]> git.ipfire.org Git - thirdparty/git.git/blob - merge-recursive.c
merge-recursive: introduce an enum for detect_directory_renames values
[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 "config.h"
8 #include "advice.h"
9 #include "lockfile.h"
10 #include "cache-tree.h"
11 #include "object-store.h"
12 #include "repository.h"
13 #include "commit.h"
14 #include "blob.h"
15 #include "builtin.h"
16 #include "tree-walk.h"
17 #include "diff.h"
18 #include "diffcore.h"
19 #include "tag.h"
20 #include "alloc.h"
21 #include "unpack-trees.h"
22 #include "string-list.h"
23 #include "xdiff-interface.h"
24 #include "ll-merge.h"
25 #include "attr.h"
26 #include "merge-recursive.h"
27 #include "dir.h"
28 #include "submodule.h"
29 #include "revision.h"
30 #include "commit-reach.h"
31
32 struct path_hashmap_entry {
33 struct hashmap_entry e;
34 char path[FLEX_ARRAY];
35 };
36
37 static int path_hashmap_cmp(const void *cmp_data,
38 const void *entry,
39 const void *entry_or_key,
40 const void *keydata)
41 {
42 const struct path_hashmap_entry *a = entry;
43 const struct path_hashmap_entry *b = entry_or_key;
44 const char *key = keydata;
45
46 if (ignore_case)
47 return strcasecmp(a->path, key ? key : b->path);
48 else
49 return strcmp(a->path, key ? key : b->path);
50 }
51
52 static unsigned int path_hash(const char *path)
53 {
54 return ignore_case ? strihash(path) : strhash(path);
55 }
56
57 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
58 char *dir)
59 {
60 struct dir_rename_entry key;
61
62 if (dir == NULL)
63 return NULL;
64 hashmap_entry_init(&key, strhash(dir));
65 key.dir = dir;
66 return hashmap_get(hashmap, &key, NULL);
67 }
68
69 static int dir_rename_cmp(const void *unused_cmp_data,
70 const void *entry,
71 const void *entry_or_key,
72 const void *unused_keydata)
73 {
74 const struct dir_rename_entry *e1 = entry;
75 const struct dir_rename_entry *e2 = entry_or_key;
76
77 return strcmp(e1->dir, e2->dir);
78 }
79
80 static void dir_rename_init(struct hashmap *map)
81 {
82 hashmap_init(map, dir_rename_cmp, NULL, 0);
83 }
84
85 static void dir_rename_entry_init(struct dir_rename_entry *entry,
86 char *directory)
87 {
88 hashmap_entry_init(entry, strhash(directory));
89 entry->dir = directory;
90 entry->non_unique_new_dir = 0;
91 strbuf_init(&entry->new_dir, 0);
92 string_list_init(&entry->possible_new_dirs, 0);
93 }
94
95 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
96 char *target_file)
97 {
98 struct collision_entry key;
99
100 hashmap_entry_init(&key, strhash(target_file));
101 key.target_file = target_file;
102 return hashmap_get(hashmap, &key, NULL);
103 }
104
105 static int collision_cmp(void *unused_cmp_data,
106 const struct collision_entry *e1,
107 const struct collision_entry *e2,
108 const void *unused_keydata)
109 {
110 return strcmp(e1->target_file, e2->target_file);
111 }
112
113 static void collision_init(struct hashmap *map)
114 {
115 hashmap_init(map, (hashmap_cmp_fn) collision_cmp, NULL, 0);
116 }
117
118 static void flush_output(struct merge_options *opt)
119 {
120 if (opt->buffer_output < 2 && opt->obuf.len) {
121 fputs(opt->obuf.buf, stdout);
122 strbuf_reset(&opt->obuf);
123 }
124 }
125
126 static int err(struct merge_options *opt, const char *err, ...)
127 {
128 va_list params;
129
130 if (opt->buffer_output < 2)
131 flush_output(opt);
132 else {
133 strbuf_complete(&opt->obuf, '\n');
134 strbuf_addstr(&opt->obuf, "error: ");
135 }
136 va_start(params, err);
137 strbuf_vaddf(&opt->obuf, err, params);
138 va_end(params);
139 if (opt->buffer_output > 1)
140 strbuf_addch(&opt->obuf, '\n');
141 else {
142 error("%s", opt->obuf.buf);
143 strbuf_reset(&opt->obuf);
144 }
145
146 return -1;
147 }
148
149 static struct tree *shift_tree_object(struct repository *repo,
150 struct tree *one, struct tree *two,
151 const char *subtree_shift)
152 {
153 struct object_id shifted;
154
155 if (!*subtree_shift) {
156 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
157 } else {
158 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
159 subtree_shift);
160 }
161 if (oideq(&two->object.oid, &shifted))
162 return two;
163 return lookup_tree(repo, &shifted);
164 }
165
166 static inline void set_commit_tree(struct commit *c, struct tree *t)
167 {
168 c->maybe_tree = t;
169 }
170
171 static struct commit *make_virtual_commit(struct repository *repo,
172 struct tree *tree,
173 const char *comment)
174 {
175 struct commit *commit = alloc_commit_node(repo);
176
177 set_merge_remote_desc(commit, comment, (struct object *)commit);
178 set_commit_tree(commit, tree);
179 commit->object.parsed = 1;
180 return commit;
181 }
182
183 /*
184 * Since we use get_tree_entry(), which does not put the read object into
185 * the object pool, we cannot rely on a == b.
186 */
187 static int oid_eq(const struct object_id *a, const struct object_id *b)
188 {
189 if (!a && !b)
190 return 2;
191 return a && b && oideq(a, b);
192 }
193
194 enum rename_type {
195 RENAME_NORMAL = 0,
196 RENAME_VIA_DIR,
197 RENAME_ADD,
198 RENAME_DELETE,
199 RENAME_ONE_FILE_TO_ONE,
200 RENAME_ONE_FILE_TO_TWO,
201 RENAME_TWO_FILES_TO_ONE
202 };
203
204 /*
205 * Since we want to write the index eventually, we cannot reuse the index
206 * for these (temporary) data.
207 */
208 struct stage_data {
209 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
210 struct rename_conflict_info *rename_conflict_info;
211 unsigned processed:1;
212 };
213
214 struct rename {
215 unsigned processed:1;
216 struct diff_filepair *pair;
217 const char *branch; /* branch that the rename occurred on */
218 /*
219 * If directory rename detection affected this rename, what was its
220 * original type ('A' or 'R') and it's original destination before
221 * the directory rename (otherwise, '\0' and NULL for these two vars).
222 */
223 char dir_rename_original_type;
224 char *dir_rename_original_dest;
225 /*
226 * Purpose of src_entry and dst_entry:
227 *
228 * If 'before' is renamed to 'after' then src_entry will contain
229 * the versions of 'before' from the merge_base, HEAD, and MERGE in
230 * stages 1, 2, and 3; dst_entry will contain the respective
231 * versions of 'after' in corresponding locations. Thus, we have a
232 * total of six modes and oids, though some will be null. (Stage 0
233 * is ignored; we're interested in handling conflicts.)
234 *
235 * Since we don't turn on break-rewrites by default, neither
236 * src_entry nor dst_entry can have all three of their stages have
237 * non-null oids, meaning at most four of the six will be non-null.
238 * Also, since this is a rename, both src_entry and dst_entry will
239 * have at least one non-null oid, meaning at least two will be
240 * non-null. Of the six oids, a typical rename will have three be
241 * non-null. Only two implies a rename/delete, and four implies a
242 * rename/add.
243 */
244 struct stage_data *src_entry;
245 struct stage_data *dst_entry;
246 };
247
248 struct rename_conflict_info {
249 enum rename_type rename_type;
250 struct rename *ren1;
251 struct rename *ren2;
252 };
253
254 static inline void setup_rename_conflict_info(enum rename_type rename_type,
255 struct merge_options *opt,
256 struct rename *ren1,
257 struct rename *ren2)
258 {
259 struct rename_conflict_info *ci;
260
261 /*
262 * When we have two renames involved, it's easiest to get the
263 * correct things into stage 2 and 3, and to make sure that the
264 * content merge puts HEAD before the other branch if we just
265 * ensure that branch1 == opt->branch1. So, simply flip arguments
266 * around if we don't have that.
267 */
268 if (ren2 && ren1->branch != opt->branch1) {
269 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
270 return;
271 }
272
273 ci = xcalloc(1, sizeof(struct rename_conflict_info));
274 ci->rename_type = rename_type;
275 ci->ren1 = ren1;
276 ci->ren2 = ren2;
277
278 ci->ren1->dst_entry->processed = 0;
279 ci->ren1->dst_entry->rename_conflict_info = ci;
280 if (ren2) {
281 ci->ren2->dst_entry->rename_conflict_info = ci;
282 }
283 }
284
285 static int show(struct merge_options *opt, int v)
286 {
287 return (!opt->call_depth && opt->verbosity >= v) || opt->verbosity >= 5;
288 }
289
290 __attribute__((format (printf, 3, 4)))
291 static void output(struct merge_options *opt, int v, const char *fmt, ...)
292 {
293 va_list ap;
294
295 if (!show(opt, v))
296 return;
297
298 strbuf_addchars(&opt->obuf, ' ', opt->call_depth * 2);
299
300 va_start(ap, fmt);
301 strbuf_vaddf(&opt->obuf, fmt, ap);
302 va_end(ap);
303
304 strbuf_addch(&opt->obuf, '\n');
305 if (!opt->buffer_output)
306 flush_output(opt);
307 }
308
309 static void output_commit_title(struct merge_options *opt, struct commit *commit)
310 {
311 struct merge_remote_desc *desc;
312
313 strbuf_addchars(&opt->obuf, ' ', opt->call_depth * 2);
314 desc = merge_remote_util(commit);
315 if (desc)
316 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
317 else {
318 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
319 DEFAULT_ABBREV);
320 strbuf_addch(&opt->obuf, ' ');
321 if (parse_commit(commit) != 0)
322 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
323 else {
324 const char *title;
325 const char *msg = get_commit_buffer(commit, NULL);
326 int len = find_commit_subject(msg, &title);
327 if (len)
328 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
329 unuse_commit_buffer(commit, msg);
330 }
331 }
332 flush_output(opt);
333 }
334
335 static int add_cacheinfo(struct merge_options *opt,
336 const struct diff_filespec *blob,
337 const char *path, int stage, int refresh, int options)
338 {
339 struct index_state *istate = opt->repo->index;
340 struct cache_entry *ce;
341 int ret;
342
343 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
344 if (!ce)
345 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
346
347 ret = add_index_entry(istate, ce, options);
348 if (refresh) {
349 struct cache_entry *nce;
350
351 nce = refresh_cache_entry(istate, ce,
352 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
353 if (!nce)
354 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
355 if (nce != ce)
356 ret = add_index_entry(istate, nce, options);
357 }
358 return ret;
359 }
360
361 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
362 {
363 parse_tree(tree);
364 init_tree_desc(desc, tree->buffer, tree->size);
365 }
366
367 static int unpack_trees_start(struct merge_options *opt,
368 struct tree *common,
369 struct tree *head,
370 struct tree *merge)
371 {
372 int rc;
373 struct tree_desc t[3];
374 struct index_state tmp_index = { NULL };
375
376 memset(&opt->unpack_opts, 0, sizeof(opt->unpack_opts));
377 if (opt->call_depth)
378 opt->unpack_opts.index_only = 1;
379 else
380 opt->unpack_opts.update = 1;
381 opt->unpack_opts.merge = 1;
382 opt->unpack_opts.head_idx = 2;
383 opt->unpack_opts.fn = threeway_merge;
384 opt->unpack_opts.src_index = opt->repo->index;
385 opt->unpack_opts.dst_index = &tmp_index;
386 opt->unpack_opts.aggressive = !merge_detect_rename(opt);
387 setup_unpack_trees_porcelain(&opt->unpack_opts, "merge");
388
389 init_tree_desc_from_tree(t+0, common);
390 init_tree_desc_from_tree(t+1, head);
391 init_tree_desc_from_tree(t+2, merge);
392
393 rc = unpack_trees(3, t, &opt->unpack_opts);
394 cache_tree_free(&opt->repo->index->cache_tree);
395
396 /*
397 * Update opt->repo->index to match the new results, AFTER saving a copy
398 * in opt->orig_index. Update src_index to point to the saved copy.
399 * (verify_uptodate() checks src_index, and the original index is
400 * the one that had the necessary modification timestamps.)
401 */
402 opt->orig_index = *opt->repo->index;
403 *opt->repo->index = tmp_index;
404 opt->unpack_opts.src_index = &opt->orig_index;
405
406 return rc;
407 }
408
409 static void unpack_trees_finish(struct merge_options *opt)
410 {
411 discard_index(&opt->orig_index);
412 clear_unpack_trees_porcelain(&opt->unpack_opts);
413 }
414
415 struct tree *write_tree_from_memory(struct merge_options *opt)
416 {
417 struct tree *result = NULL;
418 struct index_state *istate = opt->repo->index;
419
420 if (unmerged_index(istate)) {
421 int i;
422 fprintf(stderr, "BUG: There are unmerged index entries:\n");
423 for (i = 0; i < istate->cache_nr; i++) {
424 const struct cache_entry *ce = istate->cache[i];
425 if (ce_stage(ce))
426 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
427 (int)ce_namelen(ce), ce->name);
428 }
429 BUG("unmerged index entries in merge-recursive.c");
430 }
431
432 if (!istate->cache_tree)
433 istate->cache_tree = cache_tree();
434
435 if (!cache_tree_fully_valid(istate->cache_tree) &&
436 cache_tree_update(istate, 0) < 0) {
437 err(opt, _("error building trees"));
438 return NULL;
439 }
440
441 result = lookup_tree(opt->repo, &istate->cache_tree->oid);
442
443 return result;
444 }
445
446 static int save_files_dirs(const struct object_id *oid,
447 struct strbuf *base, const char *path,
448 unsigned int mode, int stage, void *context)
449 {
450 struct path_hashmap_entry *entry;
451 int baselen = base->len;
452 struct merge_options *opt = context;
453
454 strbuf_addstr(base, path);
455
456 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
457 hashmap_entry_init(entry, path_hash(entry->path));
458 hashmap_add(&opt->current_file_dir_set, entry);
459
460 strbuf_setlen(base, baselen);
461 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
462 }
463
464 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
465 {
466 struct pathspec match_all;
467 memset(&match_all, 0, sizeof(match_all));
468 read_tree_recursive(opt->repo, tree, "", 0, 0,
469 &match_all, save_files_dirs, opt);
470 }
471
472 static int get_tree_entry_if_blob(struct repository *r,
473 const struct object_id *tree,
474 const char *path,
475 struct diff_filespec *dfs)
476 {
477 int ret;
478
479 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
480 if (S_ISDIR(dfs->mode)) {
481 oidcpy(&dfs->oid, &null_oid);
482 dfs->mode = 0;
483 }
484 return ret;
485 }
486
487 /*
488 * Returns an index_entry instance which doesn't have to correspond to
489 * a real cache entry in Git's index.
490 */
491 static struct stage_data *insert_stage_data(struct repository *r,
492 const char *path,
493 struct tree *o, struct tree *a, struct tree *b,
494 struct string_list *entries)
495 {
496 struct string_list_item *item;
497 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
498 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
499 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
500 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
501 item = string_list_insert(entries, path);
502 item->util = e;
503 return e;
504 }
505
506 /*
507 * Create a dictionary mapping file names to stage_data objects. The
508 * dictionary contains one entry for every path with a non-zero stage entry.
