]> git.ipfire.org Git - thirdparty/git.git/blob - merge-recursive.c
merge-recursive: rename MERGE_RECURSIVE_* to MERGE_VARIANT_*
[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 merge_options_internal {
33 int call_depth;
34 int needed_rename_limit;
35 struct hashmap current_file_dir_set;
36 struct string_list df_conflict_file_set;
37 struct unpack_trees_options unpack_opts;
38 struct index_state orig_index;
39 };
40
41 struct path_hashmap_entry {
42 struct hashmap_entry e;
43 char path[FLEX_ARRAY];
44 };
45
46 static int path_hashmap_cmp(const void *cmp_data,
47 const void *entry,
48 const void *entry_or_key,
49 const void *keydata)
50 {
51 const struct path_hashmap_entry *a = entry;
52 const struct path_hashmap_entry *b = entry_or_key;
53 const char *key = keydata;
54
55 if (ignore_case)
56 return strcasecmp(a->path, key ? key : b->path);
57 else
58 return strcmp(a->path, key ? key : b->path);
59 }
60
61 static unsigned int path_hash(const char *path)
62 {
63 return ignore_case ? strihash(path) : strhash(path);
64 }
65
66 /*
67 * For dir_rename_entry, directory names are stored as a full path from the
68 * toplevel of the repository and do not include a trailing '/'. Also:
69 *
70 * dir: original name of directory being renamed
71 * non_unique_new_dir: if true, could not determine new_dir
72 * new_dir: final name of directory being renamed
73 * possible_new_dirs: temporary used to help determine new_dir; see comments
74 * in get_directory_renames() for details
75 */
76 struct dir_rename_entry {
77 struct hashmap_entry ent; /* must be the first member! */
78 char *dir;
79 unsigned non_unique_new_dir:1;
80 struct strbuf new_dir;
81 struct string_list possible_new_dirs;
82 };
83
84 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
85 char *dir)
86 {
87 struct dir_rename_entry key;
88
89 if (dir == NULL)
90 return NULL;
91 hashmap_entry_init(&key, strhash(dir));
92 key.dir = dir;
93 return hashmap_get(hashmap, &key, NULL);
94 }
95
96 static int dir_rename_cmp(const void *unused_cmp_data,
97 const void *entry,
98 const void *entry_or_key,
99 const void *unused_keydata)
100 {
101 const struct dir_rename_entry *e1 = entry;
102 const struct dir_rename_entry *e2 = entry_or_key;
103
104 return strcmp(e1->dir, e2->dir);
105 }
106
107 static void dir_rename_init(struct hashmap *map)
108 {
109 hashmap_init(map, dir_rename_cmp, NULL, 0);
110 }
111
112 static void dir_rename_entry_init(struct dir_rename_entry *entry,
113 char *directory)
114 {
115 hashmap_entry_init(entry, strhash(directory));
116 entry->dir = directory;
117 entry->non_unique_new_dir = 0;
118 strbuf_init(&entry->new_dir, 0);
119 string_list_init(&entry->possible_new_dirs, 0);
120 }
121
122 struct collision_entry {
123 struct hashmap_entry ent; /* must be the first member! */
124 char *target_file;
125 struct string_list source_files;
126 unsigned reported_already:1;
127 };
128
129 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
130 char *target_file)
131 {
132 struct collision_entry key;
133
134 hashmap_entry_init(&key, strhash(target_file));
135 key.target_file = target_file;
136 return hashmap_get(hashmap, &key, NULL);
137 }
138
139 static int collision_cmp(void *unused_cmp_data,
140 const struct collision_entry *e1,
141 const struct collision_entry *e2,
142 const void *unused_keydata)
143 {
144 return strcmp(e1->target_file, e2->target_file);
145 }
146
147 static void collision_init(struct hashmap *map)
148 {
149 hashmap_init(map, (hashmap_cmp_fn) collision_cmp, NULL, 0);
150 }
151
152 static void flush_output(struct merge_options *opt)
153 {
154 if (opt->buffer_output < 2 && opt->obuf.len) {
155 fputs(opt->obuf.buf, stdout);
156 strbuf_reset(&opt->obuf);
157 }
158 }
159
160 static int err(struct merge_options *opt, const char *err, ...)
161 {
162 va_list params;
163
164 if (opt->buffer_output < 2)
165 flush_output(opt);
166 else {
167 strbuf_complete(&opt->obuf, '\n');
168 strbuf_addstr(&opt->obuf, "error: ");
169 }
170 va_start(params, err);
171 strbuf_vaddf(&opt->obuf, err, params);
172 va_end(params);
173 if (opt->buffer_output > 1)
174 strbuf_addch(&opt->obuf, '\n');
175 else {
176 error("%s", opt->obuf.buf);
177 strbuf_reset(&opt->obuf);
178 }
179
180 return -1;
181 }
182
183 static struct tree *shift_tree_object(struct repository *repo,
184 struct tree *one, struct tree *two,
185 const char *subtree_shift)
186 {
187 struct object_id shifted;
188
189 if (!*subtree_shift) {
190 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
191 } else {
192 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
193 subtree_shift);
194 }
195 if (oideq(&two->object.oid, &shifted))
196 return two;
197 return lookup_tree(repo, &shifted);
198 }
199
200 static inline void set_commit_tree(struct commit *c, struct tree *t)
201 {
202 c->maybe_tree = t;
203 }
204
205 static struct commit *make_virtual_commit(struct repository *repo,
206 struct tree *tree,
207 const char *comment)
208 {
209 struct commit *commit = alloc_commit_node(repo);
210
211 set_merge_remote_desc(commit, comment, (struct object *)commit);
212 set_commit_tree(commit, tree);
213 commit->object.parsed = 1;
214 return commit;
215 }
216
217 /*
218 * Since we use get_tree_entry(), which does not put the read object into
219 * the object pool, we cannot rely on a == b.
220 */
221 static int oid_eq(const struct object_id *a, const struct object_id *b)
222 {
223 if (!a && !b)
224 return 2;
225 return a && b && oideq(a, b);
226 }
227
228 enum rename_type {
229 RENAME_NORMAL = 0,
230 RENAME_VIA_DIR,
231 RENAME_ADD,
232 RENAME_DELETE,
233 RENAME_ONE_FILE_TO_ONE,
234 RENAME_ONE_FILE_TO_TWO,
235 RENAME_TWO_FILES_TO_ONE
236 };
237
238 /*
239 * Since we want to write the index eventually, we cannot reuse the index
240 * for these (temporary) data.
241 */
242 struct stage_data {
243 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
244 struct rename_conflict_info *rename_conflict_info;
245 unsigned processed:1;
246 };
247
248 struct rename {
249 unsigned processed:1;
250 struct diff_filepair *pair;
251 const char *branch; /* branch that the rename occurred on */
252 /*
253 * If directory rename detection affected this rename, what was its
254 * original type ('A' or 'R') and it's original destination before
255 * the directory rename (otherwise, '\0' and NULL for these two vars).
256 */
257 char dir_rename_original_type;
258 char *dir_rename_original_dest;
259 /*
260 * Purpose of src_entry and dst_entry:
261 *
262 * If 'before' is renamed to 'after' then src_entry will contain
263 * the versions of 'before' from the merge_base, HEAD, and MERGE in
264 * stages 1, 2, and 3; dst_entry will contain the respective
265 * versions of 'after' in corresponding locations. Thus, we have a
266 * total of six modes and oids, though some will be null. (Stage 0
267 * is ignored; we're interested in handling conflicts.)
268 *
269 * Since we don't turn on break-rewrites by default, neither
270 * src_entry nor dst_entry can have all three of their stages have
271 * non-null oids, meaning at most four of the six will be non-null.
272 * Also, since this is a rename, both src_entry and dst_entry will
273 * have at least one non-null oid, meaning at least two will be
274 * non-null. Of the six oids, a typical rename will have three be
275 * non-null. Only two implies a rename/delete, and four implies a
276 * rename/add.
277 */
278 struct stage_data *src_entry;
279 struct stage_data *dst_entry;
280 };
281
282 struct rename_conflict_info {
283 enum rename_type rename_type;
284 struct rename *ren1;
285 struct rename *ren2;
286 };
287
288 static inline void setup_rename_conflict_info(enum rename_type rename_type,
289 struct merge_options *opt,
290 struct rename *ren1,
291 struct rename *ren2)
292 {
293 struct rename_conflict_info *ci;
294
295 /*
296 * When we have two renames involved, it's easiest to get the
297 * correct things into stage 2 and 3, and to make sure that the
298 * content merge puts HEAD before the other branch if we just
299 * ensure that branch1 == opt->branch1. So, simply flip arguments
300 * around if we don't have that.
301 */
302 if (ren2 && ren1->branch != opt->branch1) {
303 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
304 return;
305 }
306
307 ci = xcalloc(1, sizeof(struct rename_conflict_info));
308 ci->rename_type = rename_type;
309 ci->ren1 = ren1;
310 ci->ren2 = ren2;
311
312 ci->ren1->dst_entry->processed = 0;
313 ci->ren1->dst_entry->rename_conflict_info = ci;
314 if (ren2) {
315 ci->ren2->dst_entry->rename_conflict_info = ci;
316 }
317 }
318
319 static int show(struct merge_options *opt, int v)
320 {
321 return (!opt->priv->call_depth && opt->verbosity >= v) ||
322 opt->verbosity >= 5;
323 }
324
325 __attribute__((format (printf, 3, 4)))
326 static void output(struct merge_options *opt, int v, const char *fmt, ...)
327 {
328 va_list ap;
329
330 if (!show(opt, v))
331 return;
332
333 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
334
335 va_start(ap, fmt);
336 strbuf_vaddf(&opt->obuf, fmt, ap);
337 va_end(ap);
338
339 strbuf_addch(&opt->obuf, '\n');
340 if (!opt->buffer_output)
341 flush_output(opt);
342 }
343
344 static void output_commit_title(struct merge_options *opt, struct commit *commit)
345 {
346 struct merge_remote_desc *desc;
347
348 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
349 desc = merge_remote_util(commit);
350 if (desc)
351 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
352 else {
353 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
354 DEFAULT_ABBREV);
355 strbuf_addch(&opt->obuf, ' ');
356 if (parse_commit(commit) != 0)
357 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
358 else {
359 const char *title;
360 const char *msg = get_commit_buffer(commit, NULL);
361 int len = find_commit_subject(msg, &title);
362 if (len)
363 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
364 unuse_commit_buffer(commit, msg);
365 }
366 }
367 flush_output(opt);
368 }
369
370 static int add_cacheinfo(struct merge_options *opt,
371 const struct diff_filespec *blob,
372 const char *path, int stage, int refresh, int options)
373 {
374 struct index_state *istate = opt->repo->index;
375 struct cache_entry *ce;
376 int ret;
377
378 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
379 if (!ce)
380 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
381
382 ret = add_index_entry(istate, ce, options);
383 if (refresh) {
384 struct cache_entry *nce;
385
386 nce = refresh_cache_entry(istate, ce,
387 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
388 if (!nce)
389 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
390 if (nce != ce)
391 ret = add_index_entry(istate, nce, options);
392 }
393 return ret;
394 }
395
396 static inline int merge_detect_rename(struct merge_options *opt)
397 {
398 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
399 }
400
401 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
402 {
403 parse_tree(tree);
404 init_tree_desc(desc, tree->buffer, tree->size);
405 }
406
407 static int unpack_trees_start(struct merge_options *opt,
408 struct tree *common,
409 struct tree *head,
410 struct tree *merge)
411 {
412 int rc;
413 struct tree_desc t[3];
414 struct index_state tmp_index = { NULL };
415
416 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
417 if (opt->priv->call_depth)
418 opt->priv->unpack_opts.index_only = 1;
419 else
420 opt->priv->unpack_opts.update = 1;
421 opt->priv->unpack_opts.merge = 1;
422 opt->priv->unpack_opts.head_idx = 2;
423 opt->priv->unpack_opts.fn = threeway_merge;
424 opt->priv->unpack_opts.src_index = opt->repo->index;
425 opt->priv->unpack_opts.dst_index = &tmp_index;
426 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
427 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
428
429 init_tree_desc_from_tree(t+0, common);
430 init_tree_desc_from_tree(t+1, head);
431 init_tree_desc_from_tree(t+2, merge);
432
433 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
434 cache_tree_free(&opt->repo->index->cache_tree);
435
436 /*
437 * Update opt->repo->index to match the new results, AFTER saving a
438 * copy in opt->priv->orig_index. Update src_index to point to the
439 * saved copy. (verify_uptodate() checks src_index, and the original
440 * index is the one that had the necessary modification timestamps.)
441 */
442 opt->priv->orig_index = *opt->repo->index;
443 *opt->repo->index = tmp_index;
444 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
445
446 return rc;
447 }
448
449 static void unpack_trees_finish(struct merge_options *opt)
450 {
451 discard_index(&opt->priv->orig_index);
452 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
453 }
454
455 static int save_files_dirs(const struct object_id *oid,
456 struct strbuf *base, const char *path,
457 unsigned int mode, int stage, void *context)
458 {
459 struct path_hashmap_entry *entry;
460 int baselen = base->len;
461 struct merge_options *opt = context;
462
463 strbuf_addstr(base, path);
464
465 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
466 hashmap_entry_init(entry, path_hash(entry->path));
467 hashmap_add(&opt->priv->current_file_dir_set, entry);
468
469 strbuf_setlen(base, baselen);
470 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
471 }
472
473 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
474 {
475 struct pathspec match_all;
476 memset(&match_all, 0, sizeof(match_all));
477 read_tree_recursive(opt->repo, tree, "", 0, 0,
478 &match_all, save_files_dirs, opt);
479 }
480
481 static int get_tree_entry_if_blob(struct repository *r,
482 const struct object_id *tree,
483 const char *path,
484 struct diff_filespec *dfs)
485 {
486 int ret;
487
488 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
489 if (S_ISDIR(dfs->mode)) {
490 oidcpy(&dfs->oid, &null_oid);
491 dfs->mode = 0;
492 }
493 return ret;
494 }
495
496 /*
497 * Returns an index_entry instance which doesn't have to correspond to
498 * a real cache entry in Git's index.
499 */
500 static struct stage_data *insert_stage_data(struct repository *r,
501 const char *path,
502 struct tree *o, struct tree *a, struct tree *b,
503 struct string_list *entries)
504 {
505 struct string_list_item *item;
506 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
507 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
508 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
509 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
510 item = string_list_insert(entries, path);
511 item->util = e;
512 return e;
513 }
514
515 /*
516 * Create a dictionary mapping file names to stage_data objects. The
517 * dictionary contains one entry for every path with a non-zero stage entry.
518 */
519 static struct string_list *get_unmerged(struct index_state *istate)
520 {
521 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
522 int i;
523
524 unmerged->strdup_strings = 1;
525
526 for (i = 0; i < istate->cache_nr; i++) {
527 struct string_list_item *item;
528 struct stage_data *e;
529 const struct cache_entry *ce = istate->cache[i];
530 if (!ce_stage(ce))
531 continue;
532
533 item = string_list_lookup(unmerged, ce->name);
534 if (!item) {
535 item = string_list_insert(unmerged, ce->name);
536 item->util = xcalloc(1, sizeof(struct stage_data));
537 }
538 e = item->util;
539 e->stages[ce_stage(ce)].mode = ce->ce_mode;
540 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
541 }
542
543 return unmerged;
544 }
545
546 static int string_list_df_name_compare(const char *one, const char *two)
547 {
548 int onelen = strlen(one);
549 int twolen = strlen(two);
550 /*
551 * Here we only care that entries for D/F conflicts are
552 * adjacent, in particular with the file of the D/F conflict
553 * appearing before files below the corresponding directory.
554 * The order of the rest of the list is irrelevant for us.
555 *
556 * To achieve this, we sort with df_name_compare and provide
557 * the mode S_IFDIR so that D/F conflicts will sort correctly.
