]> git.ipfire.org Git - thirdparty/git.git/blob - merge-recursive.c
trace2: increment event format version
[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 return fspathcmp(a->path, key ? key : b->path);
59 }
60
61 /*
62 * For dir_rename_entry, directory names are stored as a full path from the
63 * toplevel of the repository and do not include a trailing '/'. Also:
64 *
65 * dir: original name of directory being renamed
66 * non_unique_new_dir: if true, could not determine new_dir
67 * new_dir: final name of directory being renamed
68 * possible_new_dirs: temporary used to help determine new_dir; see comments
69 * in get_directory_renames() for details
70 */
71 struct dir_rename_entry {
72 struct hashmap_entry ent;
73 char *dir;
74 unsigned non_unique_new_dir:1;
75 struct strbuf new_dir;
76 struct string_list possible_new_dirs;
77 };
78
79 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
80 char *dir)
81 {
82 struct dir_rename_entry key;
83
84 if (dir == NULL)
85 return NULL;
86 hashmap_entry_init(&key.ent, strhash(dir));
87 key.dir = dir;
88 return hashmap_get_entry(hashmap, &key, ent, NULL);
89 }
90
91 static int dir_rename_cmp(const void *unused_cmp_data,
92 const struct hashmap_entry *eptr,
93 const struct hashmap_entry *entry_or_key,
94 const void *unused_keydata)
95 {
96 const struct dir_rename_entry *e1, *e2;
97
98 e1 = container_of(eptr, const struct dir_rename_entry, ent);
99 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
100
101 return strcmp(e1->dir, e2->dir);
102 }
103
104 static void dir_rename_init(struct hashmap *map)
105 {
106 hashmap_init(map, dir_rename_cmp, NULL, 0);
107 }
108
109 static void dir_rename_entry_init(struct dir_rename_entry *entry,
110 char *directory)
111 {
112 hashmap_entry_init(&entry->ent, strhash(directory));
113 entry->dir = directory;
114 entry->non_unique_new_dir = 0;
115 strbuf_init(&entry->new_dir, 0);
116 string_list_init_nodup(&entry->possible_new_dirs);
117 }
118
119 struct collision_entry {
120 struct hashmap_entry ent;
121 char *target_file;
122 struct string_list source_files;
123 unsigned reported_already:1;
124 };
125
126 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
127 char *target_file)
128 {
129 struct collision_entry key;
130
131 hashmap_entry_init(&key.ent, strhash(target_file));
132 key.target_file = target_file;
133 return hashmap_get_entry(hashmap, &key, ent, NULL);
134 }
135
136 static int collision_cmp(const void *unused_cmp_data,
137 const struct hashmap_entry *eptr,
138 const struct hashmap_entry *entry_or_key,
139 const void *unused_keydata)
140 {
141 const struct collision_entry *e1, *e2;
142
143 e1 = container_of(eptr, const struct collision_entry, ent);
144 e2 = container_of(entry_or_key, const struct collision_entry, ent);
145
146 return strcmp(e1->target_file, e2->target_file);
147 }
148
149 static void collision_init(struct hashmap *map)
150 {
151 hashmap_init(map, collision_cmp, NULL, 0);
152 }
153
154 static void flush_output(struct merge_options *opt)
155 {
156 if (opt->buffer_output < 2 && opt->obuf.len) {
157 fputs(opt->obuf.buf, stdout);
158 strbuf_reset(&opt->obuf);
159 }
160 }
161
162 __attribute__((format (printf, 2, 3)))
163 static int err(struct merge_options *opt, const char *err, ...)
164 {
165 va_list params;
166
167 if (opt->buffer_output < 2)
168 flush_output(opt);
169 else {
170 strbuf_complete(&opt->obuf, '\n');
171 strbuf_addstr(&opt->obuf, "error: ");
172 }
173 va_start(params, err);
174 strbuf_vaddf(&opt->obuf, err, params);
175 va_end(params);
176 if (opt->buffer_output > 1)
177 strbuf_addch(&opt->obuf, '\n');
178 else {
179 error("%s", opt->obuf.buf);
180 strbuf_reset(&opt->obuf);
181 }
182
183 return -1;
184 }
185
186 static struct tree *shift_tree_object(struct repository *repo,
187 struct tree *one, struct tree *two,
188 const char *subtree_shift)
189 {
190 struct object_id shifted;
191
192 if (!*subtree_shift) {
193 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
194 } else {
195 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
196 subtree_shift);
197 }
198 if (oideq(&two->object.oid, &shifted))
199 return two;
200 return lookup_tree(repo, &shifted);
201 }
202
203 static inline void set_commit_tree(struct commit *c, struct tree *t)
204 {
205 c->maybe_tree = t;
206 }
207
208 static struct commit *make_virtual_commit(struct repository *repo,
209 struct tree *tree,
210 const char *comment)
211 {
212 struct commit *commit = alloc_commit_node(repo);
213
214 set_merge_remote_desc(commit, comment, (struct object *)commit);
215 set_commit_tree(commit, tree);
216 commit->object.parsed = 1;
217 return commit;
218 }
219
220 enum rename_type {
221 RENAME_NORMAL = 0,
222 RENAME_VIA_DIR,
223 RENAME_ADD,
224 RENAME_DELETE,
225 RENAME_ONE_FILE_TO_ONE,
226 RENAME_ONE_FILE_TO_TWO,
227 RENAME_TWO_FILES_TO_ONE
228 };
229
230 /*
231 * Since we want to write the index eventually, we cannot reuse the index
232 * for these (temporary) data.
233 */
234 struct stage_data {
235 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
236 struct rename_conflict_info *rename_conflict_info;
237 unsigned processed:1;
238 };
239
240 struct rename {
241 unsigned processed:1;
242 struct diff_filepair *pair;
243 const char *branch; /* branch that the rename occurred on */
244 /*
245 * If directory rename detection affected this rename, what was its
246 * original type ('A' or 'R') and it's original destination before
247 * the directory rename (otherwise, '\0' and NULL for these two vars).
248 */
249 char dir_rename_original_type;
250 char *dir_rename_original_dest;
251 /*
252 * Purpose of src_entry and dst_entry:
253 *
254 * If 'before' is renamed to 'after' then src_entry will contain
255 * the versions of 'before' from the merge_base, HEAD, and MERGE in
256 * stages 1, 2, and 3; dst_entry will contain the respective
257 * versions of 'after' in corresponding locations. Thus, we have a
258 * total of six modes and oids, though some will be null. (Stage 0
259 * is ignored; we're interested in handling conflicts.)
260 *
261 * Since we don't turn on break-rewrites by default, neither
262 * src_entry nor dst_entry can have all three of their stages have
263 * non-null oids, meaning at most four of the six will be non-null.
264 * Also, since this is a rename, both src_entry and dst_entry will
265 * have at least one non-null oid, meaning at least two will be
266 * non-null. Of the six oids, a typical rename will have three be
267 * non-null. Only two implies a rename/delete, and four implies a
268 * rename/add.
269 */
270 struct stage_data *src_entry;
271 struct stage_data *dst_entry;
272 };
273
274 struct rename_conflict_info {
275 enum rename_type rename_type;
276 struct rename *ren1;
277 struct rename *ren2;
278 };
279
280 static inline void setup_rename_conflict_info(enum rename_type rename_type,
281 struct merge_options *opt,
282 struct rename *ren1,
283 struct rename *ren2)
284 {
285 struct rename_conflict_info *ci;
286
287 /*
288 * When we have two renames involved, it's easiest to get the
289 * correct things into stage 2 and 3, and to make sure that the
290 * content merge puts HEAD before the other branch if we just
291 * ensure that branch1 == opt->branch1. So, simply flip arguments
292 * around if we don't have that.
293 */
294 if (ren2 && ren1->branch != opt->branch1) {
295 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
296 return;
297 }
298
299 CALLOC_ARRAY(ci, 1);
300 ci->rename_type = rename_type;
301 ci->ren1 = ren1;
302 ci->ren2 = ren2;
303
304 ci->ren1->dst_entry->processed = 0;
305 ci->ren1->dst_entry->rename_conflict_info = ci;
306 if (ren2) {
307 ci->ren2->dst_entry->rename_conflict_info = ci;
308 }
309 }
310
311 static int show(struct merge_options *opt, int v)
312 {
313 return (!opt->priv->call_depth && opt->verbosity >= v) ||
314 opt->verbosity >= 5;
315 }
316
317 __attribute__((format (printf, 3, 4)))
318 static void output(struct merge_options *opt, int v, const char *fmt, ...)
319 {
320 va_list ap;
321
322 if (!show(opt, v))
323 return;
324
325 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
326
327 va_start(ap, fmt);
328 strbuf_vaddf(&opt->obuf, fmt, ap);
329 va_end(ap);
330
331 strbuf_addch(&opt->obuf, '\n');
332 if (!opt->buffer_output)
333 flush_output(opt);
334 }
335
336 static void output_commit_title(struct merge_options *opt, struct commit *commit)
337 {
338 struct merge_remote_desc *desc;
339
340 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
341 desc = merge_remote_util(commit);
342 if (desc)
343 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
344 else {
345 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
346 DEFAULT_ABBREV);
347 strbuf_addch(&opt->obuf, ' ');
348 if (parse_commit(commit) != 0)
349 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
350 else {
351 const char *title;
352 const char *msg = get_commit_buffer(commit, NULL);
353 int len = find_commit_subject(msg, &title);
354 if (len)
355 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
356 unuse_commit_buffer(commit, msg);
357 }
358 }
359 flush_output(opt);
360 }
361
362 static int add_cacheinfo(struct merge_options *opt,
363 const struct diff_filespec *blob,
364 const char *path, int stage, int refresh, int options)
365 {
366 struct index_state *istate = opt->repo->index;
367 struct cache_entry *ce;
368 int ret;
369
370 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
371 if (!ce)
372 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
373
374 ret = add_index_entry(istate, ce, options);
375 if (refresh) {
376 struct cache_entry *nce;
377
378 nce = refresh_cache_entry(istate, ce,
379 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
380 if (!nce)
381 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
382 if (nce != ce)
383 ret = add_index_entry(istate, nce, options);
384 }
385 return ret;
386 }
387
388 static inline int merge_detect_rename(struct merge_options *opt)
389 {
390 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
391 }
392
393 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
394 {
395 parse_tree(tree);
396 init_tree_desc(desc, tree->buffer, tree->size);
397 }
398
399 static int unpack_trees_start(struct merge_options *opt,
400 struct tree *common,
401 struct tree *head,
402 struct tree *merge)
403 {
404 int rc;
405 struct tree_desc t[3];
406 struct index_state tmp_index = { NULL };
407
408 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
409 if (opt->priv->call_depth)
410 opt->priv->unpack_opts.index_only = 1;
411 else
412 opt->priv->unpack_opts.update = 1;
413 opt->priv->unpack_opts.merge = 1;
414 opt->priv->unpack_opts.head_idx = 2;
415 opt->priv->unpack_opts.fn = threeway_merge;
416 opt->priv->unpack_opts.src_index = opt->repo->index;
417 opt->priv->unpack_opts.dst_index = &tmp_index;
418 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
419 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
420
421 init_tree_desc_from_tree(t+0, common);
422 init_tree_desc_from_tree(t+1, head);
423 init_tree_desc_from_tree(t+2, merge);
424
425 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
426 cache_tree_free(&opt->repo->index->cache_tree);
427
428 /*
429 * Update opt->repo->index to match the new results, AFTER saving a
430 * copy in opt->priv->orig_index. Update src_index to point to the
431 * saved copy. (verify_uptodate() checks src_index, and the original
432 * index is the one that had the necessary modification timestamps.)
433 */
434 opt->priv->orig_index = *opt->repo->index;
435 *opt->repo->index = tmp_index;
436 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
437
438 return rc;
439 }
440
441 static void unpack_trees_finish(struct merge_options *opt)
442 {
443 discard_index(&opt->priv->orig_index);
444 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
445 }
446
447 static int save_files_dirs(const struct object_id *oid,
448 struct strbuf *base, const char *path,
449 unsigned int mode, void *context)
450 {
451 struct path_hashmap_entry *entry;
452 int baselen = base->len;
453 struct merge_options *opt = context;
454
455 strbuf_addstr(base, path);
456
457 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
458 hashmap_entry_init(&entry->e, fspathhash(entry->path));
459 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
460
461 strbuf_setlen(base, baselen);
462 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
463 }
464
465 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
466 {
467 struct pathspec match_all;
468 memset(&match_all, 0, sizeof(match_all));
469 read_tree(opt->repo, tree,
470 &match_all, save_files_dirs, opt);
471 }
472
473 static int get_tree_entry_if_blob(struct repository *r,
474 const struct object_id *tree,
475 const char *path,
476 struct diff_filespec *dfs)
477 {
478 int ret;
479
480 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
481 if (S_ISDIR(dfs->mode)) {
482 oidcpy(&dfs->oid, null_oid());
483 dfs->mode = 0;
484 }
485 return ret;
486 }
487
488 /*
489 * Returns an index_entry instance which doesn't have to correspond to
490 * a real cache entry in Git's index.
491 */
492 static struct stage_data *insert_stage_data(struct repository *r,
493 const char *path,
494 struct tree *o, struct tree *a, struct tree *b,
495 struct string_list *entries)
496 {
497 struct string_list_item *item;
498 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
499 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
500 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
501 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
502 item = string_list_insert(entries, path);
503 item->util = e;
504 return e;
505 }
506
507 /*
508 * Create a dictionary mapping file names to stage_data objects. The
509 * dictionary contains one entry for every path with a non-zero stage entry.
510 */
511 static struct string_list *get_unmerged(struct index_state *istate)
512 {
513 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
514 int i;
515
516 unmerged->strdup_strings = 1;
517
518 /* TODO: audit for interaction with sparse-index. */
519 ensure_full_index(istate);
520 for (i = 0; i < istate->cache_nr; i++) {
521 struct string_list_item *item;
522 struct stage_data *e;
523 const struct cache_entry *ce = istate->cache[i];
524 if (!ce_stage(ce))
525 continue;
526
527 item = string_list_lookup(unmerged, ce->name);
528 if (!item) {
529 item = string_list_insert(unmerged, ce->name);
530 item->util = xcalloc(1, sizeof(struct stage_data));
531 }
532 e = item->util;
533 e->stages[ce_stage(ce)].mode = ce->ce_mode;
534 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
535 }
536
537 return unmerged;
538 }
539
540 static int string_list_df_name_compare(const char *one, const char *two)
541 {
542 int onelen = strlen(one);
543 int twolen = strlen(two);
544 /*
545 * Here we only care that entries for D/F conflicts are
546 * adjacent, in particular with the file of the D/F conflict
547 * appearing before files below the corresponding directory.
548 * The order of the rest of the list is irrelevant for us.
549 *
550 * To achieve this, we sort with df_name_compare and provide
551 * the mode S_IFDIR so that D/F conflicts will sort correctly.
552 * We use the mode S_IFDIR for everything else for simplicity,
553 * since in other cases any changes in their order due to
554 * sorting cause no problems for us.
555 */
556 int cmp = df_name_compare(one, onelen, S_IFDIR,
557 two, twolen, S_IFDIR);
558 /*
559 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
560 * that 'foo' comes before 'foo/bar'.
561 */
562 if (cmp)
563 return cmp;
564 return onelen - twolen;
565 }
566
567 static void record_df_conflict_files(struct merge_options *opt,
568 struct string_list *entries)
569 {
570 /* If there is a D/F conflict and the file for such a conflict
571 * currently exists in the working tree, we want to allow it to be
572 * removed to make room for the corresponding directory if needed.
573 * The files underneath the directories of such D/F conflicts will
574 * be processed before the corresponding file involved in the D/F
575 * conflict. If the D/F directory ends up being removed by the
576 * merge, then we won't have to touch the D/F file. If the D/F
577 * directory needs to be written to the working copy, then the D/F
578 * file will simply be removed (in make_room_for_path()) to make
579 * room for the necessary paths. Note that if both the directory
580 * and the file need to be present, then the D/F file will be
581 * reinstated with a new unique name at the time it is processed.
