]> git.ipfire.org Git - thirdparty/git.git/blob - merge-recursive.c
builtin/for-each-repo: remove unnecessary argv copy to plug leak
[thirdparty/git.git] / merge-recursive.c
1 /*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6 #include "cache.h"
7 #include "merge-recursive.h"
8
9 #include "advice.h"
10 #include "alloc.h"
11 #include "attr.h"
12 #include "blob.h"
13 #include "builtin.h"
14 #include "cache-tree.h"
15 #include "commit.h"
16 #include "commit-reach.h"
17 #include "config.h"
18 #include "diff.h"
19 #include "diffcore.h"
20 #include "dir.h"
21 #include "ll-merge.h"
22 #include "lockfile.h"
23 #include "object-store.h"
24 #include "repository.h"
25 #include "revision.h"
26 #include "string-list.h"
27 #include "submodule.h"
28 #include "tag.h"
29 #include "tree-walk.h"
30 #include "unpack-trees.h"
31 #include "xdiff-interface.h"
32
33 struct merge_options_internal {
34 int call_depth;
35 int needed_rename_limit;
36 struct hashmap current_file_dir_set;
37 struct string_list df_conflict_file_set;
38 struct unpack_trees_options unpack_opts;
39 struct index_state orig_index;
40 };
41
42 struct path_hashmap_entry {
43 struct hashmap_entry e;
44 char path[FLEX_ARRAY];
45 };
46
47 static int path_hashmap_cmp(const void *cmp_data,
48 const struct hashmap_entry *eptr,
49 const struct hashmap_entry *entry_or_key,
50 const void *keydata)
51 {
52 const struct path_hashmap_entry *a, *b;
53 const char *key = keydata;
54
55 a = container_of(eptr, const struct path_hashmap_entry, e);
56 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
57
58 if (ignore_case)
59 return strcasecmp(a->path, key ? key : b->path);
60 else
61 return strcmp(a->path, key ? key : b->path);
62 }
63
64 static unsigned int path_hash(const char *path)
65 {
66 return ignore_case ? strihash(path) : strhash(path);
67 }
68
69 /*
70 * For dir_rename_entry, directory names are stored as a full path from the
71 * toplevel of the repository and do not include a trailing '/'. Also:
72 *
73 * dir: original name of directory being renamed
74 * non_unique_new_dir: if true, could not determine new_dir
75 * new_dir: final name of directory being renamed
76 * possible_new_dirs: temporary used to help determine new_dir; see comments
77 * in get_directory_renames() for details
78 */
79 struct dir_rename_entry {
80 struct hashmap_entry ent;
81 char *dir;
82 unsigned non_unique_new_dir:1;
83 struct strbuf new_dir;
84 struct string_list possible_new_dirs;
85 };
86
87 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
88 char *dir)
89 {
90 struct dir_rename_entry key;
91
92 if (dir == NULL)
93 return NULL;
94 hashmap_entry_init(&key.ent, strhash(dir));
95 key.dir = dir;
96 return hashmap_get_entry(hashmap, &key, ent, NULL);
97 }
98
99 static int dir_rename_cmp(const void *unused_cmp_data,
100 const struct hashmap_entry *eptr,
101 const struct hashmap_entry *entry_or_key,
102 const void *unused_keydata)
103 {
104 const struct dir_rename_entry *e1, *e2;
105
106 e1 = container_of(eptr, const struct dir_rename_entry, ent);
107 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
108
109 return strcmp(e1->dir, e2->dir);
110 }
111
112 static void dir_rename_init(struct hashmap *map)
113 {
114 hashmap_init(map, dir_rename_cmp, NULL, 0);
115 }
116
117 static void dir_rename_entry_init(struct dir_rename_entry *entry,
118 char *directory)
119 {
120 hashmap_entry_init(&entry->ent, strhash(directory));
121 entry->dir = directory;
122 entry->non_unique_new_dir = 0;
123 strbuf_init(&entry->new_dir, 0);
124 string_list_init_nodup(&entry->possible_new_dirs);
125 }
126
127 struct collision_entry {
128 struct hashmap_entry ent;
129 char *target_file;
130 struct string_list source_files;
131 unsigned reported_already:1;
132 };
133
134 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
135 char *target_file)
136 {
137 struct collision_entry key;
138
139 hashmap_entry_init(&key.ent, strhash(target_file));
140 key.target_file = target_file;
141 return hashmap_get_entry(hashmap, &key, ent, NULL);
142 }
143
144 static int collision_cmp(const void *unused_cmp_data,
145 const struct hashmap_entry *eptr,
146 const struct hashmap_entry *entry_or_key,
147 const void *unused_keydata)
148 {
149 const struct collision_entry *e1, *e2;
150
151 e1 = container_of(eptr, const struct collision_entry, ent);
152 e2 = container_of(entry_or_key, const struct collision_entry, ent);
153
154 return strcmp(e1->target_file, e2->target_file);
155 }
156
157 static void collision_init(struct hashmap *map)
158 {
159 hashmap_init(map, collision_cmp, NULL, 0);
160 }
161
162 static void flush_output(struct merge_options *opt)
163 {
164 if (opt->buffer_output < 2 && opt->obuf.len) {
165 fputs(opt->obuf.buf, stdout);
166 strbuf_reset(&opt->obuf);
167 }
168 }
169
170 static int err(struct merge_options *opt, const char *err, ...)
171 {
172 va_list params;
173
174 if (opt->buffer_output < 2)
175 flush_output(opt);
176 else {
177 strbuf_complete(&opt->obuf, '\n');
178 strbuf_addstr(&opt->obuf, "error: ");
179 }
180 va_start(params, err);
181 strbuf_vaddf(&opt->obuf, err, params);
182 va_end(params);
183 if (opt->buffer_output > 1)
184 strbuf_addch(&opt->obuf, '\n');
185 else {
186 error("%s", opt->obuf.buf);
187 strbuf_reset(&opt->obuf);
188 }
189
190 return -1;
191 }
192
193 static struct tree *shift_tree_object(struct repository *repo,
194 struct tree *one, struct tree *two,
195 const char *subtree_shift)
196 {
197 struct object_id shifted;
198
199 if (!*subtree_shift) {
200 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
201 } else {
202 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
203 subtree_shift);
204 }
205 if (oideq(&two->object.oid, &shifted))
206 return two;
207 return lookup_tree(repo, &shifted);
208 }
209
210 static inline void set_commit_tree(struct commit *c, struct tree *t)
211 {
212 c->maybe_tree = t;
213 }
214
215 static struct commit *make_virtual_commit(struct repository *repo,
216 struct tree *tree,
217 const char *comment)
218 {
219 struct commit *commit = alloc_commit_node(repo);
220
221 set_merge_remote_desc(commit, comment, (struct object *)commit);
222 set_commit_tree(commit, tree);
223 commit->object.parsed = 1;
224 return commit;
225 }
226
227 enum rename_type {
228 RENAME_NORMAL = 0,
229 RENAME_VIA_DIR,
230 RENAME_ADD,
231 RENAME_DELETE,
232 RENAME_ONE_FILE_TO_ONE,
233 RENAME_ONE_FILE_TO_TWO,
234 RENAME_TWO_FILES_TO_ONE
235 };
236
237 /*
238 * Since we want to write the index eventually, we cannot reuse the index
239 * for these (temporary) data.
240 */
241 struct stage_data {
242 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
243 struct rename_conflict_info *rename_conflict_info;
244 unsigned processed:1;
245 };
246
247 struct rename {
248 unsigned processed:1;
249 struct diff_filepair *pair;
250 const char *branch; /* branch that the rename occurred on */
251 /*
252 * If directory rename detection affected this rename, what was its
253 * original type ('A' or 'R') and it's original destination before
254 * the directory rename (otherwise, '\0' and NULL for these two vars).
255 */
256 char dir_rename_original_type;
257 char *dir_rename_original_dest;
258 /*
259 * Purpose of src_entry and dst_entry:
260 *
261 * If 'before' is renamed to 'after' then src_entry will contain
262 * the versions of 'before' from the merge_base, HEAD, and MERGE in
263 * stages 1, 2, and 3; dst_entry will contain the respective
264 * versions of 'after' in corresponding locations. Thus, we have a
265 * total of six modes and oids, though some will be null. (Stage 0
266 * is ignored; we're interested in handling conflicts.)
267 *
268 * Since we don't turn on break-rewrites by default, neither
269 * src_entry nor dst_entry can have all three of their stages have
270 * non-null oids, meaning at most four of the six will be non-null.
271 * Also, since this is a rename, both src_entry and dst_entry will
272 * have at least one non-null oid, meaning at least two will be
273 * non-null. Of the six oids, a typical rename will have three be
274 * non-null. Only two implies a rename/delete, and four implies a
275 * rename/add.
276 */
277 struct stage_data *src_entry;
278 struct stage_data *dst_entry;
279 };
280
281 struct rename_conflict_info {
282 enum rename_type rename_type;
283 struct rename *ren1;
284 struct rename *ren2;
285 };
286
287 static inline void setup_rename_conflict_info(enum rename_type rename_type,
288 struct merge_options *opt,
289 struct rename *ren1,
290 struct rename *ren2)
291 {
292 struct rename_conflict_info *ci;
293
294 /*
295 * When we have two renames involved, it's easiest to get the
296 * correct things into stage 2 and 3, and to make sure that the
297 * content merge puts HEAD before the other branch if we just
298 * ensure that branch1 == opt->branch1. So, simply flip arguments
299 * around if we don't have that.
300 */
301 if (ren2 && ren1->branch != opt->branch1) {
302 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
303 return;
304 }
305
306 CALLOC_ARRAY(ci, 1);
307 ci->rename_type = rename_type;
308 ci->ren1 = ren1;
309 ci->ren2 = ren2;
310
311 ci->ren1->dst_entry->processed = 0;
312 ci->ren1->dst_entry->rename_conflict_info = ci;
313 if (ren2) {
314 ci->ren2->dst_entry->rename_conflict_info = ci;
315 }
316 }
317
318 static int show(struct merge_options *opt, int v)
319 {
320 return (!opt->priv->call_depth && opt->verbosity >= v) ||
321 opt->verbosity >= 5;
322 }
323
324 __attribute__((format (printf, 3, 4)))
325 static void output(struct merge_options *opt, int v, const char *fmt, ...)
326 {
327 va_list ap;
328
329 if (!show(opt, v))
330 return;
331
332 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
333
334 va_start(ap, fmt);
335 strbuf_vaddf(&opt->obuf, fmt, ap);
336 va_end(ap);
337
338 strbuf_addch(&opt->obuf, '\n');
339 if (!opt->buffer_output)
340 flush_output(opt);
341 }
342
343 static void output_commit_title(struct merge_options *opt, struct commit *commit)
344 {
345 struct merge_remote_desc *desc;
346
347 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
348 desc = merge_remote_util(commit);
349 if (desc)
350 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
351 else {
352 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
353 DEFAULT_ABBREV);
354 strbuf_addch(&opt->obuf, ' ');
355 if (parse_commit(commit) != 0)
356 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
357 else {
358 const char *title;
359 const char *msg = get_commit_buffer(commit, NULL);
360 int len = find_commit_subject(msg, &title);
361 if (len)
362 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
363 unuse_commit_buffer(commit, msg);
364 }
365 }
366 flush_output(opt);
367 }
368
369 static int add_cacheinfo(struct merge_options *opt,
370 const struct diff_filespec *blob,
371 const char *path, int stage, int refresh, int options)
372 {
373 struct index_state *istate = opt->repo->index;
374 struct cache_entry *ce;
375 int ret;
376
377 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
378 if (!ce)
379 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
380
381 ret = add_index_entry(istate, ce, options);
382 if (refresh) {
383 struct cache_entry *nce;
384
385 nce = refresh_cache_entry(istate, ce,
386 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
387 if (!nce)
388 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
389 if (nce != ce)
390 ret = add_index_entry(istate, nce, options);
391 }
392 return ret;
393 }
394
395 static inline int merge_detect_rename(struct merge_options *opt)
396 {
397 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
398 }
399
400 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
401 {
402 parse_tree(tree);
403 init_tree_desc(desc, tree->buffer, tree->size);
404 }
405
406 static int unpack_trees_start(struct merge_options *opt,
407 struct tree *common,
408 struct tree *head,
409 struct tree *merge)
410 {
411 int rc;
412 struct tree_desc t[3];
413 struct index_state tmp_index = { NULL };
414
415 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
416 if (opt->priv->call_depth)
417 opt->priv->unpack_opts.index_only = 1;
418 else
419 opt->priv->unpack_opts.update = 1;
420 opt->priv->unpack_opts.merge = 1;
421 opt->priv->unpack_opts.head_idx = 2;
422 opt->priv->unpack_opts.fn = threeway_merge;
423 opt->priv->unpack_opts.src_index = opt->repo->index;
424 opt->priv->unpack_opts.dst_index = &tmp_index;
425 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
426 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
427
428 init_tree_desc_from_tree(t+0, common);
429 init_tree_desc_from_tree(t+1, head);
430 init_tree_desc_from_tree(t+2, merge);
431
432 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
433 cache_tree_free(&opt->repo->index->cache_tree);
434
435 /*
436 * Update opt->repo->index to match the new results, AFTER saving a
437 * copy in opt->priv->orig_index. Update src_index to point to the
438 * saved copy. (verify_uptodate() checks src_index, and the original
439 * index is the one that had the necessary modification timestamps.)
440 */
441 opt->priv->orig_index = *opt->repo->index;
442 *opt->repo->index = tmp_index;
443 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
444
445 return rc;
446 }
447
448 static void unpack_trees_finish(struct merge_options *opt)
449 {
450 discard_index(&opt->priv->orig_index);
451 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
452 }
453
454 static int save_files_dirs(const struct object_id *oid,
455 struct strbuf *base, const char *path,
456 unsigned int mode, void *context)
457 {
458 struct path_hashmap_entry *entry;
459 int baselen = base->len;
460 struct merge_options *opt = context;
461
462 strbuf_addstr(base, path);
463
464 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
465 hashmap_entry_init(&entry->e, path_hash(entry->path));
466 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
467
468 strbuf_setlen(base, baselen);
469 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
470 }
471
472 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
473 {
474 struct pathspec match_all;
475 memset(&match_all, 0, sizeof(match_all));
476 read_tree(opt->repo, tree,
477 &match_all, save_files_dirs, opt);
478 }
479
480 static int get_tree_entry_if_blob(struct repository *r,
481 const struct object_id *tree,
482 const char *path,
483 struct diff_filespec *dfs)
484 {
485 int ret;
486
487 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
488 if (S_ISDIR(dfs->mode)) {
489 oidcpy(&dfs->oid, null_oid());
490 dfs->mode = 0;
491 }
492 return ret;
493 }
494
495 /*
496 * Returns an index_entry instance which doesn't have to correspond to
497 * a real cache entry in Git's index.
498 */
499 static struct stage_data *insert_stage_data(struct repository *r,
500 const char *path,
501 struct tree *o, struct tree *a, struct tree *b,
502 struct string_list *entries)
503 {
504 struct string_list_item *item;
505 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
506 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
507 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
508 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
509 item = string_list_insert(entries, path);
510 item->util = e;
511 return e;
512 }
513
514 /*
515 * Create a dictionary mapping file names to stage_data objects. The
516 * dictionary contains one entry for every path with a non-zero stage entry.
517 */
518 static struct string_list *get_unmerged(struct index_state *istate)
519 {
520 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
521 int i;
522
523 unmerged->strdup_strings = 1;
524
525 /* TODO: audit for interaction with sparse-index. */
526 ensure_full_index(istate);
527 for (i = 0; i < istate->cache_nr; i++) {
528 struct string_list_item *item;
529 struct stage_data *e;
530 const struct cache_entry *ce = istate->cache[i];
531 if (!ce_stage(ce))
532 continue;
533
534 item = string_list_lookup(unmerged, ce->name);
535 if (!item) {
536 item = string_list_insert(unmerged, ce->name);
537 item->util = xcalloc(1, sizeof(struct stage_data));
538 }
539 e = item->util;
540 e->stages[ce_stage(ce)].mode = ce->ce_mode;
541 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
542 }
543
544 return unmerged;
545 }
546
547 static int string_list_df_name_compare(const char *one, const char *two)
548 {
549 int onelen = strlen(one);
550 int twolen = strlen(two);
551 /*
552 * Here we only care that entries for D/F conflicts are
553 * adjacent, in particular with the file of the D/F conflict
554 * appearing before files below the corresponding directory.