558 * We use the mode S_IFDIR for everything else for simplicity,
559 * since in other cases any changes in their order due to
560 * sorting cause no problems for us.
561 */
562 int cmp = df_name_compare(one, onelen, S_IFDIR,
563 two, twolen, S_IFDIR);
564 /*
565 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
566 * that 'foo' comes before 'foo/bar'.
567 */
568 if (cmp)
569 return cmp;
570 return onelen - twolen;
571 }
572
573 static void record_df_conflict_files(struct merge_options *opt,
574 struct string_list *entries)
575 {
576 /* If there is a D/F conflict and the file for such a conflict
577 * currently exists in the working tree, we want to allow it to be
578 * removed to make room for the corresponding directory if needed.
579 * The files underneath the directories of such D/F conflicts will
580 * be processed before the corresponding file involved in the D/F
581 * conflict. If the D/F directory ends up being removed by the
582 * merge, then we won't have to touch the D/F file. If the D/F
583 * directory needs to be written to the working copy, then the D/F
584 * file will simply be removed (in make_room_for_path()) to make
585 * room for the necessary paths. Note that if both the directory
586 * and the file need to be present, then the D/F file will be
587 * reinstated with a new unique name at the time it is processed.
588 */
589 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
590 const char *last_file = NULL;
591 int last_len = 0;
592 int i;
593
594 /*
595 * If we're merging merge-bases, we don't want to bother with
596 * any working directory changes.
597 */
598 if (opt->priv->call_depth)
599 return;
600
601 /* Ensure D/F conflicts are adjacent in the entries list. */
602 for (i = 0; i < entries->nr; i++) {
603 struct string_list_item *next = &entries->items[i];
604 string_list_append(&df_sorted_entries, next->string)->util =
605 next->util;
606 }
607 df_sorted_entries.cmp = string_list_df_name_compare;
608 string_list_sort(&df_sorted_entries);
609
610 string_list_clear(&opt->priv->df_conflict_file_set, 1);
611 for (i = 0; i < df_sorted_entries.nr; i++) {
612 const char *path = df_sorted_entries.items[i].string;
613 int len = strlen(path);
614 struct stage_data *e = df_sorted_entries.items[i].util;
615
616 /*
617 * Check if last_file & path correspond to a D/F conflict;
618 * i.e. whether path is last_file+'/'+<something>.
619 * If so, record that it's okay to remove last_file to make
620 * room for path and friends if needed.
621 */
622 if (last_file &&
623 len > last_len &&
624 memcmp(path, last_file, last_len) == 0 &&
625 path[last_len] == '/') {
626 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
627 }
628
629 /*
630 * Determine whether path could exist as a file in the
631 * working directory as a possible D/F conflict. This
632 * will only occur when it exists in stage 2 as a
633 * file.
634 */
635 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
636 last_file = path;
637 last_len = len;
638 } else {
639 last_file = NULL;
640 }
641 }
642 string_list_clear(&df_sorted_entries, 0);
643 }
644
645 static int update_stages(struct merge_options *opt, const char *path,
646 const struct diff_filespec *o,
647 const struct diff_filespec *a,
648 const struct diff_filespec *b)
649 {
650
651 /*
652 * NOTE: It is usually a bad idea to call update_stages on a path
653 * before calling update_file on that same path, since it can
654 * sometimes lead to spurious "refusing to lose untracked file..."
655 * messages from update_file (via make_room_for path via
656 * would_lose_untracked). Instead, reverse the order of the calls
657 * (executing update_file first and then update_stages).
658 */
659 int clear = 1;
660 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
661 if (clear)
662 if (remove_file_from_index(opt->repo->index, path))
663 return -1;
664 if (o)
665 if (add_cacheinfo(opt, o, path, 1, 0, options))
666 return -1;
667 if (a)
668 if (add_cacheinfo(opt, a, path, 2, 0, options))
669 return -1;
670 if (b)
671 if (add_cacheinfo(opt, b, path, 3, 0, options))
672 return -1;
673 return 0;
674 }
675
676 static void update_entry(struct stage_data *entry,
677 struct diff_filespec *o,
678 struct diff_filespec *a,
679 struct diff_filespec *b)
680 {
681 entry->processed = 0;
682 entry->stages[1].mode = o->mode;
683 entry->stages[2].mode = a->mode;
684 entry->stages[3].mode = b->mode;
685 oidcpy(&entry->stages[1].oid, &o->oid);
686 oidcpy(&entry->stages[2].oid, &a->oid);
687 oidcpy(&entry->stages[3].oid, &b->oid);
688 }
689
690 static int remove_file(struct merge_options *opt, int clean,
691 const char *path, int no_wd)
692 {
693 int update_cache = opt->priv->call_depth || clean;
694 int update_working_directory = !opt->priv->call_depth && !no_wd;
695
696 if (update_cache) {
697 if (remove_file_from_index(opt->repo->index, path))
698 return -1;
699 }
700 if (update_working_directory) {
701 if (ignore_case) {
702 struct cache_entry *ce;
703 ce = index_file_exists(opt->repo->index, path, strlen(path),
704 ignore_case);
705 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
706 return 0;
707 }
708 if (remove_path(path))
709 return -1;
710 }
711 return 0;
712 }
713
714 /* add a string to a strbuf, but converting "/" to "_" */
715 static void add_flattened_path(struct strbuf *out, const char *s)
716 {
717 size_t i = out->len;
718 strbuf_addstr(out, s);
719 for (; i < out->len; i++)
720 if (out->buf[i] == '/')
721 out->buf[i] = '_';
722 }
723
724 static char *unique_path(struct merge_options *opt,
725 const char *path,
726 const char *branch)
727 {
728 struct path_hashmap_entry *entry;
729 struct strbuf newpath = STRBUF_INIT;
730 int suffix = 0;
731 size_t base_len;
732
733 strbuf_addf(&newpath, "%s~", path);
734 add_flattened_path(&newpath, branch);
735
736 base_len = newpath.len;
737 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
738 path_hash(newpath.buf), newpath.buf) ||
739 (!opt->priv->call_depth && file_exists(newpath.buf))) {
740 strbuf_setlen(&newpath, base_len);
741 strbuf_addf(&newpath, "_%d", suffix++);
742 }
743
744 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
745 hashmap_entry_init(entry, path_hash(entry->path));
746 hashmap_add(&opt->priv->current_file_dir_set, entry);
747 return strbuf_detach(&newpath, NULL);
748 }
749
750 /**
751 * Check whether a directory in the index is in the way of an incoming
752 * file. Return 1 if so. If check_working_copy is non-zero, also
753 * check the working directory. If empty_ok is non-zero, also return
754 * 0 in the case where the working-tree dir exists but is empty.
755 */
756 static int dir_in_way(struct index_state *istate, const char *path,
757 int check_working_copy, int empty_ok)
758 {
759 int pos;
760 struct strbuf dirpath = STRBUF_INIT;
761 struct stat st;
762
763 strbuf_addstr(&dirpath, path);
764 strbuf_addch(&dirpath, '/');
765
766 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
767
768 if (pos < 0)
769 pos = -1 - pos;
770 if (pos < istate->cache_nr &&
771 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
772 strbuf_release(&dirpath);
773 return 1;
774 }
775
776 strbuf_release(&dirpath);
777 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
778 !(empty_ok && is_empty_dir(path));
779 }
780
781 /*
782 * Returns whether path was tracked in the index before the merge started,
783 * and its oid and mode match the specified values
784 */
785 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
786 const struct diff_filespec *blob)
787 {
788 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
789 struct cache_entry *ce;
790
791 if (0 > pos)
792 /* we were not tracking this path before the merge */
793 return 0;
794
795 /* See if the file we were tracking before matches */
796 ce = opt->priv->orig_index.cache[pos];
797 return (oid_eq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
798 }
799
800 /*
801 * Returns whether path was tracked in the index before the merge started
802 */
803 static int was_tracked(struct merge_options *opt, const char *path)
804 {
805 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
806
807 if (0 <= pos)
808 /* we were tracking this path before the merge */
809 return 1;
810
811 return 0;
812 }
813
814 static int would_lose_untracked(struct merge_options *opt, const char *path)
815 {
816 struct index_state *istate = opt->repo->index;
817
818 /*
819 * This may look like it can be simplified to:
820 * return !was_tracked(opt, path) && file_exists(path)
821 * but it can't. This function needs to know whether path was in
822 * the working tree due to EITHER having been tracked in the index
823 * before the merge OR having been put into the working copy and
824 * index by unpack_trees(). Due to that either-or requirement, we
825 * check the current index instead of the original one.
826 *
827 * Note that we do not need to worry about merge-recursive itself
828 * updating the index after unpack_trees() and before calling this
829 * function, because we strictly require all code paths in
830 * merge-recursive to update the working tree first and the index
831 * second. Doing otherwise would break
832 * update_file()/would_lose_untracked(); see every comment in this
833 * file which mentions "update_stages".
834 */
835 int pos = index_name_pos(istate, path, strlen(path));
836
837 if (pos < 0)
838 pos = -1 - pos;
839 while (pos < istate->cache_nr &&
840 !strcmp(path, istate->cache[pos]->name)) {
841 /*
842 * If stage #0, it is definitely tracked.
843 * If it has stage #2 then it was tracked
844 * before this merge started. All other
845 * cases the path was not tracked.
846 */
847 switch (ce_stage(istate->cache[pos])) {
848 case 0:
849 case 2:
850 return 0;
851 }
852 pos++;
853 }
854 return file_exists(path);
855 }
856
857 static int was_dirty(struct merge_options *opt, const char *path)
858 {
859 struct cache_entry *ce;
860 int dirty = 1;
861
862 if (opt->priv->call_depth || !was_tracked(opt, path))
863 return !dirty;
864
865 ce = index_file_exists(opt->priv->unpack_opts.src_index,
866 path, strlen(path), ignore_case);
867 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
868 return dirty;
869 }
870
871 static int make_room_for_path(struct merge_options *opt, const char *path)
872 {
873 int status, i;
874 const char *msg = _("failed to create path '%s'%s");
875
876 /* Unlink any D/F conflict files that are in the way */
877 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
878 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
879 size_t pathlen = strlen(path);
880 size_t df_pathlen = strlen(df_path);
881 if (df_pathlen < pathlen &&
882 path[df_pathlen] == '/' &&
883 strncmp(path, df_path, df_pathlen) == 0) {
884 output(opt, 3,
885 _("Removing %s to make room for subdirectory\n"),
886 df_path);
887 unlink(df_path);
888 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
889 i, 0);
890 break;
891 }
892 }
893
894 /* Make sure leading directories are created */
895 status = safe_create_leading_directories_const(path);
896 if (status) {
897 if (status == SCLD_EXISTS)
898 /* something else exists */
899 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
900 return err(opt, msg, path, "");
901 }
902
903 /*
904 * Do not unlink a file in the work tree if we are not
905 * tracking it.
906 */
907 if (would_lose_untracked(opt, path))
908 return err(opt, _("refusing to lose untracked file at '%s'"),
909 path);
910
911 /* Successful unlink is good.. */
912 if (!unlink(path))
913 return 0;
914 /* .. and so is no existing file */
915 if (errno == ENOENT)
916 return 0;
917 /* .. but not some other error (who really cares what?) */
918 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
919 }
920
921 static int update_file_flags(struct merge_options *opt,
922 const struct diff_filespec *contents,
923 const char *path,
924 int update_cache,
925 int update_wd)
926 {
927 int ret = 0;
928
929 if (opt->priv->call_depth)
930 update_wd = 0;
931
932 if (update_wd) {
933 enum object_type type;
934 void *buf;
935 unsigned long size;
936
937 if (S_ISGITLINK(contents->mode)) {
938 /*
939 * We may later decide to recursively descend into
940 * the submodule directory and update its index
941 * and/or work tree, but we do not do that now.
942 */
943 update_wd = 0;
944 goto update_index;
945 }
946
947 buf = read_object_file(&contents->oid, &type, &size);
948 if (!buf) {
949 ret = err(opt, _("cannot read object %s '%s'"),
950 oid_to_hex(&contents->oid), path);
951 goto free_buf;
952 }
953 if (type != OBJ_BLOB) {
954 ret = err(opt, _("blob expected for %s '%s'"),
955 oid_to_hex(&contents->oid), path);
956 goto free_buf;
957 }
958 if (S_ISREG(contents->mode)) {
959 struct strbuf strbuf = STRBUF_INIT;
960 if (convert_to_working_tree(opt->repo->index,
961 path, buf, size, &strbuf)) {
962 free(buf);
963 size = strbuf.len;
964 buf = strbuf_detach(&strbuf, NULL);
965 }
966 }
967
968 if (make_room_for_path(opt, path) < 0) {
969 update_wd = 0;
970 goto free_buf;
971 }
972 if (S_ISREG(contents->mode) ||
973 (!has_symlinks && S_ISLNK(contents->mode))) {
974 int fd;
975 int mode = (contents->mode & 0100 ? 0777 : 0666);
976
977 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
978 if (fd < 0) {
979 ret = err(opt, _("failed to open '%s': %s"),
980 path, strerror(errno));
981 goto free_buf;
982 }
983 write_in_full(fd, buf, size);
984 close(fd);
985 } else if (S_ISLNK(contents->mode)) {
986 char *lnk = xmemdupz(buf, size);
987 safe_create_leading_directories_const(path);
988 unlink(path);
989 if (symlink(lnk, path))
990 ret = err(opt, _("failed to symlink '%s': %s"),
991 path, strerror(errno));
992 free(lnk);
993 } else
994 ret = err(opt,
995 _("do not know what to do with %06o %s '%s'"),
996 contents->mode, oid_to_hex(&contents->oid), path);
997 free_buf:
998 free(buf);
999 }
1000 update_index:
1001 if (!ret && update_cache)
1002 if (add_cacheinfo(opt, contents, path, 0, update_wd,
1003 ADD_CACHE_OK_TO_ADD))
1004 return -1;
1005 return ret;
1006 }
1007
1008 static int update_file(struct merge_options *opt,
1009 int clean,
1010 const struct diff_filespec *contents,
1011 const char *path)
1012 {
1013 return update_file_flags(opt, contents, path,
1014 opt->priv->call_depth || clean, !opt->priv->call_depth);
1015 }
1016
1017 /* Low level file merging, update and removal */
1018
1019 struct merge_file_info {
1020 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1021 unsigned clean:1,
1022 merge:1;
1023 };
1024
1025 static int merge_3way(struct merge_options *opt,
1026 mmbuffer_t *result_buf,
1027 const struct diff_filespec *o,
1028 const struct diff_filespec *a,
1029 const struct diff_filespec *b,
1030 const char *branch1,
1031 const char *branch2,
1032 const int extra_marker_size)
1033 {
1034 mmfile_t orig, src1, src2;
1035 struct ll_merge_options ll_opts = {0};
1036 char *base, *name1, *name2;
1037 int merge_status;
1038
1039 ll_opts.renormalize = opt->renormalize;
1040 ll_opts.extra_marker_size = extra_marker_size;
1041 ll_opts.xdl_opts = opt->xdl_opts;
1042
1043 if (opt->priv->call_depth) {
1044 ll_opts.virtual_ancestor = 1;
1045 ll_opts.variant = 0;
1046 } else {
1047 switch (opt->recursive_variant) {
1048 case MERGE_VARIANT_OURS:
1049 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1050 break;
1051 case MERGE_VARIANT_THEIRS:
1052 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1053 break;
1054 default:
1055 ll_opts.variant = 0;
1056 break;
1057 }
1058 }
1059
1060 assert(a->path && b->path && o->path && opt->ancestor);
1061 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1062 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1063 name1 = mkpathdup("%s:%s", branch1, a->path);
1064 name2 = mkpathdup("%s:%s", branch2, b->path);
1065 } else {
1066 base = mkpathdup("%s", opt->ancestor);
1067 name1 = mkpathdup("%s", branch1);
1068 name2 = mkpathdup("%s", branch2);
1069 }
1070
1071 read_mmblob(&orig, &o->oid);
1072 read_mmblob(&src1, &a->oid);
1073 read_mmblob(&src2, &b->oid);
1074
1075 merge_status = ll_merge(result_buf, a->path, &orig, base,
1076 &src1, name1, &src2, name2,
1077 opt->repo->index, &ll_opts);
1078
1079 free(base);
1080 free(name1);
1081 free(name2);
1082 free(orig.ptr);
1083 free(src1.ptr);
1084 free(src2.ptr);
1085 return merge_status;
1086 }
1087
1088 static int find_first_merges(struct repository *repo,
1089 struct object_array *result, const char *path,
1090 struct commit *a, struct commit *b)
1091 {
1092 int i, j;
1093 struct object_array merges = OBJECT_ARRAY_INIT;
1094 struct commit *commit;
1095 int contains_another;
1096
1097 char merged_revision[GIT_MAX_HEXSZ + 2];
1098 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1099 "--all", merged_revision, NULL };
1100 struct rev_info revs;
1101 struct setup_revision_opt rev_opts;
1102
1103 memset(result, 0, sizeof(struct object_array));
1104 memset(&rev_opts, 0, sizeof(rev_opts));
1105
1106 /* get all revisions that merge commit a */
1107 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1108 oid_to_hex(&a->object.oid));
1109 repo_init_revisions(repo, &revs, NULL);
1110 rev_opts.submodule = path;
1111 /* FIXME: can't handle linked worktrees in submodules yet */
1112 revs.single_worktree = path != NULL;
1113 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1114
1115 /* save all revisions from the above list that contain b */
1116 if (prepare_revision_walk(&revs))
1117 die("revision walk setup failed");
1118 while ((commit = get_revision(&revs)) != NULL) {
1119 struct object *o = &(commit->object);
1120 if (in_merge_bases(b, commit))
1121 add_object_array(o, NULL, &merges);
1122 }
1123 reset_revision_walk();
1124
1125 /* Now we've got all merges that contain a and b. Prune all
1126 * merges that contain another found merge and save them in
1127 * result.