582 */
583 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
584 const char *last_file = NULL;
585 int last_len = 0;
586 int i;
587
588 /*
589 * If we're merging merge-bases, we don't want to bother with
590 * any working directory changes.
591 */
592 if (opt->priv->call_depth)
593 return;
594
595 /* Ensure D/F conflicts are adjacent in the entries list. */
596 for (i = 0; i < entries->nr; i++) {
597 struct string_list_item *next = &entries->items[i];
598 string_list_append(&df_sorted_entries, next->string)->util =
599 next->util;
600 }
601 df_sorted_entries.cmp = string_list_df_name_compare;
602 string_list_sort(&df_sorted_entries);
603
604 string_list_clear(&opt->priv->df_conflict_file_set, 1);
605 for (i = 0; i < df_sorted_entries.nr; i++) {
606 const char *path = df_sorted_entries.items[i].string;
607 int len = strlen(path);
608 struct stage_data *e = df_sorted_entries.items[i].util;
609
610 /*
611 * Check if last_file & path correspond to a D/F conflict;
612 * i.e. whether path is last_file+'/'+<something>.
613 * If so, record that it's okay to remove last_file to make
614 * room for path and friends if needed.
615 */
616 if (last_file &&
617 len > last_len &&
618 memcmp(path, last_file, last_len) == 0 &&
619 path[last_len] == '/') {
620 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
621 }
622
623 /*
624 * Determine whether path could exist as a file in the
625 * working directory as a possible D/F conflict. This
626 * will only occur when it exists in stage 2 as a
627 * file.
628 */
629 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
630 last_file = path;
631 last_len = len;
632 } else {
633 last_file = NULL;
634 }
635 }
636 string_list_clear(&df_sorted_entries, 0);
637 }
638
639 static int update_stages(struct merge_options *opt, const char *path,
640 const struct diff_filespec *o,
641 const struct diff_filespec *a,
642 const struct diff_filespec *b)
643 {
644
645 /*
646 * NOTE: It is usually a bad idea to call update_stages on a path
647 * before calling update_file on that same path, since it can
648 * sometimes lead to spurious "refusing to lose untracked file..."
649 * messages from update_file (via make_room_for path via
650 * would_lose_untracked). Instead, reverse the order of the calls
651 * (executing update_file first and then update_stages).
652 */
653 int clear = 1;
654 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
655 if (clear)
656 if (remove_file_from_index(opt->repo->index, path))
657 return -1;
658 if (o)
659 if (add_cacheinfo(opt, o, path, 1, 0, options))
660 return -1;
661 if (a)
662 if (add_cacheinfo(opt, a, path, 2, 0, options))
663 return -1;
664 if (b)
665 if (add_cacheinfo(opt, b, path, 3, 0, options))
666 return -1;
667 return 0;
668 }
669
670 static void update_entry(struct stage_data *entry,
671 struct diff_filespec *o,
672 struct diff_filespec *a,
673 struct diff_filespec *b)
674 {
675 entry->processed = 0;
676 entry->stages[1].mode = o->mode;
677 entry->stages[2].mode = a->mode;
678 entry->stages[3].mode = b->mode;
679 oidcpy(&entry->stages[1].oid, &o->oid);
680 oidcpy(&entry->stages[2].oid, &a->oid);
681 oidcpy(&entry->stages[3].oid, &b->oid);
682 }
683
684 static int remove_file(struct merge_options *opt, int clean,
685 const char *path, int no_wd)
686 {
687 int update_cache = opt->priv->call_depth || clean;
688 int update_working_directory = !opt->priv->call_depth && !no_wd;
689
690 if (update_cache) {
691 if (remove_file_from_index(opt->repo->index, path))
692 return -1;
693 }
694 if (update_working_directory) {
695 if (ignore_case) {
696 struct cache_entry *ce;
697 ce = index_file_exists(opt->repo->index, path, strlen(path),
698 ignore_case);
699 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
700 return 0;
701 }
702 if (remove_path(path))
703 return -1;
704 }
705 return 0;
706 }
707
708 /* add a string to a strbuf, but converting "/" to "_" */
709 static void add_flattened_path(struct strbuf *out, const char *s)
710 {
711 size_t i = out->len;
712 strbuf_addstr(out, s);
713 for (; i < out->len; i++)
714 if (out->buf[i] == '/')
715 out->buf[i] = '_';
716 }
717
718 static char *unique_path(struct merge_options *opt,
719 const char *path,
720 const char *branch)
721 {
722 struct path_hashmap_entry *entry;
723 struct strbuf newpath = STRBUF_INIT;
724 int suffix = 0;
725 size_t base_len;
726
727 strbuf_addf(&newpath, "%s~", path);
728 add_flattened_path(&newpath, branch);
729
730 base_len = newpath.len;
731 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
732 fspathhash(newpath.buf), newpath.buf) ||
733 (!opt->priv->call_depth && file_exists(newpath.buf))) {
734 strbuf_setlen(&newpath, base_len);
735 strbuf_addf(&newpath, "_%d", suffix++);
736 }
737
738 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
739 hashmap_entry_init(&entry->e, fspathhash(entry->path));
740 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
741 return strbuf_detach(&newpath, NULL);
742 }
743
744 /**
745 * Check whether a directory in the index is in the way of an incoming
746 * file. Return 1 if so. If check_working_copy is non-zero, also
747 * check the working directory. If empty_ok is non-zero, also return
748 * 0 in the case where the working-tree dir exists but is empty.
749 */
750 static int dir_in_way(struct index_state *istate, const char *path,
751 int check_working_copy, int empty_ok)
752 {
753 int pos;
754 struct strbuf dirpath = STRBUF_INIT;
755 struct stat st;
756
757 strbuf_addstr(&dirpath, path);
758 strbuf_addch(&dirpath, '/');
759
760 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
761
762 if (pos < 0)
763 pos = -1 - pos;
764 if (pos < istate->cache_nr &&
765 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
766 strbuf_release(&dirpath);
767 return 1;
768 }
769
770 strbuf_release(&dirpath);
771 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
772 !(empty_ok && is_empty_dir(path)) &&
773 !has_symlink_leading_path(path, strlen(path));
774 }
775
776 /*
777 * Returns whether path was tracked in the index before the merge started,
778 * and its oid and mode match the specified values
779 */
780 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
781 const struct diff_filespec *blob)
782 {
783 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
784 struct cache_entry *ce;
785
786 if (0 > pos)
787 /* we were not tracking this path before the merge */
788 return 0;
789
790 /* See if the file we were tracking before matches */
791 ce = opt->priv->orig_index.cache[pos];
792 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
793 }
794
795 /*
796 * Returns whether path was tracked in the index before the merge started
797 */
798 static int was_tracked(struct merge_options *opt, const char *path)
799 {
800 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
801
802 if (0 <= pos)
803 /* we were tracking this path before the merge */
804 return 1;
805
806 return 0;
807 }
808
809 static int would_lose_untracked(struct merge_options *opt, const char *path)
810 {
811 struct index_state *istate = opt->repo->index;
812
813 /*
814 * This may look like it can be simplified to:
815 * return !was_tracked(opt, path) && file_exists(path)
816 * but it can't. This function needs to know whether path was in
817 * the working tree due to EITHER having been tracked in the index
818 * before the merge OR having been put into the working copy and
819 * index by unpack_trees(). Due to that either-or requirement, we
820 * check the current index instead of the original one.
821 *
822 * Note that we do not need to worry about merge-recursive itself
823 * updating the index after unpack_trees() and before calling this
824 * function, because we strictly require all code paths in
825 * merge-recursive to update the working tree first and the index
826 * second. Doing otherwise would break
827 * update_file()/would_lose_untracked(); see every comment in this
828 * file which mentions "update_stages".
829 */
830 int pos = index_name_pos(istate, path, strlen(path));
831
832 if (pos < 0)
833 pos = -1 - pos;
834 while (pos < istate->cache_nr &&
835 !strcmp(path, istate->cache[pos]->name)) {
836 /*
837 * If stage #0, it is definitely tracked.
838 * If it has stage #2 then it was tracked
839 * before this merge started. All other
840 * cases the path was not tracked.
841 */
842 switch (ce_stage(istate->cache[pos])) {
843 case 0:
844 case 2:
845 return 0;
846 }
847 pos++;
848 }
849 return file_exists(path);
850 }
851
852 static int was_dirty(struct merge_options *opt, const char *path)
853 {
854 struct cache_entry *ce;
855 int dirty = 1;
856
857 if (opt->priv->call_depth || !was_tracked(opt, path))
858 return !dirty;
859
860 ce = index_file_exists(opt->priv->unpack_opts.src_index,
861 path, strlen(path), ignore_case);
862 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
863 return dirty;
864 }
865
866 static int make_room_for_path(struct merge_options *opt, const char *path)
867 {
868 int status, i;
869 const char *msg = _("failed to create path '%s'%s");
870
871 /* Unlink any D/F conflict files that are in the way */
872 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
873 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
874 size_t pathlen = strlen(path);
875 size_t df_pathlen = strlen(df_path);
876 if (df_pathlen < pathlen &&
877 path[df_pathlen] == '/' &&
878 strncmp(path, df_path, df_pathlen) == 0) {
879 output(opt, 3,
880 _("Removing %s to make room for subdirectory\n"),
881 df_path);
882 unlink(df_path);
883 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
884 i, 0);
885 break;
886 }
887 }
888
889 /* Make sure leading directories are created */
890 status = safe_create_leading_directories_const(path);
891 if (status) {
892 if (status == SCLD_EXISTS)
893 /* something else exists */
894 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
895 return err(opt, msg, path, "");
896 }
897
898 /*
899 * Do not unlink a file in the work tree if we are not
900 * tracking it.
901 */
902 if (would_lose_untracked(opt, path))
903 return err(opt, _("refusing to lose untracked file at '%s'"),
904 path);
905
906 /* Successful unlink is good.. */
907 if (!unlink(path))
908 return 0;
909 /* .. and so is no existing file */
910 if (errno == ENOENT)
911 return 0;
912 /* .. but not some other error (who really cares what?) */
913 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
914 }
915
916 static int update_file_flags(struct merge_options *opt,
917 const struct diff_filespec *contents,
918 const char *path,
919 int update_cache,
920 int update_wd)
921 {
922 int ret = 0;
923
924 if (opt->priv->call_depth)
925 update_wd = 0;
926
927 if (update_wd) {
928 enum object_type type;
929 void *buf;
930 unsigned long size;
931
932 if (S_ISGITLINK(contents->mode)) {
933 /*
934 * We may later decide to recursively descend into
935 * the submodule directory and update its index
936 * and/or work tree, but we do not do that now.
937 */
938 update_wd = 0;
939 goto update_index;
940 }
941
942 buf = read_object_file(&contents->oid, &type, &size);
943 if (!buf) {
944 ret = err(opt, _("cannot read object %s '%s'"),
945 oid_to_hex(&contents->oid), path);
946 goto free_buf;
947 }
948 if (type != OBJ_BLOB) {
949 ret = err(opt, _("blob expected for %s '%s'"),
950 oid_to_hex(&contents->oid), path);
951 goto free_buf;
952 }
953 if (S_ISREG(contents->mode)) {
954 struct strbuf strbuf = STRBUF_INIT;
955 if (convert_to_working_tree(opt->repo->index,
956 path, buf, size, &strbuf, NULL)) {
957 free(buf);
958 size = strbuf.len;
959 buf = strbuf_detach(&strbuf, NULL);
960 }
961 }
962
963 if (make_room_for_path(opt, path) < 0) {
964 update_wd = 0;
965 goto free_buf;
966 }
967 if (S_ISREG(contents->mode) ||
968 (!has_symlinks && S_ISLNK(contents->mode))) {
969 int fd;
970 int mode = (contents->mode & 0100 ? 0777 : 0666);
971
972 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
973 if (fd < 0) {
974 ret = err(opt, _("failed to open '%s': %s"),
975 path, strerror(errno));
976 goto free_buf;
977 }
978 write_in_full(fd, buf, size);
979 close(fd);
980 } else if (S_ISLNK(contents->mode)) {
981 char *lnk = xmemdupz(buf, size);
982 safe_create_leading_directories_const(path);
983 unlink(path);
984 if (symlink(lnk, path))
985 ret = err(opt, _("failed to symlink '%s': %s"),
986 path, strerror(errno));
987 free(lnk);
988 } else
989 ret = err(opt,
990 _("do not know what to do with %06o %s '%s'"),
991 contents->mode, oid_to_hex(&contents->oid), path);
992 free_buf:
993 free(buf);
994 }
995 update_index:
996 if (!ret && update_cache) {
997 int refresh = (!opt->priv->call_depth &&
998 contents->mode != S_IFGITLINK);
999 if (add_cacheinfo(opt, contents, path, 0, refresh,
1000 ADD_CACHE_OK_TO_ADD))
1001 return -1;
1002 }
1003 return ret;
1004 }
1005
1006 static int update_file(struct merge_options *opt,
1007 int clean,
1008 const struct diff_filespec *contents,
1009 const char *path)
1010 {
1011 return update_file_flags(opt, contents, path,
1012 opt->priv->call_depth || clean, !opt->priv->call_depth);
1013 }
1014
1015 /* Low level file merging, update and removal */
1016
1017 struct merge_file_info {
1018 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1019 unsigned clean:1,
1020 merge:1;
1021 };
1022
1023 static int merge_3way(struct merge_options *opt,
1024 mmbuffer_t *result_buf,
1025 const struct diff_filespec *o,
1026 const struct diff_filespec *a,
1027 const struct diff_filespec *b,
1028 const char *branch1,
1029 const char *branch2,
1030 const int extra_marker_size)
1031 {
1032 mmfile_t orig, src1, src2;
1033 struct ll_merge_options ll_opts = {0};
1034 char *base, *name1, *name2;
1035 int merge_status;
1036
1037 ll_opts.renormalize = opt->renormalize;
1038 ll_opts.extra_marker_size = extra_marker_size;
1039 ll_opts.xdl_opts = opt->xdl_opts;
1040
1041 if (opt->priv->call_depth) {
1042 ll_opts.virtual_ancestor = 1;
1043 ll_opts.variant = 0;
1044 } else {
1045 switch (opt->recursive_variant) {
1046 case MERGE_VARIANT_OURS:
1047 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1048 break;
1049 case MERGE_VARIANT_THEIRS:
1050 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1051 break;
1052 default:
1053 ll_opts.variant = 0;
1054 break;
1055 }
1056 }
1057
1058 assert(a->path && b->path && o->path && opt->ancestor);
1059 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1060 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1061 name1 = mkpathdup("%s:%s", branch1, a->path);
1062 name2 = mkpathdup("%s:%s", branch2, b->path);
1063 } else {
1064 base = mkpathdup("%s", opt->ancestor);
1065 name1 = mkpathdup("%s", branch1);
1066 name2 = mkpathdup("%s", branch2);
1067 }
1068
1069 read_mmblob(&orig, &o->oid);
1070 read_mmblob(&src1, &a->oid);
1071 read_mmblob(&src2, &b->oid);
1072
1073 /*
1074 * FIXME: Using a->path for normalization rules in ll_merge could be
1075 * wrong if we renamed from a->path to b->path. We should use the
1076 * target path for where the file will be written.
1077 */
1078 merge_status = ll_merge(result_buf, a->path, &orig, base,
1079 &src1, name1, &src2, name2,
1080 opt->repo->index, &ll_opts);
1081
1082 free(base);
1083 free(name1);
1084 free(name2);
1085 free(orig.ptr);
1086 free(src1.ptr);
1087 free(src2.ptr);
1088 return merge_status;
1089 }
1090
1091 static int find_first_merges(struct repository *repo,
1092 struct object_array *result, const char *path,
1093 struct commit *a, struct commit *b)
1094 {
1095 int i, j;
1096 struct object_array merges = OBJECT_ARRAY_INIT;
1097 struct commit *commit;
1098 int contains_another;
1099
1100 char merged_revision[GIT_MAX_HEXSZ + 2];
1101 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1102 "--all", merged_revision, NULL };
1103 struct rev_info revs;
1104 struct setup_revision_opt rev_opts;
1105
1106 memset(result, 0, sizeof(struct object_array));
1107 memset(&rev_opts, 0, sizeof(rev_opts));
1108
1109 /* get all revisions that merge commit a */
1110 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1111 oid_to_hex(&a->object.oid));
1112 repo_init_revisions(repo, &revs, NULL);
1113 rev_opts.submodule = path;
1114 /* FIXME: can't handle linked worktrees in submodules yet */
1115 revs.single_worktree = path != NULL;
1116 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1117
1118 /* save all revisions from the above list that contain b */
1119 if (prepare_revision_walk(&revs))
1120 die("revision walk setup failed");
1121 while ((commit = get_revision(&revs)) != NULL) {
1122 struct object *o = &(commit->object);
1123 if (in_merge_bases(b, commit))
1124 add_object_array(o, NULL, &merges);
1125 }
1126 reset_revision_walk();
1127
1128 /* Now we've got all merges that contain a and b. Prune all
1129 * merges that contain another found merge and save them in
1130 * result.
