->ren1->branch);
3105 if (is_dirty) {
3106 output(opt, 1, _("Refusing to lose dirty file at %s"),
3107 path);
3108 }
3109 output(opt, 1, _("Adding as %s instead"), new_path);
3110 if (update_file(opt, 0, &mfi->blob, new_path)) {
3111 free(new_path);
3112 return -1;
3113 }
3114 free(new_path);
3115 mfi->clean = 0;
3116 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3117 return -1;
3118 return !is_dirty && mfi->clean;
3119 }
3120
3121 static int handle_rename_normal(struct merge_options *opt,
3122 const char *path,
3123 const struct diff_filespec *o,
3124 const struct diff_filespec *a,
3125 const struct diff_filespec *b,
3126 struct rename_conflict_info *ci)
3127 {
3128 struct rename *ren = ci->ren1;
3129 struct merge_file_info mfi;
3130 int clean;
3131 int side = (ren->branch == opt->branch1 ? 2 : 3);
3132
3133 /* Merge the content and write it out */
3134 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3135 o, a, b, ci);
3136
3137 if (clean &&
3138 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3139 ren->dir_rename_original_dest) {
3140 if (update_stages(opt, path,
3141 NULL,
3142 side == 2 ? &mfi.blob : NULL,
3143 side == 2 ? NULL : &mfi.blob))
3144 return -1;
3145 clean = 0; /* not clean, but conflicted */
3146 }
3147 return clean;
3148 }
3149
3150 static void dir_rename_warning(const char *msg,
3151 int is_add,
3152 int clean,
3153 struct merge_options *opt,
3154 struct rename *ren)
3155 {
3156 const char *other_branch;
3157 other_branch = (ren->branch == opt->branch1 ?
3158 opt->branch2 : opt->branch1);
3159 if (is_add) {
3160 output(opt, clean ? 2 : 1, msg,
3161 ren->pair->one->path, ren->branch,
3162 other_branch, ren->pair->two->path);
3163 return;
3164 }
3165 output(opt, clean ? 2 : 1, msg,
3166 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3167 other_branch, ren->pair->two->path);
3168 }
3169 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3170 struct rename *ren)
3171 {
3172 const char *msg;
3173 int clean = 1, is_add;
3174
3175 if (!ren)
3176 return clean;
3177
3178 /* Return early if ren was not affected/created by a directory rename */
3179 if (!ren->dir_rename_original_dest)
3180 return clean;
3181
3182 /* Sanity checks */
3183 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3184 assert(ren->dir_rename_original_type == 'A' ||
3185 ren->dir_rename_original_type == 'R');
3186
3187 /* Check whether to treat directory renames as a conflict */
3188 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3189
3190 is_add = (ren->dir_rename_original_type == 'A');
3191 if (ren->dir_rename_original_type == 'A' && clean) {
3192 msg = _("Path updated: %s added in %s inside a "
3193 "directory that was renamed in %s; moving it to %s.");
3194 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3195 msg = _("CONFLICT (file location): %s added in %s "
3196 "inside a directory that was renamed in %s, "
3197 "suggesting it should perhaps be moved to %s.");
3198 } else if (ren->dir_rename_original_type == 'R' && clean) {
3199 msg = _("Path updated: %s renamed to %s in %s, inside a "
3200 "directory that was renamed in %s; moving it to %s.");
3201 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3202 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3203 "inside a directory that was renamed in %s, "
3204 "suggesting it should perhaps be moved to %s.");
3205 } else {
3206 BUG("Impossible dir_rename_original_type/clean combination");
3207 }
3208 dir_rename_warning(msg, is_add, clean, opt, ren);
3209
3210 return clean;
3211 }
3212
3213 /* Per entry merge function */
3214 static int process_entry(struct merge_options *opt,
3215 const char *path, struct stage_data *entry)
3216 {
3217 int clean_merge = 1;
3218 int normalize = opt->renormalize;
3219
3220 struct diff_filespec *o = &entry->stages[1];
3221 struct diff_filespec *a = &entry->stages[2];
3222 struct diff_filespec *b = &entry->stages[3];
3223 int o_valid = is_valid(o);
3224 int a_valid = is_valid(a);
3225 int b_valid = is_valid(b);
3226 o->path = a->path = b->path = (char*)path;
3227
3228 entry->processed = 1;
3229 if (entry->rename_conflict_info) {
3230 struct rename_conflict_info *ci = entry->rename_conflict_info;
3231 struct diff_filespec *temp;
3232 int path_clean;
3233
3234 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3235 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3236
3237 /*
3238 * For cases with a single rename, {o,a,b}->path have all been
3239 * set to the rename target path; we need to set two of these
3240 * back to the rename source.