1128 */
1129 for (i = 0; i < merges.nr; i++) {
1130 struct commit *m1 = (struct commit *) merges.objects[i].item;
1131
1132 contains_another = 0;
1133 for (j = 0; j < merges.nr; j++) {
1134 struct commit *m2 = (struct commit *) merges.objects[j].item;
1135 if (i != j && in_merge_bases(m2, m1)) {
1136 contains_another = 1;
1137 break;
1138 }
1139 }
1140
1141 if (!contains_another)
1142 add_object_array(merges.objects[i].item, NULL, result);
1143 }
1144
1145 object_array_clear(&merges);
1146 return result->nr;
1147 }
1148
1149 static void print_commit(struct commit *commit)
1150 {
1151 struct strbuf sb = STRBUF_INIT;
1152 struct pretty_print_context ctx = {0};
1153 ctx.date_mode.type = DATE_NORMAL;
1154 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1155 fprintf(stderr, "%s\n", sb.buf);
1156 strbuf_release(&sb);
1157 }
1158
1159 static int is_valid(const struct diff_filespec *dfs)
1160 {
1161 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1162 }
1163
1164 static int merge_submodule(struct merge_options *opt,
1165 struct object_id *result, const char *path,
1166 const struct object_id *base, const struct object_id *a,
1167 const struct object_id *b)
1168 {
1169 struct commit *commit_base, *commit_a, *commit_b;
1170 int parent_count;
1171 struct object_array merges;
1172
1173 int i;
1174 int search = !opt->priv->call_depth;
1175
1176 /* store a in result in case we fail */
1177 oidcpy(result, a);
1178
1179 /* we can not handle deletion conflicts */
1180 if (is_null_oid(base))
1181 return 0;
1182 if (is_null_oid(a))
1183 return 0;
1184 if (is_null_oid(b))
1185 return 0;
1186
1187 if (add_submodule_odb(path)) {
1188 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1189 return 0;
1190 }
1191
1192 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1193 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1194 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1195 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1196 return 0;
1197 }
1198
1199 /* check whether both changes are forward */
1200 if (!in_merge_bases(commit_base, commit_a) ||
1201 !in_merge_bases(commit_base, commit_b)) {
1202 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1203 return 0;
1204 }
1205
1206 /* Case #1: a is contained in b or vice versa */
1207 if (in_merge_bases(commit_a, commit_b)) {
1208 oidcpy(result, b);
1209 if (show(opt, 3)) {
1210 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1211 output_commit_title(opt, commit_b);
1212 } else if (show(opt, 2))
1213 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1214 else
1215 ; /* no output */
1216
1217 return 1;
1218 }
1219 if (in_merge_bases(commit_b, commit_a)) {
1220 oidcpy(result, a);
1221 if (show(opt, 3)) {
1222 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1223 output_commit_title(opt, commit_a);
1224 } else if (show(opt, 2))
1225 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1226 else
1227 ; /* no output */
1228
1229 return 1;
1230 }
1231
1232 /*
1233 * Case #2: There are one or more merges that contain a and b in
1234 * the submodule. If there is only one, then present it as a
1235 * suggestion to the user, but leave it marked unmerged so the
1236 * user needs to confirm the resolution.
1237 */
1238
1239 /* Skip the search if makes no sense to the calling context. */
1240 if (!search)
1241 return 0;
1242
1243 /* find commit which merges them */
1244 parent_count = find_first_merges(opt->repo, &merges, path,
1245 commit_a, commit_b);
1246 switch (parent_count) {
1247 case 0:
1248 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1249 break;
1250
1251 case 1:
1252 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1253 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1254 print_commit((struct commit *) merges.objects[0].item);
1255 output(opt, 2, _(
1256 "If this is correct simply add it to the index "
1257 "for example\n"
1258 "by using:\n\n"
1259 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1260 "which will accept this suggestion.\n"),
1261 oid_to_hex(&merges.objects[0].item->oid), path);
1262 break;
1263
1264 default:
1265 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1266 for (i = 0; i < merges.nr; i++)
1267 print_commit((struct commit *) merges.objects[i].item);
1268 }
1269
1270 object_array_clear(&merges);
1271 return 0;
1272 }
1273
1274 static int merge_mode_and_contents(struct merge_options *opt,
1275 const struct diff_filespec *o,
1276 const struct diff_filespec *a,
1277 const struct diff_filespec *b,
1278 const char *filename,
1279 const char *branch1,
1280 const char *branch2,
1281 const int extra_marker_size,
1282 struct merge_file_info *result)
1283 {
1284 if (opt->branch1 != branch1) {
1285 /*
1286 * It's weird getting a reverse merge with HEAD on the bottom
1287 * side of the conflict markers and the other branch on the
1288 * top. Fix that.
1289 */
1290 return merge_mode_and_contents(opt, o, b, a,
1291 filename,
1292 branch2, branch1,
1293 extra_marker_size, result);
1294 }
1295
1296 result->merge = 0;
1297 result->clean = 1;
1298
1299 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1300 result->clean = 0;
1301 if (S_ISREG(a->mode)) {
1302 result->blob.mode = a->mode;
1303 oidcpy(&result->blob.oid, &a->oid);
1304 } else {
1305 result->blob.mode = b->mode;
1306 oidcpy(&result->blob.oid, &b->oid);
1307 }
1308 } else {
1309 if (!oid_eq(&a->oid, &o->oid) && !oid_eq(&b->oid, &o->oid))
1310 result->merge = 1;
1311
1312 /*
1313 * Merge modes
1314 */
1315 if (a->mode == b->mode || a->mode == o->mode)
1316 result->blob.mode = b->mode;
1317 else {
1318 result->blob.mode = a->mode;
1319 if (b->mode != o->mode) {
1320 result->clean = 0;
1321 result->merge = 1;
1322 }
1323 }
1324
1325 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &o->oid))
1326 oidcpy(&result->blob.oid, &b->oid);
1327 else if (oid_eq(&b->oid, &o->oid))
1328 oidcpy(&result->blob.oid, &a->oid);
1329 else if (S_ISREG(a->mode)) {
1330 mmbuffer_t result_buf;
1331 int ret = 0, merge_status;
1332
1333 merge_status = merge_3way(opt, &result_buf, o, a, b,
1334 branch1, branch2,
1335 extra_marker_size);
1336
1337 if ((merge_status < 0) || !result_buf.ptr)
1338 ret = err(opt, _("Failed to execute internal merge"));
1339
1340 if (!ret &&
1341 write_object_file(result_buf.ptr, result_buf.size,
1342 blob_type, &result->blob.oid))
1343 ret = err(opt, _("Unable to add %s to database"),
1344 a->path);
1345
1346 free(result_buf.ptr);
1347 if (ret)
1348 return ret;
1349 result->clean = (merge_status == 0);
1350 } else if (S_ISGITLINK(a->mode)) {
1351 result->clean = merge_submodule(opt, &result->blob.oid,
1352 o->path,
1353 &o->oid,
1354 &a->oid,
1355 &b->oid);
1356 } else if (S_ISLNK(a->mode)) {
1357 switch (opt->recursive_variant) {
1358 case MERGE_VARIANT_NORMAL:
1359 oidcpy(&result->blob.oid, &a->oid);
1360 if (!oid_eq(&a->oid, &b->oid))
1361 result->clean = 0;
1362 break;
1363 case MERGE_VARIANT_OURS:
1364 oidcpy(&result->blob.oid, &a->oid);
1365 break;
1366 case MERGE_VARIANT_THEIRS:
1367 oidcpy(&result->blob.oid, &b->oid);
1368 break;
1369 }
1370 } else
1371 BUG("unsupported object type in the tree");
1372 }
1373
1374 if (result->merge)
1375 output(opt, 2, _("Auto-merging %s"), filename);
1376
1377 return 0;
1378 }
1379
1380 static int handle_rename_via_dir(struct merge_options *opt,
1381 struct rename_conflict_info *ci)
1382 {
1383 /*
1384 * Handle file adds that need to be renamed due to directory rename
1385 * detection. This differs from handle_rename_normal, because
1386 * there is no content merge to do; just move the file into the
1387 * desired final location.
1388 */
1389 const struct rename *ren = ci->ren1;
1390 const struct diff_filespec *dest = ren->pair->two;
1391 char *file_path = dest->path;
1392 int mark_conflicted = (opt->detect_directory_renames ==
1393 MERGE_DIRECTORY_RENAMES_CONFLICT);
1394 assert(ren->dir_rename_original_dest);
1395
1396 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1397 mark_conflicted = 1;
1398 file_path = unique_path(opt, dest->path, ren->branch);
1399 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1400 "writing to %s instead."),
1401 dest->path, file_path);
1402 }
1403
1404 if (mark_conflicted) {
1405 /*
1406 * Write the file in worktree at file_path. In the index,
1407 * only record the file at dest->path in the appropriate
1408 * higher stage.
1409 */
1410 if (update_file(opt, 0, dest, file_path))
1411 return -1;
1412 if (file_path != dest->path)
1413 free(file_path);
1414 if (update_stages(opt, dest->path, NULL,
1415 ren->branch == opt->branch1 ? dest : NULL,
1416 ren->branch == opt->branch1 ? NULL : dest))
1417 return -1;
1418 return 0; /* not clean, but conflicted */
1419 } else {
1420 /* Update dest->path both in index and in worktree */
1421 if (update_file(opt, 1, dest, dest->path))
1422 return -1;
1423 return 1; /* clean */
1424 }
1425 }
1426
1427 static int handle_change_delete(struct merge_options *opt,
1428 const char *path, const char *old_path,
1429 const struct diff_filespec *o,
1430 const struct diff_filespec *changed,
1431 const char *change_branch,
1432 const char *delete_branch,
1433 const char *change, const char *change_past)
1434 {
1435 char *alt_path = NULL;
1436 const char *update_path = path;
1437 int ret = 0;
1438
1439 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1440 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1441 update_path = alt_path = unique_path(opt, path, change_branch);
1442 }
1443
1444 if (opt->priv->call_depth) {
1445 /*
1446 * We cannot arbitrarily accept either a_sha or b_sha as
1447 * correct; since there is no true "middle point" between
1448 * them, simply reuse the base version for virtual merge base.
1449 */
1450 ret = remove_file_from_index(opt->repo->index, path);
1451 if (!ret)
1452 ret = update_file(opt, 0, o, update_path);
1453 } else {
1454 /*
1455 * Despite the four nearly duplicate messages and argument
1456 * lists below and the ugliness of the nested if-statements,
1457 * having complete messages makes the job easier for
1458 * translators.
1459 *
1460 * The slight variance among the cases is due to the fact
1461 * that:
1462 * 1) directory/file conflicts (in effect if
1463 * !alt_path) could cause us to need to write the
1464 * file to a different path.
1465 * 2) renames (in effect if !old_path) could mean that
1466 * there are two names for the path that the user
1467 * may know the file by.
1468 */
1469 if (!alt_path) {
1470 if (!old_path) {
1471 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1472 "and %s in %s. Version %s of %s left in tree."),
1473 change, path, delete_branch, change_past,
1474 change_branch, change_branch, path);
1475 } else {
1476 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1477 "and %s to %s in %s. Version %s of %s left in tree."),
1478 change, old_path, delete_branch, change_past, path,
1479 change_branch, change_branch, path);
1480 }
1481 } else {
1482 if (!old_path) {
1483 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1484 "and %s in %s. Version %s of %s left in tree at %s."),
1485 change, path, delete_branch, change_past,
1486 change_branch, change_branch, path, alt_path);
1487 } else {
1488 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1489 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1490 change, old_path, delete_branch, change_past, path,
1491 change_branch, change_branch, path, alt_path);
1492 }
1493 }
1494 /*
1495 * No need to call update_file() on path when change_branch ==
1496 * opt->branch1 && !alt_path, since that would needlessly touch
1497 * path. We could call update_file_flags() with update_cache=0
1498 * and update_wd=0, but that's a no-op.
1499 */
1500 if (change_branch != opt->branch1 || alt_path)
1501 ret = update_file(opt, 0, changed, update_path);
1502 }
1503 free(alt_path);
1504
1505 return ret;
1506 }
1507
1508 static int handle_rename_delete(struct merge_options *opt,
1509 struct rename_conflict_info *ci)
1510 {
1511 const struct rename *ren = ci->ren1;
1512 const struct diff_filespec *orig = ren->pair->one;
1513 const struct diff_filespec *dest = ren->pair->two;
1514 const char *rename_branch = ren->branch;
1515 const char *delete_branch = (opt->branch1 == ren->branch ?
1516 opt->branch2 : opt->branch1);
1517
1518 if (handle_change_delete(opt,
1519 opt->priv->call_depth ? orig->path : dest->path,
1520 opt->priv->call_depth ? NULL : orig->path,
1521 orig, dest,
1522 rename_branch, delete_branch,
1523 _("rename"), _("renamed")))
1524 return -1;
1525
1526 if (opt->priv->call_depth)
1527 return remove_file_from_index(opt->repo->index, dest->path);
1528 else
1529 return update_stages(opt, dest->path, NULL,
1530 rename_branch == opt->branch1 ? dest : NULL,
1531 rename_branch == opt->branch1 ? NULL : dest);
1532 }
1533
1534 static int handle_file_collision(struct merge_options *opt,
1535 const char *collide_path,
1536 const char *prev_path1,
1537 const char *prev_path2,
1538 const char *branch1, const char *branch2,
1539 struct diff_filespec *a,
1540 struct diff_filespec *b)
1541 {
1542 struct merge_file_info mfi;
1543 struct diff_filespec null;
1544 char *alt_path = NULL;
1545 const char *update_path = collide_path;
1546
1547 /*
1548 * It's easiest to get the correct things into stage 2 and 3, and
1549 * to make sure that the content merge puts HEAD before the other
1550 * branch if we just ensure that branch1 == opt->branch1. So, simply
1551 * flip arguments around if we don't have that.