1131 */
1132 for (i = 0; i < merges.nr; i++) {
1133 struct commit *m1 = (struct commit *) merges.objects[i].item;
1134
1135 contains_another = 0;
1136 for (j = 0; j < merges.nr; j++) {
1137 struct commit *m2 = (struct commit *) merges.objects[j].item;
1138 if (i != j && in_merge_bases(m2, m1)) {
1139 contains_another = 1;
1140 break;
1141 }
1142 }
1143
1144 if (!contains_another)
1145 add_object_array(merges.objects[i].item, NULL, result);
1146 }
1147
1148 object_array_clear(&merges);
1149 return result->nr;
1150 }
1151
1152 static void print_commit(struct commit *commit)
1153 {
1154 struct strbuf sb = STRBUF_INIT;
1155 struct pretty_print_context ctx = {0};
1156 ctx.date_mode.type = DATE_NORMAL;
1157 /* FIXME: Merge this with output_commit_title() */
1158 assert(!merge_remote_util(commit));
1159 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1160 fprintf(stderr, "%s\n", sb.buf);
1161 strbuf_release(&sb);
1162 }
1163
1164 static int is_valid(const struct diff_filespec *dfs)
1165 {
1166 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1167 }
1168
1169 static int merge_submodule(struct merge_options *opt,
1170 struct object_id *result, const char *path,
1171 const struct object_id *base, const struct object_id *a,
1172 const struct object_id *b)
1173 {
1174 struct commit *commit_base, *commit_a, *commit_b;
1175 int parent_count;
1176 struct object_array merges;
1177
1178 int i;
1179 int search = !opt->priv->call_depth;
1180
1181 /* store a in result in case we fail */
1182 /* FIXME: This is the WRONG resolution for the recursive case when
1183 * we need to be careful to avoid accidentally matching either side.
1184 * Should probably use o instead there, much like we do for merging
1185 * binaries.
1186 */
1187 oidcpy(result, a);
1188
1189 /* we can not handle deletion conflicts */
1190 if (is_null_oid(base))
1191 return 0;
1192 if (is_null_oid(a))
1193 return 0;
1194 if (is_null_oid(b))
1195 return 0;
1196
1197 if (add_submodule_odb(path)) {
1198 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1199 return 0;
1200 }
1201
1202 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1203 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1204 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1205 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1206 return 0;
1207 }
1208
1209 /* check whether both changes are forward */
1210 if (!in_merge_bases(commit_base, commit_a) ||
1211 !in_merge_bases(commit_base, commit_b)) {
1212 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1213 return 0;
1214 }
1215
1216 /* Case #1: a is contained in b or vice versa */
1217 if (in_merge_bases(commit_a, commit_b)) {
1218 oidcpy(result, b);
1219 if (show(opt, 3)) {
1220 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1221 output_commit_title(opt, commit_b);
1222 } else if (show(opt, 2))
1223 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1224 else
1225 ; /* no output */
1226
1227 return 1;
1228 }
1229 if (in_merge_bases(commit_b, commit_a)) {
1230 oidcpy(result, a);
1231 if (show(opt, 3)) {
1232 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1233 output_commit_title(opt, commit_a);
1234 } else if (show(opt, 2))
1235 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1236 else
1237 ; /* no output */
1238
1239 return 1;
1240 }
1241
1242 /*
1243 * Case #2: There are one or more merges that contain a and b in
1244 * the submodule. If there is only one, then present it as a
1245 * suggestion to the user, but leave it marked unmerged so the
1246 * user needs to confirm the resolution.
1247 */
1248
1249 /* Skip the search if makes no sense to the calling context. */
1250 if (!search)
1251 return 0;
1252
1253 /* find commit which merges them */
1254 parent_count = find_first_merges(opt->repo, &merges, path,
1255 commit_a, commit_b);
1256 switch (parent_count) {
1257 case 0:
1258 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1259 break;
1260
1261 case 1:
1262 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1263 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1264 print_commit((struct commit *) merges.objects[0].item);
1265 output(opt, 2, _(
1266 "If this is correct simply add it to the index "
1267 "for example\n"
1268 "by using:\n\n"
1269 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1270 "which will accept this suggestion.\n"),
1271 oid_to_hex(&merges.objects[0].item->oid), path);
1272 break;
1273
1274 default:
1275 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1276 for (i = 0; i < merges.nr; i++)
1277 print_commit((struct commit *) merges.objects[i].item);
1278 }
1279
1280 object_array_clear(&merges);
1281 return 0;
1282 }
1283
1284 static int merge_mode_and_contents(struct merge_options *opt,
1285 const struct diff_filespec *o,
1286 const struct diff_filespec *a,
1287 const struct diff_filespec *b,
1288 const char *filename,
1289 const char *branch1,
1290 const char *branch2,
1291 const int extra_marker_size,
1292 struct merge_file_info *result)
1293 {
1294 if (opt->branch1 != branch1) {
1295 /*
1296 * It's weird getting a reverse merge with HEAD on the bottom
1297 * side of the conflict markers and the other branch on the
1298 * top. Fix that.
1299 */
1300 return merge_mode_and_contents(opt, o, b, a,
1301 filename,
1302 branch2, branch1,
1303 extra_marker_size, result);
1304 }
1305
1306 result->merge = 0;
1307 result->clean = 1;
1308
1309 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1310 result->clean = 0;
1311 /*
1312 * FIXME: This is a bad resolution for recursive case; for
1313 * the recursive case we want something that is unlikely to
1314 * accidentally match either side. Also, while it makes
1315 * sense to prefer regular files over symlinks, it doesn't
1316 * make sense to prefer regular files over submodules.
1317 */
1318 if (S_ISREG(a->mode)) {
1319 result->blob.mode = a->mode;
1320 oidcpy(&result->blob.oid, &a->oid);
1321 } else {
1322 result->blob.mode = b->mode;
1323 oidcpy(&result->blob.oid, &b->oid);
1324 }
1325 } else {
1326 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1327 result->merge = 1;
1328
1329 /*
1330 * Merge modes
1331 */
1332 if (a->mode == b->mode || a->mode == o->mode)
1333 result->blob.mode = b->mode;
1334 else {
1335 result->blob.mode = a->mode;
1336 if (b->mode != o->mode) {
1337 result->clean = 0;
1338 result->merge = 1;
1339 }
1340 }
1341
1342 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1343 oidcpy(&result->blob.oid, &b->oid);
1344 else if (oideq(&b->oid, &o->oid))
1345 oidcpy(&result->blob.oid, &a->oid);
1346 else if (S_ISREG(a->mode)) {
1347 mmbuffer_t result_buf;
1348 int ret = 0, merge_status;
1349
1350 merge_status = merge_3way(opt, &result_buf, o, a, b,
1351 branch1, branch2,
1352 extra_marker_size);
1353
1354 if ((merge_status < 0) || !result_buf.ptr)
1355 ret = err(opt, _("Failed to execute internal merge"));
1356
1357 if (!ret &&
1358 write_object_file(result_buf.ptr, result_buf.size,
1359 blob_type, &result->blob.oid))
1360 ret = err(opt, _("Unable to add %s to database"),
1361 a->path);
1362
1363 free(result_buf.ptr);
1364 if (ret)
1365 return ret;
1366 /* FIXME: bug, what if modes didn't match? */
1367 result->clean = (merge_status == 0);
1368 } else if (S_ISGITLINK(a->mode)) {
1369 result->clean = merge_submodule(opt, &result->blob.oid,
1370 o->path,
1371 &o->oid,
1372 &a->oid,
1373 &b->oid);
1374 } else if (S_ISLNK(a->mode)) {
1375 switch (opt->recursive_variant) {
1376 case MERGE_VARIANT_NORMAL:
1377 oidcpy(&result->blob.oid, &a->oid);
1378 if (!oideq(&a->oid, &b->oid))
1379 result->clean = 0;
1380 break;
1381 case MERGE_VARIANT_OURS:
1382 oidcpy(&result->blob.oid, &a->oid);
1383 break;
1384 case MERGE_VARIANT_THEIRS:
1385 oidcpy(&result->blob.oid, &b->oid);
1386 break;
1387 }
1388 } else
1389 BUG("unsupported object type in the tree");
1390 }
1391
1392 if (result->merge)
1393 output(opt, 2, _("Auto-merging %s"), filename);
1394
1395 return 0;
1396 }
1397
1398 static int handle_rename_via_dir(struct merge_options *opt,
1399 struct rename_conflict_info *ci)
1400 {
1401 /*
1402 * Handle file adds that need to be renamed due to directory rename
1403 * detection. This differs from handle_rename_normal, because
1404 * there is no content merge to do; just move the file into the
1405 * desired final location.
1406 */
1407 const struct rename *ren = ci->ren1;
1408 const struct diff_filespec *dest = ren->pair->two;
1409 char *file_path = dest->path;
1410 int mark_conflicted = (opt->detect_directory_renames ==
1411 MERGE_DIRECTORY_RENAMES_CONFLICT);
1412 assert(ren->dir_rename_original_dest);
1413
1414 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1415 mark_conflicted = 1;
1416 file_path = unique_path(opt, dest->path, ren->branch);
1417 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1418 "writing to %s instead."),
1419 dest->path, file_path);
1420 }
1421
1422 if (mark_conflicted) {
1423 /*
1424 * Write the file in worktree at file_path. In the index,
1425 * only record the file at dest->path in the appropriate
1426 * higher stage.
1427 */
1428 if (update_file(opt, 0, dest, file_path))
1429 return -1;
1430 if (file_path != dest->path)
1431 free(file_path);
1432 if (update_stages(opt, dest->path, NULL,
1433 ren->branch == opt->branch1 ? dest : NULL,
1434 ren->branch == opt->branch1 ? NULL : dest))
1435 return -1;
1436 return 0; /* not clean, but conflicted */
1437 } else {
1438 /* Update dest->path both in index and in worktree */
1439 if (update_file(opt, 1, dest, dest->path))
1440 return -1;
1441 return 1; /* clean */
1442 }
1443 }
1444
1445 static int handle_change_delete(struct merge_options *opt,
1446 const char *path, const char *old_path,
1447 const struct diff_filespec *o,
1448 const struct diff_filespec *changed,
1449 const char *change_branch,
1450 const char *delete_branch,
1451 const char *change, const char *change_past)
1452 {
1453 char *alt_path = NULL;
1454 const char *update_path = path;
1455 int ret = 0;
1456
1457 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1458 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1459 update_path = alt_path = unique_path(opt, path, change_branch);
1460 }
1461
1462 if (opt->priv->call_depth) {
1463 /*
1464 * We cannot arbitrarily accept either a_sha or b_sha as
1465 * correct; since there is no true "middle point" between
1466 * them, simply reuse the base version for virtual merge base.
1467 */
1468 ret = remove_file_from_index(opt->repo->index, path);
1469 if (!ret)
1470 ret = update_file(opt, 0, o, update_path);
1471 } else {
1472 /*
1473 * Despite the four nearly duplicate messages and argument
1474 * lists below and the ugliness of the nested if-statements,
1475 * having complete messages makes the job easier for
1476 * translators.
1477 *
1478 * The slight variance among the cases is due to the fact
1479 * that:
1480 * 1) directory/file conflicts (in effect if
1481 * !alt_path) could cause us to need to write the
1482 * file to a different path.
1483 * 2) renames (in effect if !old_path) could mean that
1484 * there are two names for the path that the user
1485 * may know the file by.
1486 */
1487 if (!alt_path) {
1488 if (!old_path) {
1489 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1490 "and %s in %s. Version %s of %s left in tree."),
1491 change, path, delete_branch, change_past,
1492 change_branch, change_branch, path);
1493 } else {
1494 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1495 "and %s to %s in %s. Version %s of %s left in tree."),
1496 change, old_path, delete_branch, change_past, path,
1497 change_branch, change_branch, path);
1498 }
1499 } else {
1500 if (!old_path) {
1501 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1502 "and %s in %s. Version %s of %s left in tree at %s."),
1503 change, path, delete_branch, change_past,
1504 change_branch, change_branch, path, alt_path);
1505 } else {
1506 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1507 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1508 change, old_path, delete_branch, change_past, path,
1509 change_branch, change_branch, path, alt_path);
1510 }
1511 }
1512 /*
1513 * No need to call update_file() on path when change_branch ==
1514 * opt->branch1 && !alt_path, since that would needlessly touch
1515 * path. We could call update_file_flags() with update_cache=0
1516 * and update_wd=0, but that's a no-op.
1517 */
1518 if (change_branch != opt->branch1 || alt_path)
1519 ret = update_file(opt, 0, changed, update_path);
1520 }
1521 free(alt_path);
1522
1523 return ret;
1524 }
1525
1526 static int handle_rename_delete(struct merge_options *opt,
1527 struct rename_conflict_info *ci)
1528 {
1529 const struct rename *ren = ci->ren1;
1530 const struct diff_filespec *orig = ren->pair->one;
1531 const struct diff_filespec *dest = ren->pair->two;
1532 const char *rename_branch = ren->branch;
1533 const char *delete_branch = (opt->branch1 == ren->branch ?
1534 opt->branch2 : opt->branch1);
1535
1536 if (handle_change_delete(opt,
1537 opt->priv->call_depth ? orig->path : dest->path,
1538 opt->priv->call_depth ? NULL : orig->path,
1539 orig, dest,
1540 rename_branch, delete_branch,
1541 _("rename"), _("renamed")))
1542 return -1;
1543
1544 if (opt->priv->call_depth)
1545 return remove_file_from_index(opt->repo->index, dest->path);
1546 else
1547 return update_stages(opt, dest->path, NULL,
1548 rename_branch == opt->branch1 ? dest : NULL,
1549 rename_branch == opt->branch1 ? NULL : dest);
1550 }
1551
1552 static int handle_file_collision(struct merge_options *opt,
1553 const char *collide_path,
1554 const char *prev_path1,
1555 const char *prev_path2,
1556 const char *branch1, const char *branch2,
1557 struct diff_filespec *a,
1558 struct diff_filespec *b)
1559 {
1560 struct merge_file_info mfi;
1561 struct diff_filespec null;
1562 char *alt_path = NULL;
1563 const char *update_path = collide_path;
1564
1565 /*
1566 * It's easiest to get the correct things into stage 2 and 3, and
1567 * to make sure that the content merge puts HEAD before the other
1568 * branch if we just ensure that branch1 == opt->branch1. So, simply
1569 * flip arguments around if we don't have that.
1570 */
1571 if (branch1 != opt->branch1) {
1572 return handle_file_collision(opt, collide_path,
1573 prev_path2, prev_path1,
1574 branch2, branch1,
1575 b, a);
1576 }
1577
1578 /* Remove rename sources if rename/add or rename/rename(2to1) */
1579 if (prev_path1)
1580 remove_file(opt, 1, prev_path1,
1581 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1582 if (prev_path2)
1583 remove_file(opt, 1, prev_path2,
1584 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1585
1586 /*
1587 * Remove the collision path, if it wouldn't cause dirty contents
1588 * or an untracked file to get lost. We'll either overwrite with
1589 * merged contents, or just write out to differently named files.
1590 */
1591 if (was_dirty(opt, collide_path)) {
1592 output(opt, 1, _("Refusing to lose dirty file at %s"),
1593 collide_path);
1594 update_path = alt_path = unique_path(opt, collide_path, "merged");
1595 } else if (would_lose_untracked(opt, collide_path)) {
1596 /*
1597 * Only way we get here is if both renames were from
1598 * a directory rename AND user had an untracked file
1599 * at the location where both files end up after the
1600 * two directory renames. See testcase 10d of t6043.