3241 * For rename/rename conflicts, we'll manually fix paths below.
3242 */
3243 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3244 o->path = temp->path = ci->ren1->pair->one->path;
3245 if (ci->ren2) {
3246 assert(opt->branch1 == ci->ren1->branch);
3247 }
3248
3249 switch (ci->rename_type) {
3250 case RENAME_NORMAL:
3251 case RENAME_ONE_FILE_TO_ONE:
3252 clean_merge = handle_rename_normal(opt, path, o, a, b,
3253 ci);
3254 break;
3255 case RENAME_VIA_DIR:
3256 clean_merge = handle_rename_via_dir(opt, ci);
3257 break;
3258 case RENAME_ADD:
3259 /*
3260 * Probably unclean merge, but if the renamed file
3261 * merges cleanly and the result can then be
3262 * two-way merged cleanly with the added file, I
3263 * guess it's a clean merge?
3264 */
3265 clean_merge = handle_rename_add(opt, ci);
3266 break;
3267 case RENAME_DELETE:
3268 clean_merge = 0;
3269 if (handle_rename_delete(opt, ci))
3270 clean_merge = -1;
3271 break;
3272 case RENAME_ONE_FILE_TO_TWO:
3273 /*
3274 * Manually fix up paths; note:
3275 * ren[12]->pair->one->path are equal.
3276 */
3277 o->path = ci->ren1->pair->one->path;
3278 a->path = ci->ren1->pair->two->path;
3279 b->path = ci->ren2->pair->two->path;
3280
3281 clean_merge = 0;
3282 if (handle_rename_rename_1to2(opt, ci))
3283 clean_merge = -1;
3284 break;
3285 case RENAME_TWO_FILES_TO_ONE:
3286 /*
3287 * Manually fix up paths; note,
3288 * ren[12]->pair->two->path are actually equal.
3289 */
3290 o->path = NULL;
3291 a->path = ci->ren1->pair->two->path;
3292 b->path = ci->ren2->pair->two->path;
3293
3294 /*
3295 * Probably unclean merge, but if the two renamed
3296 * files merge cleanly and the two resulting files
3297 * can then be two-way merged cleanly, I guess it's
3298 * a clean merge?
3299 */
3300 clean_merge = handle_rename_rename_2to1(opt, ci);
3301 break;
3302 default:
3303 entry->processed = 0;
3304 break;
3305 }
3306 if (path_clean < clean_merge)
3307 clean_merge = path_clean;
3308 } else if (o_valid && (!a_valid || !b_valid)) {
3309 /* Case A: Deleted in one */
3310 if ((!a_valid && !b_valid) ||
3311 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3312 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3313 /* Deleted in both or deleted in one and
3314 * unchanged in the other */
3315 if (a_valid)
3316 output(opt, 2, _("Removing %s"), path);
3317 /* do not touch working file if it did not exist */
3318 remove_file(opt, 1, path, !a_valid);
3319 } else {
3320 /* Modify/delete; deleted side may have put a directory in the way */
3321 clean_merge = 0;
3322 if (handle_modify_delete(opt, path, o, a, b))
3323 clean_merge = -1;
3324 }
3325 } else if ((!o_valid && a_valid && !b_valid) ||
3326 (!o_valid && !a_valid && b_valid)) {
3327 /* Case B: Added in one. */
3328 /* [nothing|directory] -> ([nothing|directory], file) */
3329
3330 const char *add_branch;
3331 const char *other_branch;
3332 const char *conf;
3333 const struct diff_filespec *contents;
3334
3335 if (a_valid) {
3336 add_branch = opt->branch1;
3337 other_branch = opt->branch2;
3338 contents = a;
3339 conf = _("file/directory");
3340 } else {
3341 add_branch = opt->branch2;
3342 other_branch = opt->branch1;
3343 contents = b;
3344 conf = _("directory/file");
3345 }
3346 if (dir_in_way(opt->repo->index, path,
3347 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3348 0)) {
3349 char *new_path = unique_path(opt, path, add_branch);
3350 clean_merge = 0;
3351 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3352 "Adding %s as %s"),
3353 conf, path, other_branch, path, new_path);
3354 if (update_file(opt, 0, contents, new_path))
3355 clean_merge = -1;
3356 else if (opt->priv->call_depth)
3357 remove_file_from_index(opt->repo->index, path);
3358 free(new_path);
3359 } else {
3360 output(opt, 2, _("Adding %s"), path);
3361 /* do not overwrite file if already present */
3362 if (update_file_flags(opt, contents, path, 1, !a_valid))
3363 clean_merge = -1;
3364 }
3365 } else if (a_valid && b_valid) {
3366 if (!o_valid) {
3367 /* Case C: Added in both (check for same permissions) */
3368 output(opt, 1,
3369 _("CONFLICT (add/add): Merge conflict in %s"),
3370 path);
3371 clean_merge = handle_file_collision(opt,
3372 path, NULL, NULL,
3373 opt->branch1,
3374 opt->branch2,
3375 a, b);
3376 } else {
3377 /* case D: Modified in both, but differently. */
3378 struct merge_file_info mfi;
3379 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3380 clean_merge = handle_content_merge(&mfi, opt, path,
3381 is_dirty,
3382 o, a, b, NULL);
3383 }
3384 } else if (!o_valid && !a_valid && !b_valid) {
3385 /*
3386 * this entry was deleted altogether. a_mode == 0 means
3387 * we had that path and want to actively remove it.