509 */
510 static struct string_list *get_unmerged(struct index_state *istate)
511 {
512 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
513 int i;
514
515 unmerged->strdup_strings = 1;
516
517 for (i = 0; i < istate->cache_nr; i++) {
518 struct string_list_item *item;
519 struct stage_data *e;
520 const struct cache_entry *ce = istate->cache[i];
521 if (!ce_stage(ce))
522 continue;
523
524 item = string_list_lookup(unmerged, ce->name);
525 if (!item) {
526 item = string_list_insert(unmerged, ce->name);
527 item->util = xcalloc(1, sizeof(struct stage_data));
528 }
529 e = item->util;
530 e->stages[ce_stage(ce)].mode = ce->ce_mode;
531 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
532 }
533
534 return unmerged;
535 }
536
537 static int string_list_df_name_compare(const char *one, const char *two)
538 {
539 int onelen = strlen(one);
540 int twolen = strlen(two);
541 /*
542 * Here we only care that entries for D/F conflicts are
543 * adjacent, in particular with the file of the D/F conflict
544 * appearing before files below the corresponding directory.
545 * The order of the rest of the list is irrelevant for us.
546 *
547 * To achieve this, we sort with df_name_compare and provide
548 * the mode S_IFDIR so that D/F conflicts will sort correctly.
549 * We use the mode S_IFDIR for everything else for simplicity,
550 * since in other cases any changes in their order due to
551 * sorting cause no problems for us.
552 */
553 int cmp = df_name_compare(one, onelen, S_IFDIR,
554 two, twolen, S_IFDIR);
555 /*
556 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
557 * that 'foo' comes before 'foo/bar'.
558 */
559 if (cmp)
560 return cmp;
561 return onelen - twolen;
562 }
563
564 static void record_df_conflict_files(struct merge_options *opt,
565 struct string_list *entries)
566 {
567 /* If there is a D/F conflict and the file for such a conflict
568 * currently exists in the working tree, we want to allow it to be
569 * removed to make room for the corresponding directory if needed.
570 * The files underneath the directories of such D/F conflicts will
571 * be processed before the corresponding file involved in the D/F
572 * conflict. If the D/F directory ends up being removed by the
573 * merge, then we won't have to touch the D/F file. If the D/F
574 * directory needs to be written to the working copy, then the D/F
575 * file will simply be removed (in make_room_for_path()) to make
576 * room for the necessary paths. Note that if both the directory
577 * and the file need to be present, then the D/F file will be
578 * reinstated with a new unique name at the time it is processed.
579 */
580 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
581 const char *last_file = NULL;
582 int last_len = 0;
583 int i;
584
585 /*
586 * If we're merging merge-bases, we don't want to bother with
587 * any working directory changes.
588 */
589 if (opt->call_depth)
590 return;
591
592 /* Ensure D/F conflicts are adjacent in the entries list. */
593 for (i = 0; i < entries->nr; i++) {
594 struct string_list_item *next = &entries->items[i];
595 string_list_append(&df_sorted_entries, next->string)->util =
596 next->util;
597 }
598 df_sorted_entries.cmp = string_list_df_name_compare;
599 string_list_sort(&df_sorted_entries);
600
601 string_list_clear(&opt->df_conflict_file_set, 1);
602 for (i = 0; i < df_sorted_entries.nr; i++) {
603 const char *path = df_sorted_entries.items[i].string;
604 int len = strlen(path);
605 struct stage_data *e = df_sorted_entries.items[i].util;
606
607 /*
608 * Check if last_file & path correspond to a D/F conflict;
609 * i.e. whether path is last_file+'/'+<something>.
610 * If so, record that it's okay to remove last_file to make
611 * room for path and friends if needed.
612 */
613 if (last_file &&
614 len > last_len &&
615 memcmp(path, last_file, last_len) == 0 &&
616 path[last_len] == '/') {
617 string_list_insert(&opt->df_conflict_file_set, last_file);
618 }
619
620 /*
621 * Determine whether path could exist as a file in the
622 * working directory as a possible D/F conflict. This
623 * will only occur when it exists in stage 2 as a
624 * file.
625 */
626 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
627 last_file = path;
628 last_len = len;
629 } else {
630 last_file = NULL;
631 }
632 }
633 string_list_clear(&df_sorted_entries, 0);
634 }
635
636 static int update_stages(struct merge_options *opt, const char *path,
637 const struct diff_filespec *o,
638 const struct diff_filespec *a,
639 const struct diff_filespec *b)
640 {
641
642 /*
643 * NOTE: It is usually a bad idea to call update_stages on a path
644 * before calling update_file on that same path, since it can
645 * sometimes lead to spurious "refusing to lose untracked file..."
646 * messages from update_file (via make_room_for path via
647 * would_lose_untracked). Instead, reverse the order of the calls
648 * (executing update_file first and then update_stages).
649 */
650 int clear = 1;
651 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
652 if (clear)
653 if (remove_file_from_index(opt->repo->index, path))
654 return -1;
655 if (o)
656 if (add_cacheinfo(opt, o, path, 1, 0, options))
657 return -1;
658 if (a)
659 if (add_cacheinfo(opt, a, path, 2, 0, options))
660 return -1;
661 if (b)
662 if (add_cacheinfo(opt, b, path, 3, 0, options))
663 return -1;
664 return 0;
665 }
666
667 static void update_entry(struct stage_data *entry,
668 struct diff_filespec *o,
669 struct diff_filespec *a,
670 struct diff_filespec *b)
671 {
672 entry->processed = 0;
673 entry->stages[1].mode = o->mode;
674 entry->stages[2].mode = a->mode;
675 entry->stages[3].mode = b->mode;
676 oidcpy(&entry->stages[1].oid, &o->oid);
677 oidcpy(&entry->stages[2].oid, &a->oid);
678 oidcpy(&entry->stages[3].oid, &b->oid);
679 }
680
681 static int remove_file(struct merge_options *opt, int clean,
682 const char *path, int no_wd)
683 {
684 int update_cache = opt->call_depth || clean;
685 int update_working_directory = !opt->call_depth && !no_wd;
686
687 if (update_cache) {
688 if (remove_file_from_index(opt->repo->index, path))
689 return -1;
690 }
691 if (update_working_directory) {
692 if (ignore_case) {
693 struct cache_entry *ce;
694 ce = index_file_exists(opt->repo->index, path, strlen(path),
695 ignore_case);
696 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
697 return 0;
698 }
699 if (remove_path(path))
700 return -1;
701 }
702 return 0;
703 }
704
705 /* add a string to a strbuf, but converting "/" to "_" */
706 static void add_flattened_path(struct strbuf *out, const char *s)
707 {
708 size_t i = out->len;
709 strbuf_addstr(out, s);
710 for (; i < out->len; i++)
711 if (out->buf[i] == '/')
712 out->buf[i] = '_';
713 }
714
715 static char *unique_path(struct merge_options *opt, const char *path, const char *branch)
716 {
717 struct path_hashmap_entry *entry;
718 struct strbuf newpath = STRBUF_INIT;
719 int suffix = 0;
720 size_t base_len;
721
722 strbuf_addf(&newpath, "%s~", path);
723 add_flattened_path(&newpath, branch);
724
725 base_len = newpath.len;
726 while (hashmap_get_from_hash(&opt->current_file_dir_set,
727 path_hash(newpath.buf), newpath.buf) ||
728 (!opt->call_depth && file_exists(newpath.buf))) {
729 strbuf_setlen(&newpath, base_len);
730 strbuf_addf(&newpath, "_%d", suffix++);
731 }
732
733 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
734 hashmap_entry_init(entry, path_hash(entry->path));
735 hashmap_add(&opt->current_file_dir_set, entry);
736 return strbuf_detach(&newpath, NULL);
737 }
738
739 /**
740 * Check whether a directory in the index is in the way of an incoming
741 * file. Return 1 if so. If check_working_copy is non-zero, also
742 * check the working directory. If empty_ok is non-zero, also return
743 * 0 in the case where the working-tree dir exists but is empty.
744 */
745 static int dir_in_way(struct index_state *istate, const char *path,
746 int check_working_copy, int empty_ok)
747 {
748 int pos;
749 struct strbuf dirpath = STRBUF_INIT;
750 struct stat st;
751
752 strbuf_addstr(&dirpath, path);
753 strbuf_addch(&dirpath, '/');
754
755 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
756
757 if (pos < 0)
758 pos = -1 - pos;
759 if (pos < istate->cache_nr &&
760 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
761 strbuf_release(&dirpath);
762 return 1;
763 }
764
765 strbuf_release(&dirpath);
766 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
767 !(empty_ok && is_empty_dir(path));
768 }
769
770 /*
771 * Returns whether path was tracked in the index before the merge started,
772 * and its oid and mode match the specified values
773 */
774 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
775 const struct diff_filespec *blob)
776 {
777 int pos = index_name_pos(&opt->orig_index, path, strlen(path));
778 struct cache_entry *ce;
779
780 if (0 > pos)
781 /* we were not tracking this path before the merge */
782 return 0;
783
784 /* See if the file we were tracking before matches */
785 ce = opt->orig_index.cache[pos];
786 return (oid_eq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
787 }
788
789 /*
790 * Returns whether path was tracked in the index before the merge started
791 */
792 static int was_tracked(struct merge_options *opt, const char *path)
793 {
794 int pos = index_name_pos(&opt->orig_index, path, strlen(path));
795
796 if (0 <= pos)
797 /* we were tracking this path before the merge */
798 return 1;
799
800 return 0;
801 }
802
803 static int would_lose_untracked(struct merge_options *opt, const char *path)
804 {
805 struct index_state *istate = opt->repo->index;
806
807 /*
808 * This may look like it can be simplified to:
809 * return !was_tracked(opt, path) && file_exists(path)
810 * but it can't. This function needs to know whether path was in
811 * the working tree due to EITHER having been tracked in the index
812 * before the merge OR having been put into the working copy and
813 * index by unpack_trees(). Due to that either-or requirement, we
814 * check the current index instead of the original one.
815 *
816 * Note that we do not need to worry about merge-recursive itself
817 * updating the index after unpack_trees() and before calling this
818 * function, because we strictly require all code paths in
819 * merge-recursive to update the working tree first and the index
820 * second. Doing otherwise would break
821 * update_file()/would_lose_untracked(); see every comment in this
822 * file which mentions "update_stages".
823 */
824 int pos = index_name_pos(istate, path, strlen(path));
825
826 if (pos < 0)
827 pos = -1 - pos;
828 while (pos < istate->cache_nr &&
829 !strcmp(path, istate->cache[pos]->name)) {
830 /*
831 * If stage #0, it is definitely tracked.
832 * If it has stage #2 then it was tracked
833 * before this merge started. All other
834 * cases the path was not tracked.
835 */
836 switch (ce_stage(istate->cache[pos])) {
837 case 0:
838 case 2:
839 return 0;
840 }
841 pos++;
842 }
843 return file_exists(path);
844 }
845
846 static int was_dirty(struct merge_options *opt, const char *path)
847 {
848 struct cache_entry *ce;
849 int dirty = 1;
850
851 if (opt->call_depth || !was_tracked(opt, path))
852 return !dirty;
853
854 ce = index_file_exists(opt->unpack_opts.src_index,
855 path, strlen(path), ignore_case);
856 dirty = verify_uptodate(ce, &opt->unpack_opts) != 0;
857 return dirty;
858 }
859
860 static int make_room_for_path(struct merge_options *opt, const char *path)
861 {
862 int status, i;
863 const char *msg = _("failed to create path '%s'%s");
864
865 /* Unlink any D/F conflict files that are in the way */
866 for (i = 0; i < opt->df_conflict_file_set.nr; i++) {
867 const char *df_path = opt->df_conflict_file_set.items[i].string;
868 size_t pathlen = strlen(path);
869 size_t df_pathlen = strlen(df_path);
870 if (df_pathlen < pathlen &&
871 path[df_pathlen] == '/' &&
872 strncmp(path, df_path, df_pathlen) == 0) {
873 output(opt, 3,
874 _("Removing %s to make room for subdirectory\n"),
875 df_path);
876 unlink(df_path);
877 unsorted_string_list_delete_item(&opt->df_conflict_file_set,
878 i, 0);
879 break;
880 }
881 }
882
883 /* Make sure leading directories are created */
884 status = safe_create_leading_directories_const(path);
885 if (status) {
886 if (status == SCLD_EXISTS)
887 /* something else exists */
888 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
889 return err(opt, msg, path, "");
890 }
891
892 /*
893 * Do not unlink a file in the work tree if we are not
894 * tracking it.
895 */
896 if (would_lose_untracked(opt, path))
897 return err(opt, _("refusing to lose untracked file at '%s'"),
898 path);
899
900 /* Successful unlink is good.. */
901 if (!unlink(path))
902 return 0;
903 /* .. and so is no existing file */
904 if (errno == ENOENT)
905 return 0;
906 /* .. but not some other error (who really cares what?) */
907 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
908 }
909
910 static int update_file_flags(struct merge_options *opt,
911 const struct diff_filespec *contents,
912 const char *path,
913 int update_cache,
914 int update_wd)
915 {
916 int ret = 0;
917
918 if (opt->call_depth)
919 update_wd = 0;
920
921 if (update_wd) {
922 enum object_type type;
923 void *buf;
924 unsigned long size;
925
926 if (S_ISGITLINK(contents->mode)) {
927 /*
928 * We may later decide to recursively descend into
929 * the submodule directory and update its index
930 * and/or work tree, but we do not do that now.
931 */
932 update_wd = 0;
933 goto update_index;
934 }
935
936 buf = read_object_file(&contents->oid, &type, &size);
937 if (!buf)
938 return err(opt, _("cannot read object %s '%s'"),
939 oid_to_hex(&contents->oid), path);
940 if (type != OBJ_BLOB) {
941 ret = err(opt, _("blob expected for %s '%s'"),
942 oid_to_hex(&contents->oid), path);
943 goto free_buf;
944 }
945 if (S_ISREG(contents->mode)) {
946 struct strbuf strbuf = STRBUF_INIT;
947 if (convert_to_working_tree(opt->repo->index, path, buf, size, &strbuf)) {
948 free(buf);
949 size = strbuf.len;
950 buf = strbuf_detach(&strbuf, NULL);
951 }
952 }
953
954 if (make_room_for_path(opt, path) < 0) {
955 update_wd = 0;
956 goto free_buf;
957 }
958 if (S_ISREG(contents->mode) ||
959 (!has_symlinks && S_ISLNK(contents->mode))) {
960 int fd;
961 int mode = (contents->mode & 0100 ? 0777 : 0666);
962
963 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
964 if (fd < 0) {
965 ret = err(opt, _("failed to open '%s': %s"),
966 path, strerror(errno));
967 goto free_buf;
968 }
969 write_in_full(fd, buf, size);
970 close(fd);
971 } else if (S_ISLNK(contents->mode)) {
972 char *lnk = xmemdupz(buf, size);
973 safe_create_leading_directories_const(path);
974 unlink(path);
975 if (symlink(lnk, path))
976 ret = err(opt, _("failed to symlink '%s': %s"),
977 path, strerror(errno));
978 free(lnk);
979 } else
980 ret = err(opt,
981 _("do not know what to do with %06o %s '%s'"),
982 contents->mode, oid_to_hex(&contents->oid), path);
983 free_buf:
984 free(buf);
985 }
986 update_index:
987 if (!ret && update_cache)
988 if (add_cacheinfo(opt, contents, path, 0, update_wd,
989 ADD_CACHE_OK_TO_ADD))
990 return -1;
991 return ret;
992 }
993
994 static int update_file(struct merge_options *opt,
995 int clean,
996 const struct diff_filespec *contents,
997 const char *path)
998 {
999 return update_file_flags(opt, contents, path,
1000 opt->call_depth || clean, !opt->call_depth);
1001 }
1002
1003 /* Low level file merging, update and removal */
1004
1005 struct merge_file_info {
1006 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1007 unsigned clean:1,
1008 merge:1;
1009 };
1010
1011 static int merge_3way(struct merge_options *opt,
1012 mmbuffer_t *result_buf,
1013 const struct diff_filespec *o,
1014 const struct diff_filespec *a,
1015 const struct diff_filespec *b,
1016 const char *branch1,
1017 const char *branch2,
1018 const int extra_marker_size)
1019 {
1020 mmfile_t orig, src1, src2;
1021 struct ll_merge_options ll_opts = {0};
1022 char *base, *name1, *name2;
1023 int merge_status;
1024
1025 ll_opts.renormalize = opt->renormalize;
1026 ll_opts.extra_marker_size = extra_marker_size;
1027 ll_opts.xdl_opts = opt->xdl_opts;
1028
1029 if (opt->call_depth) {
1030 ll_opts.virtual_ancestor = 1;
1031 ll_opts.variant = 0;
1032 } else {
1033 switch (opt->recursive_variant) {
1034 case MERGE_RECURSIVE_OURS:
1035 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1036 break;
1037 case MERGE_RECURSIVE_THEIRS:
1038 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1039 break;
1040 default:
1041 ll_opts.variant = 0;
1042 break;
1043 }
1044 }
1045
1046 assert(a->path && b->path && o->path && opt->ancestor);
1047 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1048 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1049 name1 = mkpathdup("%s:%s", branch1, a->path);
1050 name2 = mkpathdup("%s:%s", branch2, b->path);
1051 } else {
1052 base = mkpathdup("%s", opt->ancestor);
1053 name1 = mkpathdup("%s", branch1);
1054 name2 = mkpathdup("%s", branch2);
1055 }
1056
1057 read_mmblob(&orig, &o->oid);
1058 read_mmblob(&src1, &a->oid);
1059 read_mmblob(&src2, &b->oid);
1060
1061 merge_status = ll_merge(result_buf, a->path, &orig, base,
1062 &src1, name1, &src2, name2,
1063 opt->repo->index, &ll_opts);
1064
1065 free(base);
1066 free(name1);
1067 free(name2);
1068 free(orig.ptr);
1069 free(src1.ptr);
1070 free(src2.ptr);
1071 return merge_status;
1072 }
1073
1074 static int find_first_merges(struct repository *repo,
1075 struct object_array *result, const char *path,
1076 struct commit *a, struct commit *b)
1077 {
1078 int i, j;
1079 struct object_array merges = OBJECT_ARRAY_INIT;
1080 struct commit *commit;
1081 int contains_another;
1082
1083 char merged_revision[GIT_MAX_HEXSZ + 2];
1084 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1085 "--all", merged_revision, NULL };
1086 struct rev_info revs;
1087 struct setup_revision_opt rev_opts;
1088
1089 memset(result, 0, sizeof(struct object_array));
1090 memset(&rev_opts, 0, sizeof(rev_opts));
1091
1092 /* get all revisions that merge commit a */
1093 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1094 oid_to_hex(&a->object.oid));
1095 repo_init_revisions(repo, &revs, NULL);
1096 rev_opts.submodule = path;
1097 /* FIXME: can't handle linked worktrees in submodules yet */
1098 revs.single_worktree = path != NULL;
1099 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1100
1101 /* save all revisions from the above list that contain b */
1102 if (prepare_revision_walk(&revs))
1103 die("revision walk setup failed");
1104 while ((commit = get_revision(&revs)) != NULL) {
1105 struct object *o = &(commit->object);
1106 if (in_merge_bases(b, commit))
1107 add_object_array(o, NULL, &merges);
1108 }
1109 reset_revision_walk();
1110
1111 /* Now we've got all merges that contain a and b. Prune all
1112 * merges that contain another found merge and save them in
1113 * result.