1601 */
1602 output(opt, 1, _("Refusing to lose untracked file at "
1603 "%s, even though it's in the way."),
1604 collide_path);
1605 update_path = alt_path = unique_path(opt, collide_path, "merged");
1606 } else {
1607 /*
1608 * FIXME: It's possible that the two files are identical
1609 * and that the current working copy happens to match, in
1610 * which case we are unnecessarily touching the working
1611 * tree file. It's not a likely enough scenario that I
1612 * want to code up the checks for it and a better fix is
1613 * available if we restructure how unpack_trees() and
1614 * merge-recursive interoperate anyway, so punting for
1615 * now...
1616 */
1617 remove_file(opt, 0, collide_path, 0);
1618 }
1619
1620 /* Store things in diff_filespecs for functions that need it */
1621 null.path = (char *)collide_path;
1622 oidcpy(&null.oid, null_oid());
1623 null.mode = 0;
1624
1625 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1626 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1627 return -1;
1628 mfi.clean &= !alt_path;
1629 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1630 return -1;
1631 if (!mfi.clean && !opt->priv->call_depth &&
1632 update_stages(opt, collide_path, NULL, a, b))
1633 return -1;
1634 free(alt_path);
1635 /*
1636 * FIXME: If both a & b both started with conflicts (only possible
1637 * if they came from a rename/rename(2to1)), but had IDENTICAL
1638 * contents including those conflicts, then in the next line we claim
1639 * it was clean. If someone cares about this case, we should have the
1640 * caller notify us if we started with conflicts.
1641 */
1642 return mfi.clean;
1643 }
1644
1645 static int handle_rename_add(struct merge_options *opt,
1646 struct rename_conflict_info *ci)
1647 {
1648 /* a was renamed to c, and a separate c was added. */
1649 struct diff_filespec *a = ci->ren1->pair->one;
1650 struct diff_filespec *c = ci->ren1->pair->two;
1651 char *path = c->path;
1652 char *prev_path_desc;
1653 struct merge_file_info mfi;
1654
1655 const char *rename_branch = ci->ren1->branch;
1656 const char *add_branch = (opt->branch1 == rename_branch ?
1657 opt->branch2 : opt->branch1);
1658 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1659
1660 output(opt, 1, _("CONFLICT (rename/add): "
1661 "Rename %s->%s in %s. Added %s in %s"),
1662 a->path, c->path, rename_branch,
1663 c->path, add_branch);
1664
1665 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1666 ci->ren1->src_entry->stages[other_stage].path = a->path;
1667 if (merge_mode_and_contents(opt, a, c,
1668 &ci->ren1->src_entry->stages[other_stage],
1669 prev_path_desc,
1670 opt->branch1, opt->branch2,
1671 1 + opt->priv->call_depth * 2, &mfi))
1672 return -1;
1673 free(prev_path_desc);
1674
1675 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1676 return handle_file_collision(opt,
1677 c->path, a->path, NULL,
1678 rename_branch, add_branch,
1679 &mfi.blob,
1680 &ci->ren1->dst_entry->stages[other_stage]);
1681 }
1682
1683 static char *find_path_for_conflict(struct merge_options *opt,
1684 const char *path,
1685 const char *branch1,
1686 const char *branch2)
1687 {
1688 char *new_path = NULL;
1689 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1690 new_path = unique_path(opt, path, branch1);
1691 output(opt, 1, _("%s is a directory in %s adding "
1692 "as %s instead"),
1693 path, branch2, new_path);
1694 } else if (would_lose_untracked(opt, path)) {
1695 new_path = unique_path(opt, path, branch1);
1696 output(opt, 1, _("Refusing to lose untracked file"
1697 " at %s; adding as %s instead"),
1698 path, new_path);
1699 }
1700
1701 return new_path;
1702 }
1703
1704 /*
1705 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1706 */
1707 static inline int flip_stage(int stage)
1708 {
1709 return (2 + 3) - stage;
1710 }
1711
1712 static int handle_rename_rename_1to2(struct merge_options *opt,
1713 struct rename_conflict_info *ci)
1714 {
1715 /* One file was renamed in both branches, but to different names. */
1716 struct merge_file_info mfi;
1717 struct diff_filespec *add;
1718 struct diff_filespec *o = ci->ren1->pair->one;
1719 struct diff_filespec *a = ci->ren1->pair->two;
1720 struct diff_filespec *b = ci->ren2->pair->two;
1721 char *path_desc;
1722
1723 output(opt, 1, _("CONFLICT (rename/rename): "
1724 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1725 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1726 o->path, a->path, ci->ren1->branch,
1727 o->path, b->path, ci->ren2->branch,
1728 opt->priv->call_depth ? _(" (left unresolved)") : "");
1729
1730 path_desc = xstrfmt("%s and %s, both renamed from %s",
1731 a->path, b->path, o->path);
1732 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1733 ci->ren1->branch, ci->ren2->branch,
1734 opt->priv->call_depth * 2, &mfi))
1735 return -1;
1736 free(path_desc);
1737
1738 if (opt->priv->call_depth)
1739 remove_file_from_index(opt->repo->index, o->path);
1740
1741 /*
1742 * For each destination path, we need to see if there is a
1743 * rename/add collision. If not, we can write the file out
1744 * to the specified location.
1745 */
1746 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1747 if (is_valid(add)) {
1748 add->path = mfi.blob.path = a->path;
1749 if (handle_file_collision(opt, a->path,
1750 NULL, NULL,
1751 ci->ren1->branch,
1752 ci->ren2->branch,
1753 &mfi.blob, add) < 0)
1754 return -1;
1755 } else {
1756 char *new_path = find_path_for_conflict(opt, a->path,
1757 ci->ren1->branch,
1758 ci->ren2->branch);
1759 if (update_file(opt, 0, &mfi.blob,
1760 new_path ? new_path : a->path))
1761 return -1;
1762 free(new_path);
1763 if (!opt->priv->call_depth &&
1764 update_stages(opt, a->path, NULL, a, NULL))
1765 return -1;
1766 }
1767
1768 if (!mfi.clean && mfi.blob.mode == a->mode &&
1769 oideq(&mfi.blob.oid, &a->oid)) {
1770 /*
1771 * Getting here means we were attempting to merge a binary
1772 * blob. Since we can't merge binaries, the merge algorithm
1773 * just takes one side. But we don't want to copy the
1774 * contents of one side to both paths; we'd rather use the
1775 * original content at the given path for each path.
1776 */
1777 oidcpy(&mfi.blob.oid, &b->oid);
1778 mfi.blob.mode = b->mode;
1779 }
1780 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1781 if (is_valid(add)) {
1782 add->path = mfi.blob.path = b->path;
1783 if (handle_file_collision(opt, b->path,
1784 NULL, NULL,
1785 ci->ren1->branch,
1786 ci->ren2->branch,
1787 add, &mfi.blob) < 0)
1788 return -1;
1789 } else {
1790 char *new_path = find_path_for_conflict(opt, b->path,
1791 ci->ren2->branch,
1792 ci->ren1->branch);
1793 if (update_file(opt, 0, &mfi.blob,
1794 new_path ? new_path : b->path))
1795 return -1;
1796 free(new_path);
1797 if (!opt->priv->call_depth &&
1798 update_stages(opt, b->path, NULL, NULL, b))
1799 return -1;
1800 }
1801
1802 return 0;
1803 }
1804
1805 static int handle_rename_rename_2to1(struct merge_options *opt,
1806 struct rename_conflict_info *ci)
1807 {
1808 /* Two files, a & b, were renamed to the same thing, c. */
1809 struct diff_filespec *a = ci->ren1->pair->one;
1810 struct diff_filespec *b = ci->ren2->pair->one;
1811 struct diff_filespec *c1 = ci->ren1->pair->two;
1812 struct diff_filespec *c2 = ci->ren2->pair->two;
1813 char *path = c1->path; /* == c2->path */
1814 char *path_side_1_desc;
1815 char *path_side_2_desc;
1816 struct merge_file_info mfi_c1;
1817 struct merge_file_info mfi_c2;
1818 int ostage1, ostage2;
1819
1820 output(opt, 1, _("CONFLICT (rename/rename): "
1821 "Rename %s->%s in %s. "
1822 "Rename %s->%s in %s"),
1823 a->path, c1->path, ci->ren1->branch,
1824 b->path, c2->path, ci->ren2->branch);
1825
1826 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1827 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1828 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1829 ostage2 = flip_stage(ostage1);
1830 ci->ren1->src_entry->stages[ostage1].path = a->path;
1831 ci->ren2->src_entry->stages[ostage2].path = b->path;
1832 if (merge_mode_and_contents(opt, a, c1,
1833 &ci->ren1->src_entry->stages[ostage1],
1834 path_side_1_desc,
1835 opt->branch1, opt->branch2,
1836 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1837 merge_mode_and_contents(opt, b,
1838 &ci->ren2->src_entry->stages[ostage2],
1839 c2, path_side_2_desc,
1840 opt->branch1, opt->branch2,
1841 1 + opt->priv->call_depth * 2, &mfi_c2))
1842 return -1;
1843 free(path_side_1_desc);
1844 free(path_side_2_desc);
1845 mfi_c1.blob.path = path;
1846 mfi_c2.blob.path = path;
1847
1848 return handle_file_collision(opt, path, a->path, b->path,
1849 ci->ren1->branch, ci->ren2->branch,
1850 &mfi_c1.blob, &mfi_c2.blob);
1851 }
1852
1853 /*
1854 * Get the diff_filepairs changed between o_tree and tree.
1855 */
1856 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1857 struct tree *o_tree,
1858 struct tree *tree)
1859 {
1860 struct diff_queue_struct *ret;
1861 struct diff_options opts;
1862
1863 repo_diff_setup(opt->repo, &opts);
1864 opts.flags.recursive = 1;
1865 opts.flags.rename_empty = 0;
1866 opts.detect_rename = merge_detect_rename(opt);
1867 /*
1868 * We do not have logic to handle the detection of copies. In
1869 * fact, it may not even make sense to add such logic: would we
1870 * really want a change to a base file to be propagated through
1871 * multiple other files by a merge?
1872 */
1873 if (opts.detect_rename > DIFF_DETECT_RENAME)
1874 opts.detect_rename = DIFF_DETECT_RENAME;
1875 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 7000;
1876 opts.rename_score = opt->rename_score;
1877 opts.show_rename_progress = opt->show_rename_progress;
1878 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1879 diff_setup_done(&opts);
1880 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1881 diffcore_std(&opts);
1882 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1883 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1884
1885 ret = xmalloc(sizeof(*ret));
1886 *ret = diff_queued_diff;
1887
1888 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1889 diff_queued_diff.nr = 0;
1890 diff_queued_diff.queue = NULL;
1891 diff_flush(&opts);
1892 return ret;
1893 }
1894
1895 static int tree_has_path(struct repository *r, struct tree *tree,
1896 const char *path)
1897 {
1898 struct object_id hashy;
1899 unsigned short mode_o;
1900
1901 return !get_tree_entry(r,
1902 &tree->object.oid, path,
1903 &hashy, &mode_o);
1904 }
1905
1906 /*
1907 * Return a new string that replaces the beginning portion (which matches
1908 * entry->dir), with entry->new_dir. In perl-speak:
1909 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1910 * NOTE:
1911 * Caller must ensure that old_path starts with entry->dir + '/'.
1912 */
1913 static char *apply_dir_rename(struct dir_rename_entry *entry,
1914 const char *old_path)
1915 {
1916 struct strbuf new_path = STRBUF_INIT;
1917 int oldlen, newlen;
1918
1919 if (entry->non_unique_new_dir)
1920 return NULL;
1921
1922 oldlen = strlen(entry->dir);
1923 if (entry->new_dir.len == 0)
1924 /*
1925 * If someone renamed/merged a subdirectory into the root
1926 * directory (e.g. 'some/subdir' -> ''), then we want to
1927 * avoid returning
1928 * '' + '/filename'
1929 * as the rename; we need to make old_path + oldlen advance
1930 * past the '/' character.
1931 */
1932 oldlen++;
1933 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1934 strbuf_grow(&new_path, newlen);
1935 strbuf_addbuf(&new_path, &entry->new_dir);
1936 strbuf_addstr(&new_path, &old_path[oldlen]);
1937
1938 return strbuf_detach(&new_path, NULL);
1939 }
1940
1941 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1942 char **old_dir, char **new_dir)
1943 {
1944 char *end_of_old, *end_of_new;
1945
1946 /* Default return values: NULL, meaning no rename */
1947 *old_dir = NULL;
1948 *new_dir = NULL;
1949
1950 /*
1951 * For
1952 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1953 * the "e/foo.c" part is the same, we just want to know that
1954 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1955 * so, for this example, this function returns "a/b/c/d" in
1956 * *old_dir and "a/b/some/thing/else" in *new_dir.
1957 */
1958
1959 /*
1960 * If the basename of the file changed, we don't care. We want
1961 * to know which portion of the directory, if any, changed.
1962 */
1963 end_of_old = strrchr(old_path, '/');
1964 end_of_new = strrchr(new_path, '/');
1965
1966 /*
1967 * If end_of_old is NULL, old_path wasn't in a directory, so there
1968 * could not be a directory rename (our rule elsewhere that a
1969 * directory which still exists is not considered to have been
1970 * renamed means the root directory can never be renamed -- because
1971 * the root directory always exists).
1972 */
1973 if (end_of_old == NULL)
1974 return; /* Note: *old_dir and *new_dir are still NULL */
1975
1976 /*
1977 * If new_path contains no directory (end_of_new is NULL), then we
1978 * have a rename of old_path's directory to the root directory.
1979 */
1980 if (end_of_new == NULL) {
1981 *old_dir = xstrndup(old_path, end_of_old - old_path);
1982 *new_dir = xstrdup("");
1983 return;
1984 }
1985
1986 /* Find the first non-matching character traversing backwards */
1987 while (*--end_of_new == *--end_of_old &&
1988 end_of_old != old_path &&
1989 end_of_new != new_path)
1990 ; /* Do nothing; all in the while loop */
1991
1992 /*
1993 * If both got back to the beginning of their strings, then the
1994 * directory didn't change at all, only the basename did.
1995 */
1996 if (end_of_old == old_path && end_of_new == new_path &&
1997 *end_of_old == *end_of_new)
1998 return; /* Note: *old_dir and *new_dir are still NULL */
1999
2000 /*
2001 * If end_of_new got back to the beginning of its string, and
2002 * end_of_old got back to the beginning of some subdirectory, then
2003 * we have a rename/merge of a subdirectory into the root, which
2004 * needs slightly special handling.
2005 *
2006 * Note: There is no need to consider the opposite case, with a
2007 * rename/merge of the root directory into some subdirectory
2008 * because as noted above the root directory always exists so it
2009 * cannot be considered to be renamed.
2010 */
2011 if (end_of_new == new_path &&
2012 end_of_old != old_path && end_of_old[-1] == '/') {
2013 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2014 *new_dir = xstrdup("");
2015 return;
2016 }
2017
2018 /*
2019 * We've found the first non-matching character in the directory
2020 * paths. That means the current characters we were looking at
2021 * were part of the first non-matching subdir name going back from
2022 * the end of the strings. Get the whole name by advancing both
2023 * end_of_old and end_of_new to the NEXT '/' character. That will
2024 * represent the entire directory rename.
2025 *
2026 * The reason for the increment is cases like
2027 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2028 * After dropping the basename and going back to the first
2029 * non-matching character, we're now comparing:
2030 * a/b/s and a/b/
2031 * and we want to be comparing:
2032 * a/b/star/ and a/b/tar/
2033 * but without the pre-increment, the one on the right would stay
2034 * a/b/.