1552 */
1553 if (branch1 != opt->branch1) {
1554 return handle_file_collision(opt, collide_path,
1555 prev_path2, prev_path1,
1556 branch2, branch1,
1557 b, a);
1558 }
1559
1560 /*
1561 * In the recursive case, we just opt to undo renames
1562 */
1563 if (opt->priv->call_depth && (prev_path1 || prev_path2)) {
1564 /* Put first file (a->oid, a->mode) in its original spot */
1565 if (prev_path1) {
1566 if (update_file(opt, 1, a, prev_path1))
1567 return -1;
1568 } else {
1569 if (update_file(opt, 1, a, collide_path))
1570 return -1;
1571 }
1572
1573 /* Put second file (b->oid, b->mode) in its original spot */
1574 if (prev_path2) {
1575 if (update_file(opt, 1, b, prev_path2))
1576 return -1;
1577 } else {
1578 if (update_file(opt, 1, b, collide_path))
1579 return -1;
1580 }
1581
1582 /* Don't leave something at collision path if unrenaming both */
1583 if (prev_path1 && prev_path2)
1584 remove_file(opt, 1, collide_path, 0);
1585
1586 return 0;
1587 }
1588
1589 /* Remove rename sources if rename/add or rename/rename(2to1) */
1590 if (prev_path1)
1591 remove_file(opt, 1, prev_path1,
1592 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1593 if (prev_path2)
1594 remove_file(opt, 1, prev_path2,
1595 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1596
1597 /*
1598 * Remove the collision path, if it wouldn't cause dirty contents
1599 * or an untracked file to get lost. We'll either overwrite with
1600 * merged contents, or just write out to differently named files.
1601 */
1602 if (was_dirty(opt, collide_path)) {
1603 output(opt, 1, _("Refusing to lose dirty file at %s"),
1604 collide_path);
1605 update_path = alt_path = unique_path(opt, collide_path, "merged");
1606 } else if (would_lose_untracked(opt, collide_path)) {
1607 /*
1608 * Only way we get here is if both renames were from
1609 * a directory rename AND user had an untracked file
1610 * at the location where both files end up after the
1611 * two directory renames. See testcase 10d of t6043.
1612 */
1613 output(opt, 1, _("Refusing to lose untracked file at "
1614 "%s, even though it's in the way."),
1615 collide_path);
1616 update_path = alt_path = unique_path(opt, collide_path, "merged");
1617 } else {
1618 /*
1619 * FIXME: It's possible that the two files are identical
1620 * and that the current working copy happens to match, in
1621 * which case we are unnecessarily touching the working
1622 * tree file. It's not a likely enough scenario that I
1623 * want to code up the checks for it and a better fix is
1624 * available if we restructure how unpack_trees() and
1625 * merge-recursive interoperate anyway, so punting for
1626 * now...
1627 */
1628 remove_file(opt, 0, collide_path, 0);
1629 }
1630
1631 /* Store things in diff_filespecs for functions that need it */
1632 null.path = (char *)collide_path;
1633 oidcpy(&null.oid, &null_oid);
1634 null.mode = 0;
1635
1636 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1637 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1638 return -1;
1639 mfi.clean &= !alt_path;
1640 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1641 return -1;
1642 if (!mfi.clean && !opt->priv->call_depth &&
1643 update_stages(opt, collide_path, NULL, a, b))
1644 return -1;
1645 free(alt_path);
1646 /*
1647 * FIXME: If both a & b both started with conflicts (only possible
1648 * if they came from a rename/rename(2to1)), but had IDENTICAL
1649 * contents including those conflicts, then in the next line we claim
1650 * it was clean. If someone cares about this case, we should have the
1651 * caller notify us if we started with conflicts.
1652 */
1653 return mfi.clean;
1654 }
1655
1656 static int handle_rename_add(struct merge_options *opt,
1657 struct rename_conflict_info *ci)
1658 {
1659 /* a was renamed to c, and a separate c was added. */
1660 struct diff_filespec *a = ci->ren1->pair->one;
1661 struct diff_filespec *c = ci->ren1->pair->two;
1662 char *path = c->path;
1663 char *prev_path_desc;
1664 struct merge_file_info mfi;
1665
1666 const char *rename_branch = ci->ren1->branch;
1667 const char *add_branch = (opt->branch1 == rename_branch ?
1668 opt->branch2 : opt->branch1);
1669 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1670
1671 output(opt, 1, _("CONFLICT (rename/add): "
1672 "Rename %s->%s in %s. Added %s in %s"),
1673 a->path, c->path, rename_branch,
1674 c->path, add_branch);
1675
1676 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1677 ci->ren1->src_entry->stages[other_stage].path = a->path;
1678 if (merge_mode_and_contents(opt, a, c,
1679 &ci->ren1->src_entry->stages[other_stage],
1680 prev_path_desc,
1681 opt->branch1, opt->branch2,
1682 1 + opt->priv->call_depth * 2, &mfi))
1683 return -1;
1684 free(prev_path_desc);
1685
1686 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1687 return handle_file_collision(opt,
1688 c->path, a->path, NULL,
1689 rename_branch, add_branch,
1690 &mfi.blob,
1691 &ci->ren1->dst_entry->stages[other_stage]);
1692 }
1693
1694 static char *find_path_for_conflict(struct merge_options *opt,
1695 const char *path,
1696 const char *branch1,
1697 const char *branch2)
1698 {
1699 char *new_path = NULL;
1700 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1701 new_path = unique_path(opt, path, branch1);
1702 output(opt, 1, _("%s is a directory in %s adding "
1703 "as %s instead"),
1704 path, branch2, new_path);
1705 } else if (would_lose_untracked(opt, path)) {
1706 new_path = unique_path(opt, path, branch1);
1707 output(opt, 1, _("Refusing to lose untracked file"
1708 " at %s; adding as %s instead"),
1709 path, new_path);
1710 }
1711
1712 return new_path;
1713 }
1714
1715 static int handle_rename_rename_1to2(struct merge_options *opt,
1716 struct rename_conflict_info *ci)
1717 {
1718 /* One file was renamed in both branches, but to different names. */
1719 struct merge_file_info mfi;
1720 struct diff_filespec *add;
1721 struct diff_filespec *o = ci->ren1->pair->one;
1722 struct diff_filespec *a = ci->ren1->pair->two;
1723 struct diff_filespec *b = ci->ren2->pair->two;
1724 char *path_desc;
1725
1726 output(opt, 1, _("CONFLICT (rename/rename): "
1727 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1728 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1729 o->path, a->path, ci->ren1->branch,
1730 o->path, b->path, ci->ren2->branch,
1731 opt->priv->call_depth ? _(" (left unresolved)") : "");
1732
1733 path_desc = xstrfmt("%s and %s, both renamed from %s",
1734 a->path, b->path, o->path);
1735 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1736 ci->ren1->branch, ci->ren2->branch,
1737 opt->priv->call_depth * 2, &mfi))
1738 return -1;
1739 free(path_desc);
1740
1741 if (opt->priv->call_depth) {
1742 /*
1743 * FIXME: For rename/add-source conflicts (if we could detect
1744 * such), this is wrong. We should instead find a unique
1745 * pathname and then either rename the add-source file to that
1746 * unique path, or use that unique path instead of src here.
1747 */
1748 if (update_file(opt, 0, &mfi.blob, o->path))
1749 return -1;
1750
1751 /*
1752 * Above, we put the merged content at the merge-base's
1753 * path. Now we usually need to delete both a->path and
1754 * b->path. However, the rename on each side of the merge
1755 * could also be involved in a rename/add conflict. In
1756 * such cases, we should keep the added file around,
1757 * resolving the conflict at that path in its favor.
1758 */
1759 add = &ci->ren1->dst_entry->stages[2 ^ 1];
1760 if (is_valid(add)) {
1761 if (update_file(opt, 0, add, a->path))
1762 return -1;
1763 }
1764 else
1765 remove_file_from_index(opt->repo->index, a->path);
1766 add = &ci->ren2->dst_entry->stages[3 ^ 1];
1767 if (is_valid(add)) {
1768 if (update_file(opt, 0, add, b->path))
1769 return -1;
1770 }
1771 else
1772 remove_file_from_index(opt->repo->index, b->path);
1773 } else {
1774 /*
1775 * For each destination path, we need to see if there is a
1776 * rename/add collision. If not, we can write the file out
1777 * to the specified location.
1778 */
1779 add = &ci->ren1->dst_entry->stages[2 ^ 1];
1780 if (is_valid(add)) {
1781 add->path = mfi.blob.path = a->path;
1782 if (handle_file_collision(opt, a->path,
1783 NULL, NULL,
1784 ci->ren1->branch,
1785 ci->ren2->branch,
1786 &mfi.blob, add) < 0)
1787 return -1;
1788 } else {
1789 char *new_path = find_path_for_conflict(opt, a->path,
1790 ci->ren1->branch,
1791 ci->ren2->branch);
1792 if (update_file(opt, 0, &mfi.blob,
1793 new_path ? new_path : a->path))
1794 return -1;
1795 free(new_path);
1796 if (update_stages(opt, a->path, NULL, a, NULL))
1797 return -1;
1798 }
1799
1800 add = &ci->ren2->dst_entry->stages[3 ^ 1];
1801 if (is_valid(add)) {
1802 add->path = mfi.blob.path = b->path;
1803 if (handle_file_collision(opt, b->path,
1804 NULL, NULL,
1805 ci->ren1->branch,
1806 ci->ren2->branch,
1807 add, &mfi.blob) < 0)
1808 return -1;
1809 } else {
1810 char *new_path = find_path_for_conflict(opt, b->path,
1811 ci->ren2->branch,
1812 ci->ren1->branch);
1813 if (update_file(opt, 0, &mfi.blob,
1814 new_path ? new_path : b->path))
1815 return -1;
1816 free(new_path);
1817 if (update_stages(opt, b->path, NULL, NULL, b))
1818 return -1;
1819 }
1820 }
1821
1822 return 0;
1823 }
1824
1825 static int handle_rename_rename_2to1(struct merge_options *opt,
1826 struct rename_conflict_info *ci)
1827 {
1828 /* Two files, a & b, were renamed to the same thing, c. */
1829 struct diff_filespec *a = ci->ren1->pair->one;
1830 struct diff_filespec *b = ci->ren2->pair->one;
1831 struct diff_filespec *c1 = ci->ren1->pair->two;
1832 struct diff_filespec *c2 = ci->ren2->pair->two;
1833 char *path = c1->path; /* == c2->path */
1834 char *path_side_1_desc;
1835 char *path_side_2_desc;
1836 struct merge_file_info mfi_c1;
1837 struct merge_file_info mfi_c2;
1838 int ostage1, ostage2;
1839
1840 output(opt, 1, _("CONFLICT (rename/rename): "
1841 "Rename %s->%s in %s. "
1842 "Rename %s->%s in %s"),
1843 a->path, c1->path, ci->ren1->branch,
1844 b->path, c2->path, ci->ren2->branch);
1845
1846 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1847 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1848 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1849 ostage2 = ostage1 ^ 1;
1850 ci->ren1->src_entry->stages[ostage1].path = a->path;
1851 ci->ren2->src_entry->stages[ostage2].path = b->path;
1852 if (merge_mode_and_contents(opt, a, c1,
1853 &ci->ren1->src_entry->stages[ostage1],
1854 path_side_1_desc,
1855 opt->branch1, opt->branch2,
1856 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1857 merge_mode_and_contents(opt, b,
1858 &ci->ren2->src_entry->stages[ostage2],
1859 c2, path_side_2_desc,
1860 opt->branch1, opt->branch2,
1861 1 + opt->priv->call_depth * 2, &mfi_c2))
1862 return -1;
1863 free(path_side_1_desc);
1864 free(path_side_2_desc);
1865 mfi_c1.blob.path = path;
1866 mfi_c2.blob.path = path;
1867
1868 return handle_file_collision(opt, path, a->path, b->path,
1869 ci->ren1->branch, ci->ren2->branch,
1870 &mfi_c1.blob, &mfi_c2.blob);
1871 }
1872
1873 /*
1874 * Get the diff_filepairs changed between o_tree and tree.
1875 */
1876 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1877 struct tree *o_tree,
1878 struct tree *tree)
1879 {
1880 struct diff_queue_struct *ret;
1881 struct diff_options opts;
1882
1883 repo_diff_setup(opt->repo, &opts);
1884 opts.flags.recursive = 1;
1885 opts.flags.rename_empty = 0;
1886 opts.detect_rename = merge_detect_rename(opt);
1887 /*
1888 * We do not have logic to handle the detection of copies. In
1889 * fact, it may not even make sense to add such logic: would we
1890 * really want a change to a base file to be propagated through
1891 * multiple other files by a merge?
1892 */
1893 if (opts.detect_rename > DIFF_DETECT_RENAME)
1894 opts.detect_rename = DIFF_DETECT_RENAME;
1895 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 1000;
1896 opts.rename_score = opt->rename_score;
1897 opts.show_rename_progress = opt->show_rename_progress;
1898 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1899 diff_setup_done(&opts);
1900 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1901 diffcore_std(&opts);
1902 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1903 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1904
1905 ret = xmalloc(sizeof(*ret));
1906 *ret = diff_queued_diff;
1907
1908 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1909 diff_queued_diff.nr = 0;
1910 diff_queued_diff.queue = NULL;
1911 diff_flush(&opts);
1912 return ret;
1913 }
1914
1915 static int tree_has_path(struct repository *r, struct tree *tree,
1916 const char *path)
1917 {
1918 struct object_id hashy;
1919 unsigned short mode_o;
1920
1921 return !get_tree_entry(r,
1922 &tree->object.oid, path,
1923 &hashy, &mode_o);
1924 }
1925
1926 /*
1927 * Return a new string that replaces the beginning portion (which matches
1928 * entry->dir), with entry->new_dir. In perl-speak:
1929 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1930 * NOTE:
1931 * Caller must ensure that old_path starts with entry->dir + '/'.
1932 */
1933 static char *apply_dir_rename(struct dir_rename_entry *entry,
1934 const char *old_path)
1935 {
1936 struct strbuf new_path = STRBUF_INIT;
1937 int oldlen, newlen;
1938
1939 if (entry->non_unique_new_dir)
1940 return NULL;
1941
1942 oldlen = strlen(entry->dir);
1943 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1944 strbuf_grow(&new_path, newlen);
1945 strbuf_addbuf(&new_path, &entry->new_dir);
1946 strbuf_addstr(&new_path, &old_path[oldlen]);
1947
1948 return strbuf_detach(&new_path, NULL);
1949 }
1950
1951 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1952 char **old_dir, char **new_dir)
1953 {
1954 char *end_of_old, *end_of_new;
1955 int old_len, new_len;
1956
1957 *old_dir = NULL;
1958 *new_dir = NULL;
1959
1960 /*
1961 * For
1962 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1963 * the "e/foo.c" part is the same, we just want to know that
1964 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1965 * so, for this example, this function returns "a/b/c/d" in
1966 * *old_dir and "a/b/some/thing/else" in *new_dir.
1967 *
1968 * Also, if the basename of the file changed, we don't care. We
1969 * want to know which portion of the directory, if any, changed.
1970 */
1971 end_of_old = strrchr(old_path, '/');
1972 end_of_new = strrchr(new_path, '/');
1973
1974 if (end_of_old == NULL || end_of_new == NULL)
1975 return;
1976 while (*--end_of_new == *--end_of_old &&
1977 end_of_old != old_path &&
1978 end_of_new != new_path)
1979 ; /* Do nothing; all in the while loop */
1980 /*
1981 * We've found the first non-matching character in the directory
1982 * paths. That means the current directory we were comparing
1983 * represents the rename. Move end_of_old and end_of_new back
1984 * to the full directory name.
1985 */
1986 if (*end_of_old == '/')
1987 end_of_old++;
1988 if (*end_of_old != '/')
1989 end_of_new++;
1990 end_of_old = strchr(end_of_old, '/');
1991 end_of_new = strchr(end_of_new, '/');
1992
1993 /*
1994 * It may have been the case that old_path and new_path were the same
1995 * directory all along. Don't claim a rename if they're the same.
