]> git.ipfire.org Git - thirdparty/git.git/blob - merge-recursive.c
Merge branch 'jk/diff-honor-wserrhighlight-in-plumbing'
[thirdparty/git.git] / merge-recursive.c
1 /*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6 #include "cache.h"
7 #include "merge-recursive.h"
8
9 #include "advice.h"
10 #include "alloc.h"
11 #include "attr.h"
12 #include "blob.h"
13 #include "builtin.h"
14 #include "cache-tree.h"
15 #include "commit.h"
16 #include "commit-reach.h"
17 #include "config.h"
18 #include "diff.h"
19 #include "diffcore.h"
20 #include "dir.h"
21 #include "ll-merge.h"
22 #include "lockfile.h"
23 #include "object-store.h"
24 #include "repository.h"
25 #include "revision.h"
26 #include "string-list.h"
27 #include "submodule.h"
28 #include "tag.h"
29 #include "tree-walk.h"
30 #include "unpack-trees.h"
31 #include "xdiff-interface.h"
32
33 struct merge_options_internal {
34 int call_depth;
35 int needed_rename_limit;
36 struct hashmap current_file_dir_set;
37 struct string_list df_conflict_file_set;
38 struct unpack_trees_options unpack_opts;
39 struct index_state orig_index;
40 };
41
42 struct path_hashmap_entry {
43 struct hashmap_entry e;
44 char path[FLEX_ARRAY];
45 };
46
47 static int path_hashmap_cmp(const void *cmp_data,
48 const struct hashmap_entry *eptr,
49 const struct hashmap_entry *entry_or_key,
50 const void *keydata)
51 {
52 const struct path_hashmap_entry *a, *b;
53 const char *key = keydata;
54
55 a = container_of(eptr, const struct path_hashmap_entry, e);
56 b = container_of(entry_or_key, const struct path_hashmap_entry, e);
57
58 if (ignore_case)
59 return strcasecmp(a->path, key ? key : b->path);
60 else
61 return strcmp(a->path, key ? key : b->path);
62 }
63
64 static unsigned int path_hash(const char *path)
65 {
66 return ignore_case ? strihash(path) : strhash(path);
67 }
68
69 /*
70 * For dir_rename_entry, directory names are stored as a full path from the
71 * toplevel of the repository and do not include a trailing '/'. Also:
72 *
73 * dir: original name of directory being renamed
74 * non_unique_new_dir: if true, could not determine new_dir
75 * new_dir: final name of directory being renamed
76 * possible_new_dirs: temporary used to help determine new_dir; see comments
77 * in get_directory_renames() for details
78 */
79 struct dir_rename_entry {
80 struct hashmap_entry ent;
81 char *dir;
82 unsigned non_unique_new_dir:1;
83 struct strbuf new_dir;
84 struct string_list possible_new_dirs;
85 };
86
87 static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
88 char *dir)
89 {
90 struct dir_rename_entry key;
91
92 if (dir == NULL)
93 return NULL;
94 hashmap_entry_init(&key.ent, strhash(dir));
95 key.dir = dir;
96 return hashmap_get_entry(hashmap, &key, ent, NULL);
97 }
98
99 static int dir_rename_cmp(const void *unused_cmp_data,
100 const struct hashmap_entry *eptr,
101 const struct hashmap_entry *entry_or_key,
102 const void *unused_keydata)
103 {
104 const struct dir_rename_entry *e1, *e2;
105
106 e1 = container_of(eptr, const struct dir_rename_entry, ent);
107 e2 = container_of(entry_or_key, const struct dir_rename_entry, ent);
108
109 return strcmp(e1->dir, e2->dir);
110 }
111
112 static void dir_rename_init(struct hashmap *map)
113 {
114 hashmap_init(map, dir_rename_cmp, NULL, 0);
115 }
116
117 static void dir_rename_entry_init(struct dir_rename_entry *entry,
118 char *directory)
119 {
120 hashmap_entry_init(&entry->ent, strhash(directory));
121 entry->dir = directory;
122 entry->non_unique_new_dir = 0;
123 strbuf_init(&entry->new_dir, 0);
124 string_list_init(&entry->possible_new_dirs, 0);
125 }
126
127 struct collision_entry {
128 struct hashmap_entry ent;
129 char *target_file;
130 struct string_list source_files;
131 unsigned reported_already:1;
132 };
133
134 static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
135 char *target_file)
136 {
137 struct collision_entry key;
138
139 hashmap_entry_init(&key.ent, strhash(target_file));
140 key.target_file = target_file;
141 return hashmap_get_entry(hashmap, &key, ent, NULL);
142 }
143
144 static int collision_cmp(const void *unused_cmp_data,
145 const struct hashmap_entry *eptr,
146 const struct hashmap_entry *entry_or_key,
147 const void *unused_keydata)
148 {
149 const struct collision_entry *e1, *e2;
150
151 e1 = container_of(eptr, const struct collision_entry, ent);
152 e2 = container_of(entry_or_key, const struct collision_entry, ent);
153
154 return strcmp(e1->target_file, e2->target_file);
155 }
156
157 static void collision_init(struct hashmap *map)
158 {
159 hashmap_init(map, collision_cmp, NULL, 0);
160 }
161
162 static void flush_output(struct merge_options *opt)
163 {
164 if (opt->buffer_output < 2 && opt->obuf.len) {
165 fputs(opt->obuf.buf, stdout);
166 strbuf_reset(&opt->obuf);
167 }
168 }
169
170 static int err(struct merge_options *opt, const char *err, ...)
171 {
172 va_list params;
173
174 if (opt->buffer_output < 2)
175 flush_output(opt);
176 else {
177 strbuf_complete(&opt->obuf, '\n');
178 strbuf_addstr(&opt->obuf, "error: ");
179 }
180 va_start(params, err);
181 strbuf_vaddf(&opt->obuf, err, params);
182 va_end(params);
183 if (opt->buffer_output > 1)
184 strbuf_addch(&opt->obuf, '\n');
185 else {
186 error("%s", opt->obuf.buf);
187 strbuf_reset(&opt->obuf);
188 }
189
190 return -1;
191 }
192
193 static struct tree *shift_tree_object(struct repository *repo,
194 struct tree *one, struct tree *two,
195 const char *subtree_shift)
196 {
197 struct object_id shifted;
198
199 if (!*subtree_shift) {
200 shift_tree(repo, &one->object.oid, &two->object.oid, &shifted, 0);
201 } else {
202 shift_tree_by(repo, &one->object.oid, &two->object.oid, &shifted,
203 subtree_shift);
204 }
205 if (oideq(&two->object.oid, &shifted))
206 return two;
207 return lookup_tree(repo, &shifted);
208 }
209
210 static inline void set_commit_tree(struct commit *c, struct tree *t)
211 {
212 c->maybe_tree = t;
213 }
214
215 static struct commit *make_virtual_commit(struct repository *repo,
216 struct tree *tree,
217 const char *comment)
218 {
219 struct commit *commit = alloc_commit_node(repo);
220
221 set_merge_remote_desc(commit, comment, (struct object *)commit);
222 set_commit_tree(commit, tree);
223 commit->object.parsed = 1;
224 return commit;
225 }
226
227 enum rename_type {
228 RENAME_NORMAL = 0,
229 RENAME_VIA_DIR,
230 RENAME_ADD,
231 RENAME_DELETE,
232 RENAME_ONE_FILE_TO_ONE,
233 RENAME_ONE_FILE_TO_TWO,
234 RENAME_TWO_FILES_TO_ONE
235 };
236
237 /*
238 * Since we want to write the index eventually, we cannot reuse the index
239 * for these (temporary) data.
240 */
241 struct stage_data {
242 struct diff_filespec stages[4]; /* mostly for oid & mode; maybe path */
243 struct rename_conflict_info *rename_conflict_info;
244 unsigned processed:1;
245 };
246
247 struct rename {
248 unsigned processed:1;
249 struct diff_filepair *pair;
250 const char *branch; /* branch that the rename occurred on */
251 /*
252 * If directory rename detection affected this rename, what was its
253 * original type ('A' or 'R') and it's original destination before
254 * the directory rename (otherwise, '\0' and NULL for these two vars).
255 */
256 char dir_rename_original_type;
257 char *dir_rename_original_dest;
258 /*
259 * Purpose of src_entry and dst_entry:
260 *
261 * If 'before' is renamed to 'after' then src_entry will contain
262 * the versions of 'before' from the merge_base, HEAD, and MERGE in
263 * stages 1, 2, and 3; dst_entry will contain the respective
264 * versions of 'after' in corresponding locations. Thus, we have a
265 * total of six modes and oids, though some will be null. (Stage 0
266 * is ignored; we're interested in handling conflicts.)
267 *
268 * Since we don't turn on break-rewrites by default, neither
269 * src_entry nor dst_entry can have all three of their stages have
270 * non-null oids, meaning at most four of the six will be non-null.
271 * Also, since this is a rename, both src_entry and dst_entry will
272 * have at least one non-null oid, meaning at least two will be
273 * non-null. Of the six oids, a typical rename will have three be
274 * non-null. Only two implies a rename/delete, and four implies a
275 * rename/add.
276 */
277 struct stage_data *src_entry;
278 struct stage_data *dst_entry;
279 };
280
281 struct rename_conflict_info {
282 enum rename_type rename_type;
283 struct rename *ren1;
284 struct rename *ren2;
285 };
286
287 static inline void setup_rename_conflict_info(enum rename_type rename_type,
288 struct merge_options *opt,
289 struct rename *ren1,
290 struct rename *ren2)
291 {
292 struct rename_conflict_info *ci;
293
294 /*
295 * When we have two renames involved, it's easiest to get the
296 * correct things into stage 2 and 3, and to make sure that the
297 * content merge puts HEAD before the other branch if we just
298 * ensure that branch1 == opt->branch1. So, simply flip arguments
299 * around if we don't have that.
300 */
301 if (ren2 && ren1->branch != opt->branch1) {
302 setup_rename_conflict_info(rename_type, opt, ren2, ren1);
303 return;
304 }
305
306 ci = xcalloc(1, sizeof(struct rename_conflict_info));
307 ci->rename_type = rename_type;
308 ci->ren1 = ren1;
309 ci->ren2 = ren2;
310
311 ci->ren1->dst_entry->processed = 0;
312 ci->ren1->dst_entry->rename_conflict_info = ci;
313 if (ren2) {
314 ci->ren2->dst_entry->rename_conflict_info = ci;
315 }
316 }
317
318 static int show(struct merge_options *opt, int v)
319 {
320 return (!opt->priv->call_depth && opt->verbosity >= v) ||
321 opt->verbosity >= 5;
322 }
323
324 __attribute__((format (printf, 3, 4)))
325 static void output(struct merge_options *opt, int v, const char *fmt, ...)