555 * The order of the rest of the list is irrelevant for us.
556 *
557 * To achieve this, we sort with df_name_compare and provide
558 * the mode S_IFDIR so that D/F conflicts will sort correctly.
559 * We use the mode S_IFDIR for everything else for simplicity,
560 * since in other cases any changes in their order due to
561 * sorting cause no problems for us.
562 */
563 int cmp = df_name_compare(one, onelen, S_IFDIR,
564 two, twolen, S_IFDIR);
565 /*
566 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
567 * that 'foo' comes before 'foo/bar'.
568 */
569 if (cmp)
570 return cmp;
571 return onelen - twolen;
572 }
573
574 static void record_df_conflict_files(struct merge_options *opt,
575 struct string_list *entries)
576 {
577 /* If there is a D/F conflict and the file for such a conflict
578 * currently exists in the working tree, we want to allow it to be
579 * removed to make room for the corresponding directory if needed.
580 * The files underneath the directories of such D/F conflicts will
581 * be processed before the corresponding file involved in the D/F
582 * conflict. If the D/F directory ends up being removed by the
583 * merge, then we won't have to touch the D/F file. If the D/F
584 * directory needs to be written to the working copy, then the D/F
585 * file will simply be removed (in make_room_for_path()) to make
586 * room for the necessary paths. Note that if both the directory
587 * and the file need to be present, then the D/F file will be
588 * reinstated with a new unique name at the time it is processed.
589 */
590 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
591 const char *last_file = NULL;
592 int last_len = 0;
593 int i;
594
595 /*
596 * If we're merging merge-bases, we don't want to bother with
597 * any working directory changes.
598 */
599 if (opt->priv->call_depth)
600 return;
601
602 /* Ensure D/F conflicts are adjacent in the entries list. */
603 for (i = 0; i < entries->nr; i++) {
604 struct string_list_item *next = &entries->items[i];
605 string_list_append(&df_sorted_entries, next->string)->util =
606 next->util;
607 }
608 df_sorted_entries.cmp = string_list_df_name_compare;
609 string_list_sort(&df_sorted_entries);
610
611 string_list_clear(&opt->priv->df_conflict_file_set, 1);
612 for (i = 0; i < df_sorted_entries.nr; i++) {
613 const char *path = df_sorted_entries.items[i].string;
614 int len = strlen(path);
615 struct stage_data *e = df_sorted_entries.items[i].util;
616
617 /*
618 * Check if last_file & path correspond to a D/F conflict;
619 * i.e. whether path is last_file+'/'+<something>.
620 * If so, record that it's okay to remove last_file to make
621 * room for path and friends if needed.
622 */
623 if (last_file &&
624 len > last_len &&
625 memcmp(path, last_file, last_len) == 0 &&
626 path[last_len] == '/') {
627 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
628 }
629
630 /*
631 * Determine whether path could exist as a file in the
632 * working directory as a possible D/F conflict. This
633 * will only occur when it exists in stage 2 as a
634 * file.
635 */
636 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
637 last_file = path;
638 last_len = len;
639 } else {
640 last_file = NULL;
641 }
642 }
643 string_list_clear(&df_sorted_entries, 0);
644 }
645
646 static int update_stages(struct merge_options *opt, const char *path,
647 const struct diff_filespec *o,
648 const struct diff_filespec *a,
649 const struct diff_filespec *b)
650 {
651
652 /*
653 * NOTE: It is usually a bad idea to call update_stages on a path
654 * before calling update_file on that same path, since it can
655 * sometimes lead to spurious "refusing to lose untracked file..."
656 * messages from update_file (via make_room_for path via
657 * would_lose_untracked). Instead, reverse the order of the calls
658 * (executing update_file first and then update_stages).
659 */
660 int clear = 1;
661 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
662 if (clear)
663 if (remove_file_from_index(opt->repo->index, path))
664 return -1;
665 if (o)
666 if (add_cacheinfo(opt, o, path, 1, 0, options))
667 return -1;
668 if (a)
669 if (add_cacheinfo(opt, a, path, 2, 0, options))
670 return -1;
671 if (b)
672 if (add_cacheinfo(opt, b, path, 3, 0, options))
673 return -1;
674 return 0;
675 }
676
677 static void update_entry(struct stage_data *entry,
678 struct diff_filespec *o,
679 struct diff_filespec *a,
680 struct diff_filespec *b)
681 {
682 entry->processed = 0;
683 entry->stages[1].mode = o->mode;
684 entry->stages[2].mode = a->mode;
685 entry->stages[3].mode = b->mode;
686 oidcpy(&entry->stages[1].oid, &o->oid);
687 oidcpy(&entry->stages[2].oid, &a->oid);
688 oidcpy(&entry->stages[3].oid, &b->oid);
689 }
690
691 static int remove_file(struct merge_options *opt, int clean,
692 const char *path, int no_wd)
693 {
694 int update_cache = opt->priv->call_depth || clean;
695 int update_working_directory = !opt->priv->call_depth && !no_wd;
696
697 if (update_cache) {
698 if (remove_file_from_index(opt->repo->index, path))
699 return -1;
700 }
701 if (update_working_directory) {
702 if (ignore_case) {
703 struct cache_entry *ce;
704 ce = index_file_exists(opt->repo->index, path, strlen(path),
705 ignore_case);
706 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
707 return 0;
708 }
709 if (remove_path(path))
710 return -1;
711 }
712 return 0;
713 }
714
715 /* add a string to a strbuf, but converting "/" to "_" */
716 static void add_flattened_path(struct strbuf *out, const char *s)
717 {
718 size_t i = out->len;
719 strbuf_addstr(out, s);
720 for (; i < out->len; i++)
721 if (out->buf[i] == '/')
722 out->buf[i] = '_';
723 }
724
725 static char *unique_path(struct merge_options *opt,
726 const char *path,
727 const char *branch)
728 {
729 struct path_hashmap_entry *entry;
730 struct strbuf newpath = STRBUF_INIT;
731 int suffix = 0;
732 size_t base_len;
733
734 strbuf_addf(&newpath, "%s~", path);
735 add_flattened_path(&newpath, branch);
736
737 base_len = newpath.len;
738 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
739 path_hash(newpath.buf), newpath.buf) ||
740 (!opt->priv->call_depth && file_exists(newpath.buf))) {
741 strbuf_setlen(&newpath, base_len);
742 strbuf_addf(&newpath, "_%d", suffix++);
743 }
744
745 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
746 hashmap_entry_init(&entry->e, path_hash(entry->path));
747 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
748 return strbuf_detach(&newpath, NULL);
749 }
750
751 /**
752 * Check whether a directory in the index is in the way of an incoming
753 * file. Return 1 if so. If check_working_copy is non-zero, also
754 * check the working directory. If empty_ok is non-zero, also return
755 * 0 in the case where the working-tree dir exists but is empty.
756 */
757 static int dir_in_way(struct index_state *istate, const char *path,
758 int check_working_copy, int empty_ok)
759 {
760 int pos;
761 struct strbuf dirpath = STRBUF_INIT;
762 struct stat st;
763
764 strbuf_addstr(&dirpath, path);
765 strbuf_addch(&dirpath, '/');
766
767 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
768
769 if (pos < 0)
770 pos = -1 - pos;
771 if (pos < istate->cache_nr &&
772 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
773 strbuf_release(&dirpath);
774 return 1;
775 }
776
777 strbuf_release(&dirpath);
778 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
779 !(empty_ok && is_empty_dir(path)) &&
780 !has_symlink_leading_path(path, strlen(path));
781 }
782
783 /*
784 * Returns whether path was tracked in the index before the merge started,
785 * and its oid and mode match the specified values
786 */
787 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
788 const struct diff_filespec *blob)
789 {
790 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
791 struct cache_entry *ce;
792
793 if (0 > pos)
794 /* we were not tracking this path before the merge */
795 return 0;
796
797 /* See if the file we were tracking before matches */
798 ce = opt->priv->orig_index.cache[pos];
799 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
800 }
801
802 /*
803 * Returns whether path was tracked in the index before the merge started
804 */
805 static int was_tracked(struct merge_options *opt, const char *path)
806 {
807 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
808
809 if (0 <= pos)
810 /* we were tracking this path before the merge */
811 return 1;
812
813 return 0;
814 }
815
816 static int would_lose_untracked(struct merge_options *opt, const char *path)
817 {
818 struct index_state *istate = opt->repo->index;
819
820 /*
821 * This may look like it can be simplified to:
822 * return !was_tracked(opt, path) && file_exists(path)
823 * but it can't. This function needs to know whether path was in
824 * the working tree due to EITHER having been tracked in the index
825 * before the merge OR having been put into the working copy and
826 * index by unpack_trees(). Due to that either-or requirement, we
827 * check the current index instead of the original one.
828 *
829 * Note that we do not need to worry about merge-recursive itself
830 * updating the index after unpack_trees() and before calling this
831 * function, because we strictly require all code paths in
832 * merge-recursive to update the working tree first and the index
833 * second. Doing otherwise would break
834 * update_file()/would_lose_untracked(); see every comment in this
835 * file which mentions "update_stages".
836 */
837 int pos = index_name_pos(istate, path, strlen(path));
838
839 if (pos < 0)
840 pos = -1 - pos;
841 while (pos < istate->cache_nr &&
842 !strcmp(path, istate->cache[pos]->name)) {
843 /*
844 * If stage #0, it is definitely tracked.
845 * If it has stage #2 then it was tracked
846 * before this merge started. All other
847 * cases the path was not tracked.
848 */
849 switch (ce_stage(istate->cache[pos])) {
850 case 0:
851 case 2:
852 return 0;
853 }
854 pos++;
855 }
856 return file_exists(path);
857 }
858
859 static int was_dirty(struct merge_options *opt, const char *path)
860 {
861 struct cache_entry *ce;
862 int dirty = 1;
863
864 if (opt->priv->call_depth || !was_tracked(opt, path))
865 return !dirty;
866
867 ce = index_file_exists(opt->priv->unpack_opts.src_index,
868 path, strlen(path), ignore_case);
869 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
870 return dirty;
871 }
872
873 static int make_room_for_path(struct merge_options *opt, const char *path)
874 {
875 int status, i;
876 const char *msg = _("failed to create path '%s'%s");
877
878 /* Unlink any D/F conflict files that are in the way */
879 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
880 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
881 size_t pathlen = strlen(path);
882 size_t df_pathlen = strlen(df_path);
883 if (df_pathlen < pathlen &&
884 path[df_pathlen] == '/' &&
885 strncmp(path, df_path, df_pathlen) == 0) {
886 output(opt, 3,
887 _("Removing %s to make room for subdirectory\n"),
888 df_path);
889 unlink(df_path);
890 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
891 i, 0);
892 break;
893 }
894 }
895
896 /* Make sure leading directories are created */
897 status = safe_create_leading_directories_const(path);
898 if (status) {
899 if (status == SCLD_EXISTS)
900 /* something else exists */
901 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
902 return err(opt, msg, path, "");
903 }
904
905 /*
906 * Do not unlink a file in the work tree if we are not
907 * tracking it.
908 */
909 if (would_lose_untracked(opt, path))
910 return err(opt, _("refusing to lose untracked file at '%s'"),
911 path);
912
913 /* Successful unlink is good.. */
914 if (!unlink(path))
915 return 0;
916 /* .. and so is no existing file */
917 if (errno == ENOENT)
918 return 0;
919 /* .. but not some other error (who really cares what?) */
920 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
921 }
922
923 static int update_file_flags(struct merge_options *opt,
924 const struct diff_filespec *contents,
925 const char *path,
926 int update_cache,
927 int update_wd)
928 {
929 int ret = 0;
930
931 if (opt->priv->call_depth)
932 update_wd = 0;
933
934 if (update_wd) {
935 enum object_type type;
936 void *buf;
937 unsigned long size;
938
939 if (S_ISGITLINK(contents->mode)) {
940 /*
941 * We may later decide to recursively descend into
942 * the submodule directory and update its index
943 * and/or work tree, but we do not do that now.
944 */
945 update_wd = 0;
946 goto update_index;
947 }
948
949 buf = read_object_file(&contents->oid, &type, &size);
950 if (!buf) {
951 ret = err(opt, _("cannot read object %s '%s'"),
952 oid_to_hex(&contents->oid), path);
953 goto free_buf;
954 }
955 if (type != OBJ_BLOB) {
956 ret = err(opt, _("blob expected for %s '%s'"),
957 oid_to_hex(&contents->oid), path);
958 goto free_buf;
959 }
960 if (S_ISREG(contents->mode)) {
961 struct strbuf strbuf = STRBUF_INIT;
962 if (convert_to_working_tree(opt->repo->index,
963 path, buf, size, &strbuf, NULL)) {
964 free(buf);
965 size = strbuf.len;
966 buf = strbuf_detach(&strbuf, NULL);
967 }
968 }
969
970 if (make_room_for_path(opt, path) < 0) {
971 update_wd = 0;
972 goto free_buf;
973 }
974 if (S_ISREG(contents->mode) ||
975 (!has_symlinks && S_ISLNK(contents->mode))) {
976 int fd;
977 int mode = (contents->mode & 0100 ? 0777 : 0666);
978
979 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
980 if (fd < 0) {
981 ret = err(opt, _("failed to open '%s': %s"),
982 path, strerror(errno));
983 goto free_buf;
984 }
985 write_in_full(fd, buf, size);
986 close(fd);
987 } else if (S_ISLNK(contents->mode)) {
988 char *lnk = xmemdupz(buf, size);
989 safe_create_leading_directories_const(path);
990 unlink(path);
991 if (symlink(lnk, path))
992 ret = err(opt, _("failed to symlink '%s': %s"),
993 path, strerror(errno));
994 free(lnk);
995 } else
996 ret = err(opt,
997 _("do not know what to do with %06o %s '%s'"),
998 contents->mode, oid_to_hex(&contents->oid), path);
999 free_buf:
1000 free(buf);
1001 }
1002 update_index:
1003 if (!ret && update_cache) {
1004 int refresh = (!opt->priv->call_depth &&
1005 contents->mode != S_IFGITLINK);
1006 if (add_cacheinfo(opt, contents, path, 0, refresh,
1007 ADD_CACHE_OK_TO_ADD))
1008 return -1;
1009 }
1010 return ret;
1011 }
1012
1013 static int update_file(struct merge_options *opt,
1014 int clean,
1015 const struct diff_filespec *contents,
1016 const char *path)
1017 {
1018 return update_file_flags(opt, contents, path,
1019 opt->priv->call_depth || clean, !opt->priv->call_depth);
1020 }
1021
1022 /* Low level file merging, update and removal */
1023
1024 struct merge_file_info {
1025 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1026 unsigned clean:1,
1027 merge:1;
1028 };
1029
1030 static int merge_3way(struct merge_options *opt,
1031 mmbuffer_t *result_buf,
1032 const struct diff_filespec *o,
1033 const struct diff_filespec *a,
1034 const struct diff_filespec *b,
1035 const char *branch1,
1036 const char *branch2,
1037 const int extra_marker_size)
1038 {
1039 mmfile_t orig, src1, src2;
1040 struct ll_merge_options ll_opts = {0};
1041 char *base, *name1, *name2;
1042 int merge_status;
1043
1044 ll_opts.renormalize = opt->renormalize;
1045 ll_opts.extra_marker_size = extra_marker_size;
1046 ll_opts.xdl_opts = opt->xdl_opts;
1047
1048 if (opt->priv->call_depth) {
1049 ll_opts.virtual_ancestor = 1;
1050 ll_opts.variant = 0;
1051 } else {
1052 switch (opt->recursive_variant) {
1053 case MERGE_VARIANT_OURS:
1054 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1055 break;
1056 case MERGE_VARIANT_THEIRS:
1057 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1058 break;
1059 default:
1060 ll_opts.variant = 0;
1061 break;
1062 }
1063 }
1064
1065 assert(a->path && b->path && o->path && opt->ancestor);
1066 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1067 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1068 name1 = mkpathdup("%s:%s", branch1, a->path);
1069 name2 = mkpathdup("%s:%s", branch2, b->path);
1070 } else {
1071 base = mkpathdup("%s", opt->ancestor);
1072 name1 = mkpathdup("%s", branch1);
1073 name2 = mkpathdup("%s", branch2);
1074 }
1075
1076 read_mmblob(&orig, &o->oid);
1077 read_mmblob(&src1, &a->oid);
1078 read_mmblob(&src2, &b->oid);
1079
1080 /*
1081 * FIXME: Using a->path for normalization rules in ll_merge could be
1082 * wrong if we renamed from a->path to b->path. We should use the
1083 * target path for where the file will be written.