2035 */
2036 end_of_old = strchr(++end_of_old, '/');
2037 end_of_new = strchr(++end_of_new, '/');
2038
2039 /* Copy the old and new directories into *old_dir and *new_dir. */
2040 *old_dir = xstrndup(old_path, end_of_old - old_path);
2041 *new_dir = xstrndup(new_path, end_of_new - new_path);
2042 }
2043
2044 static void remove_hashmap_entries(struct hashmap *dir_renames,
2045 struct string_list *items_to_remove)
2046 {
2047 int i;
2048 struct dir_rename_entry *entry;
2049
2050 for (i = 0; i < items_to_remove->nr; i++) {
2051 entry = items_to_remove->items[i].util;
2052 hashmap_remove(dir_renames, &entry->ent, NULL);
2053 }
2054 string_list_clear(items_to_remove, 0);
2055 }
2056
2057 /*
2058 * See if there is a directory rename for path, and if there are any file
2059 * level conflicts for the renamed location. If there is a rename and
2060 * there are no conflicts, return the new name. Otherwise, return NULL.
2061 */
2062 static char *handle_path_level_conflicts(struct merge_options *opt,
2063 const char *path,
2064 struct dir_rename_entry *entry,
2065 struct hashmap *collisions,
2066 struct tree *tree)
2067 {
2068 char *new_path = NULL;
2069 struct collision_entry *collision_ent;
2070 int clean = 1;
2071 struct strbuf collision_paths = STRBUF_INIT;
2072
2073 /*
2074 * entry has the mapping of old directory name to new directory name
2075 * that we want to apply to path.
2076 */
2077 new_path = apply_dir_rename(entry, path);
2078
2079 if (!new_path) {
2080 /* This should only happen when entry->non_unique_new_dir set */
2081 if (!entry->non_unique_new_dir)
2082 BUG("entry->non_unqiue_dir not set and !new_path");
2083 output(opt, 1, _("CONFLICT (directory rename split): "
2084 "Unclear where to place %s because directory "
2085 "%s was renamed to multiple other directories, "
2086 "with no destination getting a majority of the "
2087 "files."),
2088 path, entry->dir);
2089 clean = 0;
2090 return NULL;
2091 }
2092
2093 /*
2094 * The caller needs to have ensured that it has pre-populated
2095 * collisions with all paths that map to new_path. Do a quick check
2096 * to ensure that's the case.
2097 */
2098 collision_ent = collision_find_entry(collisions, new_path);
2099 if (collision_ent == NULL)
2100 BUG("collision_ent is NULL");
2101
2102 /*
2103 * Check for one-sided add/add/.../add conflicts, i.e.
2104 * where implicit renames from the other side doing
2105 * directory rename(s) can affect this side of history
2106 * to put multiple paths into the same location. Warn
2107 * and bail on directory renames for such paths.
2108 */
2109 if (collision_ent->reported_already) {
2110 clean = 0;
2111 } else if (tree_has_path(opt->repo, tree, new_path)) {
2112 collision_ent->reported_already = 1;
2113 strbuf_add_separated_string_list(&collision_paths, ", ",
2114 &collision_ent->source_files);
2115 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2116 "file/dir at %s in the way of implicit "
2117 "directory rename(s) putting the following "
2118 "path(s) there: %s."),
2119 new_path, collision_paths.buf);
2120 clean = 0;
2121 } else if (collision_ent->source_files.nr > 1) {
2122 collision_ent->reported_already = 1;
2123 strbuf_add_separated_string_list(&collision_paths, ", ",
2124 &collision_ent->source_files);
2125 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2126 "more than one path to %s; implicit directory "
2127 "renames tried to put these paths there: %s"),
2128 new_path, collision_paths.buf);
2129 clean = 0;
2130 }
2131
2132 /* Free memory we no longer need */
2133 strbuf_release(&collision_paths);
2134 if (!clean && new_path) {
2135 free(new_path);
2136 return NULL;
2137 }
2138
2139 return new_path;
2140 }
2141
2142 /*
2143 * There are a couple things we want to do at the directory level:
2144 * 1. Check for both sides renaming to the same thing, in order to avoid
2145 * implicit renaming of files that should be left in place. (See
2146 * testcase 6b in t6043 for details.)
2147 * 2. Prune directory renames if there are still files left in the
2148 * original directory. These represent a partial directory rename,
2149 * i.e. a rename where only some of the files within the directory
2150 * were renamed elsewhere. (Technically, this could be done earlier
2151 * in get_directory_renames(), except that would prevent us from
2152 * doing the previous check and thus failing testcase 6b.)
2153 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2154 * In the future, we could potentially record this info as well and
2155 * omit reporting rename/rename(1to2) conflicts for each path within
2156 * the affected directories, thus cleaning up the merge output.
2157 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2158 * directory level, because merging directories is fine. If it
2159 * causes conflicts for files within those merged directories, then
2160 * that should be detected at the individual path level.
2161 */
2162 static void handle_directory_level_conflicts(struct merge_options *opt,
2163 struct hashmap *dir_re_head,
2164 struct tree *head,
2165 struct hashmap *dir_re_merge,
2166 struct tree *merge)
2167 {
2168 struct hashmap_iter iter;
2169 struct dir_rename_entry *head_ent;
2170 struct dir_rename_entry *merge_ent;
2171
2172 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2173 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2174
2175 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2176 ent /* member name */) {
2177 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2178 if (merge_ent &&
2179 !head_ent->non_unique_new_dir &&
2180 !merge_ent->non_unique_new_dir &&
2181 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2182 /* 1. Renamed identically; remove it from both sides */
2183 string_list_append(&remove_from_head,
2184 head_ent->dir)->util = head_ent;
2185 strbuf_release(&head_ent->new_dir);
2186 string_list_append(&remove_from_merge,
2187 merge_ent->dir)->util = merge_ent;
2188 strbuf_release(&merge_ent->new_dir);
2189 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2190 /* 2. This wasn't a directory rename after all */
2191 string_list_append(&remove_from_head,
2192 head_ent->dir)->util = head_ent;
2193 strbuf_release(&head_ent->new_dir);
2194 }
2195 }
2196
2197 remove_hashmap_entries(dir_re_head, &remove_from_head);
2198 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2199
2200 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2201 ent /* member name */) {
2202 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2203 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2204 /* 2. This wasn't a directory rename after all */
2205 string_list_append(&remove_from_merge,
2206 merge_ent->dir)->util = merge_ent;
2207 } else if (head_ent &&
2208 !head_ent->non_unique_new_dir &&
2209 !merge_ent->non_unique_new_dir) {
2210 /* 3. rename/rename(1to2) */
2211 /*
2212 * We can assume it's not rename/rename(1to1) because
2213 * that was case (1), already checked above. So we
2214 * know that head_ent->new_dir and merge_ent->new_dir
2215 * are different strings.
2216 */
2217 output(opt, 1, _("CONFLICT (rename/rename): "
2218 "Rename directory %s->%s in %s. "
2219 "Rename directory %s->%s in %s"),
2220 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2221 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2222 string_list_append(&remove_from_head,
2223 head_ent->dir)->util = head_ent;
2224 strbuf_release(&head_ent->new_dir);
2225 string_list_append(&remove_from_merge,
2226 merge_ent->dir)->util = merge_ent;
2227 strbuf_release(&merge_ent->new_dir);
2228 }
2229 }
2230
2231 remove_hashmap_entries(dir_re_head, &remove_from_head);
2232 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2233 }
2234
2235 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2236 {
2237 struct hashmap *dir_renames;
2238 struct hashmap_iter iter;
2239 struct dir_rename_entry *entry;
2240 int i;
2241
2242 /*
2243 * Typically, we think of a directory rename as all files from a
2244 * certain directory being moved to a target directory. However,
2245 * what if someone first moved two files from the original
2246 * directory in one commit, and then renamed the directory
2247 * somewhere else in a later commit? At merge time, we just know
2248 * that files from the original directory went to two different
2249 * places, and that the bulk of them ended up in the same place.
2250 * We want each directory rename to represent where the bulk of the
2251 * files from that directory end up; this function exists to find
2252 * where the bulk of the files went.
2253 *
2254 * The first loop below simply iterates through the list of file
2255 * renames, finding out how often each directory rename pair
2256 * possibility occurs.
2257 */
2258 dir_renames = xmalloc(sizeof(*dir_renames));
2259 dir_rename_init(dir_renames);
2260 for (i = 0; i < pairs->nr; ++i) {
2261 struct string_list_item *item;
2262 int *count;
2263 struct diff_filepair *pair = pairs->queue[i];
2264 char *old_dir, *new_dir;
2265
2266 /* File not part of directory rename if it wasn't renamed */
2267 if (pair->status != 'R')
2268 continue;
2269
2270 get_renamed_dir_portion(pair->one->path, pair->two->path,
2271 &old_dir, &new_dir);
2272 if (!old_dir)
2273 /* Directory didn't change at all; ignore this one. */
2274 continue;
2275
2276 entry = dir_rename_find_entry(dir_renames, old_dir);
2277 if (!entry) {
2278 entry = xmalloc(sizeof(*entry));
2279 dir_rename_entry_init(entry, old_dir);
2280 hashmap_put(dir_renames, &entry->ent);
2281 } else {
2282 free(old_dir);
2283 }
2284 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2285 if (!item) {
2286 item = string_list_insert(&entry->possible_new_dirs,
2287 new_dir);
2288 item->util = xcalloc(1, sizeof(int));
2289 } else {
2290 free(new_dir);
2291 }
2292 count = item->util;
2293 *count += 1;
2294 }
2295
2296 /*
2297 * For each directory with files moved out of it, we find out which
2298 * target directory received the most files so we can declare it to
2299 * be the "winning" target location for the directory rename. This
2300 * winner gets recorded in new_dir. If there is no winner
2301 * (multiple target directories received the same number of files),
2302 * we set non_unique_new_dir. Once we've determined the winner (or
2303 * that there is no winner), we no longer need possible_new_dirs.
2304 */
2305 hashmap_for_each_entry(dir_renames, &iter, entry,
2306 ent /* member name */) {
2307 int max = 0;
2308 int bad_max = 0;
2309 char *best = NULL;
2310
2311 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2312 int *count = entry->possible_new_dirs.items[i].util;
2313
2314 if (*count == max)
2315 bad_max = max;
2316 else if (*count > max) {
2317 max = *count;
2318 best = entry->possible_new_dirs.items[i].string;
2319 }
2320 }
2321 if (bad_max == max)
2322 entry->non_unique_new_dir = 1;
2323 else {
2324 assert(entry->new_dir.len == 0);
2325 strbuf_addstr(&entry->new_dir, best);
2326 }
2327 /*
2328 * The relevant directory sub-portion of the original full
2329 * filepaths were xstrndup'ed before inserting into
2330 * possible_new_dirs, and instead of manually iterating the
2331 * list and free'ing each, just lie and tell
2332 * possible_new_dirs that it did the strdup'ing so that it
2333 * will free them for us.
2334 */
2335 entry->possible_new_dirs.strdup_strings = 1;
2336 string_list_clear(&entry->possible_new_dirs, 1);
2337 }
2338
2339 return dir_renames;
2340 }
2341
2342 static struct dir_rename_entry *check_dir_renamed(const char *path,
2343 struct hashmap *dir_renames)
2344 {
2345 char *temp = xstrdup(path);
2346 char *end;
2347 struct dir_rename_entry *entry = NULL;
2348
2349 while ((end = strrchr(temp, '/'))) {
2350 *end = '\0';
2351 entry = dir_rename_find_entry(dir_renames, temp);
2352 if (entry)
2353 break;
2354 }
2355 free(temp);
2356 return entry;
2357 }
2358
2359 static void compute_collisions(struct hashmap *collisions,
2360 struct hashmap *dir_renames,
2361 struct diff_queue_struct *pairs)
2362 {
2363 int i;
2364
2365 /*
2366 * Multiple files can be mapped to the same path due to directory
2367 * renames done by the other side of history. Since that other
2368 * side of history could have merged multiple directories into one,
2369 * if our side of history added the same file basename to each of
2370 * those directories, then all N of them would get implicitly
2371 * renamed by the directory rename detection into the same path,
2372 * and we'd get an add/add/.../add conflict, and all those adds
2373 * from *this* side of history. This is not representable in the
2374 * index, and users aren't going to easily be able to make sense of
2375 * it. So we need to provide a good warning about what's
2376 * happening, and fall back to no-directory-rename detection
2377 * behavior for those paths.
2378 *
2379 * See testcases 9e and all of section 5 from t6043 for examples.
2380 */
2381 collision_init(collisions);
2382
2383 for (i = 0; i < pairs->nr; ++i) {
2384 struct dir_rename_entry *dir_rename_ent;
2385 struct collision_entry *collision_ent;
2386 char *new_path;
2387 struct diff_filepair *pair = pairs->queue[i];
2388
2389 if (pair->status != 'A' && pair->status != 'R')
2390 continue;
2391 dir_rename_ent = check_dir_renamed(pair->two->path,
2392 dir_renames);
2393 if (!dir_rename_ent)
2394 continue;
2395
2396 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2397 if (!new_path)
2398 /*
2399 * dir_rename_ent->non_unique_new_path is true, which
2400 * means there is no directory rename for us to use,
2401 * which means it won't cause us any additional
2402 * collisions.
2403 */
2404 continue;
2405 collision_ent = collision_find_entry(collisions, new_path);
2406 if (!collision_ent) {
2407 CALLOC_ARRAY(collision_ent, 1);
2408 hashmap_entry_init(&collision_ent->ent,
2409 strhash(new_path));
2410 hashmap_put(collisions, &collision_ent->ent);
2411 collision_ent->target_file = new_path;
2412 } else {
2413 free(new_path);
2414 }
2415 string_list_insert(&collision_ent->source_files,
2416 pair->two->path);
2417 }
2418 }
2419
2420 static char *check_for_directory_rename(struct merge_options *opt,
2421 const char *path,
2422 struct tree *tree,
2423 struct hashmap *dir_renames,
2424 struct hashmap *dir_rename_exclusions,
2425 struct hashmap *collisions,
2426 int *clean_merge)
2427 {
2428 char *new_path = NULL;
2429 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2430 struct dir_rename_entry *oentry = NULL;
2431
2432 if (!entry)
2433 return new_path;
2434
2435 /*
2436 * This next part is a little weird. We do not want to do an
2437 * implicit rename into a directory we renamed on our side, because
2438 * that will result in a spurious rename/rename(1to2) conflict. An
2439 * example:
2440 * Base commit: dumbdir/afile, otherdir/bfile
2441 * Side 1: smrtdir/afile, otherdir/bfile
2442 * Side 2: dumbdir/afile, dumbdir/bfile
2443 * Here, while working on Side 1, we could notice that otherdir was
2444 * renamed/merged to dumbdir, and change the diff_filepair for
2445 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2446 * 2 will notice the rename from dumbdir to smrtdir, and do the
2447 * transitive rename to move it from dumbdir/bfile to
2448 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2449 * smrtdir, a rename/rename(1to2) conflict. We really just want
2450 * the file to end up in smrtdir. And the way to achieve that is
2451 * to not let Side1 do the rename to dumbdir, since we know that is
2452 * the source of one of our directory renames.
2453 *
2454 * That's why oentry and dir_rename_exclusions is here.
2455 *
2456 * As it turns out, this also prevents N-way transient rename
2457 * confusion; See testcases 9c and 9d of t6043.
2458 */
2459 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2460 if (oentry) {
2461 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2462 "to %s, because %s itself was renamed."),
2463 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2464 } else {
2465 new_path = handle_path_level_conflicts(opt, path, entry,
2466 collisions, tree);
2467 *clean_merge &= (new_path != NULL);
2468 }
2469
2470 return new_path;
2471 }
2472
2473 static void apply_directory_rename_modifications(struct merge_options *opt,
2474 struct diff_filepair *pair,
2475 char *new_path,
2476 struct rename *re,
2477 struct tree *tree,
2478 struct tree *o_tree,
2479 struct tree *a_tree,
2480 struct tree *b_tree,
2481 struct string_list *entries)
2482 {
2483 struct string_list_item *item;
2484 int stage = (tree == a_tree ? 2 : 3);
2485 int update_wd;
2486
2487 /*
2488 * In all cases where we can do directory rename detection,
2489 * unpack_trees() will have read pair->two->path into the
2490 * index and the working copy. We need to remove it so that
2491 * we can instead place it at new_path. It is guaranteed to
2492 * not be untracked (unpack_trees() would have errored out
2493 * saying the file would have been overwritten), but it might
2494 * be dirty, though.