326 {
327 va_list ap;
328
329 if (!show(opt, v))
330 return;
331
332 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
333
334 va_start(ap, fmt);
335 strbuf_vaddf(&opt->obuf, fmt, ap);
336 va_end(ap);
337
338 strbuf_addch(&opt->obuf, '\n');
339 if (!opt->buffer_output)
340 flush_output(opt);
341 }
342
343 static void output_commit_title(struct merge_options *opt, struct commit *commit)
344 {
345 struct merge_remote_desc *desc;
346
347 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
348 desc = merge_remote_util(commit);
349 if (desc)
350 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
351 else {
352 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
353 DEFAULT_ABBREV);
354 strbuf_addch(&opt->obuf, ' ');
355 if (parse_commit(commit) != 0)
356 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
357 else {
358 const char *title;
359 const char *msg = get_commit_buffer(commit, NULL);
360 int len = find_commit_subject(msg, &title);
361 if (len)
362 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
363 unuse_commit_buffer(commit, msg);
364 }
365 }
366 flush_output(opt);
367 }
368
369 static int add_cacheinfo(struct merge_options *opt,
370 const struct diff_filespec *blob,
371 const char *path, int stage, int refresh, int options)
372 {
373 struct index_state *istate = opt->repo->index;
374 struct cache_entry *ce;
375 int ret;
376
377 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
378 if (!ce)
379 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
380
381 ret = add_index_entry(istate, ce, options);
382 if (refresh) {
383 struct cache_entry *nce;
384
385 nce = refresh_cache_entry(istate, ce,
386 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
387 if (!nce)
388 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
389 if (nce != ce)
390 ret = add_index_entry(istate, nce, options);
391 }
392 return ret;
393 }
394
395 static inline int merge_detect_rename(struct merge_options *opt)
396 {
397 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
398 }
399
400 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
401 {
402 parse_tree(tree);
403 init_tree_desc(desc, tree->buffer, tree->size);
404 }
405
406 static int unpack_trees_start(struct merge_options *opt,
407 struct tree *common,
408 struct tree *head,
409 struct tree *merge)
410 {
411 int rc;
412 struct tree_desc t[3];
413 struct index_state tmp_index = { NULL };
414
415 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
416 if (opt->priv->call_depth)
417 opt->priv->unpack_opts.index_only = 1;
418 else
419 opt->priv->unpack_opts.update = 1;
420 opt->priv->unpack_opts.merge = 1;
421 opt->priv->unpack_opts.head_idx = 2;
422 opt->priv->unpack_opts.fn = threeway_merge;
423 opt->priv->unpack_opts.src_index = opt->repo->index;
424 opt->priv->unpack_opts.dst_index = &tmp_index;
425 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
426 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
427
428 init_tree_desc_from_tree(t+0, common);
429 init_tree_desc_from_tree(t+1, head);
430 init_tree_desc_from_tree(t+2, merge);
431
432 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
433 cache_tree_free(&opt->repo->index->cache_tree);
434
435 /*
436 * Update opt->repo->index to match the new results, AFTER saving a
437 * copy in opt->priv->orig_index. Update src_index to point to the
438 * saved copy. (verify_uptodate() checks src_index, and the original
439 * index is the one that had the necessary modification timestamps.)
440 */
441 opt->priv->orig_index = *opt->repo->index;
442 *opt->repo->index = tmp_index;
443 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
444
445 return rc;
446 }
447
448 static void unpack_trees_finish(struct merge_options *opt)
449 {
450 discard_index(&opt->priv->orig_index);
451 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
452 }
453
454 static int save_files_dirs(const struct object_id *oid,
455 struct strbuf *base, const char *path,
456 unsigned int mode, int stage, void *context)
457 {
458 struct path_hashmap_entry *entry;
459 int baselen = base->len;
460 struct merge_options *opt = context;
461
462 strbuf_addstr(base, path);
463
464 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
465 hashmap_entry_init(&entry->e, path_hash(entry->path));
466 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
467
468 strbuf_setlen(base, baselen);
469 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
470 }
471
472 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
473 {
474 struct pathspec match_all;
475 memset(&match_all, 0, sizeof(match_all));
476 read_tree_recursive(opt->repo, tree, "", 0, 0,
477 &match_all, save_files_dirs, opt);
478 }
479
480 static int get_tree_entry_if_blob(struct repository *r,
481 const struct object_id *tree,
482 const char *path,
483 struct diff_filespec *dfs)
484 {
485 int ret;
486
487 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
488 if (S_ISDIR(dfs->mode)) {
489 oidcpy(&dfs->oid, &null_oid);
490 dfs->mode = 0;
491 }
492 return ret;
493 }
494
495 /*
496 * Returns an index_entry instance which doesn't have to correspond to
497 * a real cache entry in Git's index.
498 */
499 static struct stage_data *insert_stage_data(struct repository *r,
500 const char *path,
501 struct tree *o, struct tree *a, struct tree *b,
502 struct string_list *entries)
503 {
504 struct string_list_item *item;
505 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
506 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
507 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
508 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
509 item = string_list_insert(entries, path);
510 item->util = e;
511 return e;
512 }
513
514 /*
515 * Create a dictionary mapping file names to stage_data objects. The
516 * dictionary contains one entry for every path with a non-zero stage entry.
517 */
518 static struct string_list *get_unmerged(struct index_state *istate)
519 {
520 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
521 int i;
522
523 unmerged->strdup_strings = 1;
524
525 for (i = 0; i < istate->cache_nr; i++) {
526 struct string_list_item *item;
527 struct stage_data *e;
528 const struct cache_entry *ce = istate->cache[i];
529 if (!ce_stage(ce))
530 continue;
531
532 item = string_list_lookup(unmerged, ce->name);
533 if (!item) {
534 item = string_list_insert(unmerged, ce->name);
535 item->util = xcalloc(1, sizeof(struct stage_data));
536 }
537 e = item->util;
538 e->stages[ce_stage(ce)].mode = ce->ce_mode;
539 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
540 }
541
542 return unmerged;
543 }
544
545 static int string_list_df_name_compare(const char *one, const char *two)
546 {
547 int onelen = strlen(one);
548 int twolen = strlen(two);
549 /*
550 * Here we only care that entries for D/F conflicts are
551 * adjacent, in particular with the file of the D/F conflict
552 * appearing before files below the corresponding directory.
553 * The order of the rest of the list is irrelevant for us.
554 *
555 * To achieve this, we sort with df_name_compare and provide
556 * the mode S_IFDIR so that D/F conflicts will sort correctly.
557 * We use the mode S_IFDIR for everything else for simplicity,
558 * since in other cases any changes in their order due to
559 * sorting cause no problems for us.
560 */
561 int cmp = df_name_compare(one, onelen, S_IFDIR,
562 two, twolen, S_IFDIR);
563 /*
564 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
565 * that 'foo' comes before 'foo/bar'.
566 */
567 if (cmp)
568 return cmp;
569 return onelen - twolen;
570 }
571
572 static void record_df_conflict_files(struct merge_options *opt,
573 struct string_list *entries)
574 {
575 /* If there is a D/F conflict and the file for such a conflict
576 * currently exists in the working tree, we want to allow it to be
577 * removed to make room for the corresponding directory if needed.
578 * The files underneath the directories of such D/F conflicts will
579 * be processed before the corresponding file involved in the D/F
580 * conflict. If the D/F directory ends up being removed by the
581 * merge, then we won't have to touch the D/F file. If the D/F
582 * directory needs to be written to the working copy, then the D/F
583 * file will simply be removed (in make_room_for_path()) to make
584 * room for the necessary paths. Note that if both the directory
585 * and the file need to be present, then the D/F file will be
586 * reinstated with a new unique name at the time it is processed.
587 */
588 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
589 const char *last_file = NULL;
590 int last_len = 0;
591 int i;
592
593 /*
594 * If we're merging merge-bases, we don't want to bother with
595 * any working directory changes.
596 */
597 if (opt->priv->call_depth)
598 return;
599
600 /* Ensure D/F conflicts are adjacent in the entries list. */
601 for (i = 0; i < entries->nr; i++) {
602 struct string_list_item *next = &entries->items[i];
603 string_list_append(&df_sorted_entries, next->string)->util =
604 next->util;
605 }
606 df_sorted_entries.cmp = string_list_df_name_compare;
607 string_list_sort(&df_sorted_entries);
608
609 string_list_clear(&opt->priv->df_conflict_file_set, 1);
610 for (i = 0; i < df_sorted_entries.nr; i++) {
611 const char *path = df_sorted_entries.items[i].string;
612 int len = strlen(path);
613 struct stage_data *e = df_sorted_entries.items[i].util;
614
615 /*
616 * Check if last_file & path correspond to a D/F conflict;
617 * i.e. whether path is last_file+'/'+<something>.
618 * If so, record that it's okay to remove last_file to make
619 * room for path and friends if needed.
620 */
621 if (last_file &&
622 len > last_len &&
623 memcmp(path, last_file, last_len) == 0 &&
624 path[last_len] == '/') {
625 string_list_insert(&opt->priv->df_conflict_file_set, last_file);
626 }
627
628 /*
629 * Determine whether path could exist as a file in the
630 * working directory as a possible D/F conflict. This
631 * will only occur when it exists in stage 2 as a
632 * file.
633 */
634 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
635 last_file = path;
636 last_len = len;
637 } else {
638 last_file = NULL;
639 }
640 }
641 string_list_clear(&df_sorted_entries, 0);
642 }
643
644 static int update_stages(struct merge_options *opt, const char *path,
645 const struct diff_filespec *o,
646 const struct diff_filespec *a,
647 const struct diff_filespec *b)
648 {
649
650 /*
651 * NOTE: It is usually a bad idea to call update_stages on a path
652 * before calling update_file on that same path, since it can
653 * sometimes lead to spurious "refusing to lose untracked file..."
654 * messages from update_file (via make_room_for path via
655 * would_lose_untracked). Instead, reverse the order of the calls
656 * (executing update_file first and then update_stages).