1084 */
1085 merge_status = ll_merge(result_buf, a->path, &orig, base,
1086 &src1, name1, &src2, name2,
1087 opt->repo->index, &ll_opts);
1088
1089 free(base);
1090 free(name1);
1091 free(name2);
1092 free(orig.ptr);
1093 free(src1.ptr);
1094 free(src2.ptr);
1095 return merge_status;
1096 }
1097
1098 static int find_first_merges(struct repository *repo,
1099 struct object_array *result, const char *path,
1100 struct commit *a, struct commit *b)
1101 {
1102 int i, j;
1103 struct object_array merges = OBJECT_ARRAY_INIT;
1104 struct commit *commit;
1105 int contains_another;
1106
1107 char merged_revision[GIT_MAX_HEXSZ + 2];
1108 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1109 "--all", merged_revision, NULL };
1110 struct rev_info revs;
1111 struct setup_revision_opt rev_opts;
1112
1113 memset(result, 0, sizeof(struct object_array));
1114 memset(&rev_opts, 0, sizeof(rev_opts));
1115
1116 /* get all revisions that merge commit a */
1117 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1118 oid_to_hex(&a->object.oid));
1119 repo_init_revisions(repo, &revs, NULL);
1120 rev_opts.submodule = path;
1121 /* FIXME: can't handle linked worktrees in submodules yet */
1122 revs.single_worktree = path != NULL;
1123 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1124
1125 /* save all revisions from the above list that contain b */
1126 if (prepare_revision_walk(&revs))
1127 die("revision walk setup failed");
1128 while ((commit = get_revision(&revs)) != NULL) {
1129 struct object *o = &(commit->object);
1130 if (in_merge_bases(b, commit))
1131 add_object_array(o, NULL, &merges);
1132 }
1133 reset_revision_walk();
1134
1135 /* Now we've got all merges that contain a and b. Prune all
1136 * merges that contain another found merge and save them in
1137 * result.
1138 */
1139 for (i = 0; i < merges.nr; i++) {
1140 struct commit *m1 = (struct commit *) merges.objects[i].item;
1141
1142 contains_another = 0;
1143 for (j = 0; j < merges.nr; j++) {
1144 struct commit *m2 = (struct commit *) merges.objects[j].item;
1145 if (i != j && in_merge_bases(m2, m1)) {
1146 contains_another = 1;
1147 break;
1148 }
1149 }
1150
1151 if (!contains_another)
1152 add_object_array(merges.objects[i].item, NULL, result);
1153 }
1154
1155 object_array_clear(&merges);
1156 return result->nr;
1157 }
1158
1159 static void print_commit(struct commit *commit)
1160 {
1161 struct strbuf sb = STRBUF_INIT;
1162 struct pretty_print_context ctx = {0};
1163 ctx.date_mode.type = DATE_NORMAL;
1164 /* FIXME: Merge this with output_commit_title() */
1165 assert(!merge_remote_util(commit));
1166 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1167 fprintf(stderr, "%s\n", sb.buf);
1168 strbuf_release(&sb);
1169 }
1170
1171 static int is_valid(const struct diff_filespec *dfs)
1172 {
1173 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1174 }
1175
1176 static int merge_submodule(struct merge_options *opt,
1177 struct object_id *result, const char *path,
1178 const struct object_id *base, const struct object_id *a,
1179 const struct object_id *b)
1180 {
1181 struct commit *commit_base, *commit_a, *commit_b;
1182 int parent_count;
1183 struct object_array merges;
1184
1185 int i;
1186 int search = !opt->priv->call_depth;
1187
1188 /* store a in result in case we fail */
1189 /* FIXME: This is the WRONG resolution for the recursive case when
1190 * we need to be careful to avoid accidentally matching either side.
1191 * Should probably use o instead there, much like we do for merging
1192 * binaries.
1193 */
1194 oidcpy(result, a);
1195
1196 /* we can not handle deletion conflicts */
1197 if (is_null_oid(base))
1198 return 0;
1199 if (is_null_oid(a))
1200 return 0;
1201 if (is_null_oid(b))
1202 return 0;
1203
1204 if (add_submodule_odb(path)) {
1205 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1206 return 0;
1207 }
1208
1209 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1210 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1211 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1212 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1213 return 0;
1214 }
1215
1216 /* check whether both changes are forward */
1217 if (!in_merge_bases(commit_base, commit_a) ||
1218 !in_merge_bases(commit_base, commit_b)) {
1219 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1220 return 0;
1221 }
1222
1223 /* Case #1: a is contained in b or vice versa */
1224 if (in_merge_bases(commit_a, commit_b)) {
1225 oidcpy(result, b);
1226 if (show(opt, 3)) {
1227 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1228 output_commit_title(opt, commit_b);
1229 } else if (show(opt, 2))
1230 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1231 else
1232 ; /* no output */
1233
1234 return 1;
1235 }
1236 if (in_merge_bases(commit_b, commit_a)) {
1237 oidcpy(result, a);
1238 if (show(opt, 3)) {
1239 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1240 output_commit_title(opt, commit_a);
1241 } else if (show(opt, 2))
1242 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1243 else
1244 ; /* no output */
1245
1246 return 1;
1247 }
1248
1249 /*
1250 * Case #2: There are one or more merges that contain a and b in
1251 * the submodule. If there is only one, then present it as a
1252 * suggestion to the user, but leave it marked unmerged so the
1253 * user needs to confirm the resolution.
1254 */
1255
1256 /* Skip the search if makes no sense to the calling context. */
1257 if (!search)
1258 return 0;
1259
1260 /* find commit which merges them */
1261 parent_count = find_first_merges(opt->repo, &merges, path,
1262 commit_a, commit_b);
1263 switch (parent_count) {
1264 case 0:
1265 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1266 break;
1267
1268 case 1:
1269 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1270 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1271 print_commit((struct commit *) merges.objects[0].item);
1272 output(opt, 2, _(
1273 "If this is correct simply add it to the index "
1274 "for example\n"
1275 "by using:\n\n"
1276 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1277 "which will accept this suggestion.\n"),
1278 oid_to_hex(&merges.objects[0].item->oid), path);
1279 break;
1280
1281 default:
1282 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1283 for (i = 0; i < merges.nr; i++)
1284 print_commit((struct commit *) merges.objects[i].item);
1285 }
1286
1287 object_array_clear(&merges);
1288 return 0;
1289 }
1290
1291 static int merge_mode_and_contents(struct merge_options *opt,
1292 const struct diff_filespec *o,
1293 const struct diff_filespec *a,
1294 const struct diff_filespec *b,
1295 const char *filename,
1296 const char *branch1,
1297 const char *branch2,
1298 const int extra_marker_size,
1299 struct merge_file_info *result)
1300 {
1301 if (opt->branch1 != branch1) {
1302 /*
1303 * It's weird getting a reverse merge with HEAD on the bottom
1304 * side of the conflict markers and the other branch on the
1305 * top. Fix that.
1306 */
1307 return merge_mode_and_contents(opt, o, b, a,
1308 filename,
1309 branch2, branch1,
1310 extra_marker_size, result);
1311 }
1312
1313 result->merge = 0;
1314 result->clean = 1;
1315
1316 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1317 result->clean = 0;
1318 /*
1319 * FIXME: This is a bad resolution for recursive case; for
1320 * the recursive case we want something that is unlikely to
1321 * accidentally match either side. Also, while it makes
1322 * sense to prefer regular files over symlinks, it doesn't
1323 * make sense to prefer regular files over submodules.
1324 */
1325 if (S_ISREG(a->mode)) {
1326 result->blob.mode = a->mode;
1327 oidcpy(&result->blob.oid, &a->oid);
1328 } else {
1329 result->blob.mode = b->mode;
1330 oidcpy(&result->blob.oid, &b->oid);
1331 }
1332 } else {
1333 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1334 result->merge = 1;
1335
1336 /*
1337 * Merge modes
1338 */
1339 if (a->mode == b->mode || a->mode == o->mode)
1340 result->blob.mode = b->mode;
1341 else {
1342 result->blob.mode = a->mode;
1343 if (b->mode != o->mode) {
1344 result->clean = 0;
1345 result->merge = 1;
1346 }
1347 }
1348
1349 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1350 oidcpy(&result->blob.oid, &b->oid);
1351 else if (oideq(&b->oid, &o->oid))
1352 oidcpy(&result->blob.oid, &a->oid);
1353 else if (S_ISREG(a->mode)) {
1354 mmbuffer_t result_buf;
1355 int ret = 0, merge_status;
1356
1357 merge_status = merge_3way(opt, &result_buf, o, a, b,
1358 branch1, branch2,
1359 extra_marker_size);
1360
1361 if ((merge_status < 0) || !result_buf.ptr)
1362 ret = err(opt, _("Failed to execute internal merge"));
1363
1364 if (!ret &&
1365 write_object_file(result_buf.ptr, result_buf.size,
1366 blob_type, &result->blob.oid))
1367 ret = err(opt, _("Unable to add %s to database"),
1368 a->path);
1369
1370 free(result_buf.ptr);
1371 if (ret)
1372 return ret;
1373 /* FIXME: bug, what if modes didn't match? */
1374 result->clean = (merge_status == 0);
1375 } else if (S_ISGITLINK(a->mode)) {
1376 result->clean = merge_submodule(opt, &result->blob.oid,
1377 o->path,
1378 &o->oid,
1379 &a->oid,
1380 &b->oid);
1381 } else if (S_ISLNK(a->mode)) {
1382 switch (opt->recursive_variant) {
1383 case MERGE_VARIANT_NORMAL:
1384 oidcpy(&result->blob.oid, &a->oid);
1385 if (!oideq(&a->oid, &b->oid))
1386 result->clean = 0;
1387 break;
1388 case MERGE_VARIANT_OURS:
1389 oidcpy(&result->blob.oid, &a->oid);
1390 break;
1391 case MERGE_VARIANT_THEIRS:
1392 oidcpy(&result->blob.oid, &b->oid);
1393 break;
1394 }
1395 } else
1396 BUG("unsupported object type in the tree");
1397 }
1398
1399 if (result->merge)
1400 output(opt, 2, _("Auto-merging %s"), filename);
1401
1402 return 0;
1403 }
1404
1405 static int handle_rename_via_dir(struct merge_options *opt,
1406 struct rename_conflict_info *ci)
1407 {
1408 /*
1409 * Handle file adds that need to be renamed due to directory rename
1410 * detection. This differs from handle_rename_normal, because
1411 * there is no content merge to do; just move the file into the
1412 * desired final location.
1413 */
1414 const struct rename *ren = ci->ren1;
1415 const struct diff_filespec *dest = ren->pair->two;
1416 char *file_path = dest->path;
1417 int mark_conflicted = (opt->detect_directory_renames ==
1418 MERGE_DIRECTORY_RENAMES_CONFLICT);
1419 assert(ren->dir_rename_original_dest);
1420
1421 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1422 mark_conflicted = 1;
1423 file_path = unique_path(opt, dest->path, ren->branch);
1424 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1425 "writing to %s instead."),
1426 dest->path, file_path);
1427 }
1428
1429 if (mark_conflicted) {
1430 /*
1431 * Write the file in worktree at file_path. In the index,
1432 * only record the file at dest->path in the appropriate
1433 * higher stage.
1434 */
1435 if (update_file(opt, 0, dest, file_path))
1436 return -1;
1437 if (file_path != dest->path)
1438 free(file_path);
1439 if (update_stages(opt, dest->path, NULL,
1440 ren->branch == opt->branch1 ? dest : NULL,
1441 ren->branch == opt->branch1 ? NULL : dest))
1442 return -1;
1443 return 0; /* not clean, but conflicted */
1444 } else {
1445 /* Update dest->path both in index and in worktree */
1446 if (update_file(opt, 1, dest, dest->path))
1447 return -1;
1448 return 1; /* clean */
1449 }
1450 }
1451
1452 static int handle_change_delete(struct merge_options *opt,
1453 const char *path, const char *old_path,
1454 const struct diff_filespec *o,
1455 const struct diff_filespec *changed,
1456 const char *change_branch,
1457 const char *delete_branch,
1458 const char *change, const char *change_past)
1459 {
1460 char *alt_path = NULL;
1461 const char *update_path = path;
1462 int ret = 0;
1463
1464 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1465 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1466 update_path = alt_path = unique_path(opt, path, change_branch);
1467 }
1468
1469 if (opt->priv->call_depth) {
1470 /*
1471 * We cannot arbitrarily accept either a_sha or b_sha as
1472 * correct; since there is no true "middle point" between
1473 * them, simply reuse the base version for virtual merge base.
1474 */
1475 ret = remove_file_from_index(opt->repo->index, path);
1476 if (!ret)
1477 ret = update_file(opt, 0, o, update_path);
1478 } else {
1479 /*
1480 * Despite the four nearly duplicate messages and argument
1481 * lists below and the ugliness of the nested if-statements,
1482 * having complete messages makes the job easier for
1483 * translators.
1484 *
1485 * The slight variance among the cases is due to the fact
1486 * that:
1487 * 1) directory/file conflicts (in effect if
1488 * !alt_path) could cause us to need to write the
1489 * file to a different path.
1490 * 2) renames (in effect if !old_path) could mean that
1491 * there are two names for the path that the user
1492 * may know the file by.
1493 */
1494 if (!alt_path) {
1495 if (!old_path) {
1496 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1497 "and %s in %s. Version %s of %s left in tree."),
1498 change, path, delete_branch, change_past,
1499 change_branch, change_branch, path);
1500 } else {
1501 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1502 "and %s to %s in %s. Version %s of %s left in tree."),
1503 change, old_path, delete_branch, change_past, path,
1504 change_branch, change_branch, path);
1505 }
1506 } else {
1507 if (!old_path) {
1508 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1509 "and %s in %s. Version %s of %s left in tree at %s."),
1510 change, path, delete_branch, change_past,
1511 change_branch, change_branch, path, alt_path);
1512 } else {
1513 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1514 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1515 change, old_path, delete_branch, change_past, path,
1516 change_branch, change_branch, path, alt_path);
1517 }
1518 }
1519 /*
1520 * No need to call update_file() on path when change_branch ==
1521 * opt->branch1 && !alt_path, since that would needlessly touch
1522 * path. We could call update_file_flags() with update_cache=0
1523 * and update_wd=0, but that's a no-op.
1524 */
1525 if (change_branch != opt->branch1 || alt_path)
1526 ret = update_file(opt, 0, changed, update_path);
1527 }
1528 free(alt_path);
1529
1530 return ret;
1531 }
1532
1533 static int handle_rename_delete(struct merge_options *opt,
1534 struct rename_conflict_info *ci)
1535 {
1536 const struct rename *ren = ci->ren1;
1537 const struct diff_filespec *orig = ren->pair->one;
1538 const struct diff_filespec *dest = ren->pair->two;
1539 const char *rename_branch = ren->branch;
1540 const char *delete_branch = (opt->branch1 == ren->branch ?
1541 opt->branch2 : opt->branch1);
1542
1543 if (handle_change_delete(opt,
1544 opt->priv->call_depth ? orig->path : dest->path,
1545 opt->priv->call_depth ? NULL : orig->path,
1546 orig, dest,
1547 rename_branch, delete_branch,
1548 _("rename"), _("renamed")))
1549 return -1;
1550
1551 if (opt->priv->call_depth)
1552 return remove_file_from_index(opt->repo->index, dest->path);
1553 else
1554 return update_stages(opt, dest->path, NULL,
1555 rename_branch == opt->branch1 ? dest : NULL,
1556 rename_branch == opt->branch1 ? NULL : dest);
1557 }
1558
1559 static int handle_file_collision(struct merge_options *opt,
1560 const char *collide_path,
1561 const char *prev_path1,
1562 const char *prev_path2,
1563 const char *branch1, const char *branch2,
1564 struct diff_filespec *a,
1565 struct diff_filespec *b)
1566 {
1567 struct merge_file_info mfi;
1568 struct diff_filespec null;
1569 char *alt_path = NULL;
1570 const char *update_path = collide_path;
1571
1572 /*
1573 * It's easiest to get the correct things into stage 2 and 3, and
1574 * to make sure that the content merge puts HEAD before the other
1575 * branch if we just ensure that branch1 == opt->branch1. So, simply
1576 * flip arguments around if we don't have that.