1114 */
1115 for (i = 0; i < merges.nr; i++) {
1116 struct commit *m1 = (struct commit *) merges.objects[i].item;
1117
1118 contains_another = 0;
1119 for (j = 0; j < merges.nr; j++) {
1120 struct commit *m2 = (struct commit *) merges.objects[j].item;
1121 if (i != j && in_merge_bases(m2, m1)) {
1122 contains_another = 1;
1123 break;
1124 }
1125 }
1126
1127 if (!contains_another)
1128 add_object_array(merges.objects[i].item, NULL, result);
1129 }
1130
1131 object_array_clear(&merges);
1132 return result->nr;
1133 }
1134
1135 static void print_commit(struct commit *commit)
1136 {
1137 struct strbuf sb = STRBUF_INIT;
1138 struct pretty_print_context ctx = {0};
1139 ctx.date_mode.type = DATE_NORMAL;
1140 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1141 fprintf(stderr, "%s\n", sb.buf);
1142 strbuf_release(&sb);
1143 }
1144
1145 static int is_valid(const struct diff_filespec *dfs)
1146 {
1147 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1148 }
1149
1150 static int merge_submodule(struct merge_options *opt,
1151 struct object_id *result, const char *path,
1152 const struct object_id *base, const struct object_id *a,
1153 const struct object_id *b)
1154 {
1155 struct commit *commit_base, *commit_a, *commit_b;
1156 int parent_count;
1157 struct object_array merges;
1158
1159 int i;
1160 int search = !opt->call_depth;
1161
1162 /* store a in result in case we fail */
1163 oidcpy(result, a);
1164
1165 /* we can not handle deletion conflicts */
1166 if (is_null_oid(base))
1167 return 0;
1168 if (is_null_oid(a))
1169 return 0;
1170 if (is_null_oid(b))
1171 return 0;
1172
1173 if (add_submodule_odb(path)) {
1174 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1175 return 0;
1176 }
1177
1178 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1179 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1180 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1181 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1182 return 0;
1183 }
1184
1185 /* check whether both changes are forward */
1186 if (!in_merge_bases(commit_base, commit_a) ||
1187 !in_merge_bases(commit_base, commit_b)) {
1188 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1189 return 0;
1190 }
1191
1192 /* Case #1: a is contained in b or vice versa */
1193 if (in_merge_bases(commit_a, commit_b)) {
1194 oidcpy(result, b);
1195 if (show(opt, 3)) {
1196 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1197 output_commit_title(opt, commit_b);
1198 } else if (show(opt, 2))
1199 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1200 else
1201 ; /* no output */
1202
1203 return 1;
1204 }
1205 if (in_merge_bases(commit_b, commit_a)) {
1206 oidcpy(result, a);
1207 if (show(opt, 3)) {
1208 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1209 output_commit_title(opt, commit_a);
1210 } else if (show(opt, 2))
1211 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1212 else
1213 ; /* no output */
1214
1215 return 1;
1216 }
1217
1218 /*
1219 * Case #2: There are one or more merges that contain a and b in
1220 * the submodule. If there is only one, then present it as a
1221 * suggestion to the user, but leave it marked unmerged so the
1222 * user needs to confirm the resolution.
1223 */
1224
1225 /* Skip the search if makes no sense to the calling context. */
1226 if (!search)
1227 return 0;
1228
1229 /* find commit which merges them */
1230 parent_count = find_first_merges(opt->repo, &merges, path,
1231 commit_a, commit_b);
1232 switch (parent_count) {
1233 case 0:
1234 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1235 break;
1236
1237 case 1:
1238 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1239 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1240 print_commit((struct commit *) merges.objects[0].item);
1241 output(opt, 2, _(
1242 "If this is correct simply add it to the index "
1243 "for example\n"
1244 "by using:\n\n"
1245 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1246 "which will accept this suggestion.\n"),
1247 oid_to_hex(&merges.objects[0].item->oid), path);
1248 break;
1249
1250 default:
1251 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1252 for (i = 0; i < merges.nr; i++)
1253 print_commit((struct commit *) merges.objects[i].item);
1254 }
1255
1256 object_array_clear(&merges);
1257 return 0;
1258 }
1259
1260 static int merge_mode_and_contents(struct merge_options *opt,
1261 const struct diff_filespec *o,
1262 const struct diff_filespec *a,
1263 const struct diff_filespec *b,
1264 const char *filename,
1265 const char *branch1,
1266 const char *branch2,
1267 const int extra_marker_size,
1268 struct merge_file_info *result)
1269 {
1270 if (opt->branch1 != branch1) {
1271 /*
1272 * It's weird getting a reverse merge with HEAD on the bottom
1273 * side of the conflict markers and the other branch on the
1274 * top. Fix that.
1275 */
1276 return merge_mode_and_contents(opt, o, b, a,
1277 filename,
1278 branch2, branch1,
1279 extra_marker_size, result);
1280 }
1281
1282 result->merge = 0;
1283 result->clean = 1;
1284
1285 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1286 result->clean = 0;
1287 if (S_ISREG(a->mode)) {
1288 result->blob.mode = a->mode;
1289 oidcpy(&result->blob.oid, &a->oid);
1290 } else {
1291 result->blob.mode = b->mode;
1292 oidcpy(&result->blob.oid, &b->oid);
1293 }
1294 } else {
1295 if (!oid_eq(&a->oid, &o->oid) && !oid_eq(&b->oid, &o->oid))
1296 result->merge = 1;
1297
1298 /*
1299 * Merge modes
1300 */
1301 if (a->mode == b->mode || a->mode == o->mode)
1302 result->blob.mode = b->mode;
1303 else {
1304 result->blob.mode = a->mode;
1305 if (b->mode != o->mode) {
1306 result->clean = 0;
1307 result->merge = 1;
1308 }
1309 }
1310
1311 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &o->oid))
1312 oidcpy(&result->blob.oid, &b->oid);
1313 else if (oid_eq(&b->oid, &o->oid))
1314 oidcpy(&result->blob.oid, &a->oid);
1315 else if (S_ISREG(a->mode)) {
1316 mmbuffer_t result_buf;
1317 int ret = 0, merge_status;
1318
1319 merge_status = merge_3way(opt, &result_buf, o, a, b,
1320 branch1, branch2,
1321 extra_marker_size);
1322
1323 if ((merge_status < 0) || !result_buf.ptr)
1324 ret = err(opt, _("Failed to execute internal merge"));
1325
1326 if (!ret &&
1327 write_object_file(result_buf.ptr, result_buf.size,
1328 blob_type, &result->blob.oid))
1329 ret = err(opt, _("Unable to add %s to database"),
1330 a->path);
1331
1332 free(result_buf.ptr);
1333 if (ret)
1334 return ret;
1335 result->clean = (merge_status == 0);
1336 } else if (S_ISGITLINK(a->mode)) {
1337 result->clean = merge_submodule(opt, &result->blob.oid,
1338 o->path,
1339 &o->oid,
1340 &a->oid,
1341 &b->oid);
1342 } else if (S_ISLNK(a->mode)) {
1343 switch (opt->recursive_variant) {
1344 case MERGE_RECURSIVE_NORMAL:
1345 oidcpy(&result->blob.oid, &a->oid);
1346 if (!oid_eq(&a->oid, &b->oid))
1347 result->clean = 0;
1348 break;
1349 case MERGE_RECURSIVE_OURS:
1350 oidcpy(&result->blob.oid, &a->oid);
1351 break;
1352 case MERGE_RECURSIVE_THEIRS:
1353 oidcpy(&result->blob.oid, &b->oid);
1354 break;
1355 }
1356 } else
1357 BUG("unsupported object type in the tree");
1358 }
1359
1360 if (result->merge)
1361 output(opt, 2, _("Auto-merging %s"), filename);
1362
1363 return 0;
1364 }
1365
1366 static int handle_rename_via_dir(struct merge_options *opt,
1367 struct rename_conflict_info *ci)
1368 {
1369 /*
1370 * Handle file adds that need to be renamed due to directory rename
1371 * detection. This differs from handle_rename_normal, because
1372 * there is no content merge to do; just move the file into the
1373 * desired final location.
1374 */
1375 const struct rename *ren = ci->ren1;
1376 const struct diff_filespec *dest = ren->pair->two;
1377 char *file_path = dest->path;
1378 int mark_conflicted = (opt->detect_directory_renames ==
1379 MERGE_DIRECTORY_RENAMES_CONFLICT);
1380 assert(ren->dir_rename_original_dest);
1381
1382 if (!opt->call_depth && would_lose_untracked(opt, dest->path)) {
1383 mark_conflicted = 1;
1384 file_path = unique_path(opt, dest->path, ren->branch);
1385 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1386 "writing to %s instead."),
1387 dest->path, file_path);
1388 }
1389
1390 if (mark_conflicted) {
1391 /*
1392 * Write the file in worktree at file_path. In the index,
1393 * only record the file at dest->path in the appropriate
1394 * higher stage.
1395 */
1396 if (update_file(opt, 0, dest, file_path))
1397 return -1;
1398 if (file_path != dest->path)
1399 free(file_path);
1400 if (update_stages(opt, dest->path, NULL,
1401 ren->branch == opt->branch1 ? dest : NULL,
1402 ren->branch == opt->branch1 ? NULL : dest))
1403 return -1;
1404 return 0; /* not clean, but conflicted */
1405 } else {
1406 /* Update dest->path both in index and in worktree */
1407 if (update_file(opt, 1, dest, dest->path))
1408 return -1;
1409 return 1; /* clean */
1410 }
1411 }
1412
1413 static int handle_change_delete(struct merge_options *opt,
1414 const char *path, const char *old_path,
1415 const struct diff_filespec *o,
1416 const struct diff_filespec *changed,
1417 const char *change_branch,
1418 const char *delete_branch,
1419 const char *change, const char *change_past)
1420 {
1421 char *alt_path = NULL;
1422 const char *update_path = path;
1423 int ret = 0;
1424
1425 if (dir_in_way(opt->repo->index, path, !opt->call_depth, 0) ||
1426 (!opt->call_depth && would_lose_untracked(opt, path))) {
1427 update_path = alt_path = unique_path(opt, path, change_branch);
1428 }
1429
1430 if (opt->call_depth) {
1431 /*
1432 * We cannot arbitrarily accept either a_sha or b_sha as
1433 * correct; since there is no true "middle point" between
1434 * them, simply reuse the base version for virtual merge base.
1435 */
1436 ret = remove_file_from_index(opt->repo->index, path);
1437 if (!ret)
1438 ret = update_file(opt, 0, o, update_path);
1439 } else {
1440 /*
1441 * Despite the four nearly duplicate messages and argument
1442 * lists below and the ugliness of the nested if-statements,
1443 * having complete messages makes the job easier for
1444 * translators.
1445 *
1446 * The slight variance among the cases is due to the fact
1447 * that:
1448 * 1) directory/file conflicts (in effect if
1449 * !alt_path) could cause us to need to write the
1450 * file to a different path.
1451 * 2) renames (in effect if !old_path) could mean that
1452 * there are two names for the path that the user
1453 * may know the file by.
1454 */
1455 if (!alt_path) {
1456 if (!old_path) {
1457 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1458 "and %s in %s. Version %s of %s left in tree."),
1459 change, path, delete_branch, change_past,
1460 change_branch, change_branch, path);
1461 } else {
1462 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1463 "and %s to %s in %s. Version %s of %s left in tree."),
1464 change, old_path, delete_branch, change_past, path,
1465 change_branch, change_branch, path);
1466 }
1467 } else {
1468 if (!old_path) {
1469 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1470 "and %s in %s. Version %s of %s left in tree at %s."),
1471 change, path, delete_branch, change_past,
1472 change_branch, change_branch, path, alt_path);
1473 } else {
1474 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1475 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1476 change, old_path, delete_branch, change_past, path,
1477 change_branch, change_branch, path, alt_path);
1478 }
1479 }
1480 /*
1481 * No need to call update_file() on path when change_branch ==
1482 * opt->branch1 && !alt_path, since that would needlessly touch
1483 * path. We could call update_file_flags() with update_cache=0
1484 * and update_wd=0, but that's a no-op.
1485 */
1486 if (change_branch != opt->branch1 || alt_path)
1487 ret = update_file(opt, 0, changed, update_path);
1488 }
1489 free(alt_path);
1490
1491 return ret;
1492 }
1493
1494 static int handle_rename_delete(struct merge_options *opt,
1495 struct rename_conflict_info *ci)
1496 {
1497 const struct rename *ren = ci->ren1;
1498 const struct diff_filespec *orig = ren->pair->one;
1499 const struct diff_filespec *dest = ren->pair->two;
1500 const char *rename_branch = ren->branch;
1501 const char *delete_branch = (opt->branch1 == ren->branch ?