657 */
658 int clear = 1;
659 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
660 if (clear)
661 if (remove_file_from_index(opt->repo->index, path))
662 return -1;
663 if (o)
664 if (add_cacheinfo(opt, o, path, 1, 0, options))
665 return -1;
666 if (a)
667 if (add_cacheinfo(opt, a, path, 2, 0, options))
668 return -1;
669 if (b)
670 if (add_cacheinfo(opt, b, path, 3, 0, options))
671 return -1;
672 return 0;
673 }
674
675 static void update_entry(struct stage_data *entry,
676 struct diff_filespec *o,
677 struct diff_filespec *a,
678 struct diff_filespec *b)
679 {
680 entry->processed = 0;
681 entry->stages[1].mode = o->mode;
682 entry->stages[2].mode = a->mode;
683 entry->stages[3].mode = b->mode;
684 oidcpy(&entry->stages[1].oid, &o->oid);
685 oidcpy(&entry->stages[2].oid, &a->oid);
686 oidcpy(&entry->stages[3].oid, &b->oid);
687 }
688
689 static int remove_file(struct merge_options *opt, int clean,
690 const char *path, int no_wd)
691 {
692 int update_cache = opt->priv->call_depth || clean;
693 int update_working_directory = !opt->priv->call_depth && !no_wd;
694
695 if (update_cache) {
696 if (remove_file_from_index(opt->repo->index, path))
697 return -1;
698 }
699 if (update_working_directory) {
700 if (ignore_case) {
701 struct cache_entry *ce;
702 ce = index_file_exists(opt->repo->index, path, strlen(path),
703 ignore_case);
704 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
705 return 0;
706 }
707 if (remove_path(path))
708 return -1;
709 }
710 return 0;
711 }
712
713 /* add a string to a strbuf, but converting "/" to "_" */
714 static void add_flattened_path(struct strbuf *out, const char *s)
715 {
716 size_t i = out->len;
717 strbuf_addstr(out, s);
718 for (; i < out->len; i++)
719 if (out->buf[i] == '/')
720 out->buf[i] = '_';
721 }
722
723 static char *unique_path(struct merge_options *opt,
724 const char *path,
725 const char *branch)
726 {
727 struct path_hashmap_entry *entry;
728 struct strbuf newpath = STRBUF_INIT;
729 int suffix = 0;
730 size_t base_len;
731
732 strbuf_addf(&newpath, "%s~", path);
733 add_flattened_path(&newpath, branch);
734
735 base_len = newpath.len;
736 while (hashmap_get_from_hash(&opt->priv->current_file_dir_set,
737 path_hash(newpath.buf), newpath.buf) ||
738 (!opt->priv->call_depth && file_exists(newpath.buf))) {
739 strbuf_setlen(&newpath, base_len);
740 strbuf_addf(&newpath, "_%d", suffix++);
741 }
742
743 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
744 hashmap_entry_init(&entry->e, path_hash(entry->path));
745 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
746 return strbuf_detach(&newpath, NULL);
747 }
748
749 /**
750 * Check whether a directory in the index is in the way of an incoming
751 * file. Return 1 if so. If check_working_copy is non-zero, also
752 * check the working directory. If empty_ok is non-zero, also return
753 * 0 in the case where the working-tree dir exists but is empty.
754 */
755 static int dir_in_way(struct index_state *istate, const char *path,
756 int check_working_copy, int empty_ok)
757 {
758 int pos;
759 struct strbuf dirpath = STRBUF_INIT;
760 struct stat st;
761
762 strbuf_addstr(&dirpath, path);
763 strbuf_addch(&dirpath, '/');
764
765 pos = index_name_pos(istate, dirpath.buf, dirpath.len);
766
767 if (pos < 0)
768 pos = -1 - pos;
769 if (pos < istate->cache_nr &&
770 !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
771 strbuf_release(&dirpath);
772 return 1;
773 }
774
775 strbuf_release(&dirpath);
776 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
777 !(empty_ok && is_empty_dir(path)) &&
778 !has_symlink_leading_path(path, strlen(path));
779 }
780
781 /*
782 * Returns whether path was tracked in the index before the merge started,
783 * and its oid and mode match the specified values
784 */
785 static int was_tracked_and_matches(struct merge_options *opt, const char *path,
786 const struct diff_filespec *blob)
787 {
788 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
789 struct cache_entry *ce;
790
791 if (0 > pos)
792 /* we were not tracking this path before the merge */
793 return 0;
794
795 /* See if the file we were tracking before matches */
796 ce = opt->priv->orig_index.cache[pos];
797 return (oideq(&ce->oid, &blob->oid) && ce->ce_mode == blob->mode);
798 }
799
800 /*
801 * Returns whether path was tracked in the index before the merge started
802 */
803 static int was_tracked(struct merge_options *opt, const char *path)
804 {
805 int pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
806
807 if (0 <= pos)
808 /* we were tracking this path before the merge */
809 return 1;
810
811 return 0;
812 }
813
814 static int would_lose_untracked(struct merge_options *opt, const char *path)
815 {
816 struct index_state *istate = opt->repo->index;
817
818 /*
819 * This may look like it can be simplified to:
820 * return !was_tracked(opt, path) && file_exists(path)
821 * but it can't. This function needs to know whether path was in
822 * the working tree due to EITHER having been tracked in the index
823 * before the merge OR having been put into the working copy and
824 * index by unpack_trees(). Due to that either-or requirement, we
825 * check the current index instead of the original one.
826 *
827 * Note that we do not need to worry about merge-recursive itself
828 * updating the index after unpack_trees() and before calling this
829 * function, because we strictly require all code paths in
830 * merge-recursive to update the working tree first and the index
831 * second. Doing otherwise would break
832 * update_file()/would_lose_untracked(); see every comment in this
833 * file which mentions "update_stages".
834 */
835 int pos = index_name_pos(istate, path, strlen(path));
836
837 if (pos < 0)
838 pos = -1 - pos;
839 while (pos < istate->cache_nr &&
840 !strcmp(path, istate->cache[pos]->name)) {
841 /*
842 * If stage #0, it is definitely tracked.
843 * If it has stage #2 then it was tracked
844 * before this merge started. All other
845 * cases the path was not tracked.
846 */
847 switch (ce_stage(istate->cache[pos])) {
848 case 0:
849 case 2:
850 return 0;
851 }
852 pos++;
853 }
854 return file_exists(path);
855 }
856
857 static int was_dirty(struct merge_options *opt, const char *path)
858 {
859 struct cache_entry *ce;
860 int dirty = 1;
861
862 if (opt->priv->call_depth || !was_tracked(opt, path))
863 return !dirty;
864
865 ce = index_file_exists(opt->priv->unpack_opts.src_index,
866 path, strlen(path), ignore_case);
867 dirty = verify_uptodate(ce, &opt->priv->unpack_opts) != 0;
868 return dirty;
869 }
870
871 static int make_room_for_path(struct merge_options *opt, const char *path)
872 {
873 int status, i;
874 const char *msg = _("failed to create path '%s'%s");
875
876 /* Unlink any D/F conflict files that are in the way */
877 for (i = 0; i < opt->priv->df_conflict_file_set.nr; i++) {
878 const char *df_path = opt->priv->df_conflict_file_set.items[i].string;
879 size_t pathlen = strlen(path);
880 size_t df_pathlen = strlen(df_path);
881 if (df_pathlen < pathlen &&
882 path[df_pathlen] == '/' &&
883 strncmp(path, df_path, df_pathlen) == 0) {
884 output(opt, 3,
885 _("Removing %s to make room for subdirectory\n"),
886 df_path);
887 unlink(df_path);
888 unsorted_string_list_delete_item(&opt->priv->df_conflict_file_set,
889 i, 0);
890 break;
891 }
892 }
893
894 /* Make sure leading directories are created */
895 status = safe_create_leading_directories_const(path);
896 if (status) {
897 if (status == SCLD_EXISTS)
898 /* something else exists */
899 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
900 return err(opt, msg, path, "");
901 }
902
903 /*
904 * Do not unlink a file in the work tree if we are not
905 * tracking it.
906 */
907 if (would_lose_untracked(opt, path))
908 return err(opt, _("refusing to lose untracked file at '%s'"),
909 path);
910
911 /* Successful unlink is good.. */
912 if (!unlink(path))
913 return 0;
914 /* .. and so is no existing file */
915 if (errno == ENOENT)
916 return 0;
917 /* .. but not some other error (who really cares what?) */
918 return err(opt, msg, path, _(": perhaps a D/F conflict?"));
919 }
920
921 static int update_file_flags(struct merge_options *opt,
922 const struct diff_filespec *contents,
923 const char *path,
924 int update_cache,
925 int update_wd)
926 {
927 int ret = 0;
928
929 if (opt->priv->call_depth)
930 update_wd = 0;
931
932 if (update_wd) {
933 enum object_type type;
934 void *buf;
935 unsigned long size;
936
937 if (S_ISGITLINK(contents->mode)) {
938 /*
939 * We may later decide to recursively descend into
940 * the submodule directory and update its index
941 * and/or work tree, but we do not do that now.
942 */
943 update_wd = 0;
944 goto update_index;
945 }
946
947 buf = read_object_file(&contents->oid, &type, &size);
948 if (!buf) {
949 ret = err(opt, _("cannot read object %s '%s'"),
950 oid_to_hex(&contents->oid), path);
951 goto free_buf;
952 }
953 if (type != OBJ_BLOB) {
954 ret = err(opt, _("blob expected for %s '%s'"),
955 oid_to_hex(&contents->oid), path);
956 goto free_buf;
957 }
958 if (S_ISREG(contents->mode)) {
959 struct strbuf strbuf = STRBUF_INIT;
960 if (convert_to_working_tree(opt->repo->index,
961 path, buf, size, &strbuf)) {
962 free(buf);
963 size = strbuf.len;
964 buf = strbuf_detach(&strbuf, NULL);
965 }
966 }
967
968 if (make_room_for_path(opt, path) < 0) {
969 update_wd = 0;
970 goto free_buf;
971 }
972 if (S_ISREG(contents->mode) ||
973 (!has_symlinks && S_ISLNK(contents->mode))) {
974 int fd;
975 int mode = (contents->mode & 0100 ? 0777 : 0666);
976
977 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
978 if (fd < 0) {
979 ret = err(opt, _("failed to open '%s': %s"),
980 path, strerror(errno));
981 goto free_buf;
982 }
983 write_in_full(fd, buf, size);
984 close(fd);
985 } else if (S_ISLNK(contents->mode)) {
986 char *lnk = xmemdupz(buf, size);
987 safe_create_leading_directories_const(path);
988 unlink(path);
989 if (symlink(lnk, path))
990 ret = err(opt, _("failed to symlink '%s': %s"),
991 path, strerror(errno));
992 free(lnk);
993 } else
994 ret = err(opt,
995 _("do not know what to do with %06o %s '%s'"),
996 contents->mode, oid_to_hex(&contents->oid), path);
997 free_buf:
998 free(buf);
999 }
1000 update_index:
1001 if (!ret && update_cache)
1002 if (add_cacheinfo(opt, contents, path, 0, update_wd,
1003 ADD_CACHE_OK_TO_ADD))
1004 return -1;
1005 return ret;
1006 }
1007
1008 static int update_file(struct merge_options *opt,
1009 int clean,
1010 const struct diff_filespec *contents,
1011 const char *path)
1012 {
1013 return update_file_flags(opt, contents, path,
1014 opt->priv->call_depth || clean, !opt->priv->call_depth);
1015 }
1016
1017 /* Low level file merging, update and removal */
1018
1019 struct merge_file_info {
1020 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1021 unsigned clean:1,
1022 merge:1;
1023 };
1024
1025 static int merge_3way(struct merge_options *opt,
1026 mmbuffer_t *result_buf,
1027 const struct diff_filespec *o,
1028 const struct diff_filespec *a,
1029 const struct diff_filespec *b,
1030 const char *branch1,
1031 const char *branch2,
1032 const int extra_marker_size)
1033 {
1034 mmfile_t orig, src1, src2;
1035 struct ll_merge_options ll_opts = {0};
1036 char *base, *name1, *name2;
1037 int merge_status;
1038
1039 ll_opts.renormalize = opt->renormalize;
1040 ll_opts.extra_marker_size = extra_marker_size;
1041 ll_opts.xdl_opts = opt->xdl_opts;
1042
1043 if (opt->priv->call_depth) {
1044 ll_opts.virtual_ancestor = 1;
1045 ll_opts.variant = 0;
1046 } else {
1047 switch (opt->recursive_variant) {
1048 case MERGE_VARIANT_OURS:
1049 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1050 break;
1051 case MERGE_VARIANT_THEIRS:
1052 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1053 break;
1054 default:
1055 ll_opts.variant = 0;
1056 break;
1057 }
1058 }
1059
1060 assert(a->path && b->path && o->path && opt->ancestor);
1061 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1062 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1063 name1 = mkpathdup("%s:%s", branch1, a->path);
1064 name2 = mkpathdup("%s:%s", branch2, b->path);
1065 } else {
1066 base = mkpathdup("%s", opt->ancestor);
1067 name1 = mkpathdup("%s", branch1);
1068 name2 = mkpathdup("%s", branch2);
1069 }
1070
1071 read_mmblob(&orig, &o->oid);
1072 read_mmblob(&src1, &a->oid);
1073 read_mmblob(&src2, &b->oid);
1074
1075 merge_status = ll_merge(result_buf, a->path, &orig, base,
1076 &src1, name1, &src2, name2,
1077 opt->repo->index, &ll_opts);
1078
1079 free(base);
1080 free(name1);
1081 free(name2);
1082 free(orig.ptr);
1083 free(src1.ptr);
1084 free(src2.ptr);
1085 return merge_status;
1086 }
1087
1088 static int find_first_merges(struct repository *repo,
1089 struct object_array *result, const char *path,
1090 struct commit *a, struct commit *b)
1091 {
1092 int i, j;
1093 struct object_array merges = OBJECT_ARRAY_INIT;
1094 struct commit *commit;
1095 int contains_another;
1096
1097 char merged_revision[GIT_MAX_HEXSZ + 2];
1098 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1099 "--all", merged_revision, NULL };
1100 struct rev_info revs;
1101 struct setup_revision_opt rev_opts;
1102
1103 memset(result, 0, sizeof(struct object_array));
1104 memset(&rev_opts, 0, sizeof(rev_opts));
1105
1106 /* get all revisions that merge commit a */
1107 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1108 oid_to_hex(&a->object.oid));
1109 repo_init_revisions(repo, &revs, NULL);
1110 rev_opts.submodule = path;
1111 /* FIXME: can't handle linked worktrees in submodules yet */
1112 revs.single_worktree = path != NULL;
1113 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1114
1115 /* save all revisions from the above list that contain b */
1116 if (prepare_revision_walk(&revs))
1117 die("revision walk setup failed");
1118 while ((commit = get_revision(&revs)) != NULL) {
1119 struct object *o = &(commit->object);
1120 if (in_merge_bases(b, commit))
1121 add_object_array(o, NULL, &merges);
1122 }
1123 reset_revision_walk();
1124
1125 /* Now we've got all merges that contain a and b. Prune all
1126 * merges that contain another found merge and save them in
1127 * result.