1996 */
1997 old_len = end_of_old - old_path;
1998 new_len = end_of_new - new_path;
1999
2000 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
2001 *old_dir = xstrndup(old_path, old_len);
2002 *new_dir = xstrndup(new_path, new_len);
2003 }
2004 }
2005
2006 static void remove_hashmap_entries(struct hashmap *dir_renames,
2007 struct string_list *items_to_remove)
2008 {
2009 int i;
2010 struct dir_rename_entry *entry;
2011
2012 for (i = 0; i < items_to_remove->nr; i++) {
2013 entry = items_to_remove->items[i].util;
2014 hashmap_remove(dir_renames, entry, NULL);
2015 }
2016 string_list_clear(items_to_remove, 0);
2017 }
2018
2019 /*
2020 * See if there is a directory rename for path, and if there are any file
2021 * level conflicts for the renamed location. If there is a rename and
2022 * there are no conflicts, return the new name. Otherwise, return NULL.
2023 */
2024 static char *handle_path_level_conflicts(struct merge_options *opt,
2025 const char *path,
2026 struct dir_rename_entry *entry,
2027 struct hashmap *collisions,
2028 struct tree *tree)
2029 {
2030 char *new_path = NULL;
2031 struct collision_entry *collision_ent;
2032 int clean = 1;
2033 struct strbuf collision_paths = STRBUF_INIT;
2034
2035 /*
2036 * entry has the mapping of old directory name to new directory name
2037 * that we want to apply to path.
2038 */
2039 new_path = apply_dir_rename(entry, path);
2040
2041 if (!new_path) {
2042 /* This should only happen when entry->non_unique_new_dir set */
2043 if (!entry->non_unique_new_dir)
2044 BUG("entry->non_unqiue_dir not set and !new_path");
2045 output(opt, 1, _("CONFLICT (directory rename split): "
2046 "Unclear where to place %s because directory "
2047 "%s was renamed to multiple other directories, "
2048 "with no destination getting a majority of the "
2049 "files."),
2050 path, entry->dir);
2051 clean = 0;
2052 return NULL;
2053 }
2054
2055 /*
2056 * The caller needs to have ensured that it has pre-populated
2057 * collisions with all paths that map to new_path. Do a quick check
2058 * to ensure that's the case.
2059 */
2060 collision_ent = collision_find_entry(collisions, new_path);
2061 if (collision_ent == NULL)
2062 BUG("collision_ent is NULL");
2063
2064 /*
2065 * Check for one-sided add/add/.../add conflicts, i.e.
2066 * where implicit renames from the other side doing
2067 * directory rename(s) can affect this side of history
2068 * to put multiple paths into the same location. Warn
2069 * and bail on directory renames for such paths.
2070 */
2071 if (collision_ent->reported_already) {
2072 clean = 0;
2073 } else if (tree_has_path(opt->repo, tree, new_path)) {
2074 collision_ent->reported_already = 1;
2075 strbuf_add_separated_string_list(&collision_paths, ", ",
2076 &collision_ent->source_files);
2077 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2078 "file/dir at %s in the way of implicit "
2079 "directory rename(s) putting the following "
2080 "path(s) there: %s."),
2081 new_path, collision_paths.buf);
2082 clean = 0;
2083 } else if (collision_ent->source_files.nr > 1) {
2084 collision_ent->reported_already = 1;
2085 strbuf_add_separated_string_list(&collision_paths, ", ",
2086 &collision_ent->source_files);
2087 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2088 "more than one path to %s; implicit directory "
2089 "renames tried to put these paths there: %s"),
2090 new_path, collision_paths.buf);
2091 clean = 0;
2092 }
2093
2094 /* Free memory we no longer need */
2095 strbuf_release(&collision_paths);
2096 if (!clean && new_path) {
2097 free(new_path);
2098 return NULL;
2099 }
2100
2101 return new_path;
2102 }
2103
2104 /*
2105 * There are a couple things we want to do at the directory level:
2106 * 1. Check for both sides renaming to the same thing, in order to avoid
2107 * implicit renaming of files that should be left in place. (See
2108 * testcase 6b in t6043 for details.)
2109 * 2. Prune directory renames if there are still files left in the
2110 * the original directory. These represent a partial directory rename,
2111 * i.e. a rename where only some of the files within the directory
2112 * were renamed elsewhere. (Technically, this could be done earlier
2113 * in get_directory_renames(), except that would prevent us from
2114 * doing the previous check and thus failing testcase 6b.)
2115 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2116 * In the future, we could potentially record this info as well and
2117 * omit reporting rename/rename(1to2) conflicts for each path within
2118 * the affected directories, thus cleaning up the merge output.
2119 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2120 * directory level, because merging directories is fine. If it
2121 * causes conflicts for files within those merged directories, then
2122 * that should be detected at the individual path level.
2123 */
2124 static void handle_directory_level_conflicts(struct merge_options *opt,
2125 struct hashmap *dir_re_head,
2126 struct tree *head,
2127 struct hashmap *dir_re_merge,
2128 struct tree *merge)
2129 {
2130 struct hashmap_iter iter;
2131 struct dir_rename_entry *head_ent;
2132 struct dir_rename_entry *merge_ent;
2133
2134 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2135 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2136
2137 hashmap_iter_init(dir_re_head, &iter);
2138 while ((head_ent = hashmap_iter_next(&iter))) {
2139 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2140 if (merge_ent &&
2141 !head_ent->non_unique_new_dir &&
2142 !merge_ent->non_unique_new_dir &&
2143 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2144 /* 1. Renamed identically; remove it from both sides */
2145 string_list_append(&remove_from_head,
2146 head_ent->dir)->util = head_ent;
2147 strbuf_release(&head_ent->new_dir);
2148 string_list_append(&remove_from_merge,
2149 merge_ent->dir)->util = merge_ent;
2150 strbuf_release(&merge_ent->new_dir);
2151 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2152 /* 2. This wasn't a directory rename after all */
2153 string_list_append(&remove_from_head,
2154 head_ent->dir)->util = head_ent;
2155 strbuf_release(&head_ent->new_dir);
2156 }
2157 }
2158
2159 remove_hashmap_entries(dir_re_head, &remove_from_head);
2160 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2161
2162 hashmap_iter_init(dir_re_merge, &iter);
2163 while ((merge_ent = hashmap_iter_next(&iter))) {
2164 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2165 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2166 /* 2. This wasn't a directory rename after all */
2167 string_list_append(&remove_from_merge,
2168 merge_ent->dir)->util = merge_ent;
2169 } else if (head_ent &&
2170 !head_ent->non_unique_new_dir &&
2171 !merge_ent->non_unique_new_dir) {
2172 /* 3. rename/rename(1to2) */
2173 /*
2174 * We can assume it's not rename/rename(1to1) because
2175 * that was case (1), already checked above. So we
2176 * know that head_ent->new_dir and merge_ent->new_dir
2177 * are different strings.
2178 */
2179 output(opt, 1, _("CONFLICT (rename/rename): "
2180 "Rename directory %s->%s in %s. "
2181 "Rename directory %s->%s in %s"),
2182 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2183 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2184 string_list_append(&remove_from_head,
2185 head_ent->dir)->util = head_ent;
2186 strbuf_release(&head_ent->new_dir);
2187 string_list_append(&remove_from_merge,
2188 merge_ent->dir)->util = merge_ent;
2189 strbuf_release(&merge_ent->new_dir);
2190 }
2191 }
2192
2193 remove_hashmap_entries(dir_re_head, &remove_from_head);
2194 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2195 }
2196
2197 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2198 {
2199 struct hashmap *dir_renames;
2200 struct hashmap_iter iter;
2201 struct dir_rename_entry *entry;
2202 int i;
2203
2204 /*
2205 * Typically, we think of a directory rename as all files from a
2206 * certain directory being moved to a target directory. However,
2207 * what if someone first moved two files from the original
2208 * directory in one commit, and then renamed the directory
2209 * somewhere else in a later commit? At merge time, we just know
2210 * that files from the original directory went to two different
2211 * places, and that the bulk of them ended up in the same place.
2212 * We want each directory rename to represent where the bulk of the
2213 * files from that directory end up; this function exists to find
2214 * where the bulk of the files went.
2215 *
2216 * The first loop below simply iterates through the list of file
2217 * renames, finding out how often each directory rename pair
2218 * possibility occurs.
2219 */
2220 dir_renames = xmalloc(sizeof(*dir_renames));
2221 dir_rename_init(dir_renames);
2222 for (i = 0; i < pairs->nr; ++i) {
2223 struct string_list_item *item;
2224 int *count;
2225 struct diff_filepair *pair = pairs->queue[i];
2226 char *old_dir, *new_dir;
2227
2228 /* File not part of directory rename if it wasn't renamed */
2229 if (pair->status != 'R')
2230 continue;
2231
2232 get_renamed_dir_portion(pair->one->path, pair->two->path,
2233 &old_dir, &new_dir);
2234 if (!old_dir)
2235 /* Directory didn't change at all; ignore this one. */
2236 continue;
2237
2238 entry = dir_rename_find_entry(dir_renames, old_dir);
2239 if (!entry) {
2240 entry = xmalloc(sizeof(*entry));
2241 dir_rename_entry_init(entry, old_dir);
2242 hashmap_put(dir_renames, entry);
2243 } else {
2244 free(old_dir);
2245 }
2246 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2247 if (!item) {
2248 item = string_list_insert(&entry->possible_new_dirs,
2249 new_dir);
2250 item->util = xcalloc(1, sizeof(int));
2251 } else {
2252 free(new_dir);
2253 }
2254 count = item->util;
2255 *count += 1;
2256 }
2257
2258 /*
2259 * For each directory with files moved out of it, we find out which
2260 * target directory received the most files so we can declare it to
2261 * be the "winning" target location for the directory rename. This
2262 * winner gets recorded in new_dir. If there is no winner
2263 * (multiple target directories received the same number of files),
2264 * we set non_unique_new_dir. Once we've determined the winner (or
2265 * that there is no winner), we no longer need possible_new_dirs.
2266 */
2267 hashmap_iter_init(dir_renames, &iter);
2268 while ((entry = hashmap_iter_next(&iter))) {
2269 int max = 0;
2270 int bad_max = 0;
2271 char *best = NULL;
2272
2273 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2274 int *count = entry->possible_new_dirs.items[i].util;
2275
2276 if (*count == max)
2277 bad_max = max;
2278 else if (*count > max) {
2279 max = *count;
2280 best = entry->possible_new_dirs.items[i].string;
2281 }
2282 }
2283 if (bad_max == max)
2284 entry->non_unique_new_dir = 1;
2285 else {
2286 assert(entry->new_dir.len == 0);
2287 strbuf_addstr(&entry->new_dir, best);
2288 }
2289 /*
2290 * The relevant directory sub-portion of the original full
2291 * filepaths were xstrndup'ed before inserting into
2292 * possible_new_dirs, and instead of manually iterating the
2293 * list and free'ing each, just lie and tell
2294 * possible_new_dirs that it did the strdup'ing so that it
2295 * will free them for us.
2296 */
2297 entry->possible_new_dirs.strdup_strings = 1;
2298 string_list_clear(&entry->possible_new_dirs, 1);
2299 }
2300
2301 return dir_renames;
2302 }
2303
2304 static struct dir_rename_entry *check_dir_renamed(const char *path,
2305 struct hashmap *dir_renames)
2306 {
2307 char *temp = xstrdup(path);
2308 char *end;
2309 struct dir_rename_entry *entry = NULL;
2310
2311 while ((end = strrchr(temp, '/'))) {
2312 *end = '\0';
2313 entry = dir_rename_find_entry(dir_renames, temp);
2314 if (entry)
2315 break;
2316 }
2317 free(temp);
2318 return entry;
2319 }
2320
2321 static void compute_collisions(struct hashmap *collisions,
2322 struct hashmap *dir_renames,
2323 struct diff_queue_struct *pairs)
2324 {
2325 int i;
2326
2327 /*
2328 * Multiple files can be mapped to the same path due to directory
2329 * renames done by the other side of history. Since that other
2330 * side of history could have merged multiple directories into one,
2331 * if our side of history added the same file basename to each of
2332 * those directories, then all N of them would get implicitly
2333 * renamed by the directory rename detection into the same path,
2334 * and we'd get an add/add/.../add conflict, and all those adds
2335 * from *this* side of history. This is not representable in the
2336 * index, and users aren't going to easily be able to make sense of
2337 * it. So we need to provide a good warning about what's
2338 * happening, and fall back to no-directory-rename detection
2339 * behavior for those paths.
2340 *
2341 * See testcases 9e and all of section 5 from t6043 for examples.
2342 */
2343 collision_init(collisions);
2344
2345 for (i = 0; i < pairs->nr; ++i) {
2346 struct dir_rename_entry *dir_rename_ent;
2347 struct collision_entry *collision_ent;
2348 char *new_path;
2349 struct diff_filepair *pair = pairs->queue[i];
2350
2351 if (pair->status != 'A' && pair->status != 'R')
2352 continue;
2353 dir_rename_ent = check_dir_renamed(pair->two->path,
2354 dir_renames);
2355 if (!dir_rename_ent)
2356 continue;
2357
2358 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2359 if (!new_path)
2360 /*
2361 * dir_rename_ent->non_unique_new_path is true, which
2362 * means there is no directory rename for us to use,
2363 * which means it won't cause us any additional
2364 * collisions.
2365 */
2366 continue;
2367 collision_ent = collision_find_entry(collisions, new_path);
2368 if (!collision_ent) {
2369 collision_ent = xcalloc(1,
2370 sizeof(struct collision_entry));
2371 hashmap_entry_init(collision_ent, strhash(new_path));
2372 hashmap_put(collisions, collision_ent);
2373 collision_ent->target_file = new_path;
2374 } else {
2375 free(new_path);
2376 }
2377 string_list_insert(&collision_ent->source_files,
2378 pair->two->path);
2379 }
2380 }
2381
2382 static char *check_for_directory_rename(struct merge_options *opt,
2383 const char *path,
2384 struct tree *tree,
2385 struct hashmap *dir_renames,
2386 struct hashmap *dir_rename_exclusions,
2387 struct hashmap *collisions,
2388 int *clean_merge)
2389 {
2390 char *new_path = NULL;
2391 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2392 struct dir_rename_entry *oentry = NULL;
2393
2394 if (!entry)
2395 return new_path;
2396
2397 /*
2398 * This next part is a little weird. We do not want to do an
2399 * implicit rename into a directory we renamed on our side, because
2400 * that will result in a spurious rename/rename(1to2) conflict. An
2401 * example:
2402 * Base commit: dumbdir/afile, otherdir/bfile
2403 * Side 1: smrtdir/afile, otherdir/bfile
2404 * Side 2: dumbdir/afile, dumbdir/bfile
2405 * Here, while working on Side 1, we could notice that otherdir was
2406 * renamed/merged to dumbdir, and change the diff_filepair for
2407 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2408 * 2 will notice the rename from dumbdir to smrtdir, and do the
2409 * transitive rename to move it from dumbdir/bfile to
2410 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2411 * smrtdir, a rename/rename(1to2) conflict. We really just want
2412 * the file to end up in smrtdir. And the way to achieve that is
2413 * to not let Side1 do the rename to dumbdir, since we know that is
2414 * the source of one of our directory renames.
2415 *
2416 * That's why oentry and dir_rename_exclusions is here.
2417 *
2418 * As it turns out, this also prevents N-way transient rename
2419 * confusion; See testcases 9c and 9d of t6043.