1577 */
1578 if (branch1 != opt->branch1) {
1579 return handle_file_collision(opt, collide_path,
1580 prev_path2, prev_path1,
1581 branch2, branch1,
1582 b, a);
1583 }
1584
1585 /* Remove rename sources if rename/add or rename/rename(2to1) */
1586 if (prev_path1)
1587 remove_file(opt, 1, prev_path1,
1588 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1589 if (prev_path2)
1590 remove_file(opt, 1, prev_path2,
1591 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1592
1593 /*
1594 * Remove the collision path, if it wouldn't cause dirty contents
1595 * or an untracked file to get lost. We'll either overwrite with
1596 * merged contents, or just write out to differently named files.
1597 */
1598 if (was_dirty(opt, collide_path)) {
1599 output(opt, 1, _("Refusing to lose dirty file at %s"),
1600 collide_path);
1601 update_path = alt_path = unique_path(opt, collide_path, "merged");
1602 } else if (would_lose_untracked(opt, collide_path)) {
1603 /*
1604 * Only way we get here is if both renames were from
1605 * a directory rename AND user had an untracked file
1606 * at the location where both files end up after the
1607 * two directory renames. See testcase 10d of t6043.
1608 */
1609 output(opt, 1, _("Refusing to lose untracked file at "
1610 "%s, even though it's in the way."),
1611 collide_path);
1612 update_path = alt_path = unique_path(opt, collide_path, "merged");
1613 } else {
1614 /*
1615 * FIXME: It's possible that the two files are identical
1616 * and that the current working copy happens to match, in
1617 * which case we are unnecessarily touching the working
1618 * tree file. It's not a likely enough scenario that I
1619 * want to code up the checks for it and a better fix is
1620 * available if we restructure how unpack_trees() and
1621 * merge-recursive interoperate anyway, so punting for
1622 * now...
1623 */
1624 remove_file(opt, 0, collide_path, 0);
1625 }
1626
1627 /* Store things in diff_filespecs for functions that need it */
1628 null.path = (char *)collide_path;
1629 oidcpy(&null.oid, null_oid());
1630 null.mode = 0;
1631
1632 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1633 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1634 return -1;
1635 mfi.clean &= !alt_path;
1636 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1637 return -1;
1638 if (!mfi.clean && !opt->priv->call_depth &&
1639 update_stages(opt, collide_path, NULL, a, b))
1640 return -1;
1641 free(alt_path);
1642 /*
1643 * FIXME: If both a & b both started with conflicts (only possible
1644 * if they came from a rename/rename(2to1)), but had IDENTICAL
1645 * contents including those conflicts, then in the next line we claim
1646 * it was clean. If someone cares about this case, we should have the
1647 * caller notify us if we started with conflicts.
1648 */
1649 return mfi.clean;
1650 }
1651
1652 static int handle_rename_add(struct merge_options *opt,
1653 struct rename_conflict_info *ci)
1654 {
1655 /* a was renamed to c, and a separate c was added. */
1656 struct diff_filespec *a = ci->ren1->pair->one;
1657 struct diff_filespec *c = ci->ren1->pair->two;
1658 char *path = c->path;
1659 char *prev_path_desc;
1660 struct merge_file_info mfi;
1661
1662 const char *rename_branch = ci->ren1->branch;
1663 const char *add_branch = (opt->branch1 == rename_branch ?
1664 opt->branch2 : opt->branch1);
1665 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1666
1667 output(opt, 1, _("CONFLICT (rename/add): "
1668 "Rename %s->%s in %s. Added %s in %s"),
1669 a->path, c->path, rename_branch,
1670 c->path, add_branch);
1671
1672 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1673 ci->ren1->src_entry->stages[other_stage].path = a->path;
1674 if (merge_mode_and_contents(opt, a, c,
1675 &ci->ren1->src_entry->stages[other_stage],
1676 prev_path_desc,
1677 opt->branch1, opt->branch2,
1678 1 + opt->priv->call_depth * 2, &mfi))
1679 return -1;
1680 free(prev_path_desc);
1681
1682 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1683 return handle_file_collision(opt,
1684 c->path, a->path, NULL,
1685 rename_branch, add_branch,
1686 &mfi.blob,
1687 &ci->ren1->dst_entry->stages[other_stage]);
1688 }
1689
1690 static char *find_path_for_conflict(struct merge_options *opt,
1691 const char *path,
1692 const char *branch1,
1693 const char *branch2)
1694 {
1695 char *new_path = NULL;
1696 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1697 new_path = unique_path(opt, path, branch1);
1698 output(opt, 1, _("%s is a directory in %s adding "
1699 "as %s instead"),
1700 path, branch2, new_path);
1701 } else if (would_lose_untracked(opt, path)) {
1702 new_path = unique_path(opt, path, branch1);
1703 output(opt, 1, _("Refusing to lose untracked file"
1704 " at %s; adding as %s instead"),
1705 path, new_path);
1706 }
1707
1708 return new_path;
1709 }
1710
1711 /*
1712 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1713 */
1714 static inline int flip_stage(int stage)
1715 {
1716 return (2 + 3) - stage;
1717 }
1718
1719 static int handle_rename_rename_1to2(struct merge_options *opt,
1720 struct rename_conflict_info *ci)
1721 {
1722 /* One file was renamed in both branches, but to different names. */
1723 struct merge_file_info mfi;
1724 struct diff_filespec *add;
1725 struct diff_filespec *o = ci->ren1->pair->one;
1726 struct diff_filespec *a = ci->ren1->pair->two;
1727 struct diff_filespec *b = ci->ren2->pair->two;
1728 char *path_desc;
1729
1730 output(opt, 1, _("CONFLICT (rename/rename): "
1731 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1732 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1733 o->path, a->path, ci->ren1->branch,
1734 o->path, b->path, ci->ren2->branch,
1735 opt->priv->call_depth ? _(" (left unresolved)") : "");
1736
1737 path_desc = xstrfmt("%s and %s, both renamed from %s",
1738 a->path, b->path, o->path);
1739 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1740 ci->ren1->branch, ci->ren2->branch,
1741 opt->priv->call_depth * 2, &mfi))
1742 return -1;
1743 free(path_desc);
1744
1745 if (opt->priv->call_depth)
1746 remove_file_from_index(opt->repo->index, o->path);
1747
1748 /*
1749 * For each destination path, we need to see if there is a
1750 * rename/add collision. If not, we can write the file out
1751 * to the specified location.
1752 */
1753 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1754 if (is_valid(add)) {
1755 add->path = mfi.blob.path = a->path;
1756 if (handle_file_collision(opt, a->path,
1757 NULL, NULL,
1758 ci->ren1->branch,
1759 ci->ren2->branch,
1760 &mfi.blob, add) < 0)
1761 return -1;
1762 } else {
1763 char *new_path = find_path_for_conflict(opt, a->path,
1764 ci->ren1->branch,
1765 ci->ren2->branch);
1766 if (update_file(opt, 0, &mfi.blob,
1767 new_path ? new_path : a->path))
1768 return -1;
1769 free(new_path);
1770 if (!opt->priv->call_depth &&
1771 update_stages(opt, a->path, NULL, a, NULL))
1772 return -1;
1773 }
1774
1775 if (!mfi.clean && mfi.blob.mode == a->mode &&
1776 oideq(&mfi.blob.oid, &a->oid)) {
1777 /*
1778 * Getting here means we were attempting to merge a binary
1779 * blob. Since we can't merge binaries, the merge algorithm
1780 * just takes one side. But we don't want to copy the
1781 * contents of one side to both paths; we'd rather use the
1782 * original content at the given path for each path.
1783 */
1784 oidcpy(&mfi.blob.oid, &b->oid);
1785 mfi.blob.mode = b->mode;
1786 }
1787 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1788 if (is_valid(add)) {
1789 add->path = mfi.blob.path = b->path;
1790 if (handle_file_collision(opt, b->path,
1791 NULL, NULL,
1792 ci->ren1->branch,
1793 ci->ren2->branch,
1794 add, &mfi.blob) < 0)
1795 return -1;
1796 } else {
1797 char *new_path = find_path_for_conflict(opt, b->path,
1798 ci->ren2->branch,
1799 ci->ren1->branch);
1800 if (update_file(opt, 0, &mfi.blob,
1801 new_path ? new_path : b->path))
1802 return -1;
1803 free(new_path);
1804 if (!opt->priv->call_depth &&
1805 update_stages(opt, b->path, NULL, NULL, b))
1806 return -1;
1807 }
1808
1809 return 0;
1810 }
1811
1812 static int handle_rename_rename_2to1(struct merge_options *opt,
1813 struct rename_conflict_info *ci)
1814 {
1815 /* Two files, a & b, were renamed to the same thing, c. */
1816 struct diff_filespec *a = ci->ren1->pair->one;
1817 struct diff_filespec *b = ci->ren2->pair->one;
1818 struct diff_filespec *c1 = ci->ren1->pair->two;
1819 struct diff_filespec *c2 = ci->ren2->pair->two;
1820 char *path = c1->path; /* == c2->path */
1821 char *path_side_1_desc;
1822 char *path_side_2_desc;
1823 struct merge_file_info mfi_c1;
1824 struct merge_file_info mfi_c2;
1825 int ostage1, ostage2;
1826
1827 output(opt, 1, _("CONFLICT (rename/rename): "
1828 "Rename %s->%s in %s. "
1829 "Rename %s->%s in %s"),
1830 a->path, c1->path, ci->ren1->branch,
1831 b->path, c2->path, ci->ren2->branch);
1832
1833 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1834 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1835 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1836 ostage2 = flip_stage(ostage1);
1837 ci->ren1->src_entry->stages[ostage1].path = a->path;
1838 ci->ren2->src_entry->stages[ostage2].path = b->path;
1839 if (merge_mode_and_contents(opt, a, c1,
1840 &ci->ren1->src_entry->stages[ostage1],
1841 path_side_1_desc,
1842 opt->branch1, opt->branch2,
1843 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1844 merge_mode_and_contents(opt, b,
1845 &ci->ren2->src_entry->stages[ostage2],
1846 c2, path_side_2_desc,
1847 opt->branch1, opt->branch2,
1848 1 + opt->priv->call_depth * 2, &mfi_c2))
1849 return -1;
1850 free(path_side_1_desc);
1851 free(path_side_2_desc);
1852 mfi_c1.blob.path = path;
1853 mfi_c2.blob.path = path;
1854
1855 return handle_file_collision(opt, path, a->path, b->path,
1856 ci->ren1->branch, ci->ren2->branch,
1857 &mfi_c1.blob, &mfi_c2.blob);
1858 }
1859
1860 /*
1861 * Get the diff_filepairs changed between o_tree and tree.
1862 */
1863 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1864 struct tree *o_tree,
1865 struct tree *tree)
1866 {
1867 struct diff_queue_struct *ret;
1868 struct diff_options opts;
1869
1870 repo_diff_setup(opt->repo, &opts);
1871 opts.flags.recursive = 1;
1872 opts.flags.rename_empty = 0;
1873 opts.detect_rename = merge_detect_rename(opt);
1874 /*
1875 * We do not have logic to handle the detection of copies. In
1876 * fact, it may not even make sense to add such logic: would we
1877 * really want a change to a base file to be propagated through
1878 * multiple other files by a merge?
1879 */
1880 if (opts.detect_rename > DIFF_DETECT_RENAME)
1881 opts.detect_rename = DIFF_DETECT_RENAME;
1882 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 1000;
1883 opts.rename_score = opt->rename_score;
1884 opts.show_rename_progress = opt->show_rename_progress;
1885 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1886 diff_setup_done(&opts);
1887 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1888 diffcore_std(&opts);
1889 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1890 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1891
1892 ret = xmalloc(sizeof(*ret));
1893 *ret = diff_queued_diff;
1894
1895 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1896 diff_queued_diff.nr = 0;
1897 diff_queued_diff.queue = NULL;
1898 diff_flush(&opts);
1899 return ret;
1900 }
1901
1902 static int tree_has_path(struct repository *r, struct tree *tree,
1903 const char *path)
1904 {
1905 struct object_id hashy;
1906 unsigned short mode_o;
1907
1908 return !get_tree_entry(r,
1909 &tree->object.oid, path,
1910 &hashy, &mode_o);
1911 }
1912
1913 /*
1914 * Return a new string that replaces the beginning portion (which matches
1915 * entry->dir), with entry->new_dir. In perl-speak:
1916 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1917 * NOTE:
1918 * Caller must ensure that old_path starts with entry->dir + '/'.
1919 */
1920 static char *apply_dir_rename(struct dir_rename_entry *entry,
1921 const char *old_path)
1922 {
1923 struct strbuf new_path = STRBUF_INIT;
1924 int oldlen, newlen;
1925
1926 if (entry->non_unique_new_dir)
1927 return NULL;
1928
1929 oldlen = strlen(entry->dir);
1930 if (entry->new_dir.len == 0)
1931 /*
1932 * If someone renamed/merged a subdirectory into the root
1933 * directory (e.g. 'some/subdir' -> ''), then we want to
1934 * avoid returning
1935 * '' + '/filename'
1936 * as the rename; we need to make old_path + oldlen advance
1937 * past the '/' character.
1938 */
1939 oldlen++;
1940 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1941 strbuf_grow(&new_path, newlen);
1942 strbuf_addbuf(&new_path, &entry->new_dir);
1943 strbuf_addstr(&new_path, &old_path[oldlen]);
1944
1945 return strbuf_detach(&new_path, NULL);
1946 }
1947
1948 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1949 char **old_dir, char **new_dir)
1950 {
1951 char *end_of_old, *end_of_new;
1952
1953 /* Default return values: NULL, meaning no rename */
1954 *old_dir = NULL;
1955 *new_dir = NULL;
1956
1957 /*
1958 * For
1959 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1960 * the "e/foo.c" part is the same, we just want to know that
1961 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1962 * so, for this example, this function returns "a/b/c/d" in
1963 * *old_dir and "a/b/some/thing/else" in *new_dir.
1964 */
1965
1966 /*
1967 * If the basename of the file changed, we don't care. We want
1968 * to know which portion of the directory, if any, changed.
1969 */
1970 end_of_old = strrchr(old_path, '/');
1971 end_of_new = strrchr(new_path, '/');
1972
1973 /*
1974 * If end_of_old is NULL, old_path wasn't in a directory, so there
1975 * could not be a directory rename (our rule elsewhere that a
1976 * directory which still exists is not considered to have been
1977 * renamed means the root directory can never be renamed -- because
1978 * the root directory always exists).
1979 */
1980 if (end_of_old == NULL)
1981 return; /* Note: *old_dir and *new_dir are still NULL */
1982
1983 /*
1984 * If new_path contains no directory (end_of_new is NULL), then we
1985 * have a rename of old_path's directory to the root directory.
1986 */
1987 if (end_of_new == NULL) {
1988 *old_dir = xstrndup(old_path, end_of_old - old_path);
1989 *new_dir = xstrdup("");
1990 return;
1991 }
1992
1993 /* Find the first non-matching character traversing backwards */
1994 while (*--end_of_new == *--end_of_old &&
1995 end_of_old != old_path &&
1996 end_of_new != new_path)
1997 ; /* Do nothing; all in the while loop */
1998
1999 /*
2000 * If both got back to the beginning of their strings, then the
2001 * directory didn't change at all, only the basename did.
2002 */
2003 if (end_of_old == old_path && end_of_new == new_path &&
2004 *end_of_old == *end_of_new)
2005 return; /* Note: *old_dir and *new_dir are still NULL */
2006
2007 /*
2008 * If end_of_new got back to the beginning of its string, and
2009 * end_of_old got back to the beginning of some subdirectory, then
2010 * we have a rename/merge of a subdirectory into the root, which
2011 * needs slightly special handling.
2012 *
2013 * Note: There is no need to consider the opposite case, with a
2014 * rename/merge of the root directory into some subdirectory
2015 * because as noted above the root directory always exists so it
2016 * cannot be considered to be renamed.