3388 */
3389 remove_file(opt, 1, path, !a->mode);
3390 } else
3391 BUG("fatal merge failure, shouldn't happen.");
3392
3393 return clean_merge;
3394 }
3395
3396 static int merge_trees_internal(struct merge_options *opt,
3397 struct tree *head,
3398 struct tree *merge,
3399 struct tree *merge_base,
3400 struct tree **result)
3401 {
3402 struct index_state *istate = opt->repo->index;
3403 int code, clean;
3404
3405 if (opt->subtree_shift) {
3406 merge = shift_tree_object(opt->repo, head, merge,
3407 opt->subtree_shift);
3408 merge_base = shift_tree_object(opt->repo, head, merge_base,
3409 opt->subtree_shift);
3410 }
3411
3412 if (oid_eq(&merge_base->object.oid, &merge->object.oid)) {
3413 output(opt, 0, _("Already up to date!"));
3414 *result = head;
3415 return 1;
3416 }
3417
3418 code = unpack_trees_start(opt, merge_base, head, merge);
3419
3420 if (code != 0) {
3421 if (show(opt, 4) || opt->priv->call_depth)
3422 err(opt, _("merging of trees %s and %s failed"),
3423 oid_to_hex(&head->object.oid),
3424 oid_to_hex(&merge->object.oid));
3425 unpack_trees_finish(opt);
3426 return -1;
3427 }
3428
3429 if (unmerged_index(istate)) {
3430 struct string_list *entries;
3431 struct rename_info re_info;
3432 int i;
3433 /*
3434 * Only need the hashmap while processing entries, so
3435 * initialize it here and free it when we are done running
3436 * through the entries. Keeping it in the merge_options as
3437 * opposed to decaring a local hashmap is for convenience
3438 * so that we don't have to pass it to around.
3439 */
3440 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3441 NULL, 512);
3442 get_files_dirs(opt, head);
3443 get_files_dirs(opt, merge);
3444
3445 entries = get_unmerged(opt->repo->index);
3446 clean = detect_and_process_renames(opt, merge_base, head, merge,
3447 entries, &re_info);
3448 record_df_conflict_files(opt, entries);
3449 if (clean < 0)
3450 goto cleanup;
3451 for (i = entries->nr-1; 0 <= i; i--) {
3452 const char *path = entries->items[i].string;
3453 struct stage_data *e = entries->items[i].util;
3454 if (!e->processed) {
3455 int ret = process_entry(opt, path, e);
3456 if (!ret)
3457 clean = 0;
3458 else if (ret < 0) {
3459 clean = ret;
3460 goto cleanup;
3461 }
3462 }
3463 }
3464 for (i = 0; i < entries->nr; i++) {
3465 struct stage_data *e = entries->items[i].util;
3466 if (!e->processed)
3467 BUG("unprocessed path??? %s",
3468 entries->items[i].string);
3469 }
3470
3471 cleanup:
3472 final_cleanup_renames(&re_info);
3473
3474 string_list_clear(entries, 1);
3475 free(entries);
3476
3477 hashmap_free(&opt->priv->current_file_dir_set, 1);
3478
3479 if (clean < 0) {
3480 unpack_trees_finish(opt);
3481 return clean;
3482 }
3483 }
3484 else
3485 clean = 1;
3486
3487 unpack_trees_finish(opt);
3488
3489 if (opt->priv->call_depth &&
3490 !(*result = write_in_core_index_as_tree(opt->repo)))
3491 return -1;
3492
3493 return clean;
3494 }
3495
3496 static struct commit_list *reverse_commit_list(struct commit_list *list)
3497 {
3498 struct commit_list *next = NULL, *current, *backup;
3499 for (current = list; current; current = backup) {
3500 backup = current->next;
3501 current->next = next;
3502 next = current;
3503 }
3504 return next;
3505 }
3506
3507 /*
3508 * Merge the commits h1 and h2, return the resulting virtual
3509 * commit object and a flag indicating the cleanness of the merge.