1502 opt->branch2 : opt->branch1);
1503
1504 if (handle_change_delete(opt,
1505 opt->call_depth ? orig->path : dest->path,
1506 opt->call_depth ? NULL : orig->path,
1507 orig, dest,
1508 rename_branch, delete_branch,
1509 _("rename"), _("renamed")))
1510 return -1;
1511
1512 if (opt->call_depth)
1513 return remove_file_from_index(opt->repo->index, dest->path);
1514 else
1515 return update_stages(opt, dest->path, NULL,
1516 rename_branch == opt->branch1 ? dest : NULL,
1517 rename_branch == opt->branch1 ? NULL : dest);
1518 }
1519
1520 static int handle_file_collision(struct merge_options *opt,
1521 const char *collide_path,
1522 const char *prev_path1,
1523 const char *prev_path2,
1524 const char *branch1, const char *branch2,
1525 struct diff_filespec *a,
1526 struct diff_filespec *b)
1527 {
1528 struct merge_file_info mfi;
1529 struct diff_filespec null;
1530 char *alt_path = NULL;
1531 const char *update_path = collide_path;
1532
1533 /*
1534 * It's easiest to get the correct things into stage 2 and 3, and
1535 * to make sure that the content merge puts HEAD before the other
1536 * branch if we just ensure that branch1 == opt->branch1. So, simply
1537 * flip arguments around if we don't have that.
1538 */
1539 if (branch1 != opt->branch1) {
1540 return handle_file_collision(opt, collide_path,
1541 prev_path2, prev_path1,
1542 branch2, branch1,
1543 b, a);
1544 }
1545
1546 /*
1547 * In the recursive case, we just opt to undo renames
1548 */
1549 if (opt->call_depth && (prev_path1 || prev_path2)) {
1550 /* Put first file (a->oid, a->mode) in its original spot */
1551 if (prev_path1) {
1552 if (update_file(opt, 1, a, prev_path1))
1553 return -1;
1554 } else {
1555 if (update_file(opt, 1, a, collide_path))
1556 return -1;
1557 }
1558
1559 /* Put second file (b->oid, b->mode) in its original spot */
1560 if (prev_path2) {
1561 if (update_file(opt, 1, b, prev_path2))
1562 return -1;
1563 } else {
1564 if (update_file(opt, 1, b, collide_path))
1565 return -1;
1566 }
1567
1568 /* Don't leave something at collision path if unrenaming both */
1569 if (prev_path1 && prev_path2)
1570 remove_file(opt, 1, collide_path, 0);
1571
1572 return 0;
1573 }
1574
1575 /* Remove rename sources if rename/add or rename/rename(2to1) */
1576 if (prev_path1)
1577 remove_file(opt, 1, prev_path1,
1578 opt->call_depth || would_lose_untracked(opt, prev_path1));
1579 if (prev_path2)
1580 remove_file(opt, 1, prev_path2,
1581 opt->call_depth || would_lose_untracked(opt, prev_path2));
1582
1583 /*
1584 * Remove the collision path, if it wouldn't cause dirty contents
1585 * or an untracked file to get lost. We'll either overwrite with
1586 * merged contents, or just write out to differently named files.
1587 */
1588 if (was_dirty(opt, collide_path)) {
1589 output(opt, 1, _("Refusing to lose dirty file at %s"),
1590 collide_path);
1591 update_path = alt_path = unique_path(opt, collide_path, "merged");
1592 } else if (would_lose_untracked(opt, collide_path)) {
1593 /*
1594 * Only way we get here is if both renames were from
1595 * a directory rename AND user had an untracked file
1596 * at the location where both files end up after the
1597 * two directory renames. See testcase 10d of t6043.
1598 */
1599 output(opt, 1, _("Refusing to lose untracked file at "
1600 "%s, even though it's in the way."),
1601 collide_path);
1602 update_path = alt_path = unique_path(opt, collide_path, "merged");
1603 } else {
1604 /*
1605 * FIXME: It's possible that the two files are identical
1606 * and that the current working copy happens to match, in
1607 * which case we are unnecessarily touching the working
1608 * tree file. It's not a likely enough scenario that I
1609 * want to code up the checks for it and a better fix is
1610 * available if we restructure how unpack_trees() and
1611 * merge-recursive interoperate anyway, so punting for
1612 * now...
1613 */
1614 remove_file(opt, 0, collide_path, 0);
1615 }
1616
1617 /* Store things in diff_filespecs for functions that need it */
1618 null.path = (char *)collide_path;
1619 oidcpy(&null.oid, &null_oid);
1620 null.mode = 0;
1621
1622 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1623 branch1, branch2, opt->call_depth * 2, &mfi))
1624 return -1;
1625 mfi.clean &= !alt_path;
1626 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1627 return -1;
1628 if (!mfi.clean && !opt->call_depth &&
1629 update_stages(opt, collide_path, NULL, a, b))
1630 return -1;
1631 free(alt_path);
1632 /*
1633 * FIXME: If both a & b both started with conflicts (only possible
1634 * if they came from a rename/rename(2to1)), but had IDENTICAL
1635 * contents including those conflicts, then in the next line we claim
1636 * it was clean. If someone cares about this case, we should have the
1637 * caller notify us if we started with conflicts.
1638 */
1639 return mfi.clean;
1640 }
1641
1642 static int handle_rename_add(struct merge_options *opt,
1643 struct rename_conflict_info *ci)
1644 {
1645 /* a was renamed to c, and a separate c was added. */
1646 struct diff_filespec *a = ci->ren1->pair->one;
1647 struct diff_filespec *c = ci->ren1->pair->two;
1648 char *path = c->path;
1649 char *prev_path_desc;
1650 struct merge_file_info mfi;
1651
1652 const char *rename_branch = ci->ren1->branch;
1653 const char *add_branch = (opt->branch1 == rename_branch ?
1654 opt->branch2 : opt->branch1);
1655 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1656
1657 output(opt, 1, _("CONFLICT (rename/add): "
1658 "Rename %s->%s in %s. Added %s in %s"),
1659 a->path, c->path, rename_branch,
1660 c->path, add_branch);
1661
1662 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1663 ci->ren1->src_entry->stages[other_stage].path = a->path;
1664 if (merge_mode_and_contents(opt, a, c,
1665 &ci->ren1->src_entry->stages[other_stage],
1666 prev_path_desc,
1667 opt->branch1, opt->branch2,
1668 1 + opt->call_depth * 2, &mfi))
1669 return -1;
1670 free(prev_path_desc);
1671
1672 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1673 return handle_file_collision(opt,
1674 c->path, a->path, NULL,
1675 rename_branch, add_branch,
1676 &mfi.blob,
1677 &ci->ren1->dst_entry->stages[other_stage]);
1678 }
1679
1680 static char *find_path_for_conflict(struct merge_options *opt,
1681 const char *path,
1682 const char *branch1,
1683 const char *branch2)
1684 {
1685 char *new_path = NULL;
1686 if (dir_in_way(opt->repo->index, path, !opt->call_depth, 0)) {
1687 new_path = unique_path(opt, path, branch1);
1688 output(opt, 1, _("%s is a directory in %s adding "
1689 "as %s instead"),
1690 path, branch2, new_path);
1691 } else if (would_lose_untracked(opt, path)) {
1692 new_path = unique_path(opt, path, branch1);
1693 output(opt, 1, _("Refusing to lose untracked file"
1694 " at %s; adding as %s instead"),
1695 path, new_path);
1696 }
1697
1698 return new_path;
1699 }
1700
1701 static int handle_rename_rename_1to2(struct merge_options *opt,
1702 struct rename_conflict_info *ci)
1703 {
1704 /* One file was renamed in both branches, but to different names. */
1705 struct merge_file_info mfi;
1706 struct diff_filespec *add;
1707 struct diff_filespec *o = ci->ren1->pair->one;
1708 struct diff_filespec *a = ci->ren1->pair->two;
1709 struct diff_filespec *b = ci->ren2->pair->two;
1710 char *path_desc;
1711
1712 output(opt, 1, _("CONFLICT (rename/rename): "
1713 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1714 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1715 o->path, a->path, ci->ren1->branch,
1716 o->path, b->path, ci->ren2->branch,
1717 opt->call_depth ? _(" (left unresolved)") : "");
1718
1719 path_desc = xstrfmt("%s and %s, both renamed from %s",
1720 a->path, b->path, o->path);
1721 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1722 ci->ren1->branch, ci->ren2->branch,
1723 opt->call_depth * 2, &mfi))
1724 return -1;
1725 free(path_desc);
1726
1727 if (opt->call_depth) {
1728 /*
1729 * FIXME: For rename/add-source conflicts (if we could detect
1730 * such), this is wrong. We should instead find a unique
1731 * pathname and then either rename the add-source file to that
1732 * unique path, or use that unique path instead of src here.
1733 */
1734 if (update_file(opt, 0, &mfi.blob, o->path))
1735 return -1;
1736
1737 /*
1738 * Above, we put the merged content at the merge-base's
1739 * path. Now we usually need to delete both a->path and
1740 * b->path. However, the rename on each side of the merge
1741 * could also be involved in a rename/add conflict. In
1742 * such cases, we should keep the added file around,
1743 * resolving the conflict at that path in its favor.
1744 */
1745 add = &ci->ren1->dst_entry->stages[2 ^ 1];
1746 if (is_valid(add)) {
1747 if (update_file(opt, 0, add, a->path))
1748 return -1;
1749 }
1750 else
1751 remove_file_from_index(opt->repo->index, a->path);
1752 add = &ci->ren2->dst_entry->stages[3 ^ 1];
1753 if (is_valid(add)) {
1754 if (update_file(opt, 0, add, b->path))
1755 return -1;
1756 }
1757 else
1758 remove_file_from_index(opt->repo->index, b->path);
1759 } else {
1760 /*
1761 * For each destination path, we need to see if there is a
1762 * rename/add collision. If not, we can write the file out
1763 * to the specified location.
1764 */
1765 add = &ci->ren1->dst_entry->stages[2 ^ 1];
1766 if (is_valid(add)) {
1767 add->path = mfi.blob.path = a->path;
1768 if (handle_file_collision(opt, a->path,
1769 NULL, NULL,
1770 ci->ren1->branch,
1771 ci->ren2->branch,
1772 &mfi.blob, add) < 0)
1773 return -1;
1774 } else {
1775 char *new_path = find_path_for_conflict(opt, a->path,
1776 ci->ren1->branch,
1777 ci->ren2->branch);
1778 if (update_file(opt, 0, &mfi.blob,
1779 new_path ? new_path : a->path))
1780 return -1;
1781 free(new_path);
1782 if (update_stages(opt, a->path, NULL, a, NULL))
1783 return -1;
1784 }
1785
1786 add = &ci->ren2->dst_entry->stages[3 ^ 1];
1787 if (is_valid(add)) {
1788 add->path = mfi.blob.path = b->path;
1789 if (handle_file_collision(opt, b->path,
1790 NULL, NULL,
1791 ci->ren1->branch,
1792 ci->ren2->branch,
1793 add, &mfi.blob) < 0)
1794 return -1;
1795 } else {
1796 char *new_path = find_path_for_conflict(opt, b->path,
1797 ci->ren2->branch,
1798 ci->ren1->branch);
1799 if (update_file(opt, 0, &mfi.blob,
1800 new_path ? new_path : b->path))
1801 return -1;
1802 free(new_path);
1803 if (update_stages(opt, b->path, NULL, NULL, b))
1804 return -1;
1805 }
1806 }
1807
1808 return 0;
1809 }
1810
1811 static int handle_rename_rename_2to1(struct merge_options *opt,
1812 struct rename_conflict_info *ci)
1813 {
1814 /* Two files, a & b, were renamed to the same thing, c. */
1815 struct diff_filespec *a = ci->ren1->pair->one;
1816 struct diff_filespec *b = ci->ren2->pair->one;
1817 struct diff_filespec *c1 = ci->ren1->pair->two;
1818 struct diff_filespec *c2 = ci->ren2->pair->two;
1819 char *path = c1->path; /* == c2->path */
1820 char *path_side_1_desc;
1821 char *path_side_2_desc;
1822 struct merge_file_info mfi_c1;
1823 struct merge_file_info mfi_c2;
1824 int ostage1, ostage2;
1825
1826 output(opt, 1, _("CONFLICT (rename/rename): "
1827 "Rename %s->%s in %s. "
1828 "Rename %s->%s in %s"),
1829 a->path, c1->path, ci->ren1->branch,
1830 b->path, c2->path, ci->ren2->branch);
1831
1832 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1833 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1834 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1835 ostage2 = ostage1 ^ 1;
1836 ci->ren1->src_entry->stages[ostage1].path = a->path;
1837 ci->ren2->src_entry->stages[ostage2].path = b->path;
1838 if (merge_mode_and_contents(opt, a, c1,
1839 &ci->ren1->src_entry->stages[ostage1],
1840 path_side_1_desc,
1841 opt->branch1, opt->branch2,
1842 1 + opt->call_depth * 2, &mfi_c1) ||
1843 merge_mode_and_contents(opt, b,
1844 &ci->ren2->src_entry->stages[ostage2],
1845 c2, path_side_2_desc,
1846 opt->branch1, opt->branch2,
1847 1 + opt->call_depth * 2, &mfi_c2))
1848 return -1;
1849 free(path_side_1_desc);
1850 free(path_side_2_desc);
1851 mfi_c1.blob.path = path;
1852 mfi_c2.blob.path = path;
1853
1854 return handle_file_collision(opt, path, a->path, b->path,
1855 ci->ren1->branch, ci->ren2->branch,
1856 &mfi_c1.blob, &mfi_c2.blob);
1857 }
1858
1859 /*
1860 * Get the diff_filepairs changed between o_tree and tree.
1861 */
1862 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1863 struct tree *o_tree,
1864 struct tree *tree)
1865 {
1866 struct diff_queue_struct *ret;
1867 struct diff_options opts;
1868
1869 repo_diff_setup(opt->repo, &opts);
1870 opts.flags.recursive = 1;
1871 opts.flags.rename_empty = 0;
1872 opts.detect_rename = merge_detect_rename(opt);
1873 /*
1874 * We do not have logic to handle the detection of copies. In
1875 * fact, it may not even make sense to add such logic: would we
1876 * really want a change to a base file to be propagated through
1877 * multiple other files by a merge?
1878 */
1879 if (opts.detect_rename > DIFF_DETECT_RENAME)
1880 opts.detect_rename = DIFF_DETECT_RENAME;
1881 opts.rename_limit = opt->merge_rename_limit >= 0 ? opt->merge_rename_limit :
1882 opt->diff_rename_limit >= 0 ? opt->diff_rename_limit :
1883 1000;
1884 opts.rename_score = opt->rename_score;
1885 opts.show_rename_progress = opt->show_rename_progress;
1886 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1887 diff_setup_done(&opts);
1888 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1889 diffcore_std(&opts);
1890 if (opts.needed_rename_limit > opt->needed_rename_limit)
1891 opt->needed_rename_limit = opts.needed_rename_limit;
1892
1893 ret = xmalloc(sizeof(*ret));
1894 *ret = diff_queued_diff;
1895
1896 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1897 diff_queued_diff.nr = 0;
1898 diff_queued_diff.queue = NULL;
1899 diff_flush(&opts);
1900 return ret;
1901 }
1902
1903 static int tree_has_path(struct repository *r, struct tree *tree,
1904 const char *path)
1905 {
1906 struct object_id hashy;
1907 unsigned short mode_o;
1908
1909 return !get_tree_entry(r,
1910 &tree->object.oid, path,
1911 &hashy, &mode_o);
1912 }
1913
1914 /*
1915 * Return a new string that replaces the beginning portion (which matches
1916 * entry->dir), with entry->new_dir. In perl-speak:
1917 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1918 * NOTE:
1919 * Caller must ensure that old_path starts with entry->dir + '/'.
1920 */
1921 static char *apply_dir_rename(struct dir_rename_entry *entry,
1922 const char *old_path)
1923 {
1924 struct strbuf new_path = STRBUF_INIT;
1925 int oldlen, newlen;
1926
1927 if (entry->non_unique_new_dir)
1928 return NULL;
1929
1930 oldlen = strlen(entry->dir);
1931 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1932 strbuf_grow(&new_path, newlen);
1933 strbuf_addbuf(&new_path, &entry->new_dir);
1934 strbuf_addstr(&new_path, &old_path[oldlen]);
1935
1936 return strbuf_detach(&new_path, NULL);
1937 }
1938
1939 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1940 char **old_dir, char **new_dir)
1941 {
1942 char *end_of_old, *end_of_new;
1943 int old_len, new_len;
1944
1945 *old_dir = NULL;
1946 *new_dir = NULL;
1947
1948 /*
1949 * For
1950 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1951 * the "e/foo.c" part is the same, we just want to know that
1952 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1953 * so, for this example, this function returns "a/b/c/d" in
1954 * *old_dir and "a/b/some/thing/else" in *new_dir.