2017 */
2018 if (end_of_new == new_path &&
2019 end_of_old != old_path && end_of_old[-1] == '/') {
2020 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2021 *new_dir = xstrdup("");
2022 return;
2023 }
2024
2025 /*
2026 * We've found the first non-matching character in the directory
2027 * paths. That means the current characters we were looking at
2028 * were part of the first non-matching subdir name going back from
2029 * the end of the strings. Get the whole name by advancing both
2030 * end_of_old and end_of_new to the NEXT '/' character. That will
2031 * represent the entire directory rename.
2032 *
2033 * The reason for the increment is cases like
2034 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2035 * After dropping the basename and going back to the first
2036 * non-matching character, we're now comparing:
2037 * a/b/s and a/b/
2038 * and we want to be comparing:
2039 * a/b/star/ and a/b/tar/
2040 * but without the pre-increment, the one on the right would stay
2041 * a/b/.
2042 */
2043 end_of_old = strchr(++end_of_old, '/');
2044 end_of_new = strchr(++end_of_new, '/');
2045
2046 /* Copy the old and new directories into *old_dir and *new_dir. */
2047 *old_dir = xstrndup(old_path, end_of_old - old_path);
2048 *new_dir = xstrndup(new_path, end_of_new - new_path);
2049 }
2050
2051 static void remove_hashmap_entries(struct hashmap *dir_renames,
2052 struct string_list *items_to_remove)
2053 {
2054 int i;
2055 struct dir_rename_entry *entry;
2056
2057 for (i = 0; i < items_to_remove->nr; i++) {
2058 entry = items_to_remove->items[i].util;
2059 hashmap_remove(dir_renames, &entry->ent, NULL);
2060 }
2061 string_list_clear(items_to_remove, 0);
2062 }
2063
2064 /*
2065 * See if there is a directory rename for path, and if there are any file
2066 * level conflicts for the renamed location. If there is a rename and
2067 * there are no conflicts, return the new name. Otherwise, return NULL.
2068 */
2069 static char *handle_path_level_conflicts(struct merge_options *opt,
2070 const char *path,
2071 struct dir_rename_entry *entry,
2072 struct hashmap *collisions,
2073 struct tree *tree)
2074 {
2075 char *new_path = NULL;
2076 struct collision_entry *collision_ent;
2077 int clean = 1;
2078 struct strbuf collision_paths = STRBUF_INIT;
2079
2080 /*
2081 * entry has the mapping of old directory name to new directory name
2082 * that we want to apply to path.
2083 */
2084 new_path = apply_dir_rename(entry, path);
2085
2086 if (!new_path) {
2087 /* This should only happen when entry->non_unique_new_dir set */
2088 if (!entry->non_unique_new_dir)
2089 BUG("entry->non_unqiue_dir not set and !new_path");
2090 output(opt, 1, _("CONFLICT (directory rename split): "
2091 "Unclear where to place %s because directory "
2092 "%s was renamed to multiple other directories, "
2093 "with no destination getting a majority of the "
2094 "files."),
2095 path, entry->dir);
2096 clean = 0;
2097 return NULL;
2098 }
2099
2100 /*
2101 * The caller needs to have ensured that it has pre-populated
2102 * collisions with all paths that map to new_path. Do a quick check
2103 * to ensure that's the case.
2104 */
2105 collision_ent = collision_find_entry(collisions, new_path);
2106 if (collision_ent == NULL)
2107 BUG("collision_ent is NULL");
2108
2109 /*
2110 * Check for one-sided add/add/.../add conflicts, i.e.
2111 * where implicit renames from the other side doing
2112 * directory rename(s) can affect this side of history
2113 * to put multiple paths into the same location. Warn
2114 * and bail on directory renames for such paths.
2115 */
2116 if (collision_ent->reported_already) {
2117 clean = 0;
2118 } else if (tree_has_path(opt->repo, tree, new_path)) {
2119 collision_ent->reported_already = 1;
2120 strbuf_add_separated_string_list(&collision_paths, ", ",
2121 &collision_ent->source_files);
2122 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2123 "file/dir at %s in the way of implicit "
2124 "directory rename(s) putting the following "
2125 "path(s) there: %s."),
2126 new_path, collision_paths.buf);
2127 clean = 0;
2128 } else if (collision_ent->source_files.nr > 1) {
2129 collision_ent->reported_already = 1;
2130 strbuf_add_separated_string_list(&collision_paths, ", ",
2131 &collision_ent->source_files);
2132 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2133 "more than one path to %s; implicit directory "
2134 "renames tried to put these paths there: %s"),
2135 new_path, collision_paths.buf);
2136 clean = 0;
2137 }
2138
2139 /* Free memory we no longer need */
2140 strbuf_release(&collision_paths);
2141 if (!clean && new_path) {
2142 free(new_path);
2143 return NULL;
2144 }
2145
2146 return new_path;
2147 }
2148
2149 /*
2150 * There are a couple things we want to do at the directory level:
2151 * 1. Check for both sides renaming to the same thing, in order to avoid
2152 * implicit renaming of files that should be left in place. (See
2153 * testcase 6b in t6043 for details.)
2154 * 2. Prune directory renames if there are still files left in the
2155 * original directory. These represent a partial directory rename,
2156 * i.e. a rename where only some of the files within the directory
2157 * were renamed elsewhere. (Technically, this could be done earlier
2158 * in get_directory_renames(), except that would prevent us from
2159 * doing the previous check and thus failing testcase 6b.)
2160 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2161 * In the future, we could potentially record this info as well and
2162 * omit reporting rename/rename(1to2) conflicts for each path within
2163 * the affected directories, thus cleaning up the merge output.
2164 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2165 * directory level, because merging directories is fine. If it
2166 * causes conflicts for files within those merged directories, then
2167 * that should be detected at the individual path level.
2168 */
2169 static void handle_directory_level_conflicts(struct merge_options *opt,
2170 struct hashmap *dir_re_head,
2171 struct tree *head,
2172 struct hashmap *dir_re_merge,
2173 struct tree *merge)
2174 {
2175 struct hashmap_iter iter;
2176 struct dir_rename_entry *head_ent;
2177 struct dir_rename_entry *merge_ent;
2178
2179 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2180 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2181
2182 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2183 ent /* member name */) {
2184 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2185 if (merge_ent &&
2186 !head_ent->non_unique_new_dir &&
2187 !merge_ent->non_unique_new_dir &&
2188 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2189 /* 1. Renamed identically; remove it from both sides */
2190 string_list_append(&remove_from_head,
2191 head_ent->dir)->util = head_ent;
2192 strbuf_release(&head_ent->new_dir);
2193 string_list_append(&remove_from_merge,
2194 merge_ent->dir)->util = merge_ent;
2195 strbuf_release(&merge_ent->new_dir);
2196 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2197 /* 2. This wasn't a directory rename after all */
2198 string_list_append(&remove_from_head,
2199 head_ent->dir)->util = head_ent;
2200 strbuf_release(&head_ent->new_dir);
2201 }
2202 }
2203
2204 remove_hashmap_entries(dir_re_head, &remove_from_head);
2205 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2206
2207 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2208 ent /* member name */) {
2209 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2210 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2211 /* 2. This wasn't a directory rename after all */
2212 string_list_append(&remove_from_merge,
2213 merge_ent->dir)->util = merge_ent;
2214 } else if (head_ent &&
2215 !head_ent->non_unique_new_dir &&
2216 !merge_ent->non_unique_new_dir) {
2217 /* 3. rename/rename(1to2) */
2218 /*
2219 * We can assume it's not rename/rename(1to1) because
2220 * that was case (1), already checked above. So we
2221 * know that head_ent->new_dir and merge_ent->new_dir
2222 * are different strings.
2223 */
2224 output(opt, 1, _("CONFLICT (rename/rename): "
2225 "Rename directory %s->%s in %s. "
2226 "Rename directory %s->%s in %s"),
2227 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2228 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2229 string_list_append(&remove_from_head,
2230 head_ent->dir)->util = head_ent;
2231 strbuf_release(&head_ent->new_dir);
2232 string_list_append(&remove_from_merge,
2233 merge_ent->dir)->util = merge_ent;
2234 strbuf_release(&merge_ent->new_dir);
2235 }
2236 }
2237
2238 remove_hashmap_entries(dir_re_head, &remove_from_head);
2239 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2240 }
2241
2242 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2243 {
2244 struct hashmap *dir_renames;
2245 struct hashmap_iter iter;
2246 struct dir_rename_entry *entry;
2247 int i;
2248
2249 /*
2250 * Typically, we think of a directory rename as all files from a
2251 * certain directory being moved to a target directory. However,
2252 * what if someone first moved two files from the original
2253 * directory in one commit, and then renamed the directory
2254 * somewhere else in a later commit? At merge time, we just know
2255 * that files from the original directory went to two different
2256 * places, and that the bulk of them ended up in the same place.
2257 * We want each directory rename to represent where the bulk of the
2258 * files from that directory end up; this function exists to find
2259 * where the bulk of the files went.
2260 *
2261 * The first loop below simply iterates through the list of file
2262 * renames, finding out how often each directory rename pair
2263 * possibility occurs.
2264 */
2265 dir_renames = xmalloc(sizeof(*dir_renames));
2266 dir_rename_init(dir_renames);
2267 for (i = 0; i < pairs->nr; ++i) {
2268 struct string_list_item *item;
2269 int *count;
2270 struct diff_filepair *pair = pairs->queue[i];
2271 char *old_dir, *new_dir;
2272
2273 /* File not part of directory rename if it wasn't renamed */
2274 if (pair->status != 'R')
2275 continue;
2276
2277 get_renamed_dir_portion(pair->one->path, pair->two->path,
2278 &old_dir, &new_dir);
2279 if (!old_dir)
2280 /* Directory didn't change at all; ignore this one. */
2281 continue;
2282
2283 entry = dir_rename_find_entry(dir_renames, old_dir);
2284 if (!entry) {
2285 entry = xmalloc(sizeof(*entry));
2286 dir_rename_entry_init(entry, old_dir);
2287 hashmap_put(dir_renames, &entry->ent);
2288 } else {
2289 free(old_dir);
2290 }
2291 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2292 if (!item) {
2293 item = string_list_insert(&entry->possible_new_dirs,
2294 new_dir);
2295 item->util = xcalloc(1, sizeof(int));
2296 } else {
2297 free(new_dir);
2298 }
2299 count = item->util;
2300 *count += 1;
2301 }
2302
2303 /*
2304 * For each directory with files moved out of it, we find out which
2305 * target directory received the most files so we can declare it to
2306 * be the "winning" target location for the directory rename. This
2307 * winner gets recorded in new_dir. If there is no winner
2308 * (multiple target directories received the same number of files),
2309 * we set non_unique_new_dir. Once we've determined the winner (or
2310 * that there is no winner), we no longer need possible_new_dirs.
2311 */
2312 hashmap_for_each_entry(dir_renames, &iter, entry,
2313 ent /* member name */) {
2314 int max = 0;
2315 int bad_max = 0;
2316 char *best = NULL;
2317
2318 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2319 int *count = entry->possible_new_dirs.items[i].util;
2320
2321 if (*count == max)
2322 bad_max = max;
2323 else if (*count > max) {
2324 max = *count;
2325 best = entry->possible_new_dirs.items[i].string;
2326 }
2327 }
2328 if (bad_max == max)
2329 entry->non_unique_new_dir = 1;
2330 else {
2331 assert(entry->new_dir.len == 0);
2332 strbuf_addstr(&entry->new_dir, best);
2333 }
2334 /*
2335 * The relevant directory sub-portion of the original full
2336 * filepaths were xstrndup'ed before inserting into
2337 * possible_new_dirs, and instead of manually iterating the
2338 * list and free'ing each, just lie and tell
2339 * possible_new_dirs that it did the strdup'ing so that it
2340 * will free them for us.
2341 */
2342 entry->possible_new_dirs.strdup_strings = 1;
2343 string_list_clear(&entry->possible_new_dirs, 1);
2344 }
2345
2346 return dir_renames;
2347 }
2348
2349 static struct dir_rename_entry *check_dir_renamed(const char *path,
2350 struct hashmap *dir_renames)
2351 {
2352 char *temp = xstrdup(path);
2353 char *end;
2354 struct dir_rename_entry *entry = NULL;
2355
2356 while ((end = strrchr(temp, '/'))) {
2357 *end = '\0';
2358 entry = dir_rename_find_entry(dir_renames, temp);
2359 if (entry)
2360 break;
2361 }
2362 free(temp);
2363 return entry;
2364 }
2365
2366 static void compute_collisions(struct hashmap *collisions,
2367 struct hashmap *dir_renames,
2368 struct diff_queue_struct *pairs)
2369 {
2370 int i;
2371
2372 /*
2373 * Multiple files can be mapped to the same path due to directory
2374 * renames done by the other side of history. Since that other
2375 * side of history could have merged multiple directories into one,
2376 * if our side of history added the same file basename to each of
2377 * those directories, then all N of them would get implicitly
2378 * renamed by the directory rename detection into the same path,
2379 * and we'd get an add/add/.../add conflict, and all those adds
2380 * from *this* side of history. This is not representable in the
2381 * index, and users aren't going to easily be able to make sense of
2382 * it. So we need to provide a good warning about what's
2383 * happening, and fall back to no-directory-rename detection
2384 * behavior for those paths.
2385 *
2386 * See testcases 9e and all of section 5 from t6043 for examples.
2387 */
2388 collision_init(collisions);
2389
2390 for (i = 0; i < pairs->nr; ++i) {
2391 struct dir_rename_entry *dir_rename_ent;
2392 struct collision_entry *collision_ent;
2393 char *new_path;
2394 struct diff_filepair *pair = pairs->queue[i];
2395
2396 if (pair->status != 'A' && pair->status != 'R')
2397 continue;
2398 dir_rename_ent = check_dir_renamed(pair->two->path,
2399 dir_renames);
2400 if (!dir_rename_ent)
2401 continue;
2402
2403 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2404 if (!new_path)
2405 /*
2406 * dir_rename_ent->non_unique_new_path is true, which
2407 * means there is no directory rename for us to use,
2408 * which means it won't cause us any additional
2409 * collisions.
2410 */
2411 continue;
2412 collision_ent = collision_find_entry(collisions, new_path);
2413 if (!collision_ent) {
2414 CALLOC_ARRAY(collision_ent, 1);
2415 hashmap_entry_init(&collision_ent->ent,
2416 strhash(new_path));
2417 hashmap_put(collisions, &collision_ent->ent);
2418 collision_ent->target_file = new_path;
2419 } else {
2420 free(new_path);
2421 }
2422 string_list_insert(&collision_ent->source_files,
2423 pair->two->path);
2424 }
2425 }
2426
2427 static char *check_for_directory_rename(struct merge_options *opt,
2428 const char *path,
2429 struct tree *tree,
2430 struct hashmap *dir_renames,
2431 struct hashmap *dir_rename_exclusions,
2432 struct hashmap *collisions,
2433 int *clean_merge)
2434 {
2435 char *new_path = NULL;
2436 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2437 struct dir_rename_entry *oentry = NULL;
2438
2439 if (!entry)
2440 return new_path;
2441
2442 /*
2443 * This next part is a little weird. We do not want to do an
2444 * implicit rename into a directory we renamed on our side, because
2445 * that will result in a spurious rename/rename(1to2) conflict. An
2446 * example:
2447 * Base commit: dumbdir/afile, otherdir/bfile
2448 * Side 1: smrtdir/afile, otherdir/bfile
2449 * Side 2: dumbdir/afile, dumbdir/bfile
2450 * Here, while working on Side 1, we could notice that otherdir was
2451 * renamed/merged to dumbdir, and change the diff_filepair for
2452 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2453 * 2 will notice the rename from dumbdir to smrtdir, and do the
2454 * transitive rename to move it from dumbdir/bfile to
2455 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2456 * smrtdir, a rename/rename(1to2) conflict. We really just want
2457 * the file to end up in smrtdir. And the way to achieve that is
2458 * to not let Side1 do the rename to dumbdir, since we know that is
2459 * the source of one of our directory renames.
2460 *
2461 * That's why oentry and dir_rename_exclusions is here.
2462 *
2463 * As it turns out, this also prevents N-way transient rename
2464 * confusion; See testcases 9c and 9d of t6043.