2495 */
2496 update_wd = !was_dirty(opt, pair->two->path);
2497 if (!update_wd)
2498 output(opt, 1, _("Refusing to lose dirty file at %s"),
2499 pair->two->path);
2500 remove_file(opt, 1, pair->two->path, !update_wd);
2501
2502 /* Find or create a new re->dst_entry */
2503 item = string_list_lookup(entries, new_path);
2504 if (item) {
2505 /*
2506 * Since we're renaming on this side of history, and it's
2507 * due to a directory rename on the other side of history
2508 * (which we only allow when the directory in question no
2509 * longer exists on the other side of history), the
2510 * original entry for re->dst_entry is no longer
2511 * necessary...
2512 */
2513 re->dst_entry->processed = 1;
2514
2515 /*
2516 * ...because we'll be using this new one.
2517 */
2518 re->dst_entry = item->util;
2519 } else {
2520 /*
2521 * re->dst_entry is for the before-dir-rename path, and we
2522 * need it to hold information for the after-dir-rename
2523 * path. Before creating a new entry, we need to mark the
2524 * old one as unnecessary (...unless it is shared by
2525 * src_entry, i.e. this didn't use to be a rename, in which
2526 * case we can just allow the normal processing to happen
2527 * for it).
2528 */
2529 if (pair->status == 'R')
2530 re->dst_entry->processed = 1;
2531
2532 re->dst_entry = insert_stage_data(opt->repo, new_path,
2533 o_tree, a_tree, b_tree,
2534 entries);
2535 item = string_list_insert(entries, new_path);
2536 item->util = re->dst_entry;
2537 }
2538
2539 /*
2540 * Update the stage_data with the information about the path we are
2541 * moving into place. That slot will be empty and available for us
2542 * to write to because of the collision checks in
2543 * handle_path_level_conflicts(). In other words,
2544 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2545 * open for us to write to.
2546 *
2547 * It may be tempting to actually update the index at this point as
2548 * well, using update_stages_for_stage_data(), but as per the big
2549 * "NOTE" in update_stages(), doing so will modify the current
2550 * in-memory index which will break calls to would_lose_untracked()
2551 * that we need to make. Instead, we need to just make sure that
2552 * the various handle_rename_*() functions update the index
2553 * explicitly rather than relying on unpack_trees() to have done it.
2554 */
2555 get_tree_entry(opt->repo,
2556 &tree->object.oid,
2557 pair->two->path,
2558 &re->dst_entry->stages[stage].oid,
2559 &re->dst_entry->stages[stage].mode);
2560
2561 /*
2562 * Record the original change status (or 'type' of change). If it
2563 * was originally an add ('A'), this lets us differentiate later
2564 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2565 * otherwise look the same). If it was originally a rename ('R'),
2566 * this lets us remember and report accurately about the transitive
2567 * renaming that occurred via the directory rename detection. Also,
2568 * record the original destination name.
2569 */
2570 re->dir_rename_original_type = pair->status;
2571 re->dir_rename_original_dest = pair->two->path;
2572
2573 /*
2574 * We don't actually look at pair->status again, but it seems
2575 * pedagogically correct to adjust it.
2576 */
2577 pair->status = 'R';
2578
2579 /*
2580 * Finally, record the new location.
2581 */
2582 pair->two->path = new_path;
2583 }
2584
2585 /*
2586 * Get information of all renames which occurred in 'pairs', making use of
2587 * any implicit directory renames inferred from the other side of history.
2588 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2589 * to be able to associate the correct cache entries with the rename
2590 * information; tree is always equal to either a_tree or b_tree.
2591 */
2592 static struct string_list *get_renames(struct merge_options *opt,
2593 const char *branch,
2594 struct diff_queue_struct *pairs,
2595 struct hashmap *dir_renames,
2596 struct hashmap *dir_rename_exclusions,
2597 struct tree *tree,
2598 struct tree *o_tree,
2599 struct tree *a_tree,
2600 struct tree *b_tree,
2601 struct string_list *entries,
2602 int *clean_merge)
2603 {
2604 int i;
2605 struct hashmap collisions;
2606 struct hashmap_iter iter;
2607 struct collision_entry *e;
2608 struct string_list *renames;
2609
2610 compute_collisions(&collisions, dir_renames, pairs);
2611 CALLOC_ARRAY(renames, 1);
2612
2613 for (i = 0; i < pairs->nr; ++i) {
2614 struct string_list_item *item;
2615 struct rename *re;
2616 struct diff_filepair *pair = pairs->queue[i];
2617 char *new_path; /* non-NULL only with directory renames */
2618
2619 if (pair->status != 'A' && pair->status != 'R') {
2620 diff_free_filepair(pair);
2621 continue;
2622 }
2623 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2624 dir_renames,
2625 dir_rename_exclusions,
2626 &collisions,
2627 clean_merge);
2628 if (pair->status != 'R' && !new_path) {
2629 diff_free_filepair(pair);
2630 continue;
2631 }
2632
2633 re = xmalloc(sizeof(*re));
2634 re->processed = 0;
2635 re->pair = pair;
2636 re->branch = branch;
2637 re->dir_rename_original_type = '\0';
2638 re->dir_rename_original_dest = NULL;
2639 item = string_list_lookup(entries, re->pair->one->path);
2640 if (!item)
2641 re->src_entry = insert_stage_data(opt->repo,
2642 re->pair->one->path,
2643 o_tree, a_tree, b_tree, entries);
2644 else
2645 re->src_entry = item->util;
2646
2647 item = string_list_lookup(entries, re->pair->two->path);
2648 if (!item)
2649 re->dst_entry = insert_stage_data(opt->repo,
2650 re->pair->two->path,
2651 o_tree, a_tree, b_tree, entries);
2652 else
2653 re->dst_entry = item->util;
2654 item = string_list_insert(renames, pair->one->path);
2655 item->util = re;
2656 if (new_path)
2657 apply_directory_rename_modifications(opt, pair, new_path,
2658 re, tree, o_tree,
2659 a_tree, b_tree,
2660 entries);
2661 }
2662
2663 hashmap_for_each_entry(&collisions, &iter, e,
2664 ent /* member name */) {
2665 free(e->target_file);
2666 string_list_clear(&e->source_files, 0);
2667 }
2668 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2669 return renames;
2670 }
2671
2672 static int process_renames(struct merge_options *opt,
2673 struct string_list *a_renames,
2674 struct string_list *b_renames)
2675 {
2676 int clean_merge = 1, i, j;
2677 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2678 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2679 const struct rename *sre;
2680
2681 /*
2682 * FIXME: As string-list.h notes, it's O(n^2) to build a sorted
2683 * string_list one-by-one, but O(n log n) to build it unsorted and
2684 * then sort it. Note that as we build the list, we do not need to
2685 * check if the existing destination path is already in the list,
2686 * because the structure of diffcore_rename guarantees we won't
2687 * have duplicates.
2688 */
2689 for (i = 0; i < a_renames->nr; i++) {
2690 sre = a_renames->items[i].util;
2691 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2692 = (void *)sre;
2693 }
2694 for (i = 0; i < b_renames->nr; i++) {
2695 sre = b_renames->items[i].util;
2696 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2697 = (void *)sre;
2698 }
2699
2700 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2701 struct string_list *renames1, *renames2Dst;
2702 struct rename *ren1 = NULL, *ren2 = NULL;
2703 const char *ren1_src, *ren1_dst;
2704 struct string_list_item *lookup;
2705
2706 if (i >= a_renames->nr) {
2707 ren2 = b_renames->items[j++].util;
2708 } else if (j >= b_renames->nr) {
2709 ren1 = a_renames->items[i++].util;
2710 } else {
2711 int compare = strcmp(a_renames->items[i].string,
2712 b_renames->items[j].string);
2713 if (compare <= 0)
2714 ren1 = a_renames->items[i++].util;
2715 if (compare >= 0)
2716 ren2 = b_renames->items[j++].util;
2717 }
2718
2719 /* TODO: refactor, so that 1/2 are not needed */
2720 if (ren1) {
2721 renames1 = a_renames;
2722 renames2Dst = &b_by_dst;
2723 } else {
2724 renames1 = b_renames;
2725 renames2Dst = &a_by_dst;
2726 SWAP(ren2, ren1);
2727 }
2728
2729 if (ren1->processed)
2730 continue;
2731 ren1->processed = 1;
2732 ren1->dst_entry->processed = 1;
2733 /* BUG: We should only mark src_entry as processed if we
2734 * are not dealing with a rename + add-source case.
2735 */
2736 ren1->src_entry->processed = 1;
2737
2738 ren1_src = ren1->pair->one->path;
2739 ren1_dst = ren1->pair->two->path;
2740
2741 if (ren2) {
2742 /* One file renamed on both sides */
2743 const char *ren2_src = ren2->pair->one->path;
2744 const char *ren2_dst = ren2->pair->two->path;
2745 enum rename_type rename_type;
2746 if (strcmp(ren1_src, ren2_src) != 0)
2747 BUG("ren1_src != ren2_src");
2748 ren2->dst_entry->processed = 1;
2749 ren2->processed = 1;
2750 if (strcmp(ren1_dst, ren2_dst) != 0) {
2751 rename_type = RENAME_ONE_FILE_TO_TWO;
2752 clean_merge = 0;
2753 } else {
2754 rename_type = RENAME_ONE_FILE_TO_ONE;
2755 /* BUG: We should only remove ren1_src in
2756 * the base stage (think of rename +
2757 * add-source cases).
2758 */
2759 remove_file(opt, 1, ren1_src, 1);
2760 update_entry(ren1->dst_entry,
2761 ren1->pair->one,
2762 ren1->pair->two,
2763 ren2->pair->two);
2764 }
2765 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2766 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2767 /* Two different files renamed to the same thing */
2768 char *ren2_dst;
2769 ren2 = lookup->util;
2770 ren2_dst = ren2->pair->two->path;
2771 if (strcmp(ren1_dst, ren2_dst) != 0)
2772 BUG("ren1_dst != ren2_dst");
2773
2774 clean_merge = 0;
2775 ren2->processed = 1;
2776 /*
2777 * BUG: We should only mark src_entry as processed
2778 * if we are not dealing with a rename + add-source
2779 * case.
2780 */
2781 ren2->src_entry->processed = 1;
2782
2783 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2784 opt, ren1, ren2);
2785 } else {
2786 /* Renamed in 1, maybe changed in 2 */
2787 /* we only use sha1 and mode of these */
2788 struct diff_filespec src_other, dst_other;
2789 int try_merge;
2790
2791 /*
2792 * unpack_trees loads entries from common-commit
2793 * into stage 1, from head-commit into stage 2, and
2794 * from merge-commit into stage 3. We keep track
2795 * of which side corresponds to the rename.
2796 */
2797 int renamed_stage = a_renames == renames1 ? 2 : 3;
2798 int other_stage = a_renames == renames1 ? 3 : 2;
2799
2800 /*
2801 * Directory renames have a funny corner case...
2802 */
2803 int renamed_to_self = !strcmp(ren1_src, ren1_dst);
2804
2805 /* BUG: We should only remove ren1_src in the base
2806 * stage and in other_stage (think of rename +
2807 * add-source case).
2808 */
2809 if (!renamed_to_self)
2810 remove_file(opt, 1, ren1_src,
2811 renamed_stage == 2 ||
2812 !was_tracked(opt, ren1_src));
2813
2814 oidcpy(&src_other.oid,
2815 &ren1->src_entry->stages[other_stage].oid);
2816 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2817 oidcpy(&dst_other.oid,
2818 &ren1->dst_entry->stages[other_stage].oid);
2819 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2820 try_merge = 0;
2821
2822 if (oideq(&src_other.oid, null_oid()) &&
2823 ren1->dir_rename_original_type == 'A') {
2824 setup_rename_conflict_info(RENAME_VIA_DIR,
2825 opt, ren1, NULL);
2826 } else if (renamed_to_self) {
2827 setup_rename_conflict_info(RENAME_NORMAL,
2828 opt, ren1, NULL);
2829 } else if (oideq(&src_other.oid, null_oid())) {
2830 setup_rename_conflict_info(RENAME_DELETE,
2831 opt, ren1, NULL);
2832 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2833 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2834 /*
2835 * Added file on the other side identical to
2836 * the file being renamed: clean merge.
2837 * Also, there is no need to overwrite the
2838 * file already in the working copy, so call
2839 * update_file_flags() instead of
2840 * update_file().
2841 */
2842 if (update_file_flags(opt,
2843 ren1->pair->two,
2844 ren1_dst,
2845 1, /* update_cache */
2846 0 /* update_wd */))
2847 clean_merge = -1;
2848 } else if (!oideq(&dst_other.oid, null_oid())) {
2849 /*
2850 * Probably not a clean merge, but it's
2851 * premature to set clean_merge to 0 here,
2852 * because if the rename merges cleanly and
2853 * the merge exactly matches the newly added
2854 * file, then the merge will be clean.
2855 */
2856 setup_rename_conflict_info(RENAME_ADD,
2857 opt, ren1, NULL);
2858 } else
2859 try_merge = 1;
2860
2861 if (clean_merge < 0)
2862 goto cleanup_and_return;
2863 if (try_merge) {
2864 struct diff_filespec *o, *a, *b;
2865 src_other.path = (char *)ren1_src;
2866
2867 o = ren1->pair->one;
2868 if (a_renames == renames1) {
2869 a = ren1->pair->two;
2870 b = &src_other;
2871 } else {
2872 b = ren1->pair->two;
2873 a = &src_other;
2874 }
2875 update_entry(ren1->dst_entry, o, a, b);
2876 setup_rename_conflict_info(RENAME_NORMAL,
2877 opt, ren1, NULL);
2878 }
2879 }
2880 }
2881 cleanup_and_return:
2882 string_list_clear(&a_by_dst, 0);
2883 string_list_clear(&b_by_dst, 0);
2884
2885 return clean_merge;
2886 }
2887
2888 struct rename_info {
2889 struct string_list *head_renames;
2890 struct string_list *merge_renames;
2891 };
2892
2893 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2894 struct hashmap *dir_renames)
2895 {
2896 struct hashmap_iter iter;
2897 struct dir_rename_entry *e;
2898
2899 hashmap_for_each_entry(dir_renames, &iter, e,
2900 ent /* member name */) {
2901 free(e->dir);
2902 strbuf_release(&e->new_dir);
2903 /* possible_new_dirs already cleared in get_directory_renames */
2904 }
2905 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2906 free(dir_renames);
2907
2908 free(pairs->queue);
2909 free(pairs);
2910 }
2911
2912 static int detect_and_process_renames(struct merge_options *opt,
2913 struct tree *common,
2914 struct tree *head,
2915 struct tree *merge,
2916 struct string_list *entries,
2917 struct rename_info *ri)
2918 {
2919 struct diff_queue_struct *head_pairs, *merge_pairs;
2920 struct hashmap *dir_re_head, *dir_re_merge;
2921 int clean = 1;
2922
2923 ri->head_renames = NULL;
2924 ri->merge_renames = NULL;
2925
2926 if (!merge_detect_rename(opt))
2927 return 1;
2928
2929 head_pairs = get_diffpairs(opt, common, head);
2930 merge_pairs = get_diffpairs(opt, common, merge);
2931
2932 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2933 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2934 !opt->priv->call_depth)) {
2935 dir_re_head = get_directory_renames(head_pairs);
2936 dir_re_merge = get_directory_renames(merge_pairs);
2937
2938 handle_directory_level_conflicts(opt,
2939 dir_re_head, head,
2940 dir_re_merge, merge);
2941 } else {
2942 dir_re_head = xmalloc(sizeof(*dir_re_head));
2943 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2944 dir_rename_init(dir_re_head);
2945 dir_rename_init(dir_re_merge);
2946 }
2947
2948 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2949 dir_re_merge, dir_re_head, head,
2950 common, head, merge, entries,
2951 &clean);
2952 if (clean < 0)
2953 goto cleanup;
2954 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2955 dir_re_head, dir_re_merge, merge,
2956 common, head, merge, entries,
2957 &clean);
2958 if (clean < 0)
2959 goto cleanup;
2960 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2961
2962 cleanup:
2963 /*
2964 * Some cleanup is deferred until cleanup_renames() because the
2965 * data structures are still needed and referenced in
2966 * process_entry(). But there are a few things we can free now.