1128 */
1129 for (i = 0; i < merges.nr; i++) {
1130 struct commit *m1 = (struct commit *) merges.objects[i].item;
1131
1132 contains_another = 0;
1133 for (j = 0; j < merges.nr; j++) {
1134 struct commit *m2 = (struct commit *) merges.objects[j].item;
1135 if (i != j && in_merge_bases(m2, m1)) {
1136 contains_another = 1;
1137 break;
1138 }
1139 }
1140
1141 if (!contains_another)
1142 add_object_array(merges.objects[i].item, NULL, result);
1143 }
1144
1145 object_array_clear(&merges);
1146 return result->nr;
1147 }
1148
1149 static void print_commit(struct commit *commit)
1150 {
1151 struct strbuf sb = STRBUF_INIT;
1152 struct pretty_print_context ctx = {0};
1153 ctx.date_mode.type = DATE_NORMAL;
1154 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1155 fprintf(stderr, "%s\n", sb.buf);
1156 strbuf_release(&sb);
1157 }
1158
1159 static int is_valid(const struct diff_filespec *dfs)
1160 {
1161 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1162 }
1163
1164 static int merge_submodule(struct merge_options *opt,
1165 struct object_id *result, const char *path,
1166 const struct object_id *base, const struct object_id *a,
1167 const struct object_id *b)
1168 {
1169 struct commit *commit_base, *commit_a, *commit_b;
1170 int parent_count;
1171 struct object_array merges;
1172
1173 int i;
1174 int search = !opt->priv->call_depth;
1175
1176 /* store a in result in case we fail */
1177 oidcpy(result, a);
1178
1179 /* we can not handle deletion conflicts */
1180 if (is_null_oid(base))
1181 return 0;
1182 if (is_null_oid(a))
1183 return 0;
1184 if (is_null_oid(b))
1185 return 0;
1186
1187 if (add_submodule_odb(path)) {
1188 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1189 return 0;
1190 }
1191
1192 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1193 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1194 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1195 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1196 return 0;
1197 }
1198
1199 /* check whether both changes are forward */
1200 if (!in_merge_bases(commit_base, commit_a) ||
1201 !in_merge_bases(commit_base, commit_b)) {
1202 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1203 return 0;
1204 }
1205
1206 /* Case #1: a is contained in b or vice versa */
1207 if (in_merge_bases(commit_a, commit_b)) {
1208 oidcpy(result, b);
1209 if (show(opt, 3)) {
1210 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1211 output_commit_title(opt, commit_b);
1212 } else if (show(opt, 2))
1213 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1214 else
1215 ; /* no output */
1216
1217 return 1;
1218 }
1219 if (in_merge_bases(commit_b, commit_a)) {
1220 oidcpy(result, a);
1221 if (show(opt, 3)) {
1222 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1223 output_commit_title(opt, commit_a);
1224 } else if (show(opt, 2))
1225 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1226 else
1227 ; /* no output */
1228
1229 return 1;
1230 }
1231
1232 /*
1233 * Case #2: There are one or more merges that contain a and b in
1234 * the submodule. If there is only one, then present it as a
1235 * suggestion to the user, but leave it marked unmerged so the
1236 * user needs to confirm the resolution.
1237 */
1238
1239 /* Skip the search if makes no sense to the calling context. */
1240 if (!search)
1241 return 0;
1242
1243 /* find commit which merges them */
1244 parent_count = find_first_merges(opt->repo, &merges, path,
1245 commit_a, commit_b);
1246 switch (parent_count) {
1247 case 0:
1248 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1249 break;
1250
1251 case 1:
1252 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1253 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1254 print_commit((struct commit *) merges.objects[0].item);
1255 output(opt, 2, _(
1256 "If this is correct simply add it to the index "
1257 "for example\n"
1258 "by using:\n\n"
1259 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1260 "which will accept this suggestion.\n"),
1261 oid_to_hex(&merges.objects[0].item->oid), path);
1262 break;
1263
1264 default:
1265 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1266 for (i = 0; i < merges.nr; i++)
1267 print_commit((struct commit *) merges.objects[i].item);
1268 }
1269
1270 object_array_clear(&merges);
1271 return 0;
1272 }
1273
1274 static int merge_mode_and_contents(struct merge_options *opt,
1275 const struct diff_filespec *o,
1276 const struct diff_filespec *a,
1277 const struct diff_filespec *b,
1278 const char *filename,
1279 const char *branch1,
1280 const char *branch2,
1281 const int extra_marker_size,
1282 struct merge_file_info *result)
1283 {
1284 if (opt->branch1 != branch1) {
1285 /*
1286 * It's weird getting a reverse merge with HEAD on the bottom
1287 * side of the conflict markers and the other branch on the
1288 * top. Fix that.
1289 */
1290 return merge_mode_and_contents(opt, o, b, a,
1291 filename,
1292 branch2, branch1,
1293 extra_marker_size, result);
1294 }
1295
1296 result->merge = 0;
1297 result->clean = 1;
1298
1299 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1300 result->clean = 0;
1301 if (S_ISREG(a->mode)) {
1302 result->blob.mode = a->mode;
1303 oidcpy(&result->blob.oid, &a->oid);
1304 } else {
1305 result->blob.mode = b->mode;
1306 oidcpy(&result->blob.oid, &b->oid);
1307 }
1308 } else {
1309 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1310 result->merge = 1;
1311
1312 /*
1313 * Merge modes
1314 */
1315 if (a->mode == b->mode || a->mode == o->mode)
1316 result->blob.mode = b->mode;
1317 else {
1318 result->blob.mode = a->mode;
1319 if (b->mode != o->mode) {
1320 result->clean = 0;
1321 result->merge = 1;
1322 }
1323 }
1324
1325 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1326 oidcpy(&result->blob.oid, &b->oid);
1327 else if (oideq(&b->oid, &o->oid))
1328 oidcpy(&result->blob.oid, &a->oid);
1329 else if (S_ISREG(a->mode)) {
1330 mmbuffer_t result_buf;
1331 int ret = 0, merge_status;
1332
1333 merge_status = merge_3way(opt, &result_buf, o, a, b,
1334 branch1, branch2,
1335 extra_marker_size);
1336
1337 if ((merge_status < 0) || !result_buf.ptr)
1338 ret = err(opt, _("Failed to execute internal merge"));
1339
1340 if (!ret &&
1341 write_object_file(result_buf.ptr, result_buf.size,
1342 blob_type, &result->blob.oid))
1343 ret = err(opt, _("Unable to add %s to database"),
1344 a->path);
1345
1346 free(result_buf.ptr);
1347 if (ret)
1348 return ret;
1349 result->clean = (merge_status == 0);
1350 } else if (S_ISGITLINK(a->mode)) {
1351 result->clean = merge_submodule(opt, &result->blob.oid,
1352 o->path,
1353 &o->oid,
1354 &a->oid,
1355 &b->oid);
1356 } else if (S_ISLNK(a->mode)) {
1357 switch (opt->recursive_variant) {
1358 case MERGE_VARIANT_NORMAL:
1359 oidcpy(&result->blob.oid, &a->oid);
1360 if (!oideq(&a->oid, &b->oid))
1361 result->clean = 0;
1362 break;
1363 case MERGE_VARIANT_OURS:
1364 oidcpy(&result->blob.oid, &a->oid);
1365 break;
1366 case MERGE_VARIANT_THEIRS:
1367 oidcpy(&result->blob.oid, &b->oid);
1368 break;
1369 }
1370 } else
1371 BUG("unsupported object type in the tree");
1372 }
1373
1374 if (result->merge)
1375 output(opt, 2, _("Auto-merging %s"), filename);
1376
1377 return 0;
1378 }
1379
1380 static int handle_rename_via_dir(struct merge_options *opt,
1381 struct rename_conflict_info *ci)
1382 {
1383 /*
1384 * Handle file adds that need to be renamed due to directory rename
1385 * detection. This differs from handle_rename_normal, because
1386 * there is no content merge to do; just move the file into the
1387 * desired final location.