1955 *
1956 * Also, if the basename of the file changed, we don't care. We
1957 * want to know which portion of the directory, if any, changed.
1958 */
1959 end_of_old = strrchr(old_path, '/');
1960 end_of_new = strrchr(new_path, '/');
1961
1962 if (end_of_old == NULL || end_of_new == NULL)
1963 return;
1964 while (*--end_of_new == *--end_of_old &&
1965 end_of_old != old_path &&
1966 end_of_new != new_path)
1967 ; /* Do nothing; all in the while loop */
1968 /*
1969 * We've found the first non-matching character in the directory
1970 * paths. That means the current directory we were comparing
1971 * represents the rename. Move end_of_old and end_of_new back
1972 * to the full directory name.
1973 */
1974 if (*end_of_old == '/')
1975 end_of_old++;
1976 if (*end_of_old != '/')
1977 end_of_new++;
1978 end_of_old = strchr(end_of_old, '/');
1979 end_of_new = strchr(end_of_new, '/');
1980
1981 /*
1982 * It may have been the case that old_path and new_path were the same
1983 * directory all along. Don't claim a rename if they're the same.
1984 */
1985 old_len = end_of_old - old_path;
1986 new_len = end_of_new - new_path;
1987
1988 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1989 *old_dir = xstrndup(old_path, old_len);
1990 *new_dir = xstrndup(new_path, new_len);
1991 }
1992 }
1993
1994 static void remove_hashmap_entries(struct hashmap *dir_renames,
1995 struct string_list *items_to_remove)
1996 {
1997 int i;
1998 struct dir_rename_entry *entry;
1999
2000 for (i = 0; i < items_to_remove->nr; i++) {
2001 entry = items_to_remove->items[i].util;
2002 hashmap_remove(dir_renames, entry, NULL);
2003 }
2004 string_list_clear(items_to_remove, 0);
2005 }
2006
2007 /*
2008 * See if there is a directory rename for path, and if there are any file
2009 * level conflicts for the renamed location. If there is a rename and
2010 * there are no conflicts, return the new name. Otherwise, return NULL.
2011 */
2012 static char *handle_path_level_conflicts(struct merge_options *opt,
2013 const char *path,
2014 struct dir_rename_entry *entry,
2015 struct hashmap *collisions,
2016 struct tree *tree)
2017 {
2018 char *new_path = NULL;
2019 struct collision_entry *collision_ent;
2020 int clean = 1;
2021 struct strbuf collision_paths = STRBUF_INIT;
2022
2023 /*
2024 * entry has the mapping of old directory name to new directory name
2025 * that we want to apply to path.
2026 */
2027 new_path = apply_dir_rename(entry, path);
2028
2029 if (!new_path) {
2030 /* This should only happen when entry->non_unique_new_dir set */
2031 if (!entry->non_unique_new_dir)
2032 BUG("entry->non_unqiue_dir not set and !new_path");
2033 output(opt, 1, _("CONFLICT (directory rename split): "
2034 "Unclear where to place %s because directory "
2035 "%s was renamed to multiple other directories, "
2036 "with no destination getting a majority of the "
2037 "files."),
2038 path, entry->dir);
2039 clean = 0;
2040 return NULL;
2041 }
2042
2043 /*
2044 * The caller needs to have ensured that it has pre-populated
2045 * collisions with all paths that map to new_path. Do a quick check
2046 * to ensure that's the case.
2047 */
2048 collision_ent = collision_find_entry(collisions, new_path);
2049 if (collision_ent == NULL)
2050 BUG("collision_ent is NULL");
2051
2052 /*
2053 * Check for one-sided add/add/.../add conflicts, i.e.
2054 * where implicit renames from the other side doing
2055 * directory rename(s) can affect this side of history
2056 * to put multiple paths into the same location. Warn
2057 * and bail on directory renames for such paths.
2058 */
2059 if (collision_ent->reported_already) {
2060 clean = 0;
2061 } else if (tree_has_path(opt->repo, tree, new_path)) {
2062 collision_ent->reported_already = 1;
2063 strbuf_add_separated_string_list(&collision_paths, ", ",
2064 &collision_ent->source_files);
2065 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2066 "file/dir at %s in the way of implicit "
2067 "directory rename(s) putting the following "
2068 "path(s) there: %s."),
2069 new_path, collision_paths.buf);
2070 clean = 0;
2071 } else if (collision_ent->source_files.nr > 1) {
2072 collision_ent->reported_already = 1;
2073 strbuf_add_separated_string_list(&collision_paths, ", ",
2074 &collision_ent->source_files);
2075 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2076 "more than one path to %s; implicit directory "
2077 "renames tried to put these paths there: %s"),
2078 new_path, collision_paths.buf);
2079 clean = 0;
2080 }
2081
2082 /* Free memory we no longer need */
2083 strbuf_release(&collision_paths);
2084 if (!clean && new_path) {
2085 free(new_path);
2086 return NULL;
2087 }
2088
2089 return new_path;
2090 }
2091
2092 /*
2093 * There are a couple things we want to do at the directory level:
2094 * 1. Check for both sides renaming to the same thing, in order to avoid
2095 * implicit renaming of files that should be left in place. (See
2096 * testcase 6b in t6043 for details.)
2097 * 2. Prune directory renames if there are still files left in the
2098 * the original directory. These represent a partial directory rename,
2099 * i.e. a rename where only some of the files within the directory
2100 * were renamed elsewhere. (Technically, this could be done earlier
2101 * in get_directory_renames(), except that would prevent us from
2102 * doing the previous check and thus failing testcase 6b.)
2103 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2104 * In the future, we could potentially record this info as well and
2105 * omit reporting rename/rename(1to2) conflicts for each path within
2106 * the affected directories, thus cleaning up the merge output.
2107 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2108 * directory level, because merging directories is fine. If it
2109 * causes conflicts for files within those merged directories, then
2110 * that should be detected at the individual path level.
2111 */
2112 static void handle_directory_level_conflicts(struct merge_options *opt,
2113 struct hashmap *dir_re_head,
2114 struct tree *head,
2115 struct hashmap *dir_re_merge,
2116 struct tree *merge)
2117 {
2118 struct hashmap_iter iter;
2119 struct dir_rename_entry *head_ent;
2120 struct dir_rename_entry *merge_ent;
2121
2122 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2123 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2124
2125 hashmap_iter_init(dir_re_head, &iter);
2126 while ((head_ent = hashmap_iter_next(&iter))) {
2127 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2128 if (merge_ent &&
2129 !head_ent->non_unique_new_dir &&
2130 !merge_ent->non_unique_new_dir &&
2131 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2132 /* 1. Renamed identically; remove it from both sides */
2133 string_list_append(&remove_from_head,
2134 head_ent->dir)->util = head_ent;
2135 strbuf_release(&head_ent->new_dir);
2136 string_list_append(&remove_from_merge,
2137 merge_ent->dir)->util = merge_ent;
2138 strbuf_release(&merge_ent->new_dir);
2139 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2140 /* 2. This wasn't a directory rename after all */
2141 string_list_append(&remove_from_head,
2142 head_ent->dir)->util = head_ent;
2143 strbuf_release(&head_ent->new_dir);
2144 }
2145 }
2146
2147 remove_hashmap_entries(dir_re_head, &remove_from_head);
2148 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2149
2150 hashmap_iter_init(dir_re_merge, &iter);
2151 while ((merge_ent = hashmap_iter_next(&iter))) {
2152 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2153 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2154 /* 2. This wasn't a directory rename after all */
2155 string_list_append(&remove_from_merge,
2156 merge_ent->dir)->util = merge_ent;
2157 } else if (head_ent &&
2158 !head_ent->non_unique_new_dir &&
2159 !merge_ent->non_unique_new_dir) {
2160 /* 3. rename/rename(1to2) */
2161 /*
2162 * We can assume it's not rename/rename(1to1) because
2163 * that was case (1), already checked above. So we
2164 * know that head_ent->new_dir and merge_ent->new_dir
2165 * are different strings.
2166 */
2167 output(opt, 1, _("CONFLICT (rename/rename): "
2168 "Rename directory %s->%s in %s. "
2169 "Rename directory %s->%s in %s"),
2170 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2171 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2172 string_list_append(&remove_from_head,
2173 head_ent->dir)->util = head_ent;
2174 strbuf_release(&head_ent->new_dir);
2175 string_list_append(&remove_from_merge,
2176 merge_ent->dir)->util = merge_ent;
2177 strbuf_release(&merge_ent->new_dir);
2178 }
2179 }
2180
2181 remove_hashmap_entries(dir_re_head, &remove_from_head);
2182 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2183 }
2184
2185 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2186 {
2187 struct hashmap *dir_renames;
2188 struct hashmap_iter iter;
2189 struct dir_rename_entry *entry;
2190 int i;
2191
2192 /*
2193 * Typically, we think of a directory rename as all files from a
2194 * certain directory being moved to a target directory. However,
2195 * what if someone first moved two files from the original
2196 * directory in one commit, and then renamed the directory
2197 * somewhere else in a later commit? At merge time, we just know
2198 * that files from the original directory went to two different
2199 * places, and that the bulk of them ended up in the same place.
2200 * We want each directory rename to represent where the bulk of the
2201 * files from that directory end up; this function exists to find
2202 * where the bulk of the files went.
2203 *
2204 * The first loop below simply iterates through the list of file
2205 * renames, finding out how often each directory rename pair
2206 * possibility occurs.
2207 */
2208 dir_renames = xmalloc(sizeof(*dir_renames));
2209 dir_rename_init(dir_renames);
2210 for (i = 0; i < pairs->nr; ++i) {
2211 struct string_list_item *item;
2212 int *count;
2213 struct diff_filepair *pair = pairs->queue[i];
2214 char *old_dir, *new_dir;
2215
2216 /* File not part of directory rename if it wasn't renamed */
2217 if (pair->status != 'R')
2218 continue;
2219
2220 get_renamed_dir_portion(pair->one->path, pair->two->path,
2221 &old_dir, &new_dir);
2222 if (!old_dir)
2223 /* Directory didn't change at all; ignore this one. */
2224 continue;
2225
2226 entry = dir_rename_find_entry(dir_renames, old_dir);
2227 if (!entry) {
2228 entry = xmalloc(sizeof(*entry));
2229 dir_rename_entry_init(entry, old_dir);
2230 hashmap_put(dir_renames, entry);
2231 } else {
2232 free(old_dir);
2233 }
2234 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2235 if (!item) {
2236 item = string_list_insert(&entry->possible_new_dirs,
2237 new_dir);
2238 item->util = xcalloc(1, sizeof(int));
2239 } else {
2240 free(new_dir);
2241 }
2242 count = item->util;
2243 *count += 1;
2244 }
2245
2246 /*
2247 * For each directory with files moved out of it, we find out which
2248 * target directory received the most files so we can declare it to
2249 * be the "winning" target location for the directory rename. This
2250 * winner gets recorded in new_dir. If there is no winner
2251 * (multiple target directories received the same number of files),
2252 * we set non_unique_new_dir. Once we've determined the winner (or
2253 * that there is no winner), we no longer need possible_new_dirs.
2254 */
2255 hashmap_iter_init(dir_renames, &iter);
2256 while ((entry = hashmap_iter_next(&iter))) {
2257 int max = 0;
2258 int bad_max = 0;
2259 char *best = NULL;
2260
2261 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2262 int *count = entry->possible_new_dirs.items[i].util;
2263
2264 if (*count == max)
2265 bad_max = max;
2266 else if (*count > max) {
2267 max = *count;
2268 best = entry->possible_new_dirs.items[i].string;
2269 }
2270 }
2271 if (bad_max == max)
2272 entry->non_unique_new_dir = 1;
2273 else {
2274 assert(entry->new_dir.len == 0);
2275 strbuf_addstr(&entry->new_dir, best);
2276 }
2277 /*
2278 * The relevant directory sub-portion of the original full
2279 * filepaths were xstrndup'ed before inserting into
2280 * possible_new_dirs, and instead of manually iterating the
2281 * list and free'ing each, just lie and tell
2282 * possible_new_dirs that it did the strdup'ing so that it
2283 * will free them for us.
2284 */
2285 entry->possible_new_dirs.strdup_strings = 1;
2286 string_list_clear(&entry->possible_new_dirs, 1);
2287 }
2288
2289 return dir_renames;
2290 }
2291
2292 static struct dir_rename_entry *check_dir_renamed(const char *path,
2293 struct hashmap *dir_renames)
2294 {
2295 char *temp = xstrdup(path);
2296 char *end;
2297 struct dir_rename_entry *entry = NULL;
2298
2299 while ((end = strrchr(temp, '/'))) {
2300 *end = '\0';
2301 entry = dir_rename_find_entry(dir_renames, temp);
2302 if (entry)
2303 break;
2304 }
2305 free(temp);
2306 return entry;
2307 }
2308
2309 static void compute_collisions(struct hashmap *collisions,
2310 struct hashmap *dir_renames,
2311 struct diff_queue_struct *pairs)
2312 {
2313 int i;
2314
2315 /*
2316 * Multiple files can be mapped to the same path due to directory
2317 * renames done by the other side of history. Since that other
2318 * side of history could have merged multiple directories into one,
2319 * if our side of history added the same file basename to each of
2320 * those directories, then all N of them would get implicitly
2321 * renamed by the directory rename detection into the same path,
2322 * and we'd get an add/add/.../add conflict, and all those adds
2323 * from *this* side of history. This is not representable in the
2324 * index, and users aren't going to easily be able to make sense of
2325 * it. So we need to provide a good warning about what's
2326 * happening, and fall back to no-directory-rename detection
2327 * behavior for those paths.
2328 *
2329 * See testcases 9e and all of section 5 from t6043 for examples.
2330 */
2331 collision_init(collisions);
2332
2333 for (i = 0; i < pairs->nr; ++i) {
2334 struct dir_rename_entry *dir_rename_ent;
2335 struct collision_entry *collision_ent;
2336 char *new_path;
2337 struct diff_filepair *pair = pairs->queue[i];
2338
2339 if (pair->status != 'A' && pair->status != 'R')
2340 continue;
2341 dir_rename_ent = check_dir_renamed(pair->two->path,
2342 dir_renames);
2343 if (!dir_rename_ent)
2344 continue;
2345
2346 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2347 if (!new_path)
2348 /*
2349 * dir_rename_ent->non_unique_new_path is true, which
2350 * means there is no directory rename for us to use,
2351 * which means it won't cause us any additional
2352 * collisions.
2353 */
2354 continue;
2355 collision_ent = collision_find_entry(collisions, new_path);
2356 if (!collision_ent) {
2357 collision_ent = xcalloc(1,
2358 sizeof(struct collision_entry));
2359 hashmap_entry_init(collision_ent, strhash(new_path));
2360 hashmap_put(collisions, collision_ent);
2361 collision_ent->target_file = new_path;
2362 } else {
2363 free(new_path);
2364 }
2365 string_list_insert(&collision_ent->source_files,
2366 pair->two->path);
2367 }
2368 }
2369
2370 static char *check_for_directory_rename(struct merge_options *opt,
2371 const char *path,
2372 struct tree *tree,
2373 struct hashmap *dir_renames,
2374 struct hashmap *dir_rename_exclusions,
2375 struct hashmap *collisions,
2376 int *clean_merge)
2377 {
2378 char *new_path = NULL;
2379 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2380 struct dir_rename_entry *oentry = NULL;
2381
2382 if (!entry)
2383 return new_path;
2384
2385 /*
2386 * This next part is a little weird. We do not want to do an
2387 * implicit rename into a directory we renamed on our side, because
2388 * that will result in a spurious rename/rename(1to2) conflict. An
2389 * example:
2390 * Base commit: dumbdir/afile, otherdir/bfile
2391 * Side 1: smrtdir/afile, otherdir/bfile
2392 * Side 2: dumbdir/afile, dumbdir/bfile
2393 * Here, while working on Side 1, we could notice that otherdir was
2394 * renamed/merged to dumbdir, and change the diff_filepair for
2395 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2396 * 2 will notice the rename from dumbdir to smrtdir, and do the
2397 * transitive rename to move it from dumbdir/bfile to
2398 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2399 * smrtdir, a rename/rename(1to2) conflict. We really just want
2400 * the file to end up in smrtdir. And the way to achieve that is
2401 * to not let Side1 do the rename to dumbdir, since we know that is
2402 * the source of one of our directory renames.