2967 */
2968 initial_cleanup_rename(head_pairs, dir_re_head);
2969 initial_cleanup_rename(merge_pairs, dir_re_merge);
2970
2971 return clean;
2972 }
2973
2974 static void final_cleanup_rename(struct string_list *rename)
2975 {
2976 const struct rename *re;
2977 int i;
2978
2979 if (rename == NULL)
2980 return;
2981
2982 for (i = 0; i < rename->nr; i++) {
2983 re = rename->items[i].util;
2984 diff_free_filepair(re->pair);
2985 }
2986 string_list_clear(rename, 1);
2987 free(rename);
2988 }
2989
2990 static void final_cleanup_renames(struct rename_info *re_info)
2991 {
2992 final_cleanup_rename(re_info->head_renames);
2993 final_cleanup_rename(re_info->merge_renames);
2994 }
2995
2996 static int read_oid_strbuf(struct merge_options *opt,
2997 const struct object_id *oid,
2998 struct strbuf *dst)
2999 {
3000 void *buf;
3001 enum object_type type;
3002 unsigned long size;
3003 buf = read_object_file(oid, &type, &size);
3004 if (!buf)
3005 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3006 if (type != OBJ_BLOB) {
3007 free(buf);
3008 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3009 }
3010 strbuf_attach(dst, buf, size, size + 1);
3011 return 0;
3012 }
3013
3014 static int blob_unchanged(struct merge_options *opt,
3015 const struct diff_filespec *o,
3016 const struct diff_filespec *a,
3017 int renormalize, const char *path)
3018 {
3019 struct strbuf obuf = STRBUF_INIT;
3020 struct strbuf abuf = STRBUF_INIT;
3021 int ret = 0; /* assume changed for safety */
3022 struct index_state *idx = opt->repo->index;
3023
3024 if (a->mode != o->mode)
3025 return 0;
3026 if (oideq(&o->oid, &a->oid))
3027 return 1;
3028 if (!renormalize)
3029 return 0;
3030
3031 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3032 read_oid_strbuf(opt, &a->oid, &abuf))
3033 goto error_return;
3034 /*
3035 * Note: binary | is used so that both renormalizations are
3036 * performed. Comparison can be skipped if both files are
3037 * unchanged since their sha1s have already been compared.
3038 */
3039 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3040 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3041 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3042
3043 error_return:
3044 strbuf_release(&obuf);
3045 strbuf_release(&abuf);
3046 return ret;
3047 }
3048
3049 static int handle_modify_delete(struct merge_options *opt,
3050 const char *path,
3051 const struct diff_filespec *o,
3052 const struct diff_filespec *a,
3053 const struct diff_filespec *b)
3054 {
3055 const char *modify_branch, *delete_branch;
3056 const struct diff_filespec *changed;
3057
3058 if (is_valid(a)) {
3059 modify_branch = opt->branch1;
3060 delete_branch = opt->branch2;
3061 changed = a;
3062 } else {
3063 modify_branch = opt->branch2;
3064 delete_branch = opt->branch1;
3065 changed = b;
3066 }
3067
3068 return handle_change_delete(opt,
3069 path, NULL,
3070 o, changed,
3071 modify_branch, delete_branch,
3072 _("modify"), _("modified"));
3073 }
3074
3075 static int handle_content_merge(struct merge_file_info *mfi,
3076 struct merge_options *opt,
3077 const char *path,
3078 int is_dirty,
3079 const struct diff_filespec *o,
3080 const struct diff_filespec *a,
3081 const struct diff_filespec *b,
3082 struct rename_conflict_info *ci)
3083 {
3084 const char *reason = _("content");
3085 unsigned df_conflict_remains = 0;
3086
3087 if (!is_valid(o))
3088 reason = _("add/add");
3089
3090 assert(o->path && a->path && b->path);
3091 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3092 S_ISGITLINK(ci->ren1->pair->two->mode)))
3093 df_conflict_remains = 1;
3094
3095 if (merge_mode_and_contents(opt, o, a, b, path,
3096 opt->branch1, opt->branch2,
3097 opt->priv->call_depth * 2, mfi))
3098 return -1;
3099
3100 /*
3101 * We can skip updating the working tree file iff:
3102 * a) The merge is clean
3103 * b) The merge matches what was in HEAD (content, mode, pathname)
3104 * c) The target path is usable (i.e. not involved in D/F conflict)
3105 */
3106 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3107 !df_conflict_remains) {
3108 int pos;
3109 struct cache_entry *ce;
3110
3111 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3112 if (add_cacheinfo(opt, &mfi->blob, path,
3113 0, (!opt->priv->call_depth && !is_dirty), 0))
3114 return -1;
3115 /*
3116 * However, add_cacheinfo() will delete the old cache entry
3117 * and add a new one. We need to copy over any skip_worktree
3118 * flag to avoid making the file appear as if it were
3119 * deleted by the user.
3120 */
3121 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3122 ce = opt->priv->orig_index.cache[pos];
3123 if (ce_skip_worktree(ce)) {
3124 pos = index_name_pos(opt->repo->index, path, strlen(path));
3125 ce = opt->repo->index->cache[pos];
3126 ce->ce_flags |= CE_SKIP_WORKTREE;
3127 }
3128 return mfi->clean;
3129 }
3130
3131 if (!mfi->clean) {
3132 if (S_ISGITLINK(mfi->blob.mode))
3133 reason = _("submodule");
3134 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3135 reason, path);
3136 if (ci && !df_conflict_remains)
3137 if (update_stages(opt, path, o, a, b))
3138 return -1;
3139 }
3140
3141 if (df_conflict_remains || is_dirty) {
3142 char *new_path;
3143 if (opt->priv->call_depth) {
3144 remove_file_from_index(opt->repo->index, path);
3145 } else {
3146 if (!mfi->clean) {
3147 if (update_stages(opt, path, o, a, b))
3148 return -1;
3149 } else {
3150 int file_from_stage2 = was_tracked(opt, path);
3151
3152 if (update_stages(opt, path, NULL,
3153 file_from_stage2 ? &mfi->blob : NULL,
3154 file_from_stage2 ? NULL : &mfi->blob))
3155 return -1;
3156 }
3157
3158 }
3159 new_path = unique_path(opt, path, ci->ren1->branch);
3160 if (is_dirty) {
3161 output(opt, 1, _("Refusing to lose dirty file at %s"),
3162 path);
3163 }
3164 output(opt, 1, _("Adding as %s instead"), new_path);
3165 if (update_file(opt, 0, &mfi->blob, new_path)) {
3166 free(new_path);
3167 return -1;
3168 }
3169 free(new_path);
3170 mfi->clean = 0;
3171 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3172 return -1;
3173 return !is_dirty && mfi->clean;
3174 }
3175
3176 static int handle_rename_normal(struct merge_options *opt,
3177 const char *path,
3178 const struct diff_filespec *o,
3179 const struct diff_filespec *a,
3180 const struct diff_filespec *b,
3181 struct rename_conflict_info *ci)
3182 {
3183 struct rename *ren = ci->ren1;
3184 struct merge_file_info mfi;
3185 int clean;
3186
3187 /* Merge the content and write it out */
3188 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3189 o, a, b, ci);
3190
3191 if (clean &&
3192 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3193 ren->dir_rename_original_dest) {
3194 if (update_stages(opt, path,
3195 &mfi.blob, &mfi.blob, &mfi.blob))
3196 return -1;
3197 clean = 0; /* not clean, but conflicted */
3198 }
3199 return clean;
3200 }
3201
3202 static void dir_rename_warning(const char *msg,
3203 int is_add,
3204 int clean,
3205 struct merge_options *opt,
3206 struct rename *ren)
3207 {
3208 const char *other_branch;
3209 other_branch = (ren->branch == opt->branch1 ?
3210 opt->branch2 : opt->branch1);
3211 if (is_add) {
3212 output(opt, clean ? 2 : 1, msg,
3213 ren->pair->one->path, ren->branch,
3214 other_branch, ren->pair->two->path);
3215 return;
3216 }
3217 output(opt, clean ? 2 : 1, msg,
3218 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3219 other_branch, ren->pair->two->path);
3220 }
3221 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3222 struct rename *ren)
3223 {
3224 const char *msg;
3225 int clean = 1, is_add;
3226
3227 if (!ren)
3228 return clean;
3229
3230 /* Return early if ren was not affected/created by a directory rename */
3231 if (!ren->dir_rename_original_dest)
3232 return clean;
3233
3234 /* Sanity checks */
3235 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3236 assert(ren->dir_rename_original_type == 'A' ||
3237 ren->dir_rename_original_type == 'R');
3238
3239 /* Check whether to treat directory renames as a conflict */
3240 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3241
3242 is_add = (ren->dir_rename_original_type == 'A');
3243 if (ren->dir_rename_original_type == 'A' && clean) {
3244 msg = _("Path updated: %s added in %s inside a "
3245 "directory that was renamed in %s; moving it to %s.");
3246 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3247 msg = _("CONFLICT (file location): %s added in %s "
3248 "inside a directory that was renamed in %s, "
3249 "suggesting it should perhaps be moved to %s.");
3250 } else if (ren->dir_rename_original_type == 'R' && clean) {
3251 msg = _("Path updated: %s renamed to %s in %s, inside a "
3252 "directory that was renamed in %s; moving it to %s.");
3253 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3254 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3255 "inside a directory that was renamed in %s, "
3256 "suggesting it should perhaps be moved to %s.");
3257 } else {
3258 BUG("Impossible dir_rename_original_type/clean combination");
3259 }
3260 dir_rename_warning(msg, is_add, clean, opt, ren);
3261
3262 return clean;
3263 }
3264
3265 /* Per entry merge function */
3266 static int process_entry(struct merge_options *opt,
3267 const char *path, struct stage_data *entry)
3268 {
3269 int clean_merge = 1;
3270 int normalize = opt->renormalize;
3271
3272 struct diff_filespec *o = &entry->stages[1];
3273 struct diff_filespec *a = &entry->stages[2];
3274 struct diff_filespec *b = &entry->stages[3];
3275 int o_valid = is_valid(o);
3276 int a_valid = is_valid(a);
3277 int b_valid = is_valid(b);
3278 o->path = a->path = b->path = (char*)path;
3279
3280 entry->processed = 1;
3281 if (entry->rename_conflict_info) {
3282 struct rename_conflict_info *ci = entry->rename_conflict_info;
3283 struct diff_filespec *temp;
3284 int path_clean;
3285
3286 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3287 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3288
3289 /*
3290 * For cases with a single rename, {o,a,b}->path have all been
3291 * set to the rename target path; we need to set two of these
3292 * back to the rename source.
3293 * For rename/rename conflicts, we'll manually fix paths below.
3294 */
3295 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3296 o->path = temp->path = ci->ren1->pair->one->path;
3297 if (ci->ren2) {
3298 assert(opt->branch1 == ci->ren1->branch);
3299 }
3300
3301 switch (ci->rename_type) {
3302 case RENAME_NORMAL:
3303 case RENAME_ONE_FILE_TO_ONE:
3304 clean_merge = handle_rename_normal(opt, path, o, a, b,
3305 ci);
3306 break;
3307 case RENAME_VIA_DIR:
3308 clean_merge = handle_rename_via_dir(opt, ci);
3309 break;
3310 case RENAME_ADD:
3311 /*
3312 * Probably unclean merge, but if the renamed file
3313 * merges cleanly and the result can then be
3314 * two-way merged cleanly with the added file, I
3315 * guess it's a clean merge?
3316 */
3317 clean_merge = handle_rename_add(opt, ci);
3318 break;
3319 case RENAME_DELETE:
3320 clean_merge = 0;
3321 if (handle_rename_delete(opt, ci))
3322 clean_merge = -1;
3323 break;
3324 case RENAME_ONE_FILE_TO_TWO:
3325 /*
3326 * Manually fix up paths; note:
3327 * ren[12]->pair->one->path are equal.
3328 */
3329 o->path = ci->ren1->pair->one->path;
3330 a->path = ci->ren1->pair->two->path;
3331 b->path = ci->ren2->pair->two->path;
3332
3333 clean_merge = 0;
3334 if (handle_rename_rename_1to2(opt, ci))
3335 clean_merge = -1;
3336 break;
3337 case RENAME_TWO_FILES_TO_ONE:
3338 /*
3339 * Manually fix up paths; note,
3340 * ren[12]->pair->two->path are actually equal.
3341 */
3342 o->path = NULL;
3343 a->path = ci->ren1->pair->two->path;
3344 b->path = ci->ren2->pair->two->path;
3345
3346 /*
3347 * Probably unclean merge, but if the two renamed
3348 * files merge cleanly and the two resulting files
3349 * can then be two-way merged cleanly, I guess it's
3350 * a clean merge?
3351 */
3352 clean_merge = handle_rename_rename_2to1(opt, ci);
3353 break;
3354 default:
3355 entry->processed = 0;
3356 break;
3357 }
3358 if (path_clean < clean_merge)
3359 clean_merge = path_clean;
3360 } else if (o_valid && (!a_valid || !b_valid)) {
3361 /* Case A: Deleted in one */
3362 if ((!a_valid && !b_valid) ||
3363 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3364 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3365 /* Deleted in both or deleted in one and
3366 * unchanged in the other */
3367 if (a_valid)
3368 output(opt, 2, _("Removing %s"), path);
3369 /* do not touch working file if it did not exist */
3370 remove_file(opt, 1, path, !a_valid);
3371 } else {
3372 /* Modify/delete; deleted side may have put a directory in the way */
3373 clean_merge = 0;
3374 if (handle_modify_delete(opt, path, o, a, b))
3375 clean_merge = -1;
3376 }
3377 } else if ((!o_valid && a_valid && !b_valid) ||
3378 (!o_valid && !a_valid && b_valid)) {
3379 /* Case B: Added in one. */
3380 /* [nothing|directory] -> ([nothing|directory], file) */
3381
3382 const char *add_branch;
3383 const char *other_branch;
3384 const char *conf;
3385 const struct diff_filespec *contents;
3386
3387 if (a_valid) {
3388 add_branch = opt->branch1;
3389 other_branch = opt->branch2;
3390 contents = a;
3391 conf = _("file/directory");
3392 } else {
3393 add_branch = opt->branch2;
3394 other_branch = opt->branch1;
3395 contents = b;
3396 conf = _("directory/file");
3397 }
3398 if (dir_in_way(opt->repo->index, path,
3399 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3400 0)) {
3401 char *new_path = unique_path(opt, path, add_branch);
3402 clean_merge = 0;
3403 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3404 "Adding %s as %s"),
3405 conf, path, other_branch, path, new_path);
3406 if (update_file(opt, 0, contents, new_path))
3407 clean_merge = -1;
3408 else if (opt->priv->call_depth)
3409 remove_file_from_index(opt->repo->index, path);
3410 free(new_path);
3411 } else {
3412 output(opt, 2, _("Adding %s"), path);
3413 /* do not overwrite file if already present */
3414 if (update_file_flags(opt, contents, path, 1, !a_valid))
3415 clean_merge = -1;
3416 }
3417 } else if (a_valid && b_valid) {
3418 if (!o_valid) {
3419 /* Case C: Added in both (check for same permissions) */
3420 output(opt, 1,
3421 _("CONFLICT (add/add): Merge conflict in %s"),
3422 path);
3423 clean_merge = handle_file_collision(opt,
3424 path, NULL, NULL,
3425 opt->branch1,
3426 opt->branch2,
3427 a, b);
3428 } else {
3429 /* case D: Modified in both, but differently. */
3430 struct merge_file_info mfi;
3431 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3432 clean_merge = handle_content_merge(&mfi, opt, path,
3433 is_dirty,
3434 o, a, b, NULL);
3435 }
3436 } else if (!o_valid && !a_valid && !b_valid) {
3437 /*
3438 * this entry was deleted altogether. a_mode == 0 means
3439 * we had that path and want to actively remove it.