1388 */
1389 const struct rename *ren = ci->ren1;
1390 const struct diff_filespec *dest = ren->pair->two;
1391 char *file_path = dest->path;
1392 int mark_conflicted = (opt->detect_directory_renames ==
1393 MERGE_DIRECTORY_RENAMES_CONFLICT);
1394 assert(ren->dir_rename_original_dest);
1395
1396 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1397 mark_conflicted = 1;
1398 file_path = unique_path(opt, dest->path, ren->branch);
1399 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1400 "writing to %s instead."),
1401 dest->path, file_path);
1402 }
1403
1404 if (mark_conflicted) {
1405 /*
1406 * Write the file in worktree at file_path. In the index,
1407 * only record the file at dest->path in the appropriate
1408 * higher stage.
1409 */
1410 if (update_file(opt, 0, dest, file_path))
1411 return -1;
1412 if (file_path != dest->path)
1413 free(file_path);
1414 if (update_stages(opt, dest->path, NULL,
1415 ren->branch == opt->branch1 ? dest : NULL,
1416 ren->branch == opt->branch1 ? NULL : dest))
1417 return -1;
1418 return 0; /* not clean, but conflicted */
1419 } else {
1420 /* Update dest->path both in index and in worktree */
1421 if (update_file(opt, 1, dest, dest->path))
1422 return -1;
1423 return 1; /* clean */
1424 }
1425 }
1426
1427 static int handle_change_delete(struct merge_options *opt,
1428 const char *path, const char *old_path,
1429 const struct diff_filespec *o,
1430 const struct diff_filespec *changed,
1431 const char *change_branch,
1432 const char *delete_branch,
1433 const char *change, const char *change_past)
1434 {
1435 char *alt_path = NULL;
1436 const char *update_path = path;
1437 int ret = 0;
1438
1439 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1440 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1441 update_path = alt_path = unique_path(opt, path, change_branch);
1442 }
1443
1444 if (opt->priv->call_depth) {
1445 /*
1446 * We cannot arbitrarily accept either a_sha or b_sha as
1447 * correct; since there is no true "middle point" between
1448 * them, simply reuse the base version for virtual merge base.
1449 */
1450 ret = remove_file_from_index(opt->repo->index, path);
1451 if (!ret)
1452 ret = update_file(opt, 0, o, update_path);
1453 } else {
1454 /*
1455 * Despite the four nearly duplicate messages and argument
1456 * lists below and the ugliness of the nested if-statements,
1457 * having complete messages makes the job easier for
1458 * translators.
1459 *
1460 * The slight variance among the cases is due to the fact
1461 * that:
1462 * 1) directory/file conflicts (in effect if
1463 * !alt_path) could cause us to need to write the
1464 * file to a different path.
1465 * 2) renames (in effect if !old_path) could mean that
1466 * there are two names for the path that the user
1467 * may know the file by.
1468 */
1469 if (!alt_path) {
1470 if (!old_path) {
1471 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1472 "and %s in %s. Version %s of %s left in tree."),
1473 change, path, delete_branch, change_past,
1474 change_branch, change_branch, path);
1475 } else {
1476 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1477 "and %s to %s in %s. Version %s of %s left in tree."),
1478 change, old_path, delete_branch, change_past, path,
1479 change_branch, change_branch, path);
1480 }
1481 } else {
1482 if (!old_path) {
1483 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1484 "and %s in %s. Version %s of %s left in tree at %s."),
1485 change, path, delete_branch, change_past,
1486 change_branch, change_branch, path, alt_path);
1487 } else {
1488 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1489 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1490 change, old_path, delete_branch, change_past, path,
1491 change_branch, change_branch, path, alt_path);
1492 }
1493 }
1494 /*
1495 * No need to call update_file() on path when change_branch ==
1496 * opt->branch1 && !alt_path, since that would needlessly touch
1497 * path. We could call update_file_flags() with update_cache=0
1498 * and update_wd=0, but that's a no-op.
1499 */
1500 if (change_branch != opt->branch1 || alt_path)
1501 ret = update_file(opt, 0, changed, update_path);
1502 }
1503 free(alt_path);
1504
1505 return ret;
1506 }
1507
1508 static int handle_rename_delete(struct merge_options *opt,
1509 struct rename_conflict_info *ci)
1510 {
1511 const struct rename *ren = ci->ren1;
1512 const struct diff_filespec *orig = ren->pair->one;
1513 const struct diff_filespec *dest = ren->pair->two;
1514 const char *rename_branch = ren->branch;
1515 const char *delete_branch = (opt->branch1 == ren->branch ?
1516 opt->branch2 : opt->branch1);
1517
1518 if (handle_change_delete(opt,
1519 opt->priv->call_depth ? orig->path : dest->path,
1520 opt->priv->call_depth ? NULL : orig->path,
1521 orig, dest,
1522 rename_branch, delete_branch,
1523 _("rename"), _("renamed")))
1524 return -1;
1525
1526 if (opt->priv->call_depth)
1527 return remove_file_from_index(opt->repo->index, dest->path);
1528 else
1529 return update_stages(opt, dest->path, NULL,
1530 rename_branch == opt->branch1 ? dest : NULL,
1531 rename_branch == opt->branch1 ? NULL : dest);
1532 }
1533
1534 static int handle_file_collision(struct merge_options *opt,
1535 const char *collide_path,
1536 const char *prev_path1,
1537 const char *prev_path2,
1538 const char *branch1, const char *branch2,
1539 struct diff_filespec *a,
1540 struct diff_filespec *b)
1541 {
1542 struct merge_file_info mfi;
1543 struct diff_filespec null;
1544 char *alt_path = NULL;
1545 const char *update_path = collide_path;
1546
1547 /*
1548 * It's easiest to get the correct things into stage 2 and 3, and
1549 * to make sure that the content merge puts HEAD before the other
1550 * branch if we just ensure that branch1 == opt->branch1. So, simply
1551 * flip arguments around if we don't have that.
1552 */
1553 if (branch1 != opt->branch1) {
1554 return handle_file_collision(opt, collide_path,
1555 prev_path2, prev_path1,
1556 branch2, branch1,
1557 b, a);
1558 }
1559
1560 /*
1561 * In the recursive case, we just opt to undo renames
1562 */
1563 if (opt->priv->call_depth && (prev_path1 || prev_path2)) {
1564 /* Put first file (a->oid, a->mode) in its original spot */
1565 if (prev_path1) {
1566 if (update_file(opt, 1, a, prev_path1))
1567 return -1;
1568 } else {
1569 if (update_file(opt, 1, a, collide_path))
1570 return -1;
1571 }
1572
1573 /* Put second file (b->oid, b->mode) in its original spot */
1574 if (prev_path2) {
1575 if (update_file(opt, 1, b, prev_path2))
1576 return -1;
1577 } else {
1578 if (update_file(opt, 1, b, collide_path))
1579 return -1;
1580 }
1581
1582 /* Don't leave something at collision path if unrenaming both */
1583 if (prev_path1 && prev_path2)
1584 remove_file(opt, 1, collide_path, 0);
1585
1586 return 0;
1587 }
1588
1589 /* Remove rename sources if rename/add or rename/rename(2to1) */
1590 if (prev_path1)
1591 remove_file(opt, 1, prev_path1,
1592 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1593 if (prev_path2)
1594 remove_file(opt, 1, prev_path2,
1595 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1596
1597 /*
1598 * Remove the collision path, if it wouldn't cause dirty contents
1599 * or an untracked file to get lost. We'll either overwrite with
1600 * merged contents, or just write out to differently named files.
1601 */
1602 if (was_dirty(opt, collide_path)) {
1603 output(opt, 1, _("Refusing to lose dirty file at %s"),
1604 collide_path);
1605 update_path = alt_path = unique_path(opt, collide_path, "merged");
1606 } else if (would_lose_untracked(opt, collide_path)) {
1607 /*
1608 * Only way we get here is if both renames were from
1609 * a directory rename AND user had an untracked file
1610 * at the location where both files end up after the
1611 * two directory renames. See testcase 10d of t6043.
1612 */
1613 output(opt, 1, _("Refusing to lose untracked file at "
1614 "%s, even though it's in the way."),
1615 collide_path);
1616 update_path = alt_path = unique_path(opt, collide_path, "merged");
1617 } else {
1618 /*
1619 * FIXME: It's possible that the two files are identical
1620 * and that the current working copy happens to match, in
1621 * which case we are unnecessarily touching the working
1622 * tree file. It's not a likely enough scenario that I
1623 * want to code up the checks for it and a better fix is
1624 * available if we restructure how unpack_trees() and
1625 * merge-recursive interoperate anyway, so punting for
1626 * now...
1627 */
1628 remove_file(opt, 0, collide_path, 0);
1629 }
1630
1631 /* Store things in diff_filespecs for functions that need it */
1632 null.path = (char *)collide_path;
1633 oidcpy(&null.oid, &null_oid);
1634 null.mode = 0;
1635
1636 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1637 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1638 return -1;
1639 mfi.clean &= !alt_path;
1640 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1641 return -1;
1642 if (!mfi.clean && !opt->priv->call_depth &&
1643 update_stages(opt, collide_path, NULL, a, b))
1644 return -1;
1645 free(alt_path);
1646 /*
1647 * FIXME: If both a & b both started with conflicts (only possible
1648 * if they came from a rename/rename(2to1)), but had IDENTICAL
1649 * contents including those conflicts, then in the next line we claim
1650 * it was clean. If someone cares about this case, we should have the
1651 * caller notify us if we started with conflicts.
1652 */
1653 return mfi.clean;
1654 }
1655
1656 static int handle_rename_add(struct merge_options *opt,
1657 struct rename_conflict_info *ci)
1658 {
1659 /* a was renamed to c, and a separate c was added. */
1660 struct diff_filespec *a = ci->ren1->pair->one;
1661 struct diff_filespec *c = ci->ren1->pair->two;
1662 char *path = c->path;
1663 char *prev_path_desc;
1664 struct merge_file_info mfi;
1665
1666 const char *rename_branch = ci->ren1->branch;
1667 const char *add_branch = (opt->branch1 == rename_branch ?