2420 */
2421 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2422 if (oentry) {
2423 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2424 "to %s, because %s itself was renamed."),
2425 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2426 } else {
2427 new_path = handle_path_level_conflicts(opt, path, entry,
2428 collisions, tree);
2429 *clean_merge &= (new_path != NULL);
2430 }
2431
2432 return new_path;
2433 }
2434
2435 static void apply_directory_rename_modifications(struct merge_options *opt,
2436 struct diff_filepair *pair,
2437 char *new_path,
2438 struct rename *re,
2439 struct tree *tree,
2440 struct tree *o_tree,
2441 struct tree *a_tree,
2442 struct tree *b_tree,
2443 struct string_list *entries)
2444 {
2445 struct string_list_item *item;
2446 int stage = (tree == a_tree ? 2 : 3);
2447 int update_wd;
2448
2449 /*
2450 * In all cases where we can do directory rename detection,
2451 * unpack_trees() will have read pair->two->path into the
2452 * index and the working copy. We need to remove it so that
2453 * we can instead place it at new_path. It is guaranteed to
2454 * not be untracked (unpack_trees() would have errored out
2455 * saying the file would have been overwritten), but it might
2456 * be dirty, though.
2457 */
2458 update_wd = !was_dirty(opt, pair->two->path);
2459 if (!update_wd)
2460 output(opt, 1, _("Refusing to lose dirty file at %s"),
2461 pair->two->path);
2462 remove_file(opt, 1, pair->two->path, !update_wd);
2463
2464 /* Find or create a new re->dst_entry */
2465 item = string_list_lookup(entries, new_path);
2466 if (item) {
2467 /*
2468 * Since we're renaming on this side of history, and it's
2469 * due to a directory rename on the other side of history
2470 * (which we only allow when the directory in question no
2471 * longer exists on the other side of history), the
2472 * original entry for re->dst_entry is no longer
2473 * necessary...
2474 */
2475 re->dst_entry->processed = 1;
2476
2477 /*
2478 * ...because we'll be using this new one.
2479 */
2480 re->dst_entry = item->util;
2481 } else {
2482 /*
2483 * re->dst_entry is for the before-dir-rename path, and we
2484 * need it to hold information for the after-dir-rename
2485 * path. Before creating a new entry, we need to mark the
2486 * old one as unnecessary (...unless it is shared by
2487 * src_entry, i.e. this didn't use to be a rename, in which
2488 * case we can just allow the normal processing to happen
2489 * for it).
2490 */
2491 if (pair->status == 'R')
2492 re->dst_entry->processed = 1;
2493
2494 re->dst_entry = insert_stage_data(opt->repo, new_path,
2495 o_tree, a_tree, b_tree,
2496 entries);
2497 item = string_list_insert(entries, new_path);
2498 item->util = re->dst_entry;
2499 }
2500
2501 /*
2502 * Update the stage_data with the information about the path we are
2503 * moving into place. That slot will be empty and available for us
2504 * to write to because of the collision checks in
2505 * handle_path_level_conflicts(). In other words,
2506 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2507 * open for us to write to.
2508 *
2509 * It may be tempting to actually update the index at this point as
2510 * well, using update_stages_for_stage_data(), but as per the big
2511 * "NOTE" in update_stages(), doing so will modify the current
2512 * in-memory index which will break calls to would_lose_untracked()
2513 * that we need to make. Instead, we need to just make sure that
2514 * the various handle_rename_*() functions update the index
2515 * explicitly rather than relying on unpack_trees() to have done it.
2516 */
2517 get_tree_entry(opt->repo,
2518 &tree->object.oid,
2519 pair->two->path,
2520 &re->dst_entry->stages[stage].oid,
2521 &re->dst_entry->stages[stage].mode);
2522
2523 /*
2524 * Record the original change status (or 'type' of change). If it
2525 * was originally an add ('A'), this lets us differentiate later
2526 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2527 * otherwise look the same). If it was originally a rename ('R'),
2528 * this lets us remember and report accurately about the transitive
2529 * renaming that occurred via the directory rename detection. Also,
2530 * record the original destination name.
2531 */
2532 re->dir_rename_original_type = pair->status;
2533 re->dir_rename_original_dest = pair->two->path;
2534
2535 /*
2536 * We don't actually look at pair->status again, but it seems
2537 * pedagogically correct to adjust it.
2538 */
2539 pair->status = 'R';
2540
2541 /*
2542 * Finally, record the new location.
2543 */
2544 pair->two->path = new_path;
2545 }
2546
2547 /*
2548 * Get information of all renames which occurred in 'pairs', making use of
2549 * any implicit directory renames inferred from the other side of history.
2550 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2551 * to be able to associate the correct cache entries with the rename
2552 * information; tree is always equal to either a_tree or b_tree.
2553 */
2554 static struct string_list *get_renames(struct merge_options *opt,
2555 const char *branch,
2556 struct diff_queue_struct *pairs,
2557 struct hashmap *dir_renames,
2558 struct hashmap *dir_rename_exclusions,
2559 struct tree *tree,
2560 struct tree *o_tree,
2561 struct tree *a_tree,
2562 struct tree *b_tree,
2563 struct string_list *entries,
2564 int *clean_merge)
2565 {
2566 int i;
2567 struct hashmap collisions;
2568 struct hashmap_iter iter;
2569 struct collision_entry *e;
2570 struct string_list *renames;
2571
2572 compute_collisions(&collisions, dir_renames, pairs);
2573 renames = xcalloc(1, sizeof(struct string_list));
2574
2575 for (i = 0; i < pairs->nr; ++i) {
2576 struct string_list_item *item;
2577 struct rename *re;
2578 struct diff_filepair *pair = pairs->queue[i];
2579 char *new_path; /* non-NULL only with directory renames */
2580
2581 if (pair->status != 'A' && pair->status != 'R') {
2582 diff_free_filepair(pair);
2583 continue;
2584 }
2585 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2586 dir_renames,
2587 dir_rename_exclusions,
2588 &collisions,
2589 clean_merge);
2590 if (pair->status != 'R' && !new_path) {
2591 diff_free_filepair(pair);
2592 continue;
2593 }
2594
2595 re = xmalloc(sizeof(*re));
2596 re->processed = 0;
2597 re->pair = pair;
2598 re->branch = branch;
2599 re->dir_rename_original_type = '\0';
2600 re->dir_rename_original_dest = NULL;
2601 item = string_list_lookup(entries, re->pair->one->path);
2602 if (!item)
2603 re->src_entry = insert_stage_data(opt->repo,
2604 re->pair->one->path,
2605 o_tree, a_tree, b_tree, entries);
2606 else
2607 re->src_entry = item->util;
2608
2609 item = string_list_lookup(entries, re->pair->two->path);
2610 if (!item)
2611 re->dst_entry = insert_stage_data(opt->repo,
2612 re->pair->two->path,
2613 o_tree, a_tree, b_tree, entries);
2614 else
2615 re->dst_entry = item->util;
2616 item = string_list_insert(renames, pair->one->path);
2617 item->util = re;
2618 if (new_path)
2619 apply_directory_rename_modifications(opt, pair, new_path,
2620 re, tree, o_tree,
2621 a_tree, b_tree,
2622 entries);
2623 }
2624
2625 hashmap_iter_init(&collisions, &iter);
2626 while ((e = hashmap_iter_next(&iter))) {
2627 free(e->target_file);
2628 string_list_clear(&e->source_files, 0);
2629 }
2630 hashmap_free(&collisions, 1);
2631 return renames;
2632 }
2633
2634 static int process_renames(struct merge_options *opt,
2635 struct string_list *a_renames,
2636 struct string_list *b_renames)
2637 {
2638 int clean_merge = 1, i, j;
2639 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2640 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2641 const struct rename *sre;
2642
2643 for (i = 0; i < a_renames->nr; i++) {
2644 sre = a_renames->items[i].util;
2645 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2646 = (void *)sre;
2647 }
2648 for (i = 0; i < b_renames->nr; i++) {
2649 sre = b_renames->items[i].util;
2650 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2651 = (void *)sre;
2652 }
2653
2654 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2655 struct string_list *renames1, *renames2Dst;
2656 struct rename *ren1 = NULL, *ren2 = NULL;
2657 const char *ren1_src, *ren1_dst;
2658 struct string_list_item *lookup;
2659
2660 if (i >= a_renames->nr) {
2661 ren2 = b_renames->items[j++].util;
2662 } else if (j >= b_renames->nr) {
2663 ren1 = a_renames->items[i++].util;
2664 } else {
2665 int compare = strcmp(a_renames->items[i].string,
2666 b_renames->items[j].string);
2667 if (compare <= 0)
2668 ren1 = a_renames->items[i++].util;
2669 if (compare >= 0)
2670 ren2 = b_renames->items[j++].util;
2671 }
2672
2673 /* TODO: refactor, so that 1/2 are not needed */
2674 if (ren1) {
2675 renames1 = a_renames;
2676 renames2Dst = &b_by_dst;
2677 } else {
2678 renames1 = b_renames;
2679 renames2Dst = &a_by_dst;
2680 SWAP(ren2, ren1);
2681 }
2682
2683 if (ren1->processed)
2684 continue;
2685 ren1->processed = 1;
2686 ren1->dst_entry->processed = 1;
2687 /* BUG: We should only mark src_entry as processed if we
2688 * are not dealing with a rename + add-source case.
2689 */
2690 ren1->src_entry->processed = 1;
2691
2692 ren1_src = ren1->pair->one->path;
2693 ren1_dst = ren1->pair->two->path;
2694
2695 if (ren2) {
2696 /* One file renamed on both sides */
2697 const char *ren2_src = ren2->pair->one->path;
2698 const char *ren2_dst = ren2->pair->two->path;
2699 enum rename_type rename_type;
2700 if (strcmp(ren1_src, ren2_src) != 0)
2701 BUG("ren1_src != ren2_src");
2702 ren2->dst_entry->processed = 1;
2703 ren2->processed = 1;
2704 if (strcmp(ren1_dst, ren2_dst) != 0) {
2705 rename_type = RENAME_ONE_FILE_TO_TWO;
2706 clean_merge = 0;
2707 } else {
2708 rename_type = RENAME_ONE_FILE_TO_ONE;
2709 /* BUG: We should only remove ren1_src in
2710 * the base stage (think of rename +
2711 * add-source cases).
2712 */
2713 remove_file(opt, 1, ren1_src, 1);
2714 update_entry(ren1->dst_entry,
2715 ren1->pair->one,
2716 ren1->pair->two,
2717 ren2->pair->two);
2718 }
2719 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2720 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2721 /* Two different files renamed to the same thing */
2722 char *ren2_dst;
2723 ren2 = lookup->util;
2724 ren2_dst = ren2->pair->two->path;
2725 if (strcmp(ren1_dst, ren2_dst) != 0)
2726 BUG("ren1_dst != ren2_dst");
2727
2728 clean_merge = 0;
2729 ren2->processed = 1;
2730 /*
2731 * BUG: We should only mark src_entry as processed
2732 * if we are not dealing with a rename + add-source
2733 * case.
2734 */
2735 ren2->src_entry->processed = 1;
2736
2737 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2738 opt, ren1, ren2);
2739 } else {
2740 /* Renamed in 1, maybe changed in 2 */
2741 /* we only use sha1 and mode of these */
2742 struct diff_filespec src_other, dst_other;
2743 int try_merge;
2744
2745 /*
2746 * unpack_trees loads entries from common-commit
2747 * into stage 1, from head-commit into stage 2, and
2748 * from merge-commit into stage 3. We keep track
2749 * of which side corresponds to the rename.
2750 */
2751 int renamed_stage = a_renames == renames1 ? 2 : 3;
2752 int other_stage = a_renames == renames1 ? 3 : 2;
2753
2754 /* BUG: We should only remove ren1_src in the base
2755 * stage and in other_stage (think of rename +
2756 * add-source case).
2757 */
2758 remove_file(opt, 1, ren1_src,
2759 renamed_stage == 2 || !was_tracked(opt, ren1_src));
2760
2761 oidcpy(&src_other.oid,
2762 &ren1->src_entry->stages[other_stage].oid);
2763 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2764 oidcpy(&dst_other.oid,
2765 &ren1->dst_entry->stages[other_stage].oid);
2766 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2767 try_merge = 0;
2768
2769 if (oid_eq(&src_other.oid, &null_oid) &&
2770 ren1->dir_rename_original_type == 'A') {
2771 setup_rename_conflict_info(RENAME_VIA_DIR,
2772 opt, ren1, NULL);
2773 } else if (oid_eq(&src_other.oid, &null_oid)) {
2774 setup_rename_conflict_info(RENAME_DELETE,
2775 opt, ren1, NULL);
2776 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2777 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2778 /*
2779 * Added file on the other side identical to
2780 * the file being renamed: clean merge.
2781 * Also, there is no need to overwrite the
2782 * file already in the working copy, so call
2783 * update_file_flags() instead of
2784 * update_file().
2785 */
2786 if (update_file_flags(opt,
2787 ren1->pair->two,
2788 ren1_dst,
2789 1, /* update_cache */
2790 0 /* update_wd */))
2791 clean_merge = -1;
2792 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2793 /*
2794 * Probably not a clean merge, but it's
2795 * premature to set clean_merge to 0 here,
2796 * because if the rename merges cleanly and
2797 * the merge exactly matches the newly added
2798 * file, then the merge will be clean.
2799 */
2800 setup_rename_conflict_info(RENAME_ADD,
2801 opt, ren1, NULL);
2802 } else
2803 try_merge = 1;
2804
2805 if (clean_merge < 0)
2806 goto cleanup_and_return;
2807 if (try_merge) {
2808 struct diff_filespec *o, *a, *b;
2809 src_other.path = (char *)ren1_src;
2810
2811 o = ren1->pair->one;
2812 if (a_renames == renames1) {
2813 a = ren1->pair->two;
2814 b = &src_other;
2815 } else {
2816 b = ren1->pair->two;
2817 a = &src_other;
2818 }
2819 update_entry(ren1->dst_entry, o, a, b);
2820 setup_rename_conflict_info(RENAME_NORMAL,
2821 opt, ren1, NULL);
2822 }
2823 }
2824 }
2825 cleanup_and_return:
2826 string_list_clear(&a_by_dst, 0);
2827 string_list_clear(&b_by_dst, 0);
2828
2829 return clean_merge;
2830 }
2831
2832 struct rename_info {
2833 struct string_list *head_renames;
2834 struct string_list *merge_renames;
2835 };
2836
2837 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2838 struct hashmap *dir_renames)
2839 {
2840 struct hashmap_iter iter;
2841 struct dir_rename_entry *e;
2842
2843 hashmap_iter_init(dir_renames, &iter);
2844 while ((e = hashmap_iter_next(&iter))) {
2845 free(e->dir);
2846 strbuf_release(&e->new_dir);
2847 /* possible_new_dirs already cleared in get_directory_renames */
2848 }
2849 hashmap_free(dir_renames, 1);
2850 free(dir_renames);
2851
2852 free(pairs->queue);
2853 free(pairs);
2854 }
2855
2856 static int detect_and_process_renames(struct merge_options *opt,
2857 struct tree *common,
2858 struct tree *head,
2859 struct tree *merge,
2860 struct string_list *entries,
2861 struct rename_info *ri)
2862 {
2863 struct diff_queue_struct *head_pairs, *merge_pairs;
2864 struct hashmap *dir_re_head, *dir_re_merge;
2865 int clean = 1;
2866
2867 ri->head_renames = NULL;
2868 ri->merge_renames = NULL;
2869
2870 if (!merge_detect_rename(opt))
2871 return 1;
2872
2873 head_pairs = get_diffpairs(opt, common, head);
2874 merge_pairs = get_diffpairs(opt, common, merge);
2875
2876 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2877 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2878 !opt->priv->call_depth)) {
2879 dir_re_head = get_directory_renames(head_pairs);
2880 dir_re_merge = get_directory_renames(merge_pairs);
2881
2882 handle_directory_level_conflicts(opt,
2883 dir_re_head, head,
2884 dir_re_merge, merge);
2885 } else {
2886 dir_re_head = xmalloc(sizeof(*dir_re_head));
2887 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2888 dir_rename_init(dir_re_head);
2889 dir_rename_init(dir_re_merge);
2890 }
2891
2892 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2893 dir_re_merge, dir_re_head, head,
2894 common, head, merge, entries,
2895 &clean);
2896 if (clean < 0)
2897 goto cleanup;
2898 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2899 dir_re_head, dir_re_merge, merge,
2900 common, head, merge, entries,
2901 &clean);
2902 if (clean < 0)
2903 goto cleanup;
2904 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2905
2906 cleanup:
2907 /*
2908 * Some cleanup is deferred until cleanup_renames() because the
2909 * data structures are still needed and referenced in
2910 * process_entry(). But there are a few things we can free now.