3510 */
3511 static int merge_recursive_internal(struct merge_options *opt,
3512 struct commit *h1,
3513 struct commit *h2,
3514 struct commit_list *merge_bases,
3515 struct commit **result)
3516 {
3517 struct commit_list *iter;
3518 struct commit *merged_merge_bases;
3519 struct tree *result_tree;
3520 int clean;
3521 const char *ancestor_name;
3522 struct strbuf merge_base_abbrev = STRBUF_INIT;
3523
3524 if (show(opt, 4)) {
3525 output(opt, 4, _("Merging:"));
3526 output_commit_title(opt, h1);
3527 output_commit_title(opt, h2);
3528 }
3529
3530 if (!merge_bases) {
3531 merge_bases = get_merge_bases(h1, h2);
3532 merge_bases = reverse_commit_list(merge_bases);
3533 }
3534
3535 if (show(opt, 5)) {
3536 unsigned cnt = commit_list_count(merge_bases);
3537
3538 output(opt, 5, Q_("found %u common ancestor:",
3539 "found %u common ancestors:", cnt), cnt);
3540 for (iter = merge_bases; iter; iter = iter->next)
3541 output_commit_title(opt, iter->item);
3542 }
3543
3544 merged_merge_bases = pop_commit(&merge_bases);
3545 if (merged_merge_bases == NULL) {
3546 /* if there is no common ancestor, use an empty tree */
3547 struct tree *tree;
3548
3549 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3550 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3551 "ancestor");
3552 ancestor_name = "empty tree";
3553 } else if (opt->ancestor) {
3554 ancestor_name = opt->ancestor;
3555 } else if (merge_bases) {
3556 ancestor_name = "merged common ancestors";
3557 } else {
3558 strbuf_add_unique_abbrev(&merge_base_abbrev,
3559 &merged_merge_bases->object.oid,
3560 DEFAULT_ABBREV);
3561 ancestor_name = merge_base_abbrev.buf;
3562 }
3563
3564 for (iter = merge_bases; iter; iter = iter->next) {
3565 const char *saved_b1, *saved_b2;
3566 opt->priv->call_depth++;
3567 /*
3568 * When the merge fails, the result contains files
3569 * with conflict markers. The cleanness flag is
3570 * ignored (unless indicating an error), it was never
3571 * actually used, as result of merge_trees has always
3572 * overwritten it: the committed "conflicts" were
3573 * already resolved.