2465 */
2466 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2467 if (oentry) {
2468 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2469 "to %s, because %s itself was renamed."),
2470 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2471 } else {
2472 new_path = handle_path_level_conflicts(opt, path, entry,
2473 collisions, tree);
2474 *clean_merge &= (new_path != NULL);
2475 }
2476
2477 return new_path;
2478 }
2479
2480 static void apply_directory_rename_modifications(struct merge_options *opt,
2481 struct diff_filepair *pair,
2482 char *new_path,
2483 struct rename *re,
2484 struct tree *tree,
2485 struct tree *o_tree,
2486 struct tree *a_tree,
2487 struct tree *b_tree,
2488 struct string_list *entries)
2489 {
2490 struct string_list_item *item;
2491 int stage = (tree == a_tree ? 2 : 3);
2492 int update_wd;
2493
2494 /*
2495 * In all cases where we can do directory rename detection,
2496 * unpack_trees() will have read pair->two->path into the
2497 * index and the working copy. We need to remove it so that
2498 * we can instead place it at new_path. It is guaranteed to
2499 * not be untracked (unpack_trees() would have errored out
2500 * saying the file would have been overwritten), but it might
2501 * be dirty, though.
2502 */
2503 update_wd = !was_dirty(opt, pair->two->path);
2504 if (!update_wd)
2505 output(opt, 1, _("Refusing to lose dirty file at %s"),
2506 pair->two->path);
2507 remove_file(opt, 1, pair->two->path, !update_wd);
2508
2509 /* Find or create a new re->dst_entry */
2510 item = string_list_lookup(entries, new_path);
2511 if (item) {
2512 /*
2513 * Since we're renaming on this side of history, and it's
2514 * due to a directory rename on the other side of history
2515 * (which we only allow when the directory in question no
2516 * longer exists on the other side of history), the
2517 * original entry for re->dst_entry is no longer
2518 * necessary...
2519 */
2520 re->dst_entry->processed = 1;
2521
2522 /*
2523 * ...because we'll be using this new one.
2524 */
2525 re->dst_entry = item->util;
2526 } else {
2527 /*
2528 * re->dst_entry is for the before-dir-rename path, and we
2529 * need it to hold information for the after-dir-rename
2530 * path. Before creating a new entry, we need to mark the
2531 * old one as unnecessary (...unless it is shared by
2532 * src_entry, i.e. this didn't use to be a rename, in which
2533 * case we can just allow the normal processing to happen
2534 * for it).
2535 */
2536 if (pair->status == 'R')
2537 re->dst_entry->processed = 1;
2538
2539 re->dst_entry = insert_stage_data(opt->repo, new_path,
2540 o_tree, a_tree, b_tree,
2541 entries);
2542 item = string_list_insert(entries, new_path);
2543 item->util = re->dst_entry;
2544 }
2545
2546 /*
2547 * Update the stage_data with the information about the path we are
2548 * moving into place. That slot will be empty and available for us
2549 * to write to because of the collision checks in
2550 * handle_path_level_conflicts(). In other words,
2551 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2552 * open for us to write to.
2553 *
2554 * It may be tempting to actually update the index at this point as
2555 * well, using update_stages_for_stage_data(), but as per the big
2556 * "NOTE" in update_stages(), doing so will modify the current
2557 * in-memory index which will break calls to would_lose_untracked()
2558 * that we need to make. Instead, we need to just make sure that
2559 * the various handle_rename_*() functions update the index
2560 * explicitly rather than relying on unpack_trees() to have done it.
2561 */
2562 get_tree_entry(opt->repo,
2563 &tree->object.oid,
2564 pair->two->path,
2565 &re->dst_entry->stages[stage].oid,
2566 &re->dst_entry->stages[stage].mode);
2567
2568 /*
2569 * Record the original change status (or 'type' of change). If it
2570 * was originally an add ('A'), this lets us differentiate later
2571 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2572 * otherwise look the same). If it was originally a rename ('R'),
2573 * this lets us remember and report accurately about the transitive
2574 * renaming that occurred via the directory rename detection. Also,
2575 * record the original destination name.
2576 */
2577 re->dir_rename_original_type = pair->status;
2578 re->dir_rename_original_dest = pair->two->path;
2579
2580 /*
2581 * We don't actually look at pair->status again, but it seems
2582 * pedagogically correct to adjust it.
2583 */
2584 pair->status = 'R';
2585
2586 /*
2587 * Finally, record the new location.
2588 */
2589 pair->two->path = new_path;
2590 }
2591
2592 /*
2593 * Get information of all renames which occurred in 'pairs', making use of
2594 * any implicit directory renames inferred from the other side of history.
2595 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2596 * to be able to associate the correct cache entries with the rename
2597 * information; tree is always equal to either a_tree or b_tree.
2598 */
2599 static struct string_list *get_renames(struct merge_options *opt,
2600 const char *branch,
2601 struct diff_queue_struct *pairs,
2602 struct hashmap *dir_renames,
2603 struct hashmap *dir_rename_exclusions,
2604 struct tree *tree,
2605 struct tree *o_tree,
2606 struct tree *a_tree,
2607 struct tree *b_tree,
2608 struct string_list *entries,
2609 int *clean_merge)
2610 {
2611 int i;
2612 struct hashmap collisions;
2613 struct hashmap_iter iter;
2614 struct collision_entry *e;
2615 struct string_list *renames;
2616
2617 compute_collisions(&collisions, dir_renames, pairs);
2618 CALLOC_ARRAY(renames, 1);
2619
2620 for (i = 0; i < pairs->nr; ++i) {
2621 struct string_list_item *item;
2622 struct rename *re;
2623 struct diff_filepair *pair = pairs->queue[i];
2624 char *new_path; /* non-NULL only with directory renames */
2625
2626 if (pair->status != 'A' && pair->status != 'R') {
2627 diff_free_filepair(pair);
2628 continue;
2629 }
2630 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2631 dir_renames,
2632 dir_rename_exclusions,
2633 &collisions,
2634 clean_merge);
2635 if (pair->status != 'R' && !new_path) {
2636 diff_free_filepair(pair);
2637 continue;
2638 }
2639
2640 re = xmalloc(sizeof(*re));
2641 re->processed = 0;
2642 re->pair = pair;
2643 re->branch = branch;
2644 re->dir_rename_original_type = '\0';
2645 re->dir_rename_original_dest = NULL;
2646 item = string_list_lookup(entries, re->pair->one->path);
2647 if (!item)
2648 re->src_entry = insert_stage_data(opt->repo,
2649 re->pair->one->path,
2650 o_tree, a_tree, b_tree, entries);
2651 else
2652 re->src_entry = item->util;
2653
2654 item = string_list_lookup(entries, re->pair->two->path);
2655 if (!item)
2656 re->dst_entry = insert_stage_data(opt->repo,
2657 re->pair->two->path,
2658 o_tree, a_tree, b_tree, entries);
2659 else
2660 re->dst_entry = item->util;
2661 item = string_list_insert(renames, pair->one->path);
2662 item->util = re;
2663 if (new_path)
2664 apply_directory_rename_modifications(opt, pair, new_path,
2665 re, tree, o_tree,
2666 a_tree, b_tree,
2667 entries);
2668 }
2669
2670 hashmap_for_each_entry(&collisions, &iter, e,
2671 ent /* member name */) {
2672 free(e->target_file);
2673 string_list_clear(&e->source_files, 0);
2674 }
2675 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2676 return renames;
2677 }
2678
2679 static int process_renames(struct merge_options *opt,
2680 struct string_list *a_renames,
2681 struct string_list *b_renames)
2682 {
2683 int clean_merge = 1, i, j;
2684 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2685 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2686 const struct rename *sre;
2687
2688 /*
2689 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2690 * string_list one-by-one, but O(n log n) to build it unsorted and
2691 * then sort it. Note that as we build the list, we do not need to
2692 * check if the existing destination path is already in the list,
2693 * because the structure of diffcore_rename guarantees we won't
2694 * have duplicates.
2695 */
2696 for (i = 0; i < a_renames->nr; i++) {
2697 sre = a_renames->items[i].util;
2698 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2699 = (void *)sre;
2700 }
2701 for (i = 0; i < b_renames->nr; i++) {
2702 sre = b_renames->items[i].util;
2703 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2704 = (void *)sre;
2705 }
2706
2707 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2708 struct string_list *renames1, *renames2Dst;
2709 struct rename *ren1 = NULL, *ren2 = NULL;
2710 const char *ren1_src, *ren1_dst;
2711 struct string_list_item *lookup;
2712
2713 if (i >= a_renames->nr) {
2714 ren2 = b_renames->items[j++].util;
2715 } else if (j >= b_renames->nr) {
2716 ren1 = a_renames->items[i++].util;
2717 } else {
2718 int compare = strcmp(a_renames->items[i].string,
2719 b_renames->items[j].string);
2720 if (compare <= 0)
2721 ren1 = a_renames->items[i++].util;
2722 if (compare >= 0)
2723 ren2 = b_renames->items[j++].util;
2724 }
2725
2726 /* TODO: refactor, so that 1/2 are not needed */
2727 if (ren1) {
2728 renames1 = a_renames;
2729 renames2Dst = &b_by_dst;
2730 } else {
2731 renames1 = b_renames;
2732 renames2Dst = &a_by_dst;
2733 SWAP(ren2, ren1);
2734 }
2735
2736 if (ren1->processed)
2737 continue;
2738 ren1->processed = 1;
2739 ren1->dst_entry->processed = 1;
2740 /* BUG: We should only mark src_entry as processed if we
2741 * are not dealing with a rename + add-source case.
2742 */
2743 ren1->src_entry->processed = 1;
2744
2745 ren1_src = ren1->pair->one->path;
2746 ren1_dst = ren1->pair->two->path;
2747
2748 if (ren2) {
2749 /* One file renamed on both sides */
2750 const char *ren2_src = ren2->pair->one->path;
2751 const char *ren2_dst = ren2->pair->two->path;
2752 enum rename_type rename_type;
2753 if (strcmp(ren1_src, ren2_src) != 0)
2754 BUG("ren1_src != ren2_src");
2755 ren2->dst_entry->processed = 1;
2756 ren2->processed = 1;
2757 if (strcmp(ren1_dst, ren2_dst) != 0) {
2758 rename_type = RENAME_ONE_FILE_TO_TWO;
2759 clean_merge = 0;
2760 } else {
2761 rename_type = RENAME_ONE_FILE_TO_ONE;
2762 /* BUG: We should only remove ren1_src in
2763 * the base stage (think of rename +
2764 * add-source cases).
2765 */
2766 remove_file(opt, 1, ren1_src, 1);
2767 update_entry(ren1->dst_entry,
2768 ren1->pair->one,
2769 ren1->pair->two,
2770 ren2->pair->two);
2771 }
2772 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2773 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2774 /* Two different files renamed to the same thing */
2775 char *ren2_dst;
2776 ren2 = lookup->util;
2777 ren2_dst = ren2->pair->two->path;
2778 if (strcmp(ren1_dst, ren2_dst) != 0)
2779 BUG("ren1_dst != ren2_dst");
2780
2781 clean_merge = 0;
2782 ren2->processed = 1;
2783 /*
2784 * BUG: We should only mark src_entry as processed
2785 * if we are not dealing with a rename + add-source
2786 * case.
2787 */
2788 ren2->src_entry->processed = 1;
2789
2790 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2791 opt, ren1, ren2);
2792 } else {
2793 /* Renamed in 1, maybe changed in 2 */
2794 /* we only use sha1 and mode of these */
2795 struct diff_filespec src_other, dst_other;
2796 int try_merge;
2797
2798 /*
2799 * unpack_trees loads entries from common-commit
2800 * into stage 1, from head-commit into stage 2, and
2801 * from merge-commit into stage 3. We keep track
2802 * of which side corresponds to the rename.
2803 */
2804 int renamed_stage = a_renames == renames1 ? 2 : 3;
2805 int other_stage = a_renames == renames1 ? 3 : 2;
2806
2807 /*
2808 * Directory renames have a funny corner case...
2809 */
2810 int renamed_to_self = !strcmp(ren1_src, ren1_dst);
2811
2812 /* BUG: We should only remove ren1_src in the base
2813 * stage and in other_stage (think of rename +
2814 * add-source case).
2815 */
2816 if (!renamed_to_self)
2817 remove_file(opt, 1, ren1_src,
2818 renamed_stage == 2 ||
2819 !was_tracked(opt, ren1_src));
2820
2821 oidcpy(&src_other.oid,
2822 &ren1->src_entry->stages[other_stage].oid);
2823 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2824 oidcpy(&dst_other.oid,
2825 &ren1->dst_entry->stages[other_stage].oid);
2826 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2827 try_merge = 0;
2828
2829 if (oideq(&src_other.oid, null_oid()) &&
2830 ren1->dir_rename_original_type == 'A') {
2831 setup_rename_conflict_info(RENAME_VIA_DIR,
2832 opt, ren1, NULL);
2833 } else if (renamed_to_self) {
2834 setup_rename_conflict_info(RENAME_NORMAL,
2835 opt, ren1, NULL);
2836 } else if (oideq(&src_other.oid, null_oid())) {
2837 setup_rename_conflict_info(RENAME_DELETE,
2838 opt, ren1, NULL);
2839 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2840 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2841 /*
2842 * Added file on the other side identical to
2843 * the file being renamed: clean merge.
2844 * Also, there is no need to overwrite the
2845 * file already in the working copy, so call
2846 * update_file_flags() instead of
2847 * update_file().
2848 */
2849 if (update_file_flags(opt,
2850 ren1->pair->two,
2851 ren1_dst,
2852 1, /* update_cache */
2853 0 /* update_wd */))
2854 clean_merge = -1;
2855 } else if (!oideq(&dst_other.oid, null_oid())) {
2856 /*
2857 * Probably not a clean merge, but it's
2858 * premature to set clean_merge to 0 here,
2859 * because if the rename merges cleanly and
2860 * the merge exactly matches the newly added
2861 * file, then the merge will be clean.
2862 */
2863 setup_rename_conflict_info(RENAME_ADD,
2864 opt, ren1, NULL);
2865 } else
2866 try_merge = 1;
2867
2868 if (clean_merge < 0)
2869 goto cleanup_and_return;
2870 if (try_merge) {
2871 struct diff_filespec *o, *a, *b;
2872 src_other.path = (char *)ren1_src;
2873
2874 o = ren1->pair->one;
2875 if (a_renames == renames1) {
2876 a = ren1->pair->two;
2877 b = &src_other;
2878 } else {
2879 b = ren1->pair->two;
2880 a = &src_other;
2881 }
2882 update_entry(ren1->dst_entry, o, a, b);
2883 setup_rename_conflict_info(RENAME_NORMAL,
2884 opt, ren1, NULL);
2885 }
2886 }
2887 }
2888 cleanup_and_return:
2889 string_list_clear(&a_by_dst, 0);
2890 string_list_clear(&b_by_dst, 0);
2891
2892 return clean_merge;
2893 }
2894
2895 struct rename_info {
2896 struct string_list *head_renames;
2897 struct string_list *merge_renames;
2898 };
2899
2900 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2901 struct hashmap *dir_renames)
2902 {
2903 struct hashmap_iter iter;
2904 struct dir_rename_entry *e;
2905
2906 hashmap_for_each_entry(dir_renames, &iter, e,
2907 ent /* member name */) {
2908 free(e->dir);
2909 strbuf_release(&e->new_dir);
2910 /* possible_new_dirs already cleared in get_directory_renames */
2911 }
2912 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2913 free(dir_renames);
2914
2915 free(pairs->queue);
2916 free(pairs);
2917 }
2918
2919 static int detect_and_process_renames(struct merge_options *opt,
2920 struct tree *common,
2921 struct tree *head,
2922 struct tree *merge,
2923 struct string_list *entries,
2924 struct rename_info *ri)
2925 {
2926 struct diff_queue_struct *head_pairs, *merge_pairs;
2927 struct hashmap *dir_re_head, *dir_re_merge;
2928 int clean = 1;
2929
2930 ri->head_renames = NULL;
2931 ri->merge_renames = NULL;
2932
2933 if (!merge_detect_rename(opt))
2934 return 1;
2935
2936 head_pairs = get_diffpairs(opt, common, head);
2937 merge_pairs = get_diffpairs(opt, common, merge);
2938
2939 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2940 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2941 !opt->priv->call_depth)) {
2942 dir_re_head = get_directory_renames(head_pairs);
2943 dir_re_merge = get_directory_renames(merge_pairs);
2944
2945 handle_directory_level_conflicts(opt,
2946 dir_re_head, head,
2947 dir_re_merge, merge);
2948 } else {
2949 dir_re_head = xmalloc(sizeof(*dir_re_head));
2950 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2951 dir_rename_init(dir_re_head);
2952 dir_rename_init(dir_re_merge);
2953 }
2954
2955 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2956 dir_re_merge, dir_re_head, head,
2957 common, head, merge, entries,
2958 &clean);
2959 if (clean < 0)
2960 goto cleanup;
2961 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2962 dir_re_head, dir_re_merge, merge,
2963 common, head, merge, entries,
2964 &clean);
2965 if (clean < 0)
2966 goto cleanup;
2967 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2968
2969 cleanup:
2970 /*
2971 * Some cleanup is deferred until cleanup_renames() because the
2972 * data structures are still needed and referenced in
2973 * process_entry(). But there are a few things we can free now.