2911 */
2912 initial_cleanup_rename(head_pairs, dir_re_head);
2913 initial_cleanup_rename(merge_pairs, dir_re_merge);
2914
2915 return clean;
2916 }
2917
2918 static void final_cleanup_rename(struct string_list *rename)
2919 {
2920 const struct rename *re;
2921 int i;
2922
2923 if (rename == NULL)
2924 return;
2925
2926 for (i = 0; i < rename->nr; i++) {
2927 re = rename->items[i].util;
2928 diff_free_filepair(re->pair);
2929 }
2930 string_list_clear(rename, 1);
2931 free(rename);
2932 }
2933
2934 static void final_cleanup_renames(struct rename_info *re_info)
2935 {
2936 final_cleanup_rename(re_info->head_renames);
2937 final_cleanup_rename(re_info->merge_renames);
2938 }
2939
2940 static int read_oid_strbuf(struct merge_options *opt,
2941 const struct object_id *oid,
2942 struct strbuf *dst)
2943 {
2944 void *buf;
2945 enum object_type type;
2946 unsigned long size;
2947 buf = read_object_file(oid, &type, &size);
2948 if (!buf)
2949 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
2950 if (type != OBJ_BLOB) {
2951 free(buf);
2952 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
2953 }
2954 strbuf_attach(dst, buf, size, size + 1);
2955 return 0;
2956 }
2957
2958 static int blob_unchanged(struct merge_options *opt,
2959 const struct diff_filespec *o,
2960 const struct diff_filespec *a,
2961 int renormalize, const char *path)
2962 {
2963 struct strbuf obuf = STRBUF_INIT;
2964 struct strbuf abuf = STRBUF_INIT;
2965 int ret = 0; /* assume changed for safety */
2966 const struct index_state *idx = opt->repo->index;
2967
2968 if (a->mode != o->mode)
2969 return 0;
2970 if (oid_eq(&o->oid, &a->oid))
2971 return 1;
2972 if (!renormalize)
2973 return 0;
2974
2975 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
2976 read_oid_strbuf(opt, &a->oid, &abuf))
2977 goto error_return;
2978 /*
2979 * Note: binary | is used so that both renormalizations are
2980 * performed. Comparison can be skipped if both files are
2981 * unchanged since their sha1s have already been compared.
2982 */
2983 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
2984 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
2985 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
2986
2987 error_return:
2988 strbuf_release(&obuf);
2989 strbuf_release(&abuf);
2990 return ret;
2991 }
2992
2993 static int handle_modify_delete(struct merge_options *opt,
2994 const char *path,
2995 const struct diff_filespec *o,
2996 const struct diff_filespec *a,
2997 const struct diff_filespec *b)
2998 {
2999 const char *modify_branch, *delete_branch;
3000 const struct diff_filespec *changed;
3001
3002 if (is_valid(a)) {
3003 modify_branch = opt->branch1;
3004 delete_branch = opt->branch2;
3005 changed = a;
3006 } else {
3007 modify_branch = opt->branch2;
3008 delete_branch = opt->branch1;
3009 changed = b;
3010 }
3011
3012 return handle_change_delete(opt,
3013 path, NULL,
3014 o, changed,
3015 modify_branch, delete_branch,
3016 _("modify"), _("modified"));
3017 }
3018
3019 static int handle_content_merge(struct merge_file_info *mfi,
3020 struct merge_options *opt,
3021 const char *path,
3022 int is_dirty,
3023 const struct diff_filespec *o,
3024 const struct diff_filespec *a,
3025 const struct diff_filespec *b,
3026 struct rename_conflict_info *ci)
3027 {
3028 const char *reason = _("content");
3029 unsigned df_conflict_remains = 0;
3030
3031 if (!is_valid(o))
3032 reason = _("add/add");
3033
3034 assert(o->path && a->path && b->path);
3035 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3036 S_ISGITLINK(ci->ren1->pair->two->mode)))
3037 df_conflict_remains = 1;
3038
3039 if (merge_mode_and_contents(opt, o, a, b, path,
3040 opt->branch1, opt->branch2,
3041 opt->priv->call_depth * 2, mfi))
3042 return -1;
3043
3044 /*
3045 * We can skip updating the working tree file iff:
3046 * a) The merge is clean
3047 * b) The merge matches what was in HEAD (content, mode, pathname)
3048 * c) The target path is usable (i.e. not involved in D/F conflict)
3049 */
3050 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3051 !df_conflict_remains) {
3052 int pos;
3053 struct cache_entry *ce;
3054
3055 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3056 if (add_cacheinfo(opt, &mfi->blob, path,
3057 0, (!opt->priv->call_depth && !is_dirty), 0))
3058 return -1;
3059 /*
3060 * However, add_cacheinfo() will delete the old cache entry
3061 * and add a new one. We need to copy over any skip_worktree
3062 * flag to avoid making the file appear as if it were
3063 * deleted by the user.
3064 */
3065 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3066 ce = opt->priv->orig_index.cache[pos];
3067 if (ce_skip_worktree(ce)) {
3068 pos = index_name_pos(opt->repo->index, path, strlen(path));
3069 ce = opt->repo->index->cache[pos];
3070 ce->ce_flags |= CE_SKIP_WORKTREE;
3071 }
3072 return mfi->clean;
3073 }
3074
3075 if (!mfi->clean) {
3076 if (S_ISGITLINK(mfi->blob.mode))
3077 reason = _("submodule");
3078 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3079 reason, path);
3080 if (ci && !df_conflict_remains)
3081 if (update_stages(opt, path, o, a, b))
3082 return -1;
3083 }
3084
3085 if (df_conflict_remains || is_dirty) {
3086 char *new_path;
3087 if (opt->priv->call_depth) {
3088 remove_file_from_index(opt->repo->index, path);
3089 } else {
3090 if (!mfi->clean) {
3091 if (update_stages(opt, path, o, a, b))
3092 return -1;
3093 } else {
3094 int file_from_stage2 = was_tracked(opt, path);
3095
3096 if (update_stages(opt, path, NULL,
3097 file_from_stage2 ? &mfi->blob : NULL,
3098 file_from_stage2 ? NULL : &mfi->blob))
3099 return -1;
3100 }
3101
3102 }
3103 new_path = unique_path(opt, path, ci->ren1->branch);
3104 if (is_dirty) {
3105 output(opt, 1, _("Refusing to lose dirty file at %s"),
3106 path);
3107 }
3108 output(opt, 1, _("Adding as %s instead"), new_path);
3109 if (update_file(opt, 0, &mfi->blob, new_path)) {
3110 free(new_path);
3111 return -1;
3112 }
3113 free(new_path);
3114 mfi->clean = 0;
3115 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3116 return -1;
3117 return !is_dirty && mfi->clean;
3118 }
3119
3120 static int handle_rename_normal(struct merge_options *opt,
3121 const char *path,
3122 const struct diff_filespec *o,
3123 const struct diff_filespec *a,
3124 const struct diff_filespec *b,
3125 struct rename_conflict_info *ci)
3126 {
3127 struct rename *ren = ci->ren1;
3128 struct merge_file_info mfi;
3129 int clean;
3130 int side = (ren->branch == opt->branch1 ? 2 : 3);
3131
3132 /* Merge the content and write it out */
3133 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3134 o, a, b, ci);
3135
3136 if (clean &&
3137 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3138 ren->dir_rename_original_dest) {
3139 if (update_stages(opt, path,
3140 NULL,
3141 side == 2 ? &mfi.blob : NULL,
3142 side == 2 ? NULL : &mfi.blob))
3143 return -1;
3144 clean = 0; /* not clean, but conflicted */
3145 }
3146 return clean;
3147 }
3148
3149 static void dir_rename_warning(const char *msg,
3150 int is_add,
3151 int clean,
3152 struct merge_options *opt,
3153 struct rename *ren)
3154 {
3155 const char *other_branch;
3156 other_branch = (ren->branch == opt->branch1 ?
3157 opt->branch2 : opt->branch1);
3158 if (is_add) {
3159 output(opt, clean ? 2 : 1, msg,
3160 ren->pair->one->path, ren->branch,
3161 other_branch, ren->pair->two->path);
3162 return;
3163 }
3164 output(opt, clean ? 2 : 1, msg,
3165 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3166 other_branch, ren->pair->two->path);
3167 }
3168 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3169 struct rename *ren)
3170 {
3171 const char *msg;
3172 int clean = 1, is_add;
3173
3174 if (!ren)
3175 return clean;
3176
3177 /* Return early if ren was not affected/created by a directory rename */
3178 if (!ren->dir_rename_original_dest)
3179 return clean;
3180
3181 /* Sanity checks */
3182 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3183 assert(ren->dir_rename_original_type == 'A' ||
3184 ren->dir_rename_original_type == 'R');
3185
3186 /* Check whether to treat directory renames as a conflict */
3187 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3188
3189 is_add = (ren->dir_rename_original_type == 'A');
3190 if (ren->dir_rename_original_type == 'A' && clean) {
3191 msg = _("Path updated: %s added in %s inside a "
3192 "directory that was renamed in %s; moving it to %s.");
3193 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3194 msg = _("CONFLICT (file location): %s added in %s "
3195 "inside a directory that was renamed in %s, "
3196 "suggesting it should perhaps be moved to %s.");
3197 } else if (ren->dir_rename_original_type == 'R' && clean) {
3198 msg = _("Path updated: %s renamed to %s in %s, inside a "
3199 "directory that was renamed in %s; moving it to %s.");
3200 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3201 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3202 "inside a directory that was renamed in %s, "
3203 "suggesting it should perhaps be moved to %s.");
3204 } else {
3205 BUG("Impossible dir_rename_original_type/clean combination");
3206 }
3207 dir_rename_warning(msg, is_add, clean, opt, ren);
3208
3209 return clean;
3210 }
3211
3212 /* Per entry merge function */
3213 static int process_entry(struct merge_options *opt,
3214 const char *path, struct stage_data *entry)
3215 {
3216 int clean_merge = 1;
3217 int normalize = opt->renormalize;
3218
3219 struct diff_filespec *o = &entry->stages[1];
3220 struct diff_filespec *a = &entry->stages[2];
3221 struct diff_filespec *b = &entry->stages[3];
3222 int o_valid = is_valid(o);
3223 int a_valid = is_valid(a);
3224 int b_valid = is_valid(b);
3225 o->path = a->path = b->path = (char*)path;
3226
3227 entry->processed = 1;
3228 if (entry->rename_conflict_info) {
3229 struct rename_conflict_info *ci = entry->rename_conflict_info;
3230 struct diff_filespec *temp;
3231 int path_clean;
3232
3233 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3234 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3235
3236 /*
3237 * For cases with a single rename, {o,a,b}->path have all been
3238 * set to the rename target path; we need to set two of these
3239 * back to the rename source.
3240 * For rename/rename conflicts, we'll manually fix paths below.
3241 */
3242 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3243 o->path = temp->path = ci->ren1->pair->one->path;
3244 if (ci->ren2) {
3245 assert(opt->branch1 == ci->ren1->branch);
3246 }
3247
3248 switch (ci->rename_type) {
3249 case RENAME_NORMAL:
3250 case RENAME_ONE_FILE_TO_ONE:
3251 clean_merge = handle_rename_normal(opt, path, o, a, b,
3252 ci);
3253 break;
3254 case RENAME_VIA_DIR:
3255 clean_merge = handle_rename_via_dir(opt, ci);
3256 break;
3257 case RENAME_ADD:
3258 /*
3259 * Probably unclean merge, but if the renamed file
3260 * merges cleanly and the result can then be
3261 * two-way merged cleanly with the added file, I
3262 * guess it's a clean merge?
3263 */
3264 clean_merge = handle_rename_add(opt, ci);
3265 break;
3266 case RENAME_DELETE:
3267 clean_merge = 0;
3268 if (handle_rename_delete(opt, ci))
3269 clean_merge = -1;
3270 break;
3271 case RENAME_ONE_FILE_TO_TWO:
3272 /*
3273 * Manually fix up paths; note:
3274 * ren[12]->pair->one->path are equal.
3275 */
3276 o->path = ci->ren1->pair->one->path;
3277 a->path = ci->ren1->pair->two->path;
3278 b->path = ci->ren2->pair->two->path;
3279
3280 clean_merge = 0;
3281 if (handle_rename_rename_1to2(opt, ci))
3282 clean_merge = -1;
3283 break;
3284 case RENAME_TWO_FILES_TO_ONE:
3285 /*
3286 * Manually fix up paths; note,
3287 * ren[12]->pair->two->path are actually equal.
3288 */
3289 o->path = NULL;
3290 a->path = ci->ren1->pair->two->path;
3291 b->path = ci->ren2->pair->two->path;
3292
3293 /*
3294 * Probably unclean merge, but if the two renamed
3295 * files merge cleanly and the two resulting files
3296 * can then be two-way merged cleanly, I guess it's
3297 * a clean merge?
3298 */
3299 clean_merge = handle_rename_rename_2to1(opt, ci);
3300 break;
3301 default:
3302 entry->processed = 0;
3303 break;
3304 }
3305 if (path_clean < clean_merge)
3306 clean_merge = path_clean;
3307 } else if (o_valid && (!a_valid || !b_valid)) {
3308 /* Case A: Deleted in one */
3309 if ((!a_valid && !b_valid) ||
3310 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3311 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3312 /* Deleted in both or deleted in one and
3313 * unchanged in the other */
3314 if (a_valid)
3315 output(opt, 2, _("Removing %s"), path);
3316 /* do not touch working file if it did not exist */
3317 remove_file(opt, 1, path, !a_valid);
3318 } else {
3319 /* Modify/delete; deleted side may have put a directory in the way */
3320 clean_merge = 0;
3321 if (handle_modify_delete(opt, path, o, a, b))
3322 clean_merge = -1;
3323 }
3324 } else if ((!o_valid && a_valid && !b_valid) ||
3325 (!o_valid && !a_valid && b_valid)) {
3326 /* Case B: Added in one. */
3327 /* [nothing|directory] -> ([nothing|directory], file) */
3328
3329 const char *add_branch;
3330 const char *other_branch;
3331 const char *conf;
3332 const struct diff_filespec *contents;
3333
3334 if (a_valid) {
3335 add_branch = opt->branch1;
3336 other_branch = opt->branch2;
3337 contents = a;
3338 conf = _("file/directory");
3339 } else {
3340 add_branch = opt->branch2;
3341 other_branch = opt->branch1;
3342 contents = b;
3343 conf = _("directory/file");
3344 }
3345 if (dir_in_way(opt->repo->index, path,
3346 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3347 0)) {
3348 char *new_path = unique_path(opt, path, add_branch);
3349 clean_merge = 0;
3350 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3351 "Adding %s as %s"),
3352 conf, path, other_branch, path, new_path);
3353 if (update_file(opt, 0, contents, new_path))
3354 clean_merge = -1;
3355 else if (opt->priv->call_depth)
3356 remove_file_from_index(opt->repo->index, path);
3357 free(new_path);
3358 } else {
3359 output(opt, 2, _("Adding %s"), path);
3360 /* do not overwrite file if already present */
3361 if (update_file_flags(opt, contents, path, 1, !a_valid))
3362 clean_merge = -1;
3363 }
3364 } else if (a_valid && b_valid) {
3365 if (!o_valid) {
3366 /* Case C: Added in both (check for same permissions) */
3367 output(opt, 1,
3368 _("CONFLICT (add/add): Merge conflict in %s"),
3369 path);
3370 clean_merge = handle_file_collision(opt,
3371 path, NULL, NULL,
3372 opt->branch1,
3373 opt->branch2,
3374 a, b);
3375 } else {
3376 /* case D: Modified in both, but differently. */
3377 struct merge_file_info mfi;
3378 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3379 clean_merge = handle_content_merge(&mfi, opt, path,
3380 is_dirty,
3381 o, a, b, NULL);
3382 }
3383 } else if (!o_valid && !a_valid && !b_valid) {
3384 /*
3385 * this entry was deleted altogether. a_mode == 0 means
3386 * we had that path and want to actively remove it.