3574 */
3575 discard_index(opt->repo->index);
3576 saved_b1 = opt->branch1;
3577 saved_b2 = opt->branch2;
3578 opt->branch1 = "Temporary merge branch 1";
3579 opt->branch2 = "Temporary merge branch 2";
3580 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3581 NULL, &merged_merge_bases) < 0)
3582 return -1;
3583 opt->branch1 = saved_b1;
3584 opt->branch2 = saved_b2;
3585 opt->priv->call_depth--;
3586
3587 if (!merged_merge_bases)
3588 return err(opt, _("merge returned no commit"));
3589 }
3590
3591 discard_index(opt->repo->index);
3592 if (!opt->priv->call_depth)
3593 repo_read_index(opt->repo);
3594
3595 opt->ancestor = ancestor_name;
3596 clean = merge_trees_internal(opt,
3597 repo_get_commit_tree(opt->repo, h1),
3598 repo_get_commit_tree(opt->repo, h2),
3599 repo_get_commit_tree(opt->repo,
3600 merged_merge_bases),
3601 &result_tree);
3602 strbuf_release(&merge_base_abbrev);
3603 if (clean < 0) {
3604 flush_output(opt);
3605 return clean;
3606 }
3607
3608 if (opt->priv->call_depth) {
3609 *result = make_virtual_commit(opt->repo, result_tree,
3610 "merged tree");
3611 commit_list_insert(h1, &(*result)->parents);
3612 commit_list_insert(h2, &(*result)->parents->next);
3613 }
3614 return clean;
3615 }
3616
3617 static int merge_start(struct merge_options *opt, struct tree *head)
3618 {
3619 struct strbuf sb = STRBUF_INIT;
3620
3621 /* Sanity checks on opt */
3622 assert(opt->repo);
3623
3624 assert(opt->branch1 && opt->branch2);
3625
3626 assert(opt->detect_renames >= -1 &&
3627 opt->detect_renames <= DIFF_DETECT_COPY);
3628 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3629 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3630 assert(opt->rename_limit >= -1);
3631 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3632 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3633
3634 assert(opt->xdl_opts >= 0);
3635 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3636 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3637
3638 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3639 assert(opt->buffer_output <= 2);
3640 assert(opt->obuf.len == 0);
3641
3642 assert(opt->priv == NULL);
3643
3644 /* Sanity check on repo state; index must match head */
3645 if (repo_index_has_changes(opt->repo, head, &sb)) {
3646 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3647 sb.buf);
3648 strbuf_release(&sb);
3649 return -1;
3650 }
3651
3652 opt->priv = xcalloc(1, sizeof(*opt->priv));
3653 string_list_init(&opt->priv->df_conflict_file_set, 1);
3654 return 0;
3655 }
3656
3657 static void merge_finalize(struct merge_options *opt)
3658 {
3659 flush_output(opt);
3660 if (!opt->priv->call_depth && opt->buffer_output < 2)
3661 strbuf_release(&opt->obuf);
3662 if (show(opt, 2))
3663 diff_warn_rename_limit("merge.renamelimit",
3664 opt->priv->needed_rename_limit, 0);
3665 FREE_AND_NULL(opt->priv);
3666 }
3667
3668 int merge_trees(struct merge_options *opt,
3669 struct tree *head,
3670 struct tree *merge,
3671 struct tree *merge_base)
3672 {
3673 int clean;
3674 struct tree *ignored;
3675
3676 assert(opt->ancestor != NULL);
3677
3678 if (merge_start(opt, head))
3679 return -1;
3680 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3681 merge_finalize(opt);
3682
3683 return clean;
3684 }
3685
3686 int merge_recursive(struct merge_options *opt,
3687 struct commit *h1,
3688 struct commit *h2,
3689 struct commit_list *merge_bases,
3690 struct commit **result)
3691 {
3692 int clean;
3693
3694 assert(opt->ancestor == NULL ||
3695 !strcmp(opt->ancestor, "constructed merge base"));
3696
3697 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3698 return -1;
3699 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3700 merge_finalize(opt);
3701
3702 return clean;
3703 }
3704
3705 static struct commit *get_ref(struct repository *repo,
3706 const struct object_id *oid,
3707 const char *name)
3708 {
3709 struct object *object;
3710
3711 object = deref_tag(repo, parse_object(repo, oid),
3712 name, strlen(name));
3713 if (!object)
3714 return NULL;
3715 if (object->type == OBJ_TREE)
3716 return make_virtual_commit(repo, (struct tree*)object, name);
3717 if (object->type != OBJ_COMMIT)
3718 return NULL;
3719 if (parse_commit((struct commit *)object))
3720 return NULL;
3721 return (struct commit *)object;
3722 }
3723
3724 int merge_recursive_generic(struct merge_options *opt,
3725 const struct object_id *head,
3726 const struct object_id *merge,
3727 int num_merge_bases,
3728 const struct object_id **merge_bases,
3729 struct commit **result)
3730 {
3731 int clean;
3732 struct lock_file lock = LOCK_INIT;
3733 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3734 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3735 struct commit_list *ca = NULL;
3736
3737 if (merge_bases) {
3738 int i;
3739 for (i = 0; i < num_merge_bases; ++i) {
3740 struct commit *base;
3741 if (!(base = get_ref(opt->repo, merge_bases[i],
3742 oid_to_hex(merge_bases[i]))))
3743 return err(opt, _("Could not parse object '%s'"),
3744 oid_to_hex(merge_bases[i]));
3745 commit_list_insert(base, &ca);
3746 }
3747 if (num_merge_bases == 1)
3748 opt->ancestor = "constructed merge base";
3749 }
3750
3751 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3752 clean = merge_recursive(opt, head_commit, next_commit, ca,
3753 result);
3754 if (clean < 0) {
3755 rollback_lock_file(&lock);
3756 return clean;
3757 }
3758
3759 if (write_locked_index(opt->repo->index, &lock,
3760 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3761 return err(opt, _("Unable to write index."));
3762
3763 return clean ? 0 : 1;
3764 }
3765
3766 static void merge_recursive_config(struct merge_options *opt)
3767 {
3768 char *value = NULL;
3769 git_config_get_int("merge.verbosity", &opt->verbosity);
3770 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3771 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3772 if (!git_config_get_string("diff.renames", &value)) {
3773 opt->detect_renames = git_config_rename("diff.renames", value);
3774 free(value);
3775 }
3776 if (!git_config_get_string("merge.renames", &value)) {
3777 opt->detect_renames = git_config_rename("merge.renames", value);
3778 free(value);
3779 }
3780 if (!git_config_get_string("merge.directoryrenames", &value)) {
3781 int boolval = git_parse_maybe_bool(value);
3782 if (0 <= boolval) {
3783 opt->detect_directory_renames = boolval ?
