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