2974 */
2975 initial_cleanup_rename(head_pairs, dir_re_head);
2976 initial_cleanup_rename(merge_pairs, dir_re_merge);
2977
2978 return clean;
2979 }
2980
2981 static void final_cleanup_rename(struct string_list *rename)
2982 {
2983 const struct rename *re;
2984 int i;
2985
2986 if (rename == NULL)
2987 return;
2988
2989 for (i = 0; i < rename->nr; i++) {
2990 re = rename->items[i].util;
2991 diff_free_filepair(re->pair);
2992 }
2993 string_list_clear(rename, 1);
2994 free(rename);
2995 }
2996
2997 static void final_cleanup_renames(struct rename_info *re_info)
2998 {
2999 final_cleanup_rename(re_info->head_renames);
3000 final_cleanup_rename(re_info->merge_renames);
3001 }
3002
3003 static int read_oid_strbuf(struct merge_options *opt,
3004 const struct object_id *oid,
3005 struct strbuf *dst)
3006 {
3007 void *buf;
3008 enum object_type type;
3009 unsigned long size;
3010 buf = read_object_file(oid, &type, &size);
3011 if (!buf)
3012 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3013 if (type != OBJ_BLOB) {
3014 free(buf);
3015 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3016 }
3017 strbuf_attach(dst, buf, size, size + 1);
3018 return 0;
3019 }
3020
3021 static int blob_unchanged(struct merge_options *opt,
3022 const struct diff_filespec *o,
3023 const struct diff_filespec *a,
3024 int renormalize, const char *path)
3025 {
3026 struct strbuf obuf = STRBUF_INIT;
3027 struct strbuf abuf = STRBUF_INIT;
3028 int ret = 0; /* assume changed for safety */
3029 struct index_state *idx = opt->repo->index;
3030
3031 if (a->mode != o->mode)
3032 return 0;
3033 if (oideq(&o->oid, &a->oid))
3034 return 1;
3035 if (!renormalize)
3036 return 0;
3037
3038 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3039 read_oid_strbuf(opt, &a->oid, &abuf))
3040 goto error_return;
3041 /*
3042 * Note: binary | is used so that both renormalizations are
3043 * performed. Comparison can be skipped if both files are
3044 * unchanged since their sha1s have already been compared.
3045 */
3046 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3047 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3048 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3049
3050 error_return:
3051 strbuf_release(&obuf);
3052 strbuf_release(&abuf);
3053 return ret;
3054 }
3055
3056 static int handle_modify_delete(struct merge_options *opt,
3057 const char *path,
3058 const struct diff_filespec *o,
3059 const struct diff_filespec *a,
3060 const struct diff_filespec *b)
3061 {
3062 const char *modify_branch, *delete_branch;
3063 const struct diff_filespec *changed;
3064
3065 if (is_valid(a)) {
3066 modify_branch = opt->branch1;
3067 delete_branch = opt->branch2;
3068 changed = a;
3069 } else {
3070 modify_branch = opt->branch2;
3071 delete_branch = opt->branch1;
3072 changed = b;
3073 }
3074
3075 return handle_change_delete(opt,
3076 path, NULL,
3077 o, changed,
3078 modify_branch, delete_branch,
3079 _("modify"), _("modified"));
3080 }
3081
3082 static int handle_content_merge(struct merge_file_info *mfi,
3083 struct merge_options *opt,
3084 const char *path,
3085 int is_dirty,
3086 const struct diff_filespec *o,
3087 const struct diff_filespec *a,
3088 const struct diff_filespec *b,
3089 struct rename_conflict_info *ci)
3090 {
3091 const char *reason = _("content");
3092 unsigned df_conflict_remains = 0;
3093
3094 if (!is_valid(o))
3095 reason = _("add/add");
3096
3097 assert(o->path && a->path && b->path);
3098 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3099 S_ISGITLINK(ci->ren1->pair->two->mode)))
3100 df_conflict_remains = 1;
3101
3102 if (merge_mode_and_contents(opt, o, a, b, path,
3103 opt->branch1, opt->branch2,
3104 opt->priv->call_depth * 2, mfi))
3105 return -1;
3106
3107 /*
3108 * We can skip updating the working tree file iff:
3109 * a) The merge is clean
3110 * b) The merge matches what was in HEAD (content, mode, pathname)
3111 * c) The target path is usable (i.e. not involved in D/F conflict)
3112 */
3113 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3114 !df_conflict_remains) {
3115 int pos;
3116 struct cache_entry *ce;
3117
3118 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3119 if (add_cacheinfo(opt, &mfi->blob, path,
3120 0, (!opt->priv->call_depth && !is_dirty), 0))
3121 return -1;
3122 /*
3123 * However, add_cacheinfo() will delete the old cache entry
3124 * and add a new one. We need to copy over any skip_worktree
3125 * flag to avoid making the file appear as if it were
3126 * deleted by the user.
3127 */
3128 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3129 ce = opt->priv->orig_index.cache[pos];
3130 if (ce_skip_worktree(ce)) {
3131 pos = index_name_pos(opt->repo->index, path, strlen(path));
3132 ce = opt->repo->index->cache[pos];
3133 ce->ce_flags |= CE_SKIP_WORKTREE;
3134 }
3135 return mfi->clean;
3136 }
3137
3138 if (!mfi->clean) {
3139 if (S_ISGITLINK(mfi->blob.mode))
3140 reason = _("submodule");
3141 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3142 reason, path);
3143 if (ci && !df_conflict_remains)
3144 if (update_stages(opt, path, o, a, b))
3145 return -1;
3146 }
3147
3148 if (df_conflict_remains || is_dirty) {
3149 char *new_path;
3150 if (opt->priv->call_depth) {
3151 remove_file_from_index(opt->repo->index, path);
3152 } else {
3153 if (!mfi->clean) {
3154 if (update_stages(opt, path, o, a, b))
3155 return -1;
3156 } else {
3157 int file_from_stage2 = was_tracked(opt, path);
3158
3159 if (update_stages(opt, path, NULL,
3160 file_from_stage2 ? &mfi->blob : NULL,
3161 file_from_stage2 ? NULL : &mfi->blob))
3162 return -1;
3163 }
3164
3165 }
3166 new_path = unique_path(opt, path, ci->ren1->branch);
3167 if (is_dirty) {
3168 output(opt, 1, _("Refusing to lose dirty file at %s"),
3169 path);
3170 }
3171 output(opt, 1, _("Adding as %s instead"), new_path);
3172 if (update_file(opt, 0, &mfi->blob, new_path)) {
3173 free(new_path);
3174 return -1;
3175 }
3176 free(new_path);
3177 mfi->clean = 0;
3178 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3179 return -1;
3180 return !is_dirty && mfi->clean;
3181 }
3182
3183 static int handle_rename_normal(struct merge_options *opt,
3184 const char *path,
3185 const struct diff_filespec *o,
3186 const struct diff_filespec *a,
3187 const struct diff_filespec *b,
3188 struct rename_conflict_info *ci)
3189 {
3190 struct rename *ren = ci->ren1;
3191 struct merge_file_info mfi;
3192 int clean;
3193
3194 /* Merge the content and write it out */
3195 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3196 o, a, b, ci);
3197
3198 if (clean &&
3199 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3200 ren->dir_rename_original_dest) {
3201 if (update_stages(opt, path,
3202 &mfi.blob, &mfi.blob, &mfi.blob))
3203 return -1;
3204 clean = 0; /* not clean, but conflicted */
3205 }
3206 return clean;
3207 }
3208
3209 static void dir_rename_warning(const char *msg,
3210 int is_add,
3211 int clean,
3212 struct merge_options *opt,
3213 struct rename *ren)
3214 {
3215 const char *other_branch;
3216 other_branch = (ren->branch == opt->branch1 ?
3217 opt->branch2 : opt->branch1);
3218 if (is_add) {
3219 output(opt, clean ? 2 : 1, msg,
3220 ren->pair->one->path, ren->branch,
3221 other_branch, ren->pair->two->path);
3222 return;
3223 }
3224 output(opt, clean ? 2 : 1, msg,
3225 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3226 other_branch, ren->pair->two->path);
3227 }
3228 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3229 struct rename *ren)
3230 {
3231 const char *msg;
3232 int clean = 1, is_add;
3233
3234 if (!ren)
3235 return clean;
3236
3237 /* Return early if ren was not affected/created by a directory rename */
3238 if (!ren->dir_rename_original_dest)
3239 return clean;
3240
3241 /* Sanity checks */
3242 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3243 assert(ren->dir_rename_original_type == 'A' ||
3244 ren->dir_rename_original_type == 'R');
3245
3246 /* Check whether to treat directory renames as a conflict */
3247 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3248
3249 is_add = (ren->dir_rename_original_type == 'A');
3250 if (ren->dir_rename_original_type == 'A' && clean) {
3251 msg = _("Path updated: %s added in %s inside a "
3252 "directory that was renamed in %s; moving it to %s.");
3253 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3254 msg = _("CONFLICT (file location): %s added in %s "
3255 "inside a directory that was renamed in %s, "
3256 "suggesting it should perhaps be moved to %s.");
3257 } else if (ren->dir_rename_original_type == 'R' && clean) {
3258 msg = _("Path updated: %s renamed to %s in %s, inside a "
3259 "directory that was renamed in %s; moving it to %s.");
3260 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3261 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3262 "inside a directory that was renamed in %s, "
3263 "suggesting it should perhaps be moved to %s.");
3264 } else {
3265 BUG("Impossible dir_rename_original_type/clean combination");
3266 }
3267 dir_rename_warning(msg, is_add, clean, opt, ren);
3268
3269 return clean;
3270 }
3271
3272 /* Per entry merge function */
3273 static int process_entry(struct merge_options *opt,
3274 const char *path, struct stage_data *entry)
3275 {
3276 int clean_merge = 1;
3277 int normalize = opt->renormalize;
3278
3279 struct diff_filespec *o = &entry->stages[1];
3280 struct diff_filespec *a = &entry->stages[2];
3281 struct diff_filespec *b = &entry->stages[3];
3282 int o_valid = is_valid(o);
3283 int a_valid = is_valid(a);
3284 int b_valid = is_valid(b);
3285 o->path = a->path = b->path = (char*)path;
3286
3287 entry->processed = 1;
3288 if (entry->rename_conflict_info) {
3289 struct rename_conflict_info *ci = entry->rename_conflict_info;
3290 struct diff_filespec *temp;
3291 int path_clean;
3292
3293 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3294 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3295
3296 /*
3297 * For cases with a single rename, {o,a,b}->path have all been
3298 * set to the rename target path; we need to set two of these
3299 * back to the rename source.
3300 * For rename/rename conflicts, we'll manually fix paths below.
3301 */
3302 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3303 o->path = temp->path = ci->ren1->pair->one->path;
3304 if (ci->ren2) {
3305 assert(opt->branch1 == ci->ren1->branch);
3306 }
3307
3308 switch (ci->rename_type) {
3309 case RENAME_NORMAL:
3310 case RENAME_ONE_FILE_TO_ONE:
3311 clean_merge = handle_rename_normal(opt, path, o, a, b,
3312 ci);
3313 break;
3314 case RENAME_VIA_DIR:
3315 clean_merge = handle_rename_via_dir(opt, ci);
3316 break;
3317 case RENAME_ADD:
3318 /*
3319 * Probably unclean merge, but if the renamed file
3320 * merges cleanly and the result can then be
3321 * two-way merged cleanly with the added file, I
3322 * guess it's a clean merge?
3323 */
3324 clean_merge = handle_rename_add(opt, ci);
3325 break;
3326 case RENAME_DELETE:
3327 clean_merge = 0;
3328 if (handle_rename_delete(opt, ci))
3329 clean_merge = -1;
3330 break;
3331 case RENAME_ONE_FILE_TO_TWO:
3332 /*
3333 * Manually fix up paths; note:
3334 * ren[12]->pair->one->path are equal.
3335 */
3336 o->path = ci->ren1->pair->one->path;
3337 a->path = ci->ren1->pair->two->path;
3338 b->path = ci->ren2->pair->two->path;
3339
3340 clean_merge = 0;
3341 if (handle_rename_rename_1to2(opt, ci))
3342 clean_merge = -1;
3343 break;
3344 case RENAME_TWO_FILES_TO_ONE:
3345 /*
3346 * Manually fix up paths; note,
3347 * ren[12]->pair->two->path are actually equal.
3348 */
3349 o->path = NULL;
3350 a->path = ci->ren1->pair->two->path;
3351 b->path = ci->ren2->pair->two->path;
3352
3353 /*
3354 * Probably unclean merge, but if the two renamed
3355 * files merge cleanly and the two resulting files
3356 * can then be two-way merged cleanly, I guess it's
3357 * a clean merge?
3358 */
3359 clean_merge = handle_rename_rename_2to1(opt, ci);
3360 break;
3361 default:
3362 entry->processed = 0;
3363 break;
3364 }
3365 if (path_clean < clean_merge)
3366 clean_merge = path_clean;
3367 } else if (o_valid && (!a_valid || !b_valid)) {
3368 /* Case A: Deleted in one */
3369 if ((!a_valid && !b_valid) ||
3370 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3371 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3372 /* Deleted in both or deleted in one and
3373 * unchanged in the other */
3374 if (a_valid)
3375 output(opt, 2, _("Removing %s"), path);
3376 /* do not touch working file if it did not exist */
3377 remove_file(opt, 1, path, !a_valid);
3378 } else {
3379 /* Modify/delete; deleted side may have put a directory in the way */
3380 clean_merge = 0;
3381 if (handle_modify_delete(opt, path, o, a, b))
3382 clean_merge = -1;
3383 }
3384 } else if ((!o_valid && a_valid && !b_valid) ||
3385 (!o_valid && !a_valid && b_valid)) {
3386 /* Case B: Added in one. */
3387 /* [nothing|directory] -> ([nothing|directory], file) */
3388
3389 const char *add_branch;
3390 const char *other_branch;
3391 const char *conf;
3392 const struct diff_filespec *contents;
3393
3394 if (a_valid) {
3395 add_branch = opt->branch1;
3396 other_branch = opt->branch2;
3397 contents = a;
3398 conf = _("file/directory");
3399 } else {
3400 add_branch = opt->branch2;
3401 other_branch = opt->branch1;
3402 contents = b;
3403 conf = _("directory/file");
3404 }
3405 if (dir_in_way(opt->repo->index, path,
3406 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3407 0)) {
3408 char *new_path = unique_path(opt, path, add_branch);
3409 clean_merge = 0;
3410 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3411 "Adding %s as %s"),
3412 conf, path, other_branch, path, new_path);
3413 if (update_file(opt, 0, contents, new_path))
3414 clean_merge = -1;
3415 else if (opt->priv->call_depth)
3416 remove_file_from_index(opt->repo->index, path);
3417 free(new_path);
3418 } else {
3419 output(opt, 2, _("Adding %s"), path);
3420 /* do not overwrite file if already present */
3421 if (update_file_flags(opt, contents, path, 1, !a_valid))
3422 clean_merge = -1;
3423 }
3424 } else if (a_valid && b_valid) {
3425 if (!o_valid) {
3426 /* Case C: Added in both (check for same permissions) */
3427 output(opt, 1,
3428 _("CONFLICT (add/add): Merge conflict in %s"),
3429 path);
3430 clean_merge = handle_file_collision(opt,
3431 path, NULL, NULL,
3432 opt->branch1,
3433 opt->branch2,
3434 a, b);
3435 } else {
3436 /* case D: Modified in both, but differently. */
3437 struct merge_file_info mfi;
3438 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3439 clean_merge = handle_content_merge(&mfi, opt, path,
3440 is_dirty,
3441 o, a, b, NULL);
3442 }
3443 } else if (!o_valid && !a_valid && !b_valid) {
3444 /*
3445 * this entry was deleted altogether. a_mode == 0 means
3446 * we had that path and want to actively remove it.