3440 */
3441 remove_file(opt, 1, path, !a->mode);
3442 } else
3443 BUG("fatal merge failure, shouldn't happen.");
3444
3445 return clean_merge;
3446 }
3447
3448 static int merge_trees_internal(struct merge_options *opt,
3449 struct tree *head,
3450 struct tree *merge,
3451 struct tree *merge_base,
3452 struct tree **result)
3453 {
3454 struct index_state *istate = opt->repo->index;
3455 int code, clean;
3456
3457 if (opt->subtree_shift) {
3458 merge = shift_tree_object(opt->repo, head, merge,
3459 opt->subtree_shift);
3460 merge_base = shift_tree_object(opt->repo, head, merge_base,
3461 opt->subtree_shift);
3462 }
3463
3464 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3465 output(opt, 0, _("Already up to date."));
3466 *result = head;
3467 return 1;
3468 }
3469
3470 code = unpack_trees_start(opt, merge_base, head, merge);
3471
3472 if (code != 0) {
3473 if (show(opt, 4) || opt->priv->call_depth)
3474 err(opt, _("merging of trees %s and %s failed"),
3475 oid_to_hex(&head->object.oid),
3476 oid_to_hex(&merge->object.oid));
3477 unpack_trees_finish(opt);
3478 return -1;
3479 }
3480
3481 if (unmerged_index(istate)) {
3482 struct string_list *entries;
3483 struct rename_info re_info;
3484 int i;
3485 /*
3486 * Only need the hashmap while processing entries, so
3487 * initialize it here and free it when we are done running
3488 * through the entries. Keeping it in the merge_options as
3489 * opposed to decaring a local hashmap is for convenience
3490 * so that we don't have to pass it to around.
3491 */
3492 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3493 NULL, 512);
3494 get_files_dirs(opt, head);
3495 get_files_dirs(opt, merge);
3496
3497 entries = get_unmerged(opt->repo->index);
3498 clean = detect_and_process_renames(opt, merge_base, head, merge,
3499 entries, &re_info);
3500 record_df_conflict_files(opt, entries);
3501 if (clean < 0)
3502 goto cleanup;
3503 for (i = entries->nr-1; 0 <= i; i--) {
3504 const char *path = entries->items[i].string;
3505 struct stage_data *e = entries->items[i].util;
3506 if (!e->processed) {
3507 int ret = process_entry(opt, path, e);
3508 if (!ret)
3509 clean = 0;
3510 else if (ret < 0) {
3511 clean = ret;
3512 goto cleanup;
3513 }
3514 }
3515 }
3516 for (i = 0; i < entries->nr; i++) {
3517 struct stage_data *e = entries->items[i].util;
3518 if (!e->processed)
3519 BUG("unprocessed path??? %s",
3520 entries->items[i].string);
3521 }
3522
3523 cleanup:
3524 final_cleanup_renames(&re_info);
3525
3526 string_list_clear(entries, 1);
3527 free(entries);
3528
3529 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3530 struct path_hashmap_entry, e);
3531
3532 if (clean < 0) {
3533 unpack_trees_finish(opt);
3534 return clean;
3535 }
3536 }
3537 else
3538 clean = 1;
3539
3540 unpack_trees_finish(opt);
3541
3542 if (opt->priv->call_depth &&
3543 !(*result = write_in_core_index_as_tree(opt->repo)))
3544 return -1;
3545
3546 return clean;
3547 }
3548
3549 /*
3550 * Merge the commits h1 and h2, returning a flag (int) indicating the
3551 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3552 * virtual commit and write its location to *result.
3553 */
3554 static int merge_recursive_internal(struct merge_options *opt,
3555 struct commit *h1,
3556 struct commit *h2,
3557 struct commit_list *merge_bases,
3558 struct commit **result)
3559 {
3560 struct commit_list *iter;
3561 struct commit *merged_merge_bases;
3562 struct tree *result_tree;
3563 int clean;
3564 const char *ancestor_name;
3565 struct strbuf merge_base_abbrev = STRBUF_INIT;
3566
3567 if (show(opt, 4)) {
3568 output(opt, 4, _("Merging:"));
3569 output_commit_title(opt, h1);
3570 output_commit_title(opt, h2);
3571 }
3572
3573 if (!merge_bases) {
3574 merge_bases = get_merge_bases(h1, h2);
3575 merge_bases = reverse_commit_list(merge_bases);
3576 }
3577
3578 if (show(opt, 5)) {
3579 unsigned cnt = commit_list_count(merge_bases);
3580
3581 output(opt, 5, Q_("found %u common ancestor:",
3582 "found %u common ancestors:", cnt), cnt);
3583 for (iter = merge_bases; iter; iter = iter->next)
3584 output_commit_title(opt, iter->item);
3585 }
3586
3587 merged_merge_bases = pop_commit(&merge_bases);
3588 if (merged_merge_bases == NULL) {
3589 /* if there is no common ancestor, use an empty tree */
3590 struct tree *tree;
3591
3592 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3593 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3594 "ancestor");
3595 ancestor_name = "empty tree";
3596 } else if (opt->ancestor && !opt->priv->call_depth) {
3597 ancestor_name = opt->ancestor;
3598 } else if (merge_bases) {
3599 ancestor_name = "merged common ancestors";
3600 } else {
3601 strbuf_add_unique_abbrev(&merge_base_abbrev,
3602 &merged_merge_bases->object.oid,
3603 DEFAULT_ABBREV);
3604 ancestor_name = merge_base_abbrev.buf;
3605 }
3606
3607 for (iter = merge_bases; iter; iter = iter->next) {
3608 const char *saved_b1, *saved_b2;
3609 opt->priv->call_depth++;
3610 /*
3611 * When the merge fails, the result contains files
3612 * with conflict markers. The cleanness flag is
3613 * ignored (unless indicating an error), it was never
3614 * actually used, as result of merge_trees has always
3615 * overwritten it: the committed "conflicts" were
3616 * already resolved.
3617 */
3618 discard_index(opt->repo->index);
3619 saved_b1 = opt->branch1;
3620 saved_b2 = opt->branch2;
3621 opt->branch1 = "Temporary merge branch 1";
3622 opt->branch2 = "Temporary merge branch 2";
3623 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3624 NULL, &merged_merge_bases) < 0)
3625 return -1;
3626 opt->branch1 = saved_b1;
3627 opt->branch2 = saved_b2;
3628 opt->priv->call_depth--;
3629
3630 if (!merged_merge_bases)
3631 return err(opt, _("merge returned no commit"));
3632 }
3633
3634 /*
3635 * FIXME: Since merge_recursive_internal() is only ever called by
3636 * places that ensure the index is loaded first
3637 * (e.g. builtin/merge.c, rebase/sequencer, etc.), in the common
3638 * case where the merge base was unique that means when we get here
3639 * we immediately discard the index and re-read it, which is a
3640 * complete waste of time. We should only be discarding and
3641 * re-reading if we were forced to recurse.
3642 */
3643 discard_index(opt->repo->index);
3644 if (!opt->priv->call_depth)
3645 repo_read_index(opt->repo);
3646
3647 opt->ancestor = ancestor_name;
3648 clean = merge_trees_internal(opt,
3649 repo_get_commit_tree(opt->repo, h1),
3650 repo_get_commit_tree(opt->repo, h2),
3651 repo_get_commit_tree(opt->repo,
3652 merged_merge_bases),
3653 &result_tree);
3654 strbuf_release(&merge_base_abbrev);
3655 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3656 if (clean < 0) {
3657 flush_output(opt);
3658 return clean;
3659 }
3660
3661 if (opt->priv->call_depth) {
3662 *result = make_virtual_commit(opt->repo, result_tree,
3663 "merged tree");
3664 commit_list_insert(h1, &(*result)->parents);
3665 commit_list_insert(h2, &(*result)->parents->next);
3666 }
3667 return clean;
3668 }
3669
3670 static int merge_start(struct merge_options *opt, struct tree *head)
3671 {
3672 struct strbuf sb = STRBUF_INIT;
3673
3674 /* Sanity checks on opt */
3675 assert(opt->repo);
3676
3677 assert(opt->branch1 && opt->branch2);
3678
3679 assert(opt->detect_renames >= -1 &&
3680 opt->detect_renames <= DIFF_DETECT_COPY);
3681 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3682 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3683 assert(opt->rename_limit >= -1);
3684 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3685 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3686
3687 assert(opt->xdl_opts >= 0);
3688 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3689 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3690
3691 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3692 assert(opt->buffer_output <= 2);
3693 assert(opt->obuf.len == 0);
3694
3695 assert(opt->priv == NULL);
3696
3697 /* Sanity check on repo state; index must match head */
3698 if (repo_index_has_changes(opt->repo, head, &sb)) {
3699 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3700 sb.buf);
3701 strbuf_release(&sb);
3702 return -1;
3703 }
3704
3705 CALLOC_ARRAY(opt->priv, 1);
3706 string_list_init_dup(&opt->priv->df_conflict_file_set);
3707 return 0;
3708 }
3709
3710 static void merge_finalize(struct merge_options *opt)
3711 {
3712 flush_output(opt);
3713 if (!opt->priv->call_depth && opt->buffer_output < 2)
3714 strbuf_release(&opt->obuf);
3715 if (show(opt, 2))
3716 diff_warn_rename_limit("merge.renamelimit",
3717 opt->priv->needed_rename_limit, 0);
3718 FREE_AND_NULL(opt->priv);
3719 }
3720
3721 int merge_trees(struct merge_options *opt,
3722 struct tree *head,
3723 struct tree *merge,
3724 struct tree *merge_base)
3725 {
3726 int clean;
3727 struct tree *ignored;
3728
3729 assert(opt->ancestor != NULL);
3730
3731 if (merge_start(opt, head))
3732 return -1;
3733 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3734 merge_finalize(opt);
3735
3736 return clean;
3737 }
3738
3739 int merge_recursive(struct merge_options *opt,
3740 struct commit *h1,
3741 struct commit *h2,
3742 struct commit_list *merge_bases,
3743 struct commit **result)
3744 {
3745 int clean;
3746
3747 assert(opt->ancestor == NULL ||
3748 !strcmp(opt->ancestor, "constructed merge base"));
3749
3750 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3751 return -1;
3752 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3753 merge_finalize(opt);
3754
3755 return clean;
3756 }
3757
3758 static struct commit *get_ref(struct repository *repo,
3759 const struct object_id *oid,
3760 const char *name)
3761 {
3762 struct object *object;
3763
3764 object = deref_tag(repo, parse_object(repo, oid),
3765 name, strlen(name));
3766 if (!object)
3767 return NULL;
3768 if (object->type == OBJ_TREE)
3769 return make_virtual_commit(repo, (struct tree*)object, name);
3770 if (object->type != OBJ_COMMIT)
3771 return NULL;
3772 if (parse_commit((struct commit *)object))
3773 return NULL;
3774 return (struct commit *)object;
3775 }
3776
3777 int merge_recursive_generic(struct merge_options *opt,
3778 const struct object_id *head,
3779 const struct object_id *merge,
3780 int num_merge_bases,
3781 const struct object_id **merge_bases,
3782 struct commit **result)
3783 {
3784 int clean;
3785 struct lock_file lock = LOCK_INIT;
3786 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3787 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3788 struct commit_list *ca = NULL;
3789
3790 if (merge_bases) {
3791 int i;
3792 for (i = 0; i < num_merge_bases; ++i) {
3793 struct commit *base;
3794 if (!(base = get_ref(opt->repo, merge_bases[i],
3795 oid_to_hex(merge_bases[i]))))
3796 return err(opt, _("Could not parse object '%s'"),
3797 oid_to_hex(merge_bases[i]));
3798 commit_list_insert(base, &ca);
3799 }
3800 if (num_merge_bases == 1)
3801 opt->ancestor = "constructed merge base";
3802 }
3803
3804 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3805 clean = merge_recursive(opt, head_commit, next_commit, ca,
3806 result);
3807 if (clean < 0) {
3808 rollback_lock_file(&lock);
3809 return clean;
3810 }
3811
3812 if (write_locked_index(opt->repo->index, &lock,
3813 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3814 return err(opt, _("Unable to write index."));
3815
3816 return clean ? 0 : 1;
3817 }
3818
3819 static void merge_recursive_config(struct merge_options *opt)
3820 {
3821 char *value = NULL;
3822 int renormalize = 0;
3823 git_config_get_int("merge.verbosity", &opt->verbosity);
3824 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3825 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3826 git_config_get_bool("merge.renormalize", &renormalize);
3827 opt->renormalize = renormalize;
3828 if (!git_config_get_string("diff.renames", &value)) {
3829 opt->detect_renames = git_config_rename("diff.renames", value);
3830 free(value);
3831 }
3832 if (!git_config_get_string("merge.renames", &value)) {
3833 opt->detect_renames = git_config_rename("merge.renames", value);
3834 free(value);
3835 }
3836 if (!git_config_get_string("merge.directoryrenames", &value)) {
3837 int boolval = git_parse_maybe_bool(value);
3838 if (0 <= boolval) {
3839 opt->detect_directory_renames = boolval ?
3840 MERGE_DIRECTORY_RENAMES_TRUE :
3841 MERGE_DIRECTORY_RENAMES_NONE;
3842 } else if (!strcasecmp(value, "conflict")) {
3843 opt->detect_directory_renames =
3844 MERGE_DIRECTORY_RENAMES_CONFLICT;
3845 } /* avoid erroring on values from future versions of git */
3846 free(value);
3847 }
3848 git_config(git_xmerge_config, NULL);
3849 }
3850
3851 void init_merge_options(struct merge_options *opt,
3852 struct repository *repo)
3853 {
3854 const char *merge_verbosity;
3855 memset(opt, 0, sizeof(struct merge_options));
3856
3857 opt->repo = repo;
3858
3859 opt->detect_renames = -1;
3860 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3861 opt->rename_limit = -1;
3862
3863 opt->verbosity = 2;
3864 opt->buffer_output = 1;
3865 strbuf_init(&opt->obuf, 0);
3866
3867 opt->renormalize = 0;
3868
3869 merge_recursive_config(opt);
3870 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3871 if (merge_verbosity)
3872 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3873 if (opt->verbosity >= 5)
3874 opt->buffer_output = 0;
3875 }
3876
3877 int parse_merge_opt(struct merge_options *opt, const char *s)
3878 {
3879 const char *arg;
3880
3881 if (!s || !*s)
3882 return -1;
3883 if (!strcmp(s, "ours"))
3884 opt->recursive_variant = MERGE_VARIANT_OURS;
3885 else if (!strcmp(s, "theirs"))
3886 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3887 else if (!strcmp(s, "subtree"))
3888 opt->subtree_shift = "";
3889 else if (skip_prefix(s, "subtree=", &arg))
3890 opt->subtree_shift = arg;
3891 else if (!strcmp(s, "patience"))
3892 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3893 else if (!strcmp(s, "histogram"))
3894 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3895 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3896 long value = parse_algorithm_value(arg);
3897 if (value < 0)
3898 return -1;
3899 /* clear out previous settings */
3900 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3901 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3902 opt->xdl_opts |= value;
3903 }
3904 else if (!strcmp(s, "ignore-space-change"))
3905 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3906 else if (!strcmp(s, "ignore-all-space"))
3907 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3908 else if (!strcmp(s, "ignore-space-at-eol"))
3909 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3910 else if (!strcmp(s, "ignore-cr-at-eol"))
3911 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3912 else if (!strcmp(s, "renormalize"))
3913 opt->renormalize = 1;
3914 else if (!strcmp(s, "no-renormalize"))
3915 opt->renormalize = 0;
3916 else if (!strcmp(s, "no-renames"))
3917 opt->detect_renames = 0;
3918 else if (!strcmp(s, "find-renames")) {
3919 opt->detect_renames = 1;
3920 opt->rename_score = 0;
3921 }
3922 else if (skip_prefix(s, "find-renames=", &arg) ||
3923 skip_prefix(s, "rename-threshold=", &arg)) {
3924 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3925 return -1;
3926 opt->detect_renames = 1;
3927 }
3928 /*
3929 * Please update $__git_merge_strategy_options in
3930 * git-completion.bash when you add new options
3931 */
3932 else
3933 return -1;
3934 return 0;
3935 }