1668 opt->branch2 : opt->branch1);
1669 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1670
1671 output(opt, 1, _("CONFLICT (rename/add): "
1672 "Rename %s->%s in %s. Added %s in %s"),
1673 a->path, c->path, rename_branch,
1674 c->path, add_branch);
1675
1676 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1677 ci->ren1->src_entry->stages[other_stage].path = a->path;
1678 if (merge_mode_and_contents(opt, a, c,
1679 &ci->ren1->src_entry->stages[other_stage],
1680 prev_path_desc,
1681 opt->branch1, opt->branch2,
1682 1 + opt->priv->call_depth * 2, &mfi))
1683 return -1;
1684 free(prev_path_desc);
1685
1686 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1687 return handle_file_collision(opt,
1688 c->path, a->path, NULL,
1689 rename_branch, add_branch,
1690 &mfi.blob,
1691 &ci->ren1->dst_entry->stages[other_stage]);
1692 }
1693
1694 static char *find_path_for_conflict(struct merge_options *opt,
1695 const char *path,
1696 const char *branch1,
1697 const char *branch2)
1698 {
1699 char *new_path = NULL;
1700 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1701 new_path = unique_path(opt, path, branch1);
1702 output(opt, 1, _("%s is a directory in %s adding "
1703 "as %s instead"),
1704 path, branch2, new_path);
1705 } else if (would_lose_untracked(opt, path)) {
1706 new_path = unique_path(opt, path, branch1);
1707 output(opt, 1, _("Refusing to lose untracked file"
1708 " at %s; adding as %s instead"),
1709 path, new_path);
1710 }
1711
1712 return new_path;
1713 }
1714
1715 /*
1716 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1717 */
1718 static inline int flip_stage(int stage)
1719 {
1720 return (2 + 3) - stage;
1721 }
1722
1723 static int handle_rename_rename_1to2(struct merge_options *opt,
1724 struct rename_conflict_info *ci)
1725 {
1726 /* One file was renamed in both branches, but to different names. */
1727 struct merge_file_info mfi;
1728 struct diff_filespec *add;
1729 struct diff_filespec *o = ci->ren1->pair->one;
1730 struct diff_filespec *a = ci->ren1->pair->two;
1731 struct diff_filespec *b = ci->ren2->pair->two;
1732 char *path_desc;
1733
1734 output(opt, 1, _("CONFLICT (rename/rename): "
1735 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1736 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1737 o->path, a->path, ci->ren1->branch,
1738 o->path, b->path, ci->ren2->branch,
1739 opt->priv->call_depth ? _(" (left unresolved)") : "");
1740
1741 path_desc = xstrfmt("%s and %s, both renamed from %s",
1742 a->path, b->path, o->path);
1743 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1744 ci->ren1->branch, ci->ren2->branch,
1745 opt->priv->call_depth * 2, &mfi))
1746 return -1;
1747 free(path_desc);
1748
1749 if (opt->priv->call_depth) {
1750 /*
1751 * FIXME: For rename/add-source conflicts (if we could detect
1752 * such), this is wrong. We should instead find a unique
1753 * pathname and then either rename the add-source file to that
1754 * unique path, or use that unique path instead of src here.
1755 */
1756 if (update_file(opt, 0, &mfi.blob, o->path))
1757 return -1;
1758
1759 /*
1760 * Above, we put the merged content at the merge-base's
1761 * path. Now we usually need to delete both a->path and
1762 * b->path. However, the rename on each side of the merge
1763 * could also be involved in a rename/add conflict. In
1764 * such cases, we should keep the added file around,
1765 * resolving the conflict at that path in its favor.
1766 */
1767 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1768 if (is_valid(add)) {
1769 if (update_file(opt, 0, add, a->path))
1770 return -1;
1771 }
1772 else
1773 remove_file_from_index(opt->repo->index, a->path);
1774 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1775 if (is_valid(add)) {
1776 if (update_file(opt, 0, add, b->path))
1777 return -1;
1778 }
1779 else
1780 remove_file_from_index(opt->repo->index, b->path);
1781 } else {
1782 /*
1783 * For each destination path, we need to see if there is a
1784 * rename/add collision. If not, we can write the file out
1785 * to the specified location.
1786 */
1787 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1788 if (is_valid(add)) {
1789 add->path = mfi.blob.path = a->path;
1790 if (handle_file_collision(opt, a->path,
1791 NULL, NULL,
1792 ci->ren1->branch,
1793 ci->ren2->branch,
1794 &mfi.blob, add) < 0)
1795 return -1;
1796 } else {
1797 char *new_path = find_path_for_conflict(opt, a->path,
1798 ci->ren1->branch,
1799 ci->ren2->branch);
1800 if (update_file(opt, 0, &mfi.blob,
1801 new_path ? new_path : a->path))
1802 return -1;
1803 free(new_path);
1804 if (update_stages(opt, a->path, NULL, a, NULL))
1805 return -1;
1806 }
1807
1808 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1809 if (is_valid(add)) {
1810 add->path = mfi.blob.path = b->path;
1811 if (handle_file_collision(opt, b->path,
1812 NULL, NULL,
1813 ci->ren1->branch,
1814 ci->ren2->branch,
1815 add, &mfi.blob) < 0)
1816 return -1;
1817 } else {
1818 char *new_path = find_path_for_conflict(opt, b->path,
1819 ci->ren2->branch,
1820 ci->ren1->branch);
1821 if (update_file(opt, 0, &mfi.blob,
1822 new_path ? new_path : b->path))
1823 return -1;
1824 free(new_path);
1825 if (update_stages(opt, b->path, NULL, NULL, b))
1826 return -1;
1827 }
1828 }
1829
1830 return 0;
1831 }
1832
1833 static int handle_rename_rename_2to1(struct merge_options *opt,
1834 struct rename_conflict_info *ci)
1835 {
1836 /* Two files, a & b, were renamed to the same thing, c. */
1837 struct diff_filespec *a = ci->ren1->pair->one;
1838 struct diff_filespec *b = ci->ren2->pair->one;
1839 struct diff_filespec *c1 = ci->ren1->pair->two;
1840 struct diff_filespec *c2 = ci->ren2->pair->two;
1841 char *path = c1->path; /* == c2->path */
1842 char *path_side_1_desc;
1843 char *path_side_2_desc;
1844 struct merge_file_info mfi_c1;
1845 struct merge_file_info mfi_c2;
1846 int ostage1, ostage2;
1847
1848 output(opt, 1, _("CONFLICT (rename/rename): "
1849 "Rename %s->%s in %s. "
1850 "Rename %s->%s in %s"),
1851 a->path, c1->path, ci->ren1->branch,
1852 b->path, c2->path, ci->ren2->branch);
1853
1854 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1855 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1856 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1857 ostage2 = flip_stage(ostage1);
1858 ci->ren1->src_entry->stages[ostage1].path = a->path;
1859 ci->ren2->src_entry->stages[ostage2].path = b->path;
1860 if (merge_mode_and_contents(opt, a, c1,
1861 &ci->ren1->src_entry->stages[ostage1],
1862 path_side_1_desc,
1863 opt->branch1, opt->branch2,
1864 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1865 merge_mode_and_contents(opt, b,
1866 &ci->ren2->src_entry->stages[ostage2],
1867 c2, path_side_2_desc,
1868 opt->branch1, opt->branch2,
1869 1 + opt->priv->call_depth * 2, &mfi_c2))
1870 return -1;
1871 free(path_side_1_desc);
1872 free(path_side_2_desc);
1873 mfi_c1.blob.path = path;
1874 mfi_c2.blob.path = path;
1875
1876 return handle_file_collision(opt, path, a->path, b->path,
1877 ci->ren1->branch, ci->ren2->branch,
1878 &mfi_c1.blob, &mfi_c2.blob);
1879 }
1880
1881 /*
1882 * Get the diff_filepairs changed between o_tree and tree.
1883 */
1884 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1885 struct tree *o_tree,
1886 struct tree *tree)
1887 {
1888 struct diff_queue_struct *ret;
1889 struct diff_options opts;
1890
1891 repo_diff_setup(opt->repo, &opts);
1892 opts.flags.recursive = 1;
1893 opts.flags.rename_empty = 0;
1894 opts.detect_rename = merge_detect_rename(opt);
1895 /*
1896 * We do not have logic to handle the detection of copies. In
1897 * fact, it may not even make sense to add such logic: would we
1898 * really want a change to a base file to be propagated through
1899 * multiple other files by a merge?
1900 */
1901 if (opts.detect_rename > DIFF_DETECT_RENAME)
1902 opts.detect_rename = DIFF_DETECT_RENAME;
1903 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 1000;
1904 opts.rename_score = opt->rename_score;
1905 opts.show_rename_progress = opt->show_rename_progress;
1906 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1907 diff_setup_done(&opts);
1908 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1909 diffcore_std(&opts);
1910 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1911 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1912
1913 ret = xmalloc(sizeof(*ret));
1914 *ret = diff_queued_diff;
1915
1916 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1917 diff_queued_diff.nr = 0;
1918 diff_queued_diff.queue = NULL;
1919 diff_flush(&opts);
1920 return ret;
1921 }
1922
1923 static int tree_has_path(struct repository *r, struct tree *tree,
1924 const char *path)
1925 {
1926 struct object_id hashy;
1927 unsigned short mode_o;
1928
1929 return !get_tree_entry(r,
1930 &tree->object.oid, path,
1931 &hashy, &mode_o);
1932 }
1933
1934 /*
1935 * Return a new string that replaces the beginning portion (which matches
1936 * entry->dir), with entry->new_dir. In perl-speak:
1937 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1938 * NOTE:
1939 * Caller must ensure that old_path starts with entry->dir + '/'.
1940 */
1941 static char *apply_dir_rename(struct dir_rename_entry *entry,
1942 const char *old_path)
1943 {
1944 struct strbuf new_path = STRBUF_INIT;
1945 int oldlen, newlen;
1946
1947 if (entry->non_unique_new_dir)
1948 return NULL;
1949
1950 oldlen = strlen(entry->dir);
1951 if (entry->new_dir.len == 0)
1952 /*
1953 * If someone renamed/merged a subdirectory into the root
1954 * directory (e.g. 'some/subdir' -> ''), then we want to
1955 * avoid returning
1956 * '' + '/filename'
1957 * as the rename; we need to make old_path + oldlen advance
1958 * past the '/' character.
1959 */
1960 oldlen++;
1961 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1962 strbuf_grow(&new_path, newlen);
1963 strbuf_addbuf(&new_path, &entry->new_dir);
1964 strbuf_addstr(&new_path, &old_path[oldlen]);
1965
1966 return strbuf_detach(&new_path, NULL);
1967 }
1968
1969 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1970 char **old_dir, char **new_dir)
1971 {
1972 char *end_of_old, *end_of_new;
1973
1974 /* Default return values: NULL, meaning no rename */
1975 *old_dir = NULL;
1976 *new_dir = NULL;
1977
1978 /*
1979 * For
1980 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1981 * the "e/foo.c" part is the same, we just want to know that
1982 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1983 * so, for this example, this function returns "a/b/c/d" in
1984 * *old_dir and "a/b/some/thing/else" in *new_dir.
1985 */
1986
1987 /*
1988 * If the basename of the file changed, we don't care. We want
1989 * to know which portion of the directory, if any, changed.
1990 */
1991 end_of_old = strrchr(old_path, '/');
1992 end_of_new = strrchr(new_path, '/');
1993
1994 /*
1995 * If end_of_old is NULL, old_path wasn't in a directory, so there
1996 * could not be a directory rename (our rule elsewhere that a
1997 * directory which still exists is not considered to have been
1998 * renamed means the root directory can never be renamed -- because
1999 * the root directory always exists).
2000 */
2001 if (end_of_old == NULL)
2002 return; /* Note: *old_dir and *new_dir are still NULL */
2003
2004 /*
2005 * If new_path contains no directory (end_of_new is NULL), then we
2006 * have a rename of old_path's directory to the root directory.