2403 *
2404 * That's why oentry and dir_rename_exclusions is here.
2405 *
2406 * As it turns out, this also prevents N-way transient rename
2407 * confusion; See testcases 9c and 9d of t6043.
2408 */
2409 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2410 if (oentry) {
2411 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2412 "to %s, because %s itself was renamed."),
2413 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2414 } else {
2415 new_path = handle_path_level_conflicts(opt, path, entry,
2416 collisions, tree);
2417 *clean_merge &= (new_path != NULL);
2418 }
2419
2420 return new_path;
2421 }
2422
2423 static void apply_directory_rename_modifications(struct merge_options *opt,
2424 struct diff_filepair *pair,
2425 char *new_path,
2426 struct rename *re,
2427 struct tree *tree,
2428 struct tree *o_tree,
2429 struct tree *a_tree,
2430 struct tree *b_tree,
2431 struct string_list *entries)
2432 {
2433 struct string_list_item *item;
2434 int stage = (tree == a_tree ? 2 : 3);
2435 int update_wd;
2436
2437 /*
2438 * In all cases where we can do directory rename detection,
2439 * unpack_trees() will have read pair->two->path into the
2440 * index and the working copy. We need to remove it so that
2441 * we can instead place it at new_path. It is guaranteed to
2442 * not be untracked (unpack_trees() would have errored out
2443 * saying the file would have been overwritten), but it might
2444 * be dirty, though.
2445 */
2446 update_wd = !was_dirty(opt, pair->two->path);
2447 if (!update_wd)
2448 output(opt, 1, _("Refusing to lose dirty file at %s"),
2449 pair->two->path);
2450 remove_file(opt, 1, pair->two->path, !update_wd);
2451
2452 /* Find or create a new re->dst_entry */
2453 item = string_list_lookup(entries, new_path);
2454 if (item) {
2455 /*
2456 * Since we're renaming on this side of history, and it's
2457 * due to a directory rename on the other side of history
2458 * (which we only allow when the directory in question no
2459 * longer exists on the other side of history), the
2460 * original entry for re->dst_entry is no longer
2461 * necessary...
2462 */
2463 re->dst_entry->processed = 1;
2464
2465 /*
2466 * ...because we'll be using this new one.
2467 */
2468 re->dst_entry = item->util;
2469 } else {
2470 /*
2471 * re->dst_entry is for the before-dir-rename path, and we
2472 * need it to hold information for the after-dir-rename
2473 * path. Before creating a new entry, we need to mark the
2474 * old one as unnecessary (...unless it is shared by
2475 * src_entry, i.e. this didn't use to be a rename, in which
2476 * case we can just allow the normal processing to happen
2477 * for it).
2478 */
2479 if (pair->status == 'R')
2480 re->dst_entry->processed = 1;
2481
2482 re->dst_entry = insert_stage_data(opt->repo, new_path,
2483 o_tree, a_tree, b_tree,
2484 entries);
2485 item = string_list_insert(entries, new_path);
2486 item->util = re->dst_entry;
2487 }
2488
2489 /*
2490 * Update the stage_data with the information about the path we are
2491 * moving into place. That slot will be empty and available for us
2492 * to write to because of the collision checks in
2493 * handle_path_level_conflicts(). In other words,
2494 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2495 * open for us to write to.
2496 *
2497 * It may be tempting to actually update the index at this point as
2498 * well, using update_stages_for_stage_data(), but as per the big
2499 * "NOTE" in update_stages(), doing so will modify the current
2500 * in-memory index which will break calls to would_lose_untracked()
2501 * that we need to make. Instead, we need to just make sure that
2502 * the various handle_rename_*() functions update the index
2503 * explicitly rather than relying on unpack_trees() to have done it.
2504 */
2505 get_tree_entry(opt->repo,
2506 &tree->object.oid,
2507 pair->two->path,
2508 &re->dst_entry->stages[stage].oid,
2509 &re->dst_entry->stages[stage].mode);
2510
2511 /*
2512 * Record the original change status (or 'type' of change). If it
2513 * was originally an add ('A'), this lets us differentiate later
2514 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2515 * otherwise look the same). If it was originally a rename ('R'),
2516 * this lets us remember and report accurately about the transitive
2517 * renaming that occurred via the directory rename detection. Also,
2518 * record the original destination name.
2519 */
2520 re->dir_rename_original_type = pair->status;
2521 re->dir_rename_original_dest = pair->two->path;
2522
2523 /*
2524 * We don't actually look at pair->status again, but it seems
2525 * pedagogically correct to adjust it.
2526 */
2527 pair->status = 'R';
2528
2529 /*
2530 * Finally, record the new location.
2531 */
2532 pair->two->path = new_path;
2533 }
2534
2535 /*
2536 * Get information of all renames which occurred in 'pairs', making use of
2537 * any implicit directory renames inferred from the other side of history.
2538 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2539 * to be able to associate the correct cache entries with the rename
2540 * information; tree is always equal to either a_tree or b_tree.
2541 */
2542 static struct string_list *get_renames(struct merge_options *opt,
2543 const char *branch,
2544 struct diff_queue_struct *pairs,
2545 struct hashmap *dir_renames,
2546 struct hashmap *dir_rename_exclusions,
2547 struct tree *tree,
2548 struct tree *o_tree,
2549 struct tree *a_tree,
2550 struct tree *b_tree,
2551 struct string_list *entries,
2552 int *clean_merge)
2553 {
2554 int i;
2555 struct hashmap collisions;
2556 struct hashmap_iter iter;
2557 struct collision_entry *e;
2558 struct string_list *renames;
2559
2560 compute_collisions(&collisions, dir_renames, pairs);
2561 renames = xcalloc(1, sizeof(struct string_list));
2562
2563 for (i = 0; i < pairs->nr; ++i) {
2564 struct string_list_item *item;
2565 struct rename *re;
2566 struct diff_filepair *pair = pairs->queue[i];
2567 char *new_path; /* non-NULL only with directory renames */
2568
2569 if (pair->status != 'A' && pair->status != 'R') {
2570 diff_free_filepair(pair);
2571 continue;
2572 }
2573 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2574 dir_renames,
2575 dir_rename_exclusions,
2576 &collisions,
2577 clean_merge);
2578 if (pair->status != 'R' && !new_path) {
2579 diff_free_filepair(pair);
2580 continue;
2581 }
2582
2583 re = xmalloc(sizeof(*re));
2584 re->processed = 0;
2585 re->pair = pair;
2586 re->branch = branch;
2587 re->dir_rename_original_type = '\0';
2588 re->dir_rename_original_dest = NULL;
2589 item = string_list_lookup(entries, re->pair->one->path);
2590 if (!item)
2591 re->src_entry = insert_stage_data(opt->repo,
2592 re->pair->one->path,
2593 o_tree, a_tree, b_tree, entries);
2594 else
2595 re->src_entry = item->util;
2596
2597 item = string_list_lookup(entries, re->pair->two->path);
2598 if (!item)
2599 re->dst_entry = insert_stage_data(opt->repo,
2600 re->pair->two->path,
2601 o_tree, a_tree, b_tree, entries);
2602 else
2603 re->dst_entry = item->util;
2604 item = string_list_insert(renames, pair->one->path);
2605 item->util = re;
2606 if (new_path)
2607 apply_directory_rename_modifications(opt, pair, new_path,
2608 re, tree, o_tree,
2609 a_tree, b_tree,
2610 entries);
2611 }
2612
2613 hashmap_iter_init(&collisions, &iter);
2614 while ((e = hashmap_iter_next(&iter))) {
2615 free(e->target_file);
2616 string_list_clear(&e->source_files, 0);
2617 }
2618 hashmap_free(&collisions, 1);
2619 return renames;
2620 }
2621
2622 static int process_renames(struct merge_options *opt,
2623 struct string_list *a_renames,
2624 struct string_list *b_renames)
2625 {
2626 int clean_merge = 1, i, j;
2627 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2628 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2629 const struct rename *sre;
2630
2631 for (i = 0; i < a_renames->nr; i++) {
2632 sre = a_renames->items[i].util;
2633 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2634 = (void *)sre;
2635 }
2636 for (i = 0; i < b_renames->nr; i++) {
2637 sre = b_renames->items[i].util;
2638 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2639 = (void *)sre;
2640 }
2641
2642 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2643 struct string_list *renames1, *renames2Dst;
2644 struct rename *ren1 = NULL, *ren2 = NULL;
2645 const char *ren1_src, *ren1_dst;
2646 struct string_list_item *lookup;
2647
2648 if (i >= a_renames->nr) {
2649 ren2 = b_renames->items[j++].util;
2650 } else if (j >= b_renames->nr) {
2651 ren1 = a_renames->items[i++].util;
2652 } else {
2653 int compare = strcmp(a_renames->items[i].string,
2654 b_renames->items[j].string);
2655 if (compare <= 0)
2656 ren1 = a_renames->items[i++].util;
2657 if (compare >= 0)
2658 ren2 = b_renames->items[j++].util;
2659 }
2660
2661 /* TODO: refactor, so that 1/2 are not needed */
2662 if (ren1) {
2663 renames1 = a_renames;
2664 renames2Dst = &b_by_dst;
2665 } else {
2666 renames1 = b_renames;
2667 renames2Dst = &a_by_dst;
2668 SWAP(ren2, ren1);
2669 }
2670
2671 if (ren1->processed)
2672 continue;
2673 ren1->processed = 1;
2674 ren1->dst_entry->processed = 1;
2675 /* BUG: We should only mark src_entry as processed if we
2676 * are not dealing with a rename + add-source case.
2677 */
2678 ren1->src_entry->processed = 1;
2679
2680 ren1_src = ren1->pair->one->path;
2681 ren1_dst = ren1->pair->two->path;
2682
2683 if (ren2) {
2684 /* One file renamed on both sides */
2685 const char *ren2_src = ren2->pair->one->path;
2686 const char *ren2_dst = ren2->pair->two->path;
2687 enum rename_type rename_type;
2688 if (strcmp(ren1_src, ren2_src) != 0)
2689 BUG("ren1_src != ren2_src");
2690 ren2->dst_entry->processed = 1;
2691 ren2->processed = 1;
2692 if (strcmp(ren1_dst, ren2_dst) != 0) {
2693 rename_type = RENAME_ONE_FILE_TO_TWO;
2694 clean_merge = 0;
2695 } else {
2696 rename_type = RENAME_ONE_FILE_TO_ONE;
2697 /* BUG: We should only remove ren1_src in
2698 * the base stage (think of rename +
2699 * add-source cases).
2700 */
2701 remove_file(opt, 1, ren1_src, 1);
2702 update_entry(ren1->dst_entry,
2703 ren1->pair->one,
2704 ren1->pair->two,
2705 ren2->pair->two);
2706 }
2707 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2708 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2709 /* Two different files renamed to the same thing */
2710 char *ren2_dst;
2711 ren2 = lookup->util;
2712 ren2_dst = ren2->pair->two->path;
2713 if (strcmp(ren1_dst, ren2_dst) != 0)
2714 BUG("ren1_dst != ren2_dst");
2715
2716 clean_merge = 0;
2717 ren2->processed = 1;
2718 /*
2719 * BUG: We should only mark src_entry as processed
2720 * if we are not dealing with a rename + add-source
2721 * case.
2722 */
2723 ren2->src_entry->processed = 1;
2724
2725 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2726 opt, ren1, ren2);
2727 } else {
2728 /* Renamed in 1, maybe changed in 2 */
2729 /* we only use sha1 and mode of these */
2730 struct diff_filespec src_other, dst_other;
2731 int try_merge;
2732
2733 /*
2734 * unpack_trees loads entries from common-commit
2735 * into stage 1, from head-commit into stage 2, and
2736 * from merge-commit into stage 3. We keep track
2737 * of which side corresponds to the rename.
2738 */
2739 int renamed_stage = a_renames == renames1 ? 2 : 3;
2740 int other_stage = a_renames == renames1 ? 3 : 2;
2741
2742 /* BUG: We should only remove ren1_src in the base
2743 * stage and in other_stage (think of rename +
2744 * add-source case).
2745 */
2746 remove_file(opt, 1, ren1_src,
2747 renamed_stage == 2 || !was_tracked(opt, ren1_src));
2748
2749 oidcpy(&src_other.oid,
2750 &ren1->src_entry->stages[other_stage].oid);
2751 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2752 oidcpy(&dst_other.oid,
2753 &ren1->dst_entry->stages[other_stage].oid);
2754 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2755 try_merge = 0;
2756
2757 if (oid_eq(&src_other.oid, &null_oid) &&
2758 ren1->dir_rename_original_type == 'A') {
2759 setup_rename_conflict_info(RENAME_VIA_DIR,
2760 opt, ren1, NULL);
2761 } else if (oid_eq(&src_other.oid, &null_oid)) {
2762 setup_rename_conflict_info(RENAME_DELETE,
2763 opt, ren1, NULL);
2764 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2765 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2766 /*
2767 * Added file on the other side identical to
2768 * the file being renamed: clean merge.
2769 * Also, there is no need to overwrite the
2770 * file already in the working copy, so call
2771 * update_file_flags() instead of
2772 * update_file().
2773 */
2774 if (update_file_flags(opt,
2775 ren1->pair->two,
2776 ren1_dst,
2777 1, /* update_cache */
2778 0 /* update_wd */))
2779 clean_merge = -1;
2780 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2781 /*
2782 * Probably not a clean merge, but it's
2783 * premature to set clean_merge to 0 here,
2784 * because if the rename merges cleanly and
2785 * the merge exactly matches the newly added
2786 * file, then the merge will be clean.