3387 */
3388 remove_file(opt, 1, path, !a->mode);
3389 } else
3390 BUG("fatal merge failure, shouldn't happen.");
3391
3392 return clean_merge;
3393 }
3394
3395 static int merge_trees_internal(struct merge_options *opt,
3396 struct tree *head,
3397 struct tree *merge,
3398 struct tree *merge_base,
3399 struct tree **result)
3400 {
3401 struct index_state *istate = opt->repo->index;
3402 int code, clean;
3403
3404 if (opt->subtree_shift) {
3405 merge = shift_tree_object(opt->repo, head, merge,
3406 opt->subtree_shift);
3407 merge_base = shift_tree_object(opt->repo, head, merge_base,
3408 opt->subtree_shift);
3409 }
3410
3411 if (oid_eq(&merge_base->object.oid, &merge->object.oid)) {
3412 output(opt, 0, _("Already up to date!"));
3413 *result = head;
3414 return 1;
3415 }
3416
3417 code = unpack_trees_start(opt, merge_base, head, merge);
3418
3419 if (code != 0) {
3420 if (show(opt, 4) || opt->priv->call_depth)
3421 err(opt, _("merging of trees %s and %s failed"),
3422 oid_to_hex(&head->object.oid),
3423 oid_to_hex(&merge->object.oid));
3424 unpack_trees_finish(opt);
3425 return -1;
3426 }
3427
3428 if (unmerged_index(istate)) {
3429 struct string_list *entries;
3430 struct rename_info re_info;
3431 int i;
3432 /*
3433 * Only need the hashmap while processing entries, so
3434 * initialize it here and free it when we are done running
3435 * through the entries. Keeping it in the merge_options as
3436 * opposed to decaring a local hashmap is for convenience
3437 * so that we don't have to pass it to around.
3438 */
3439 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3440 NULL, 512);
3441 get_files_dirs(opt, head);
3442 get_files_dirs(opt, merge);
3443
3444 entries = get_unmerged(opt->repo->index);
3445 clean = detect_and_process_renames(opt, merge_base, head, merge,
3446 entries, &re_info);
3447 record_df_conflict_files(opt, entries);
3448 if (clean < 0)
3449 goto cleanup;
3450 for (i = entries->nr-1; 0 <= i; i--) {
3451 const char *path = entries->items[i].string;
3452 struct stage_data *e = entries->items[i].util;
3453 if (!e->processed) {
3454 int ret = process_entry(opt, path, e);
3455 if (!ret)
3456 clean = 0;
3457 else if (ret < 0) {
3458 clean = ret;
3459 goto cleanup;
3460 }
3461 }
3462 }
3463 for (i = 0; i < entries->nr; i++) {
3464 struct stage_data *e = entries->items[i].util;
3465 if (!e->processed)
3466 BUG("unprocessed path??? %s",
3467 entries->items[i].string);
3468 }
3469
3470 cleanup:
3471 final_cleanup_renames(&re_info);
3472
3473 string_list_clear(entries, 1);
3474 free(entries);
3475
3476 hashmap_free(&opt->priv->current_file_dir_set, 1);
3477
3478 if (clean < 0) {
3479 unpack_trees_finish(opt);
3480 return clean;
3481 }
3482 }
3483 else
3484 clean = 1;
3485
3486 unpack_trees_finish(opt);
3487
3488 if (opt->priv->call_depth &&
3489 !(*result = write_in_core_index_as_tree(opt->repo)))
3490 return -1;
3491
3492 return clean;
3493 }
3494
3495 static struct commit_list *reverse_commit_list(struct commit_list *list)
3496 {
3497 struct commit_list *next = NULL, *current, *backup;
3498 for (current = list; current; current = backup) {
3499 backup = current->next;
3500 current->next = next;
3501 next = current;
3502 }
3503 return next;
3504 }
3505
3506 /*
3507 * Merge the commits h1 and h2, return the resulting virtual
3508 * commit object and a flag indicating the cleanness of the merge.
3509 */
3510 static int merge_recursive_internal(struct merge_options *opt,
3511 struct commit *h1,
3512 struct commit *h2,
3513 struct commit_list *merge_bases,
3514 struct commit **result)
3515 {
3516 struct commit_list *iter;
3517 struct commit *merged_merge_bases;
3518 struct tree *result_tree;
3519 int clean;
3520 const char *ancestor_name;
3521 struct strbuf merge_base_abbrev = STRBUF_INIT;
3522
3523 if (show(opt, 4)) {
3524 output(opt, 4, _("Merging:"));
3525 output_commit_title(opt, h1);
3526 output_commit_title(opt, h2);
3527 }
3528
3529 if (!merge_bases) {
3530 merge_bases = get_merge_bases(h1, h2);
3531 merge_bases = reverse_commit_list(merge_bases);
3532 }
3533
3534 if (show(opt, 5)) {
3535 unsigned cnt = commit_list_count(merge_bases);
3536
3537 output(opt, 5, Q_("found %u common ancestor:",
3538 "found %u common ancestors:", cnt), cnt);
3539 for (iter = merge_bases; iter; iter = iter->next)
3540 output_commit_title(opt, iter->item);
3541 }
3542
3543 merged_merge_bases = pop_commit(&merge_bases);
3544 if (merged_merge_bases == NULL) {
3545 /* if there is no common ancestor, use an empty tree */
3546 struct tree *tree;
3547
3548 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3549 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3550 "ancestor");
3551 ancestor_name = "empty tree";
3552 } else if (merge_bases) {
3553 ancestor_name = "merged common ancestors";
3554 } else {
3555 strbuf_add_unique_abbrev(&merge_base_abbrev,
3556 &merged_merge_bases->object.oid,
3557 DEFAULT_ABBREV);
3558 ancestor_name = merge_base_abbrev.buf;
3559 }
3560
3561 for (iter = merge_bases; iter; iter = iter->next) {
3562 const char *saved_b1, *saved_b2;
3563 opt->priv->call_depth++;
3564 /*
3565 * When the merge fails, the result contains files
3566 * with conflict markers. The cleanness flag is
3567 * ignored (unless indicating an error), it was never
3568 * actually used, as result of merge_trees has always
3569 * overwritten it: the committed "conflicts" were
3570 * already resolved.
3571 */
3572 discard_index(opt->repo->index);
3573 saved_b1 = opt->branch1;
3574 saved_b2 = opt->branch2;
3575 opt->branch1 = "Temporary merge branch 1";
3576 opt->branch2 = "Temporary merge branch 2";
3577 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3578 NULL, &merged_merge_bases) < 0)
3579 return -1;
3580 opt->branch1 = saved_b1;
3581 opt->branch2 = saved_b2;
3582 opt->priv->call_depth--;
3583
3584 if (!merged_merge_bases)
3585 return err(opt, _("merge returned no commit"));
3586 }
3587
3588 discard_index(opt->repo->index);
3589 if (!opt->priv->call_depth)
3590 repo_read_index(opt->repo);
3591
3592 opt->ancestor = ancestor_name;
3593 clean = merge_trees_internal(opt,
3594 repo_get_commit_tree(opt->repo, h1),
3595 repo_get_commit_tree(opt->repo, h2),
3596 repo_get_commit_tree(opt->repo,
3597 merged_merge_bases),
3598 &result_tree);
3599 strbuf_release(&merge_base_abbrev);
3600 if (clean < 0) {
3601 flush_output(opt);
3602 return clean;
3603 }
3604
3605 if (opt->priv->call_depth) {
3606 *result = make_virtual_commit(opt->repo, result_tree,
3607 "merged tree");
3608 commit_list_insert(h1, &(*result)->parents);
3609 commit_list_insert(h2, &(*result)->parents->next);
3610 }
3611 return clean;
3612 }
3613
3614 static int merge_start(struct merge_options *opt, struct tree *head)
3615 {
3616 struct strbuf sb = STRBUF_INIT;
3617
3618 if (repo_index_has_changes(opt->repo, head, &sb)) {
3619 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3620 sb.buf);
3621 strbuf_release(&sb);
3622 return -1;
3623 }
3624
3625 opt->priv = xcalloc(1, sizeof(*opt->priv));
3626 string_list_init(&opt->priv->df_conflict_file_set, 1);
3627 return 0;
3628 }
3629
3630 static void merge_finalize(struct merge_options *opt)
3631 {
3632 flush_output(opt);
3633 if (!opt->priv->call_depth && opt->buffer_output < 2)
3634 strbuf_release(&opt->obuf);
3635 if (show(opt, 2))
3636 diff_warn_rename_limit("merge.renamelimit",
3637 opt->priv->needed_rename_limit, 0);
3638 FREE_AND_NULL(opt->priv);
3639 }
3640
3641 int merge_trees(struct merge_options *opt,
3642 struct tree *head,
3643 struct tree *merge,
3644 struct tree *merge_base)
3645 {
3646 int clean;
3647 struct tree *ignored;
3648
3649 assert(opt->ancestor != NULL);
3650
3651 if (merge_start(opt, head))
3652 return -1;
3653 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3654 merge_finalize(opt);
3655
3656 return clean;
3657 }
3658
3659 int merge_recursive(struct merge_options *opt,
3660 struct commit *h1,
3661 struct commit *h2,
3662 struct commit_list *merge_bases,
3663 struct commit **result)
3664 {
3665 int clean;
3666
3667 assert(opt->ancestor == NULL);
3668
3669 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3670 return -1;
3671 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3672 merge_finalize(opt);
3673
3674 return clean;
3675 }
3676
3677 static struct commit *get_ref(struct repository *repo,
3678 const struct object_id *oid,
3679 const char *name)
3680 {
3681 struct object *object;
3682
3683 object = deref_tag(repo, parse_object(repo, oid),
3684 name, strlen(name));
3685 if (!object)
3686 return NULL;
3687 if (object->type == OBJ_TREE)
3688 return make_virtual_commit(repo, (struct tree*)object, name);
3689 if (object->type != OBJ_COMMIT)
3690 return NULL;
3691 if (parse_commit((struct commit *)object))
3692 return NULL;
3693 return (struct commit *)object;
3694 }
3695
3696 int merge_recursive_generic(struct merge_options *opt,
3697 const struct object_id *head,
3698 const struct object_id *merge,
3699 int num_merge_bases,
3700 const struct object_id **merge_bases,
3701 struct commit **result)
3702 {
3703 int clean;
3704 struct lock_file lock = LOCK_INIT;
3705 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3706 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3707 struct commit_list *ca = NULL;
3708
3709 if (merge_bases) {
3710 int i;
3711 for (i = 0; i < num_merge_bases; ++i) {
3712 struct commit *base;
3713 if (!(base = get_ref(opt->repo, merge_bases[i],
3714 oid_to_hex(merge_bases[i]))))
3715 return err(opt, _("Could not parse object '%s'"),
3716 oid_to_hex(merge_bases[i]));
3717 commit_list_insert(base, &ca);
3718 }
3719 }
3720
3721 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3722 clean = merge_recursive(opt, head_commit, next_commit, ca,
3723 result);
3724 if (clean < 0) {
3725 rollback_lock_file(&lock);
3726 return clean;
3727 }
3728
3729 if (write_locked_index(opt->repo->index, &lock,
3730 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3731 return err(opt, _("Unable to write index."));
3732
3733 return clean ? 0 : 1;
3734 }
3735
3736 static void merge_recursive_config(struct merge_options *opt)
3737 {
3738 char *value = NULL;
3739 git_config_get_int("merge.verbosity", &opt->verbosity);
3740 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3741 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3742 if (!git_config_get_string("diff.renames", &value)) {
3743 opt->detect_renames = git_config_rename("diff.renames", value);
3744 free(value);
3745 }
3746 if (!git_config_get_string("merge.renames", &value)) {
3747 opt->detect_renames = git_config_rename("merge.renames", value);
3748 free(value);
3749 }
3750 if (!git_config_get_string("merge.directoryrenames", &value)) {
3751 int boolval = git_parse_maybe_bool(value);
3752 if (0 <= boolval) {
3753 opt->detect_directory_renames = boolval ?
3754 MERGE_DIRECTORY_RENAMES_TRUE :
3755 MERGE_DIRECTORY_RENAMES_NONE;
3756 } else if (!strcasecmp(value, "conflict")) {
3757 opt->detect_directory_renames =
3758 MERGE_DIRECTORY_RENAMES_CONFLICT;
3759 } /* avoid erroring on values from future versions of git */
3760 free(value);
3761 }
3762 git_config(git_xmerge_config, NULL);
3763 }
3764
3765 void init_merge_options(struct merge_options *opt,
3766 struct repository *repo)
3767 {
3768 const char *merge_verbosity;
3769 memset(opt, 0, sizeof(struct merge_options));
3770
3771 opt->repo = repo;
3772
3773 opt->detect_renames = -1;
3774 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3775 opt->rename_limit = -1;
3776
3777 opt->verbosity = 2;
3778 opt->buffer_output = 1;
3779 strbuf_init(&opt->obuf, 0);
3780
3781 opt->renormalize = 0;
3782
3783 merge_recursive_config(opt);
3784 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3785 if (merge_verbosity)
3786 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3787 if (opt->verbosity >= 5)
3788 opt->buffer_output = 0;
3789 }
3790
3791 int parse_merge_opt(struct merge_options *opt, const char *s)
3792 {
3793 const char *arg;
3794
3795 if (!s || !*s)
3796 return -1;
3797 if (!strcmp(s, "ours"))
3798 opt->recursive_variant = MERGE_VARIANT_OURS;
3799 else if (!strcmp(s, "theirs"))
3800 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3801 else if (!strcmp(s, "subtree"))
3802 opt->subtree_shift = "";
3803 else if (skip_prefix(s, "subtree=", &arg))
3804 opt->subtree_shift = arg;
3805 else if (!strcmp(s, "patience"))
3806 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3807 else if (!strcmp(s, "histogram"))
3808 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3809 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3810 long value = parse_algorithm_value(arg);
3811 if (value < 0)
3812 return -1;
3813 /* clear out previous settings */
3814 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3815 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3816 opt->xdl_opts |= value;
3817 }
3818 else if (!strcmp(s, "ignore-space-change"))
3819 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3820 else if (!strcmp(s, "ignore-all-space"))
3821 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3822 else if (!strcmp(s, "ignore-space-at-eol"))
3823 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3824 else if (!strcmp(s, "ignore-cr-at-eol"))
3825 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3826 else if (!strcmp(s, "renormalize"))
3827 opt->renormalize = 1;
3828 else if (!strcmp(s, "no-renormalize"))
3829 opt->renormalize = 0;
3830 else if (!strcmp(s, "no-renames"))
3831 opt->detect_renames = 0;
3832 else if (!strcmp(s, "find-renames")) {
3833 opt->detect_renames = 1;
3834 opt->rename_score = 0;
3835 }
3836 else if (skip_prefix(s, "find-renames=", &arg) ||
3837 skip_prefix(s, "rename-threshold=", &arg)) {
3838 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3839 return -1;
3840 opt->detect_renames = 1;
3841 }
3842 /*
3843 * Please update $__git_merge_strategy_options in
3844 * git-completion.bash when you add new options
3845 */
3846 else
3847 return -1;
3848 return 0;
3849 }