2007 */
2008 if (end_of_new == NULL) {
2009 *old_dir = xstrndup(old_path, end_of_old - old_path);
2010 *new_dir = xstrdup("");
2011 return;
2012 }
2013
2014 /* Find the first non-matching character traversing backwards */
2015 while (*--end_of_new == *--end_of_old &&
2016 end_of_old != old_path &&
2017 end_of_new != new_path)
2018 ; /* Do nothing; all in the while loop */
2019
2020 /*
2021 * If both got back to the beginning of their strings, then the
2022 * directory didn't change at all, only the basename did.
2023 */
2024 if (end_of_old == old_path && end_of_new == new_path &&
2025 *end_of_old == *end_of_new)
2026 return; /* Note: *old_dir and *new_dir are still NULL */
2027
2028 /*
2029 * If end_of_new got back to the beginning of its string, and
2030 * end_of_old got back to the beginning of some subdirectory, then
2031 * we have a rename/merge of a subdirectory into the root, which
2032 * needs slightly special handling.
2033 *
2034 * Note: There is no need to consider the opposite case, with a
2035 * rename/merge of the root directory into some subdirectory
2036 * because as noted above the root directory always exists so it
2037 * cannot be considered to be renamed.
2038 */
2039 if (end_of_new == new_path &&
2040 end_of_old != old_path && end_of_old[-1] == '/') {
2041 *old_dir = xstrndup(old_path, --end_of_old - old_path);
2042 *new_dir = xstrdup("");
2043 return;
2044 }
2045
2046 /*
2047 * We've found the first non-matching character in the directory
2048 * paths. That means the current characters we were looking at
2049 * were part of the first non-matching subdir name going back from
2050 * the end of the strings. Get the whole name by advancing both
2051 * end_of_old and end_of_new to the NEXT '/' character. That will
2052 * represent the entire directory rename.
2053 *
2054 * The reason for the increment is cases like
2055 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2056 * After dropping the basename and going back to the first
2057 * non-matching character, we're now comparing:
2058 * a/b/s and a/b/
2059 * and we want to be comparing:
2060 * a/b/star/ and a/b/tar/
2061 * but without the pre-increment, the one on the right would stay
2062 * a/b/.
2063 */
2064 end_of_old = strchr(++end_of_old, '/');
2065 end_of_new = strchr(++end_of_new, '/');
2066
2067 /* Copy the old and new directories into *old_dir and *new_dir. */
2068 *old_dir = xstrndup(old_path, end_of_old - old_path);
2069 *new_dir = xstrndup(new_path, end_of_new - new_path);
2070 }
2071
2072 static void remove_hashmap_entries(struct hashmap *dir_renames,
2073 struct string_list *items_to_remove)
2074 {
2075 int i;
2076 struct dir_rename_entry *entry;
2077
2078 for (i = 0; i < items_to_remove->nr; i++) {
2079 entry = items_to_remove->items[i].util;
2080 hashmap_remove(dir_renames, &entry->ent, NULL);
2081 }
2082 string_list_clear(items_to_remove, 0);
2083 }
2084
2085 /*
2086 * See if there is a directory rename for path, and if there are any file
2087 * level conflicts for the renamed location. If there is a rename and
2088 * there are no conflicts, return the new name. Otherwise, return NULL.
2089 */
2090 static char *handle_path_level_conflicts(struct merge_options *opt,
2091 const char *path,
2092 struct dir_rename_entry *entry,
2093 struct hashmap *collisions,
2094 struct tree *tree)
2095 {
2096 char *new_path = NULL;
2097 struct collision_entry *collision_ent;
2098 int clean = 1;
2099 struct strbuf collision_paths = STRBUF_INIT;
2100
2101 /*
2102 * entry has the mapping of old directory name to new directory name
2103 * that we want to apply to path.
2104 */
2105 new_path = apply_dir_rename(entry, path);
2106
2107 if (!new_path) {
2108 /* This should only happen when entry->non_unique_new_dir set */
2109 if (!entry->non_unique_new_dir)
2110 BUG("entry->non_unqiue_dir not set and !new_path");
2111 output(opt, 1, _("CONFLICT (directory rename split): "
2112 "Unclear where to place %s because directory "
2113 "%s was renamed to multiple other directories, "
2114 "with no destination getting a majority of the "
2115 "files."),
2116 path, entry->dir);
2117 clean = 0;
2118 return NULL;
2119 }
2120
2121 /*
2122 * The caller needs to have ensured that it has pre-populated
2123 * collisions with all paths that map to new_path. Do a quick check
2124 * to ensure that's the case.
2125 */
2126 collision_ent = collision_find_entry(collisions, new_path);
2127 if (collision_ent == NULL)
2128 BUG("collision_ent is NULL");
2129
2130 /*
2131 * Check for one-sided add/add/.../add conflicts, i.e.
2132 * where implicit renames from the other side doing
2133 * directory rename(s) can affect this side of history
2134 * to put multiple paths into the same location. Warn
2135 * and bail on directory renames for such paths.
2136 */
2137 if (collision_ent->reported_already) {
2138 clean = 0;
2139 } else if (tree_has_path(opt->repo, tree, new_path)) {
2140 collision_ent->reported_already = 1;
2141 strbuf_add_separated_string_list(&collision_paths, ", ",
2142 &collision_ent->source_files);
2143 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2144 "file/dir at %s in the way of implicit "
2145 "directory rename(s) putting the following "
2146 "path(s) there: %s."),
2147 new_path, collision_paths.buf);
2148 clean = 0;
2149 } else if (collision_ent->source_files.nr > 1) {
2150 collision_ent->reported_already = 1;
2151 strbuf_add_separated_string_list(&collision_paths, ", ",
2152 &collision_ent->source_files);
2153 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2154 "more than one path to %s; implicit directory "
2155 "renames tried to put these paths there: %s"),
2156 new_path, collision_paths.buf);
2157 clean = 0;
2158 }
2159
2160 /* Free memory we no longer need */
2161 strbuf_release(&collision_paths);
2162 if (!clean && new_path) {
2163 free(new_path);
2164 return NULL;
2165 }
2166
2167 return new_path;
2168 }
2169
2170 /*
2171 * There are a couple things we want to do at the directory level:
2172 * 1. Check for both sides renaming to the same thing, in order to avoid
2173 * implicit renaming of files that should be left in place. (See
2174 * testcase 6b in t6043 for details.)
2175 * 2. Prune directory renames if there are still files left in the
2176 * the original directory. These represent a partial directory rename,
2177 * i.e. a rename where only some of the files within the directory
2178 * were renamed elsewhere. (Technically, this could be done earlier
2179 * in get_directory_renames(), except that would prevent us from
2180 * doing the previous check and thus failing testcase 6b.)
2181 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2182 * In the future, we could potentially record this info as well and
2183 * omit reporting rename/rename(1to2) conflicts for each path within
2184 * the affected directories, thus cleaning up the merge output.
2185 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2186 * directory level, because merging directories is fine. If it
2187 * causes conflicts for files within those merged directories, then
2188 * that should be detected at the individual path level.
2189 */
2190 static void handle_directory_level_conflicts(struct merge_options *opt,
2191 struct hashmap *dir_re_head,
2192 struct tree *head,
2193 struct hashmap *dir_re_merge,
2194 struct tree *merge)
2195 {
2196 struct hashmap_iter iter;
2197 struct dir_rename_entry *head_ent;
2198 struct dir_rename_entry *merge_ent;
2199
2200 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2201 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2202
2203 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2204 ent /* member name */) {
2205 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2206 if (merge_ent &&
2207 !head_ent->non_unique_new_dir &&
2208 !merge_ent->non_unique_new_dir &&
2209 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2210 /* 1. Renamed identically; remove it from both sides */
2211 string_list_append(&remove_from_head,
2212 head_ent->dir)->util = head_ent;
2213 strbuf_release(&head_ent->new_dir);
2214 string_list_append(&remove_from_merge,
2215 merge_ent->dir)->util = merge_ent;
2216 strbuf_release(&merge_ent->new_dir);
2217 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2218 /* 2. This wasn't a directory rename after all */
2219 string_list_append(&remove_from_head,
2220 head_ent->dir)->util = head_ent;
2221 strbuf_release(&head_ent->new_dir);
2222 }
2223 }
2224
2225 remove_hashmap_entries(dir_re_head, &remove_from_head);
2226 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2227
2228 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2229 ent /* member name */) {
2230 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2231 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2232 /* 2. This wasn't a directory rename after all */
2233 string_list_append(&remove_from_merge,
2234 merge_ent->dir)->util = merge_ent;
2235 } else if (head_ent &&
2236 !head_ent->non_unique_new_dir &&
2237 !merge_ent->non_unique_new_dir) {
2238 /* 3. rename/rename(1to2) */
2239 /*
2240 * We can assume it's not rename/rename(1to1) because
2241 * that was case (1), already checked above. So we
2242 * know that head_ent->new_dir and merge_ent->new_dir
2243 * are different strings.
2244 */
2245 output(opt, 1, _("CONFLICT (rename/rename): "
2246 "Rename directory %s->%s in %s. "
2247 "Rename directory %s->%s in %s"),
2248 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2249 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2250 string_list_append(&remove_from_head,
2251 head_ent->dir)->util = head_ent;
2252 strbuf_release(&head_ent->new_dir);
2253 string_list_append(&remove_from_merge,
2254 merge_ent->dir)->util = merge_ent;
2255 strbuf_release(&merge_ent->new_dir);
2256 }
2257 }
2258
2259 remove_hashmap_entries(dir_re_head, &remove_from_head);
2260 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2261 }
2262
2263 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2264 {
2265 struct hashmap *dir_renames;
2266 struct hashmap_iter iter;
2267 struct dir_rename_entry *entry;
2268 int i;
2269
2270 /*
2271 * Typically, we think of a directory rename as all files from a
2272 * certain directory being moved to a target directory. However,
2273 * what if someone first moved two files from the original
2274 * directory in one commit, and then renamed the directory
2275 * somewhere else in a later commit? At merge time, we just know
2276 * that files from the original directory went to two different
2277 * places, and that the bulk of them ended up in the same place.
2278 * We want each directory rename to represent where the bulk of the
2279 * files from that directory end up; this function exists to find
2280 * where the bulk of the files went.
2281 *
2282 * The first loop below simply iterates through the list of file
2283 * renames, finding out how often each directory rename pair
2284 * possibility occurs.