2787 */
2788 setup_rename_conflict_info(RENAME_ADD,
2789 opt, ren1, NULL);
2790 } else
2791 try_merge = 1;
2792
2793 if (clean_merge < 0)
2794 goto cleanup_and_return;
2795 if (try_merge) {
2796 struct diff_filespec *o, *a, *b;
2797 src_other.path = (char *)ren1_src;
2798
2799 o = ren1->pair->one;
2800 if (a_renames == renames1) {
2801 a = ren1->pair->two;
2802 b = &src_other;
2803 } else {
2804 b = ren1->pair->two;
2805 a = &src_other;
2806 }
2807 update_entry(ren1->dst_entry, o, a, b);
2808 setup_rename_conflict_info(RENAME_NORMAL,
2809 opt, ren1, NULL);
2810 }
2811 }
2812 }
2813 cleanup_and_return:
2814 string_list_clear(&a_by_dst, 0);
2815 string_list_clear(&b_by_dst, 0);
2816
2817 return clean_merge;
2818 }
2819
2820 struct rename_info {
2821 struct string_list *head_renames;
2822 struct string_list *merge_renames;
2823 };
2824
2825 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2826 struct hashmap *dir_renames)
2827 {
2828 struct hashmap_iter iter;
2829 struct dir_rename_entry *e;
2830
2831 hashmap_iter_init(dir_renames, &iter);
2832 while ((e = hashmap_iter_next(&iter))) {
2833 free(e->dir);
2834 strbuf_release(&e->new_dir);
2835 /* possible_new_dirs already cleared in get_directory_renames */
2836 }
2837 hashmap_free(dir_renames, 1);
2838 free(dir_renames);
2839
2840 free(pairs->queue);
2841 free(pairs);
2842 }
2843
2844 static int detect_and_process_renames(struct merge_options *opt,
2845 struct tree *common,
2846 struct tree *head,
2847 struct tree *merge,
2848 struct string_list *entries,
2849 struct rename_info *ri)
2850 {
2851 struct diff_queue_struct *head_pairs, *merge_pairs;
2852 struct hashmap *dir_re_head, *dir_re_merge;
2853 int clean = 1;
2854
2855 ri->head_renames = NULL;
2856 ri->merge_renames = NULL;
2857
2858 if (!merge_detect_rename(opt))
2859 return 1;
2860
2861 head_pairs = get_diffpairs(opt, common, head);
2862 merge_pairs = get_diffpairs(opt, common, merge);
2863
2864 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2865 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2866 !opt->call_depth)) {
2867 dir_re_head = get_directory_renames(head_pairs);
2868 dir_re_merge = get_directory_renames(merge_pairs);
2869
2870 handle_directory_level_conflicts(opt,
2871 dir_re_head, head,
2872 dir_re_merge, merge);
2873 } else {
2874 dir_re_head = xmalloc(sizeof(*dir_re_head));
2875 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2876 dir_rename_init(dir_re_head);
2877 dir_rename_init(dir_re_merge);
2878 }
2879
2880 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2881 dir_re_merge, dir_re_head, head,
2882 common, head, merge, entries,
2883 &clean);
2884 if (clean < 0)
2885 goto cleanup;
2886 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2887 dir_re_head, dir_re_merge, merge,
2888 common, head, merge, entries,
2889 &clean);
2890 if (clean < 0)
2891 goto cleanup;
2892 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2893
2894 cleanup:
2895 /*
2896 * Some cleanup is deferred until cleanup_renames() because the
2897 * data structures are still needed and referenced in
2898 * process_entry(). But there are a few things we can free now.
2899 */
2900 initial_cleanup_rename(head_pairs, dir_re_head);
2901 initial_cleanup_rename(merge_pairs, dir_re_merge);
2902
2903 return clean;
2904 }
2905
2906 static void final_cleanup_rename(struct string_list *rename)
2907 {
2908 const struct rename *re;
2909 int i;
2910
2911 if (rename == NULL)
2912 return;
2913
2914 for (i = 0; i < rename->nr; i++) {
2915 re = rename->items[i].util;
2916 diff_free_filepair(re->pair);
2917 }
2918 string_list_clear(rename, 1);
2919 free(rename);
2920 }
2921
2922 static void final_cleanup_renames(struct rename_info *re_info)
2923 {
2924 final_cleanup_rename(re_info->head_renames);
2925 final_cleanup_rename(re_info->merge_renames);
2926 }
2927
2928 static int read_oid_strbuf(struct merge_options *opt,
2929 const struct object_id *oid,
2930 struct strbuf *dst)
2931 {
2932 void *buf;
2933 enum object_type type;
2934 unsigned long size;
2935 buf = read_object_file(oid, &type, &size);
2936 if (!buf)
2937 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
2938 if (type != OBJ_BLOB) {
2939 free(buf);
2940 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
2941 }
2942 strbuf_attach(dst, buf, size, size + 1);
2943 return 0;
2944 }
2945
2946 static int blob_unchanged(struct merge_options *opt,
2947 const struct diff_filespec *o,
2948 const struct diff_filespec *a,
2949 int renormalize, const char *path)
2950 {
2951 struct strbuf obuf = STRBUF_INIT;
2952 struct strbuf abuf = STRBUF_INIT;
2953 int ret = 0; /* assume changed for safety */
2954 const struct index_state *idx = opt->repo->index;
2955
2956 if (a->mode != o->mode)
2957 return 0;
2958 if (oid_eq(&o->oid, &a->oid))
2959 return 1;
2960 if (!renormalize)
2961 return 0;
2962
2963 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
2964 read_oid_strbuf(opt, &a->oid, &abuf))
2965 goto error_return;
2966 /*
2967 * Note: binary | is used so that both renormalizations are
2968 * performed. Comparison can be skipped if both files are
2969 * unchanged since their sha1s have already been compared.
2970 */
2971 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
2972 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
2973 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
2974
2975 error_return:
2976 strbuf_release(&obuf);
2977 strbuf_release(&abuf);
2978 return ret;
2979 }
2980
2981 static int handle_modify_delete(struct merge_options *opt,
2982 const char *path,
2983 const struct diff_filespec *o,
2984 const struct diff_filespec *a,
2985 const struct diff_filespec *b)
2986 {
2987 const char *modify_branch, *delete_branch;
2988 const struct diff_filespec *changed;
2989
2990 if (is_valid(a)) {
2991 modify_branch = opt->branch1;
2992 delete_branch = opt->branch2;
2993 changed = a;
2994 } else {
2995 modify_branch = opt->branch2;
2996 delete_branch = opt->branch1;
2997 changed = b;
2998 }
2999
3000 return handle_change_delete(opt,
3001 path, NULL,
3002 o, changed,
3003 modify_branch, delete_branch,
3004 _("modify"), _("modified"));
3005 }
3006
3007 static int handle_content_merge(struct merge_file_info *mfi,
3008 struct merge_options *opt,
3009 const char *path,
3010 int is_dirty,
3011 const struct diff_filespec *o,
3012 const struct diff_filespec *a,
3013 const struct diff_filespec *b,
3014 struct rename_conflict_info *ci)
3015 {
3016 const char *reason = _("content");
3017 unsigned df_conflict_remains = 0;
3018
3019 if (!is_valid(o))
3020 reason = _("add/add");
3021
3022 assert(o->path && a->path && b->path);
3023 if (ci && dir_in_way(opt->repo->index, path, !opt->call_depth,
3024 S_ISGITLINK(ci->ren1->pair->two->mode)))
3025 df_conflict_remains = 1;
3026
3027 if (merge_mode_and_contents(opt, o, a, b, path,
3028 opt->branch1, opt->branch2,
3029 opt->call_depth * 2, mfi))
3030 return -1;
3031
3032 /*
3033 * We can skip updating the working tree file iff:
3034 * a) The merge is clean
3035 * b) The merge matches what was in HEAD (content, mode, pathname)
3036 * c) The target path is usable (i.e. not involved in D/F conflict)
3037 */
3038 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3039 !df_conflict_remains) {
3040 int pos;
3041 struct cache_entry *ce;
3042
3043 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3044 if (add_cacheinfo(opt, &mfi->blob, path,
3045 0, (!opt->call_depth && !is_dirty), 0))
3046 return -1;
3047 /*
3048 * However, add_cacheinfo() will delete the old cache entry
3049 * and add a new one. We need to copy over any skip_worktree
3050 * flag to avoid making the file appear as if it were
3051 * deleted by the user.
3052 */
3053 pos = index_name_pos(&opt->orig_index, path, strlen(path));
3054 ce = opt->orig_index.cache[pos];
3055 if (ce_skip_worktree(ce)) {
3056 pos = index_name_pos(opt->repo->index, path, strlen(path));
3057 ce = opt->repo->index->cache[pos];
3058 ce->ce_flags |= CE_SKIP_WORKTREE;
3059 }
3060 return mfi->clean;
3061 }
3062
3063 if (!mfi->clean) {
3064 if (S_ISGITLINK(mfi->blob.mode))
3065 reason = _("submodule");
3066 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3067 reason, path);
3068 if (ci && !df_conflict_remains)
3069 if (update_stages(opt, path, o, a, b))
3070 return -1;
3071 }
3072
3073 if (df_conflict_remains || is_dirty) {
3074 char *new_path;
3075 if (opt->call_depth) {
3076 remove_file_from_index(opt->repo->index, path);
3077 } else {
3078 if (!mfi->clean) {
3079 if (update_stages(opt, path, o, a, b))
3080 return -1;
3081 } else {
3082 int file_from_stage2 = was_tracked(opt, path);
3083
3084 if (update_stages(opt, path, NULL,
3085 file_from_stage2 ? &mfi->blob : NULL,
3086 file_from_stage2 ? NULL : &mfi->blob))
3087 return -1;
3088 }
3089
3090 }
3091 new_path = unique_path(opt, path, ci->ren1->branch);
3092 if (is_dirty) {
3093 output(opt, 1, _("Refusing to lose dirty file at %s"),
3094 path);
3095 }
3096 output(opt, 1, _("Adding as %s instead"), new_path);
3097 if (update_file(opt, 0, &mfi->blob, new_path)) {
3098 free(new_path);
3099 return -1;
3100 }
3101 free(new_path);
3102 mfi->clean = 0;
3103 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3104 return -1;
3105 return !is_dirty && mfi->clean;
3106 }
3107
3108 static int handle_rename_normal(struct merge_options *opt,
3109 const char *path,
3110 const struct diff_filespec *o,
3111 const struct diff_filespec *a,
3112 const struct diff_filespec *b,
3113 struct rename_conflict_info *ci)
3114 {
3115 struct rename *ren = ci->ren1;
3116 struct merge_file_info mfi;
3117 int clean;
3118 int side = (ren->branch == opt->branch1 ? 2 : 3);
3119
3120 /* Merge the content and write it out */
3121 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3122 o, a, b, ci);
3123
3124 if (clean &&
3125 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3126 ren->dir_rename_original_dest) {
3127 if (update_stages(opt, path,
3128 NULL,
3129 side == 2 ? &mfi.blob : NULL,
3130 side == 2 ? NULL : &mfi.blob))
3131 return -1;
3132 clean = 0; /* not clean, but conflicted */
3133 }
3134 return clean;
3135 }
3136
3137 static void dir_rename_warning(const char *msg,
3138 int is_add,
3139 int clean,
3140 struct merge_options *opt,
3141 struct rename *ren)
3142 {
3143 const char *other_branch;
3144 other_branch = (ren->branch == opt->branch1 ?
3145 opt->branch2 : opt->branch1);
3146 if (is_add) {
3147 output(opt, clean ? 2 : 1, msg,
3148 ren->pair->one->path, ren->branch,
3149 other_branch, ren->pair->two->path);
3150 return;
3151 }
3152 output(opt, clean ? 2 : 1, msg,
3153 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3154 other_branch, ren->pair->two->path);
3155 }
3156 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3157 struct rename *ren)
3158 {
3159 const char *msg;
3160 int clean = 1, is_add;
3161
3162 if (!ren)
3163 return clean;
3164
3165 /* Return early if ren was not affected/created by a directory rename */
3166 if (!ren->dir_rename_original_dest)
3167 return clean;
3168
3169 /* Sanity checks */
3170 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3171 assert(ren->dir_rename_original_type == 'A' ||
3172 ren->dir_rename_original_type == 'R');
3173
3174 /* Check whether to treat directory renames as a conflict */
3175 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3176
3177 is_add = (ren->dir_rename_original_type == 'A');
3178 if (ren->dir_rename_original_type == 'A' && clean) {
3179 msg = _("Path updated: %s added in %s inside a "
3180 "directory that was renamed in %s; moving it to %s.");
3181 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3182 msg = _("CONFLICT (file location): %s added in %s "
3183 "inside a directory that was renamed in %s, "
3184 "suggesting it should perhaps be moved to %s.");
3185 } else if (ren->dir_rename_original_type == 'R' && clean) {
3186 msg = _("Path updated: %s renamed to %s in %s, inside a "
3187 "directory that was renamed in %s; moving it to %s.");
3188 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3189 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3190 "inside a directory that was renamed in %s, "
3191 "suggesting it should perhaps be moved to %s.");
3192 } else {
3193 BUG("Impossible dir_rename_original_type/clean combination");
3194 }
3195 dir_rename_warning(msg, is_add, clean, opt, ren);
3196
3197 return clean;
3198 }
3199
3200 /* Per entry merge function */
3201 static int process_entry(struct merge_options *opt,
3202 const char *path, struct stage_data *entry)
3203 {
3204 int clean_merge = 1;
3205 int normalize = opt->renormalize;
3206
3207 struct diff_filespec *o = &entry->stages[1];
3208 struct diff_filespec *a = &entry->stages[2];
3209 struct diff_filespec *b = &entry->stages[3];
3210 int o_valid = is_valid(o);
3211 int a_valid = is_valid(a);
3212 int b_valid = is_valid(b);
3213 o->path = a->path = b->path = (char*)path;
3214
3215 entry->processed = 1;
3216 if (entry->rename_conflict_info) {
3217 struct rename_conflict_info *ci = entry->rename_conflict_info;
3218 struct diff_filespec *temp;
3219 int path_clean;
3220
3221 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3222 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3223
3224 /*
3225 * For cases with a single rename, {o,a,b}->path have all been
3226 * set to the rename target path; we need to set two of these
3227 * back to the rename source.
3228 * For rename/rename conflicts, we'll manually fix paths below.
3229 */
3230 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3231 o->path = temp->path = ci->ren1->pair->one->path;
3232 if (ci->ren2) {
3233 assert(opt->branch1 == ci->ren1->branch);
3234 }
3235
3236 switch (ci->rename_type) {
3237 case RENAME_NORMAL:
3238 case RENAME_ONE_FILE_TO_ONE:
3239 clean_merge = handle_rename_normal(opt, path, o, a, b,
3240 ci);
3241 break;
3242 case RENAME_VIA_DIR:
3243 clean_merge = handle_rename_via_dir(opt, ci);
3244 break;
3245 case RENAME_ADD:
3246 /*
3247 * Probably unclean merge, but if the renamed file
3248 * merges cleanly and the result can then be
3249 * two-way merged cleanly with the added file, I
3250 * guess it's a clean merge?
3251 */
3252 clean_merge = handle_rename_add(opt, ci);
3253 break;
3254 case RENAME_DELETE:
3255 clean_merge = 0;
3256 if (handle_rename_delete(opt, ci))
3257 clean_merge = -1;
3258 break;
3259 case RENAME_ONE_FILE_TO_TWO:
3260 /*
3261 * Manually fix up paths; note:
3262 * ren[12]->pair->one->path are equal.
3263 */
3264 o->path = ci->ren1->pair->one->path;
3265 a->path = ci->ren1->pair->two->path;
3266 b->path = ci->ren2->pair->two->path;
3267
3268 clean_merge = 0;
3269 if (handle_rename_rename_1to2(opt, ci))
3270 clean_merge = -1;
3271 break;
3272 case RENAME_TWO_FILES_TO_ONE:
3273 /*
3274 * Manually fix up paths; note,
3275 * ren[12]->pair->two->path are actually equal.
3276 */
3277 o->path = NULL;
3278 a->path = ci->ren1->pair->two->path;
3279 b->path = ci->ren2->pair->two->path;
3280
3281 /*
3282 * Probably unclean merge, but if the two renamed
3283 * files merge cleanly and the two resulting files
3284 * can then be two-way merged cleanly, I guess it's
3285 * a clean merge?