3784 MERGE_DIRECTORY_RENAMES_TRUE :
3785 MERGE_DIRECTORY_RENAMES_NONE;
3786 } else if (!strcasecmp(value, "conflict")) {
3787 opt->detect_directory_renames =
3788 MERGE_DIRECTORY_RENAMES_CONFLICT;
3789 } /* avoid erroring on values from future versions of git */
3790 free(value);
3791 }
3792 git_config(git_xmerge_config, NULL);
3793 }
3794
3795 void init_merge_options(struct merge_options *opt,
3796 struct repository *repo)
3797 {
3798 const char *merge_verbosity;
3799 memset(opt, 0, sizeof(struct merge_options));
3800
3801 opt->repo = repo;
3802
3803 opt->detect_renames = -1;
3804 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3805 opt->rename_limit = -1;
3806
3807 opt->verbosity = 2;
3808 opt->buffer_output = 1;
3809 strbuf_init(&opt->obuf, 0);
3810
3811 opt->renormalize = 0;
3812
3813 merge_recursive_config(opt);
3814 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3815 if (merge_verbosity)
3816 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3817 if (opt->verbosity >= 5)
3818 opt->buffer_output = 0;
3819 }
3820
3821 int parse_merge_opt(struct merge_options *opt, const char *s)
3822 {
3823 const char *arg;
3824
3825 if (!s || !*s)
3826 return -1;
3827 if (!strcmp(s, "ours"))
3828 opt->recursive_variant = MERGE_VARIANT_OURS;
3829 else if (!strcmp(s, "theirs"))
3830 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3831 else if (!strcmp(s, "subtree"))
3832 opt->subtree_shift = "";
3833 else if (skip_prefix(s, "subtree=", &arg))
3834 opt->subtree_shift = arg;
3835 else if (!strcmp(s, "patience"))
3836 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3837 else if (!strcmp(s, "histogram"))
3838 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3839 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3840 long value = parse_algorithm_value(arg);
3841 if (value < 0)
3842 return -1;
3843 /* clear out previous settings */
3844 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3845 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3846 opt->xdl_opts |= value;
3847 }
3848 else if (!strcmp(s, "ignore-space-change"))
3849 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3850 else if (!strcmp(s, "ignore-all-space"))
3851 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3852 else if (!strcmp(s, "ignore-space-at-eol"))
3853 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3854 else if (!strcmp(s, "ignore-cr-at-eol"))
3855 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3856 else if (!strcmp(s, "renormalize"))
3857 opt->renormalize = 1;
3858 else if (!strcmp(s, "no-renormalize"))
3859 opt->renormalize = 0;
3860 else if (!strcmp(s, "no-renames"))
3861 opt->detect_renames = 0;
3862 else if (!strcmp(s, "find-renames")) {
3863 opt->detect_renames = 1;
3864 opt->rename_score = 0;
3865 }
3866 else if (skip_prefix(s, "find-renames=", &arg) ||
3867 skip_prefix(s, "rename-threshold=", &arg)) {
3868 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3869 return -1;
3870 opt->detect_renames = 1;
3871 }
3872 /*
3873 * Please update $__git_merge_strategy_options in
3874 * git-completion.bash when you add new options
3875 */
3876 else
3877 return -1;
3878 return 0;
3879 }