3447 */
3448 remove_file(opt, 1, path, !a->mode);
3449 } else
3450 BUG("fatal merge failure, shouldn't happen.");
3451
3452 return clean_merge;
3453 }
3454
3455 static int merge_trees_internal(struct merge_options *opt,
3456 struct tree *head,
3457 struct tree *merge,
3458 struct tree *merge_base,
3459 struct tree **result)
3460 {
3461 struct index_state *istate = opt->repo->index;
3462 int code, clean;
3463
3464 if (opt->subtree_shift) {
3465 merge = shift_tree_object(opt->repo, head, merge,
3466 opt->subtree_shift);
3467 merge_base = shift_tree_object(opt->repo, head, merge_base,
3468 opt->subtree_shift);
3469 }
3470
3471 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3472 output(opt, 0, _("Already up to date."));
3473 *result = head;
3474 return 1;
3475 }
3476
3477 code = unpack_trees_start(opt, merge_base, head, merge);
3478
3479 if (code != 0) {
3480 if (show(opt, 4) || opt->priv->call_depth)
3481 err(opt, _("merging of trees %s and %s failed"),
3482 oid_to_hex(&head->object.oid),
3483 oid_to_hex(&merge->object.oid));
3484 unpack_trees_finish(opt);
3485 return -1;
3486 }
3487
3488 if (unmerged_index(istate)) {
3489 struct string_list *entries;
3490 struct rename_info re_info;
3491 int i;
3492 /*
3493 * Only need the hashmap while processing entries, so
3494 * initialize it here and free it when we are done running
3495 * through the entries. Keeping it in the merge_options as
3496 * opposed to decaring a local hashmap is for convenience
3497 * so that we don't have to pass it to around.
3498 */
3499 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3500 NULL, 512);
3501 get_files_dirs(opt, head);
3502 get_files_dirs(opt, merge);
3503
3504 entries = get_unmerged(opt->repo->index);
3505 clean = detect_and_process_renames(opt, merge_base, head, merge,
3506 entries, &re_info);
3507 record_df_conflict_files(opt, entries);
3508 if (clean < 0)
3509 goto cleanup;
3510 for (i = entries->nr-1; 0 <= i; i--) {
3511 const char *path = entries->items[i].string;
3512 struct stage_data *e = entries->items[i].util;
3513 if (!e->processed) {
3514 int ret = process_entry(opt, path, e);
3515 if (!ret)
3516 clean = 0;
3517 else if (ret < 0) {
3518 clean = ret;
3519 goto cleanup;
3520 }
3521 }
3522 }
3523 for (i = 0; i < entries->nr; i++) {
3524 struct stage_data *e = entries->items[i].util;
3525 if (!e->processed)
3526 BUG("unprocessed path??? %s",
3527 entries->items[i].string);
3528 }
3529
3530 cleanup:
3531 final_cleanup_renames(&re_info);
3532
3533 string_list_clear(entries, 1);
3534 free(entries);
3535
3536 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3537 struct path_hashmap_entry, e);
3538
3539 if (clean < 0) {
3540 unpack_trees_finish(opt);
3541 return clean;
3542 }
3543 }
3544 else
3545 clean = 1;
3546
3547 unpack_trees_finish(opt);
3548
3549 if (opt->priv->call_depth &&
3550 !(*result = write_in_core_index_as_tree(opt->repo)))
3551 return -1;
3552
3553 return clean;
3554 }
3555
3556 /*
3557 * Merge the commits h1 and h2, returning a flag (int) indicating the
3558 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3559 * virtual commit and write its location to *result.
3560 */
3561 static int merge_recursive_internal(struct merge_options *opt,
3562 struct commit *h1,
3563 struct commit *h2,
3564 struct commit_list *merge_bases,
3565 struct commit **result)
3566 {
3567 struct commit_list *iter;
3568 struct commit *merged_merge_bases;
3569 struct tree *result_tree;
3570 int clean;
3571 const char *ancestor_name;
3572 struct strbuf merge_base_abbrev = STRBUF_INIT;
3573
3574 if (show(opt, 4)) {
3575 output(opt, 4, _("Merging:"));
3576 output_commit_title(opt, h1);
3577 output_commit_title(opt, h2);
3578 }
3579
3580 if (!merge_bases) {
3581 merge_bases = get_merge_bases(h1, h2);
3582 merge_bases = reverse_commit_list(merge_bases);
3583 }
3584
3585 if (show(opt, 5)) {
3586 unsigned cnt = commit_list_count(merge_bases);
3587
3588 output(opt, 5, Q_("found %u common ancestor:",
3589 "found %u common ancestors:", cnt), cnt);
3590 for (iter = merge_bases; iter; iter = iter->next)
3591 output_commit_title(opt, iter->item);
3592 }
3593
3594 merged_merge_bases = pop_commit(&merge_bases);
3595 if (merged_merge_bases == NULL) {
3596 /* if there is no common ancestor, use an empty tree */
3597 struct tree *tree;
3598
3599 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3600 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3601 "ancestor");
3602 ancestor_name = "empty tree";
3603 } else if (opt->ancestor && !opt->priv->call_depth) {
3604 ancestor_name = opt->ancestor;
3605 } else if (merge_bases) {
3606 ancestor_name = "merged common ancestors";
3607 } else {
3608 strbuf_add_unique_abbrev(&merge_base_abbrev,
3609 &merged_merge_bases->object.oid,
3610 DEFAULT_ABBREV);
3611 ancestor_name = merge_base_abbrev.buf;
3612 }
3613
3614 for (iter = merge_bases; iter; iter = iter->next) {
3615 const char *saved_b1, *saved_b2;
3616 opt->priv->call_depth++;
3617 /*
3618 * When the merge fails, the result contains files
3619 * with conflict markers. The cleanness flag is
3620 * ignored (unless indicating an error), it was never
3621 * actually used, as result of merge_trees has always
3622 * overwritten it: the committed "conflicts" were
3623 * already resolved.
3624 */
3625 discard_index(opt->repo->index);
3626 saved_b1 = opt->branch1;
3627 saved_b2 = opt->branch2;
3628 opt->branch1 = "Temporary merge branch 1";
3629 opt->branch2 = "Temporary merge branch 2";
3630 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3631 NULL, &merged_merge_bases) < 0)
3632 return -1;
3633 opt->branch1 = saved_b1;
3634 opt->branch2 = saved_b2;
3635 opt->priv->call_depth--;
3636
3637 if (!merged_merge_bases)
3638 return err(opt, _("merge returned no commit"));
3639 }
3640
3641 /*
3642 * FIXME: Since merge_recursive_internal() is only ever called by
3643 * places that ensure the index is loaded first
3644 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3645 * case where the merge base was unique that means when we get here
3646 * we immediately discard the index and re-read it, which is a
3647 * complete waste of time. We should only be discarding and
3648 * re-reading if we were forced to recurse.
3649 */
3650 discard_index(opt->repo->index);
3651 if (!opt->priv->call_depth)
3652 repo_read_index(opt->repo);
3653
3654 opt->ancestor = ancestor_name;
3655 clean = merge_trees_internal(opt,
3656 repo_get_commit_tree(opt->repo, h1),
3657 repo_get_commit_tree(opt->repo, h2),
3658 repo_get_commit_tree(opt->repo,
3659 merged_merge_bases),
3660 &result_tree);
3661 strbuf_release(&merge_base_abbrev);
3662 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3663 if (clean < 0) {
3664 flush_output(opt);
3665 return clean;
3666 }
3667
3668 if (opt->priv->call_depth) {
3669 *result = make_virtual_commit(opt->repo, result_tree,
3670 "merged tree");
3671 commit_list_insert(h1, &(*result)->parents);
3672 commit_list_insert(h2, &(*result)->parents->next);
3673 }
3674 return clean;
3675 }
3676
3677 static int merge_start(struct merge_options *opt, struct tree *head)
3678 {
3679 struct strbuf sb = STRBUF_INIT;
3680
3681 /* Sanity checks on opt */
3682 assert(opt->repo);
3683
3684 assert(opt->branch1 && opt->branch2);
3685
3686 assert(opt->detect_renames >= -1 &&
3687 opt->detect_renames <= DIFF_DETECT_COPY);
3688 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3689 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3690 assert(opt->rename_limit >= -1);
3691 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3692 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3693
3694 assert(opt->xdl_opts >= 0);
3695 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3696 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3697
3698 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3699 assert(opt->buffer_output <= 2);
3700 assert(opt->obuf.len == 0);
3701
3702 assert(opt->priv == NULL);
3703
3704 /* Sanity check on repo state; index must match head */
3705 if (repo_index_has_changes(opt->repo, head, &sb)) {
3706 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3707 sb.buf);
3708 strbuf_release(&sb);
3709 return -1;
3710 }
3711
3712 CALLOC_ARRAY(opt->priv, 1);
3713 string_list_init_dup(&opt->priv->df_conflict_file_set);
3714 return 0;
3715 }
3716
3717 static void merge_finalize(struct merge_options *opt)
3718 {
3719 flush_output(opt);
3720 if (!opt->priv->call_depth && opt->buffer_output < 2)
3721 strbuf_release(&opt->obuf);
3722 if (show(opt, 2))
3723 diff_warn_rename_limit("merge.renamelimit",
3724 opt->priv->needed_rename_limit, 0);
3725 FREE_AND_NULL(opt->priv);
3726 }
3727
3728 int merge_trees(struct merge_options *opt,
3729 struct tree *head,
3730 struct tree *merge,
3731 struct tree *merge_base)
3732 {
3733 int clean;
3734 struct tree *ignored;
3735
3736 assert(opt->ancestor != NULL);
3737
3738 if (merge_start(opt, head))
3739 return -1;
3740 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3741 merge_finalize(opt);
3742
3743 return clean;
3744 }
3745
3746 int merge_recursive(struct merge_options *opt,
3747 struct commit *h1,
3748 struct commit *h2,
3749 struct commit_list *merge_bases,
3750 struct commit **result)
3751 {
3752 int clean;
3753
3754 assert(opt->ancestor == NULL ||
3755 !strcmp(opt->ancestor, "constructed merge base"));
3756
3757 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3758 return -1;
3759 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3760 merge_finalize(opt);
3761
3762 return clean;
3763 }
3764
3765 static struct commit *get_ref(struct repository *repo,
3766 const struct object_id *oid,
3767 const char *name)
3768 {
3769 struct object *object;
3770
3771 object = deref_tag(repo, parse_object(repo, oid),
3772 name, strlen(name));
3773 if (!object)
3774 return NULL;
3775 if (object->type == OBJ_TREE)
3776 return make_virtual_commit(repo, (struct tree*)object, name);
3777 if (object->type != OBJ_COMMIT)
3778 return NULL;
3779 if (parse_commit((struct commit *)object))
3780 return NULL;
3781 return (struct commit *)object;
3782 }
3783
3784 int merge_recursive_generic(struct merge_options *opt,
3785 const struct object_id *head,
3786 const struct object_id *merge,
3787 int num_merge_bases,
3788 const struct object_id **merge_bases,
3789 struct commit **result)
3790 {
3791 int clean;
3792 struct lock_file lock = LOCK_INIT;
3793 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3794 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3795 struct commit_list *ca = NULL;
3796
3797 if (merge_bases) {
3798 int i;
3799 for (i = 0; i < num_merge_bases; ++i) {
3800 struct commit *base;
3801 if (!(base = get_ref(opt->repo, merge_bases[i],
3802 oid_to_hex(merge_bases[i]))))
3803 return err(opt, _("Could not parse object '%s'"),
3804 oid_to_hex(merge_bases[i]));
3805 commit_list_insert(base, &ca);
3806 }
3807 if (num_merge_bases == 1)
3808 opt->ancestor = "constructed merge base";
3809 }
3810
3811 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3812 clean = merge_recursive(opt, head_commit, next_commit, ca,
3813 result);
3814 if (clean < 0) {
3815 rollback_lock_file(&lock);
3816 return clean;
3817 }
3818
3819 if (write_locked_index(opt->repo->index, &lock,
3820 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3821 return err(opt, _("Unable to write index."));
3822
3823 return clean ? 0 : 1;
3824 }
3825
3826 static void merge_recursive_config(struct merge_options *opt)
3827 {
3828 char *value = NULL;
3829 int renormalize = 0;
3830 git_config_get_int("merge.verbosity", &opt->verbosity);
3831 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3832 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3833 git_config_get_bool("merge.renormalize", &renormalize);
3834 opt->renormalize = renormalize;
3835 if (!git_config_get_string("diff.renames", &value)) {
3836 opt->detect_renames = git_config_rename("diff.renames", value);
3837 free(value);
3838 }
3839 if (!git_config_get_string("merge.renames", &value)) {
3840 opt->detect_renames = git_config_rename("merge.renames", value);
3841 free(value);
3842 }
3843 if (!git_config_get_string("merge.directoryrenames", &value)) {
3844 int boolval = git_parse_maybe_bool(value);
3845 if (0 <= boolval) {
3846 opt->detect_directory_renames = boolval ?
3847 MERGE_DIRECTORY_RENAMES_TRUE :
3848 MERGE_DIRECTORY_RENAMES_NONE;
3849 } else if (!strcasecmp(value, "conflict")) {
3850 opt->detect_directory_renames =
3851 MERGE_DIRECTORY_RENAMES_CONFLICT;
3852 } /* avoid erroring on values from future versions of git */
3853 free(value);
3854 }
3855 git_config(git_xmerge_config, NULL);
3856 }
3857
3858 void init_merge_options(struct merge_options *opt,
3859 struct repository *repo)
3860 {
3861 const char *merge_verbosity;
3862 memset(opt, 0, sizeof(struct merge_options));
3863
3864 opt->repo = repo;
3865
3866 opt->detect_renames = -1;
3867 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3868 opt->rename_limit = -1;
3869
3870 opt->verbosity = 2;
3871 opt->buffer_output = 1;
3872 strbuf_init(&opt->obuf, 0);
3873
3874 opt->renormalize = 0;
3875
3876 merge_recursive_config(opt);
3877 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3878 if (merge_verbosity)
3879 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3880 if (opt->verbosity >= 5)
3881 opt->buffer_output = 0;
3882 }
3883
3884 int parse_merge_opt(struct merge_options *opt, const char *s)
3885 {
3886 const char *arg;
3887
3888 if (!s || !*s)
3889 return -1;
3890 if (!strcmp(s, "ours"))
3891 opt->recursive_variant = MERGE_VARIANT_OURS;
3892 else if (!strcmp(s, "theirs"))
3893 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3894 else if (!strcmp(s, "subtree"))
3895 opt->subtree_shift = "";
3896 else if (skip_prefix(s, "subtree=", &arg))
3897 opt->subtree_shift = arg;
3898 else if (!strcmp(s, "patience"))
3899 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3900 else if (!strcmp(s, "histogram"))
3901 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3902 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3903 long value = parse_algorithm_value(arg);
3904 if (value < 0)
3905 return -1;
3906 /* clear out previous settings */
3907 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3908 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3909 opt->xdl_opts |= value;
3910 }
3911 else if (!strcmp(s, "ignore-space-change"))
3912 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3913 else if (!strcmp(s, "ignore-all-space"))
3914 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3915 else if (!strcmp(s, "ignore-space-at-eol"))
3916 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3917 else if (!strcmp(s, "ignore-cr-at-eol"))
3918 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3919 else if (!strcmp(s, "renormalize"))
3920 opt->renormalize = 1;
3921 else if (!strcmp(s, "no-renormalize"))
3922 opt->renormalize = 0;
3923 else if (!strcmp(s, "no-renames"))
3924 opt->detect_renames = 0;
3925 else if (!strcmp(s, "find-renames")) {
3926 opt->detect_renames = 1;
3927 opt->rename_score = 0;
3928 }
3929 else if (skip_prefix(s, "find-renames=", &arg) ||
3930 skip_prefix(s, "rename-threshold=", &arg)) {
3931 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3932 return -1;
3933 opt->detect_renames = 1;
3934 }
3935 /*
3936 * Please update $__git_merge_strategy_options in
3937 * git-completion.bash when you add new options
3938 */
3939 else
3940 return -1;
3941 return 0;
3942 }