3286 */
3287 clean_merge = handle_rename_rename_2to1(opt, ci);
3288 break;
3289 default:
3290 entry->processed = 0;
3291 break;
3292 }
3293 if (path_clean < clean_merge)
3294 clean_merge = path_clean;
3295 } else if (o_valid && (!a_valid || !b_valid)) {
3296 /* Case A: Deleted in one */
3297 if ((!a_valid && !b_valid) ||
3298 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3299 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3300 /* Deleted in both or deleted in one and
3301 * unchanged in the other */
3302 if (a_valid)
3303 output(opt, 2, _("Removing %s"), path);
3304 /* do not touch working file if it did not exist */
3305 remove_file(opt, 1, path, !a_valid);
3306 } else {
3307 /* Modify/delete; deleted side may have put a directory in the way */
3308 clean_merge = 0;
3309 if (handle_modify_delete(opt, path, o, a, b))
3310 clean_merge = -1;
3311 }
3312 } else if ((!o_valid && a_valid && !b_valid) ||
3313 (!o_valid && !a_valid && b_valid)) {
3314 /* Case B: Added in one. */
3315 /* [nothing|directory] -> ([nothing|directory], file) */
3316
3317 const char *add_branch;
3318 const char *other_branch;
3319 const char *conf;
3320 const struct diff_filespec *contents;
3321
3322 if (a_valid) {
3323 add_branch = opt->branch1;
3324 other_branch = opt->branch2;
3325 contents = a;
3326 conf = _("file/directory");
3327 } else {
3328 add_branch = opt->branch2;
3329 other_branch = opt->branch1;
3330 contents = b;
3331 conf = _("directory/file");
3332 }
3333 if (dir_in_way(opt->repo->index, path,
3334 !opt->call_depth && !S_ISGITLINK(a->mode),
3335 0)) {
3336 char *new_path = unique_path(opt, path, add_branch);
3337 clean_merge = 0;
3338 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3339 "Adding %s as %s"),
3340 conf, path, other_branch, path, new_path);
3341 if (update_file(opt, 0, contents, new_path))
3342 clean_merge = -1;
3343 else if (opt->call_depth)
3344 remove_file_from_index(opt->repo->index, path);
3345 free(new_path);
3346 } else {
3347 output(opt, 2, _("Adding %s"), path);
3348 /* do not overwrite file if already present */
3349 if (update_file_flags(opt, contents, path, 1, !a_valid))
3350 clean_merge = -1;
3351 }
3352 } else if (a_valid && b_valid) {
3353 if (!o_valid) {
3354 /* Case C: Added in both (check for same permissions) */
3355 output(opt, 1,
3356 _("CONFLICT (add/add): Merge conflict in %s"),
3357 path);
3358 clean_merge = handle_file_collision(opt,
3359 path, NULL, NULL,
3360 opt->branch1,
3361 opt->branch2,
3362 a, b);
3363 } else {
3364 /* case D: Modified in both, but differently. */
3365 struct merge_file_info mfi;
3366 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3367 clean_merge = handle_content_merge(&mfi, opt, path,
3368 is_dirty,
3369 o, a, b, NULL);
3370 }
3371 } else if (!o_valid && !a_valid && !b_valid) {
3372 /*
3373 * this entry was deleted altogether. a_mode == 0 means
3374 * we had that path and want to actively remove it.
3375 */
3376 remove_file(opt, 1, path, !a->mode);
3377 } else
3378 BUG("fatal merge failure, shouldn't happen.");
3379
3380 return clean_merge;
3381 }
3382
3383 int merge_trees(struct merge_options *opt,
3384 struct tree *head,
3385 struct tree *merge,
3386 struct tree *common,
3387 struct tree **result)
3388 {
3389 struct index_state *istate = opt->repo->index;
3390 int code, clean;
3391 struct strbuf sb = STRBUF_INIT;
3392
3393 assert(opt->ancestor != NULL);
3394
3395 if (!opt->call_depth && repo_index_has_changes(opt->repo, head, &sb)) {
3396 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3397 sb.buf);
3398 return -1;
3399 }
3400
3401 if (opt->subtree_shift) {
3402 merge = shift_tree_object(opt->repo, head, merge, opt->subtree_shift);
3403 common = shift_tree_object(opt->repo, head, common, opt->subtree_shift);
3404 }
3405
3406 if (oid_eq(&common->object.oid, &merge->object.oid)) {
3407 output(opt, 0, _("Already up to date!"));
3408 *result = head;
3409 return 1;
3410 }
3411
3412 code = unpack_trees_start(opt, common, head, merge);
3413
3414 if (code != 0) {
3415 if (show(opt, 4) || opt->call_depth)
3416 err(opt, _("merging of trees %s and %s failed"),
3417 oid_to_hex(&head->object.oid),
3418 oid_to_hex(&merge->object.oid));
3419 unpack_trees_finish(opt);
3420 return -1;
3421 }
3422
3423 if (unmerged_index(istate)) {
3424 struct string_list *entries;
3425 struct rename_info re_info;
3426 int i;
3427 /*
3428 * Only need the hashmap while processing entries, so
3429 * initialize it here and free it when we are done running
3430 * through the entries. Keeping it in the merge_options as
3431 * opposed to decaring a local hashmap is for convenience
3432 * so that we don't have to pass it to around.
3433 */
3434 hashmap_init(&opt->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3435 get_files_dirs(opt, head);
3436 get_files_dirs(opt, merge);
3437
3438 entries = get_unmerged(opt->repo->index);
3439 clean = detect_and_process_renames(opt, common, head, merge,
3440 entries, &re_info);
3441 record_df_conflict_files(opt, entries);
3442 if (clean < 0)
3443 goto cleanup;
3444 for (i = entries->nr-1; 0 <= i; i--) {
3445 const char *path = entries->items[i].string;
3446 struct stage_data *e = entries->items[i].util;
3447 if (!e->processed) {
3448 int ret = process_entry(opt, path, e);
3449 if (!ret)
3450 clean = 0;
3451 else if (ret < 0) {
3452 clean = ret;
3453 goto cleanup;
3454 }
3455 }
3456 }
3457 for (i = 0; i < entries->nr; i++) {
3458 struct stage_data *e = entries->items[i].util;
3459 if (!e->processed)
3460 BUG("unprocessed path??? %s",
3461 entries->items[i].string);
3462 }
3463
3464 cleanup:
3465 final_cleanup_renames(&re_info);
3466
3467 string_list_clear(entries, 1);
3468 free(entries);
3469
3470 hashmap_free(&opt->current_file_dir_set, 1);
3471
3472 if (clean < 0) {
3473 unpack_trees_finish(opt);
3474 return clean;
3475 }
3476 }
3477 else
3478 clean = 1;
3479
3480 unpack_trees_finish(opt);
3481
3482 if (opt->call_depth && !(*result = write_tree_from_memory(opt)))
3483 return -1;
3484
3485 return clean;
3486 }
3487
3488 static struct commit_list *reverse_commit_list(struct commit_list *list)
3489 {
3490 struct commit_list *next = NULL, *current, *backup;
3491 for (current = list; current; current = backup) {
3492 backup = current->next;
3493 current->next = next;
3494 next = current;
3495 }
3496 return next;
3497 }
3498
3499 /*
3500 * Merge the commits h1 and h2, return the resulting virtual
3501 * commit object and a flag indicating the cleanness of the merge.
3502 */
3503 int merge_recursive(struct merge_options *opt,
3504 struct commit *h1,
3505 struct commit *h2,
3506 struct commit_list *ca,
3507 struct commit **result)
3508 {
3509 struct commit_list *iter;
3510 struct commit *merged_common_ancestors;
3511 struct tree *mrtree;
3512 int clean;
3513 const char *ancestor_name;
3514 struct strbuf merge_base_abbrev = STRBUF_INIT;
3515
3516 if (!opt->call_depth)
3517 assert(opt->ancestor == NULL);
3518
3519 if (show(opt, 4)) {
3520 output(opt, 4, _("Merging:"));
3521 output_commit_title(opt, h1);
3522 output_commit_title(opt, h2);
3523 }
3524
3525 if (!ca) {
3526 ca = get_merge_bases(h1, h2);
3527 ca = reverse_commit_list(ca);
3528 }
3529
3530 if (show(opt, 5)) {
3531 unsigned cnt = commit_list_count(ca);
3532
3533 output(opt, 5, Q_("found %u common ancestor:",
3534 "found %u common ancestors:", cnt), cnt);
3535 for (iter = ca; iter; iter = iter->next)
3536 output_commit_title(opt, iter->item);
3537 }
3538
3539 merged_common_ancestors = pop_commit(&ca);
3540 if (merged_common_ancestors == NULL) {
3541 /* if there is no common ancestor, use an empty tree */
3542 struct tree *tree;
3543
3544 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3545 merged_common_ancestors = make_virtual_commit(opt->repo, tree, "ancestor");
3546 ancestor_name = "empty tree";
3547 } else if (ca) {
3548 ancestor_name = "merged common ancestors";
3549 } else {
3550 strbuf_add_unique_abbrev(&merge_base_abbrev,
3551 &merged_common_ancestors->object.oid,
3552 DEFAULT_ABBREV);
3553 ancestor_name = merge_base_abbrev.buf;
3554 }
3555
3556 for (iter = ca; iter; iter = iter->next) {
3557 const char *saved_b1, *saved_b2;
3558 opt->call_depth++;
3559 /*
3560 * When the merge fails, the result contains files
3561 * with conflict markers. The cleanness flag is
3562 * ignored (unless indicating an error), it was never
3563 * actually used, as result of merge_trees has always
3564 * overwritten it: the committed "conflicts" were
3565 * already resolved.
3566 */
3567 discard_index(opt->repo->index);
3568 saved_b1 = opt->branch1;
3569 saved_b2 = opt->branch2;
3570 opt->branch1 = "Temporary merge branch 1";
3571 opt->branch2 = "Temporary merge branch 2";
3572 if (merge_recursive(opt, merged_common_ancestors, iter->item,
3573 NULL, &merged_common_ancestors) < 0)
3574 return -1;
3575 opt->branch1 = saved_b1;
3576 opt->branch2 = saved_b2;
3577 opt->call_depth--;
3578
3579 if (!merged_common_ancestors)
3580 return err(opt, _("merge returned no commit"));
3581 }
3582
3583 discard_index(opt->repo->index);
3584 if (!opt->call_depth)
3585 repo_read_index(opt->repo);
3586
3587 opt->ancestor = ancestor_name;
3588 clean = merge_trees(opt, get_commit_tree(h1), get_commit_tree(h2),
3589 get_commit_tree(merged_common_ancestors),
3590 &mrtree);
3591 strbuf_release(&merge_base_abbrev);
3592 if (clean < 0) {
3593 flush_output(opt);
3594 return clean;
3595 }
3596
3597 if (opt->call_depth) {
3598 *result = make_virtual_commit(opt->repo, mrtree, "merged tree");
3599 commit_list_insert(h1, &(*result)->parents);
3600 commit_list_insert(h2, &(*result)->parents->next);
3601 }
3602 flush_output(opt);
3603 if (!opt->call_depth && opt->buffer_output < 2)
3604 strbuf_release(&opt->obuf);
3605 if (show(opt, 2))
3606 diff_warn_rename_limit("merge.renamelimit",
3607 opt->needed_rename_limit, 0);
3608 return clean;
3609 }
3610
3611 static struct commit *get_ref(struct repository *repo, const struct object_id *oid,
3612 const char *name)
3613 {
3614 struct object *object;
3615
3616 object = deref_tag(repo, parse_object(repo, oid),
3617 name, strlen(name));
3618 if (!object)
3619 return NULL;
3620 if (object->type == OBJ_TREE)
3621 return make_virtual_commit(repo, (struct tree*)object, name);
3622 if (object->type != OBJ_COMMIT)
3623 return NULL;
3624 if (parse_commit((struct commit *)object))
3625 return NULL;
3626 return (struct commit *)object;
3627 }
3628
3629 int merge_recursive_generic(struct merge_options *opt,
3630 const struct object_id *head,
3631 const struct object_id *merge,
3632 int num_base_list,
3633 const struct object_id **base_list,
3634 struct commit **result)
3635 {
3636 int clean;
3637 struct lock_file lock = LOCK_INIT;
3638 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3639 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3640 struct commit_list *ca = NULL;
3641
3642 if (base_list) {
3643 int i;
3644 for (i = 0; i < num_base_list; ++i) {
3645 struct commit *base;
3646 if (!(base = get_ref(opt->repo, base_list[i], oid_to_hex(base_list[i]))))
3647 return err(opt, _("Could not parse object '%s'"),
3648 oid_to_hex(base_list[i]));
3649 commit_list_insert(base, &ca);
3650 }
3651 }
3652
3653 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3654 clean = merge_recursive(opt, head_commit, next_commit, ca,
3655 result);
3656 if (clean < 0) {
3657 rollback_lock_file(&lock);
3658 return clean;
3659 }
3660
3661 if (write_locked_index(opt->repo->index, &lock,
3662 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3663 return err(opt, _("Unable to write index."));
3664
3665 return clean ? 0 : 1;
3666 }
3667
3668 static void merge_recursive_config(struct merge_options *opt)
3669 {
3670 char *value = NULL;
3671 git_config_get_int("merge.verbosity", &opt->verbosity);
3672 git_config_get_int("diff.renamelimit", &opt->diff_rename_limit);
3673 git_config_get_int("merge.renamelimit", &opt->merge_rename_limit);
3674 if (!git_config_get_string("diff.renames", &value)) {
3675 opt->diff_detect_rename = git_config_rename("diff.renames", value);
3676 free(value);
3677 }
3678 if (!git_config_get_string("merge.renames", &value)) {
3679 opt->merge_detect_rename = git_config_rename("merge.renames", value);
3680 free(value);
3681 }
3682 if (!git_config_get_string("merge.directoryrenames", &value)) {
3683 int boolval = git_parse_maybe_bool(value);
3684 if (0 <= boolval) {
3685 opt->detect_directory_renames = boolval ?
3686 MERGE_DIRECTORY_RENAMES_TRUE :
3687 MERGE_DIRECTORY_RENAMES_NONE;
3688 } else if (!strcasecmp(value, "conflict")) {
3689 opt->detect_directory_renames =
3690 MERGE_DIRECTORY_RENAMES_CONFLICT;
3691 } /* avoid erroring on values from future versions of git */
3692 free(value);
3693 }
3694 git_config(git_xmerge_config, NULL);
3695 }
3696
3697 void init_merge_options(struct merge_options *opt,
3698 struct repository *repo)
3699 {
3700 const char *merge_verbosity;
3701 memset(opt, 0, sizeof(struct merge_options));
3702 opt->repo = repo;
3703 opt->verbosity = 2;
3704 opt->buffer_output = 1;
3705 opt->diff_rename_limit = -1;
3706 opt->merge_rename_limit = -1;
3707 opt->renormalize = 0;
3708 opt->diff_detect_rename = -1;
3709 opt->merge_detect_rename = -1;
3710 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3711 merge_recursive_config(opt);
3712 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3713 if (merge_verbosity)
3714 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3715 if (opt->verbosity >= 5)
3716 opt->buffer_output = 0;
3717 strbuf_init(&opt->obuf, 0);
3718 string_list_init(&opt->df_conflict_file_set, 1);
3719 }
3720
3721 int parse_merge_opt(struct merge_options *opt, const char *s)
3722 {
3723 const char *arg;
3724
3725 if (!s || !*s)
3726 return -1;
3727 if (!strcmp(s, "ours"))
3728 opt->recursive_variant = MERGE_RECURSIVE_OURS;
3729 else if (!strcmp(s, "theirs"))
3730 opt->recursive_variant = MERGE_RECURSIVE_THEIRS;
3731 else if (!strcmp(s, "subtree"))
3732 opt->subtree_shift = "";
3733 else if (skip_prefix(s, "subtree=", &arg))
3734 opt->subtree_shift = arg;
3735 else if (!strcmp(s, "patience"))
3736 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3737 else if (!strcmp(s, "histogram"))
3738 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3739 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3740 long value = parse_algorithm_value(arg);
3741 if (value < 0)
3742 return -1;
3743 /* clear out previous settings */
3744 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3745 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3746 opt->xdl_opts |= value;
3747 }
3748 else if (!strcmp(s, "ignore-space-change"))
3749 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3750 else if (!strcmp(s, "ignore-all-space"))
3751 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3752 else if (!strcmp(s, "ignore-space-at-eol"))
3753 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3754 else if (!strcmp(s, "ignore-cr-at-eol"))
3755 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3756 else if (!strcmp(s, "renormalize"))
3757 opt->renormalize = 1;
3758 else if (!strcmp(s, "no-renormalize"))
3759 opt->renormalize = 0;
3760 else if (!strcmp(s, "no-renames"))
3761 opt->merge_detect_rename = 0;
3762 else if (!strcmp(s, "find-renames")) {
3763 opt->merge_detect_rename = 1;
3764 opt->rename_score = 0;
3765 }
3766 else if (skip_prefix(s, "find-renames=", &arg) ||
3767 skip_prefix(s, "rename-threshold=", &arg)) {
3768 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3769 return -1;
3770 opt->merge_detect_rename = 1;
3771 }
3772 /*
3773 * Please update $__git_merge_strategy_options in
3774 * git-completion.bash when you add new options
3775 */
3776 else
3777 return -1;
3778 return 0;
3779 }