2285 */
2286 dir_renames = xmalloc(sizeof(*dir_renames));
2287 dir_rename_init(dir_renames);
2288 for (i = 0; i < pairs->nr; ++i) {
2289 struct string_list_item *item;
2290 int *count;
2291 struct diff_filepair *pair = pairs->queue[i];
2292 char *old_dir, *new_dir;
2293
2294 /* File not part of directory rename if it wasn't renamed */
2295 if (pair->status != 'R')
2296 continue;
2297
2298 get_renamed_dir_portion(pair->one->path, pair->two->path,
2299 &old_dir, &new_dir);
2300 if (!old_dir)
2301 /* Directory didn't change at all; ignore this one. */
2302 continue;
2303
2304 entry = dir_rename_find_entry(dir_renames, old_dir);
2305 if (!entry) {
2306 entry = xmalloc(sizeof(*entry));
2307 dir_rename_entry_init(entry, old_dir);
2308 hashmap_put(dir_renames, &entry->ent);
2309 } else {
2310 free(old_dir);
2311 }
2312 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2313 if (!item) {
2314 item = string_list_insert(&entry->possible_new_dirs,
2315 new_dir);
2316 item->util = xcalloc(1, sizeof(int));
2317 } else {
2318 free(new_dir);
2319 }
2320 count = item->util;
2321 *count += 1;
2322 }
2323
2324 /*
2325 * For each directory with files moved out of it, we find out which
2326 * target directory received the most files so we can declare it to
2327 * be the "winning" target location for the directory rename. This
2328 * winner gets recorded in new_dir. If there is no winner
2329 * (multiple target directories received the same number of files),
2330 * we set non_unique_new_dir. Once we've determined the winner (or
2331 * that there is no winner), we no longer need possible_new_dirs.
2332 */
2333 hashmap_for_each_entry(dir_renames, &iter, entry,
2334 ent /* member name */) {
2335 int max = 0;
2336 int bad_max = 0;
2337 char *best = NULL;
2338
2339 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2340 int *count = entry->possible_new_dirs.items[i].util;
2341
2342 if (*count == max)
2343 bad_max = max;
2344 else if (*count > max) {
2345 max = *count;
2346 best = entry->possible_new_dirs.items[i].string;
2347 }
2348 }
2349 if (bad_max == max)
2350 entry->non_unique_new_dir = 1;
2351 else {
2352 assert(entry->new_dir.len == 0);
2353 strbuf_addstr(&entry->new_dir, best);
2354 }
2355 /*
2356 * The relevant directory sub-portion of the original full
2357 * filepaths were xstrndup'ed before inserting into
2358 * possible_new_dirs, and instead of manually iterating the
2359 * list and free'ing each, just lie and tell
2360 * possible_new_dirs that it did the strdup'ing so that it
2361 * will free them for us.
2362 */
2363 entry->possible_new_dirs.strdup_strings = 1;
2364 string_list_clear(&entry->possible_new_dirs, 1);
2365 }
2366
2367 return dir_renames;
2368 }
2369
2370 static struct dir_rename_entry *check_dir_renamed(const char *path,
2371 struct hashmap *dir_renames)
2372 {
2373 char *temp = xstrdup(path);
2374 char *end;
2375 struct dir_rename_entry *entry = NULL;
2376
2377 while ((end = strrchr(temp, '/'))) {
2378 *end = '\0';
2379 entry = dir_rename_find_entry(dir_renames, temp);
2380 if (entry)
2381 break;
2382 }
2383 free(temp);
2384 return entry;
2385 }
2386
2387 static void compute_collisions(struct hashmap *collisions,
2388 struct hashmap *dir_renames,
2389 struct diff_queue_struct *pairs)
2390 {
2391 int i;
2392
2393 /*
2394 * Multiple files can be mapped to the same path due to directory
2395 * renames done by the other side of history. Since that other
2396 * side of history could have merged multiple directories into one,
2397 * if our side of history added the same file basename to each of
2398 * those directories, then all N of them would get implicitly
2399 * renamed by the directory rename detection into the same path,
2400 * and we'd get an add/add/.../add conflict, and all those adds
2401 * from *this* side of history. This is not representable in the
2402 * index, and users aren't going to easily be able to make sense of
2403 * it. So we need to provide a good warning about what's
2404 * happening, and fall back to no-directory-rename detection
2405 * behavior for those paths.
2406 *
2407 * See testcases 9e and all of section 5 from t6043 for examples.
2408 */
2409 collision_init(collisions);
2410
2411 for (i = 0; i < pairs->nr; ++i) {
2412 struct dir_rename_entry *dir_rename_ent;
2413 struct collision_entry *collision_ent;
2414 char *new_path;
2415 struct diff_filepair *pair = pairs->queue[i];
2416
2417 if (pair->status != 'A' && pair->status != 'R')
2418 continue;
2419 dir_rename_ent = check_dir_renamed(pair->two->path,
2420 dir_renames);
2421 if (!dir_rename_ent)
2422 continue;
2423
2424 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2425 if (!new_path)
2426 /*
2427 * dir_rename_ent->non_unique_new_path is true, which
2428 * means there is no directory rename for us to use,
2429 * which means it won't cause us any additional
2430 * collisions.
2431 */
2432 continue;
2433 collision_ent = collision_find_entry(collisions, new_path);
2434 if (!collision_ent) {
2435 collision_ent = xcalloc(1,
2436 sizeof(struct collision_entry));
2437 hashmap_entry_init(&collision_ent->ent,
2438 strhash(new_path));
2439 hashmap_put(collisions, &collision_ent->ent);
2440 collision_ent->target_file = new_path;
2441 } else {
2442 free(new_path);
2443 }
2444 string_list_insert(&collision_ent->source_files,
2445 pair->two->path);
2446 }
2447 }
2448
2449 static char *check_for_directory_rename(struct merge_options *opt,
2450 const char *path,
2451 struct tree *tree,
2452 struct hashmap *dir_renames,
2453 struct hashmap *dir_rename_exclusions,
2454 struct hashmap *collisions,
2455 int *clean_merge)
2456 {
2457 char *new_path = NULL;
2458 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2459 struct dir_rename_entry *oentry = NULL;
2460
2461 if (!entry)
2462 return new_path;
2463
2464 /*
2465 * This next part is a little weird. We do not want to do an
2466 * implicit rename into a directory we renamed on our side, because
2467 * that will result in a spurious rename/rename(1to2) conflict. An
2468 * example:
2469 * Base commit: dumbdir/afile, otherdir/bfile
2470 * Side 1: smrtdir/afile, otherdir/bfile
2471 * Side 2: dumbdir/afile, dumbdir/bfile
2472 * Here, while working on Side 1, we could notice that otherdir was
2473 * renamed/merged to dumbdir, and change the diff_filepair for
2474 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2475 * 2 will notice the rename from dumbdir to smrtdir, and do the
2476 * transitive rename to move it from dumbdir/bfile to
2477 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2478 * smrtdir, a rename/rename(1to2) conflict. We really just want
2479 * the file to end up in smrtdir. And the way to achieve that is
2480 * to not let Side1 do the rename to dumbdir, since we know that is
2481 * the source of one of our directory renames.
2482 *
2483 * That's why oentry and dir_rename_exclusions is here.
2484 *
2485 * As it turns out, this also prevents N-way transient rename
2486 * confusion; See testcases 9c and 9d of t6043.
2487 */
2488 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2489 if (oentry) {
2490 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2491 "to %s, because %s itself was renamed."),
2492 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2493 } else {
2494 new_path = handle_path_level_conflicts(opt, path, entry,
2495 collisions, tree);
2496 *clean_merge &= (new_path != NULL);
2497 }
2498
2499 return new_path;
2500 }
2501
2502 static void apply_directory_rename_modifications(struct merge_options *opt,
2503 struct diff_filepair *pair,
2504 char *new_path,
2505 struct rename *re,
2506 struct tree *tree,
2507 struct tree *o_tree,
2508 struct tree *a_tree,
2509 struct tree *b_tree,
2510 struct string_list *entries)
2511 {
2512 struct string_list_item *item;
2513 int stage = (tree == a_tree ? 2 : 3);
2514 int update_wd;
2515
2516 /*
2517 * In all cases where we can do directory rename detection,
2518 * unpack_trees() will have read pair->two->path into the
2519 * index and the working copy. We need to remove it so that
2520 * we can instead place it at new_path. It is guaranteed to
2521 * not be untracked (unpack_trees() would have errored out
2522 * saying the file would have been overwritten), but it might
2523 * be dirty, though.
2524 */
2525 update_wd = !was_dirty(opt, pair->two->path);
2526 if (!update_wd)
2527 output(opt, 1, _("Refusing to lose dirty file at %s"),
2528 pair->two->path);
2529 remove_file(opt, 1, pair->two->path, !update_wd);
2530
2531 /* Find or create a new re->dst_entry */
2532 item = string_list_lookup(entries, new_path);
2533 if (item) {
2534 /*
2535 * Since we're renaming on this side of history, and it's
2536 * due to a directory rename on the other side of history
2537 * (which we only allow when the directory in question no
2538 * longer exists on the other side of history), the
2539 * original entry for re->dst_entry is no longer
2540 * necessary...
2541 */
2542 re->dst_entry->processed = 1;
2543
2544 /*
2545 * ...because we'll be using this new one.
2546 */
2547 re->dst_entry = item->util;
2548 } else {
2549 /*
2550 * re->dst_entry is for the before-dir-rename path, and we
2551 * need it to hold information for the after-dir-rename
2552 * path. Before creating a new entry, we need to mark the
2553 * old one as unnecessary (...unless it is shared by
2554 * src_entry, i.e. this didn't use to be a rename, in which
2555 * case we can just allow the normal processing to happen
2556 * for it).
2557 */
2558 if (pair->status == 'R')
2559 re->dst_entry->processed = 1;
2560
2561 re->dst_entry = insert_stage_data(opt->repo, new_path,
2562 o_tree, a_tree, b_tree,
2563 entries);
2564 item = string_list_insert(entries, new_path);
2565 item->util = re->dst_entry;
2566 }
2567
2568 /*
2569 * Update the stage_data with the information about the path we are
2570 * moving into place. That slot will be empty and available for us
2571 * to write to because of the collision checks in
2572 * handle_path_level_conflicts(). In other words,
2573 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2574 * open for us to write to.
2575 *
2576 * It may be tempting to actually update the index at this point as
2577 * well, using update_stages_for_stage_data(), but as per the big
2578 * "NOTE" in update_stages(), doing so will modify the current
2579 * in-memory index which will break calls to would_lose_untracked()
2580 * that we need to make. Instead, we need to just make sure that
2581 * the various handle_rename_*() functions update the index
2582 * explicitly rather than relying on unpack_trees() to have done it.
2583 */
2584 get_tree_entry(opt->repo,
2585 &tree->object.oid,
2586 pair->two->path,
2587 &re->dst_entry->stages[stage].oid,
2588 &re->dst_entry->stages[stage].mode);
2589
2590 /*
2591 * Record the original change status (or 'type' of change). If it
2592 * was originally an add ('A'), this lets us differentiate later
2593 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2594 * otherwise look the same). If it was originally a rename ('R'),
2595 * this lets us remember and report accurately about the transitive
2596 * renaming that occurred via the directory rename detection. Also,
2597 * record the original destination name.
2598 */
2599 re->dir_rename_original_type = pair->status;
2600 re->dir_rename_original_dest = pair->two->path;
2601
2602 /*
2603 * We don't actually look at pair->status again, but it seems
2604 * pedagogically correct to adjust it.
2605 */
2606 pair->status = 'R';
2607
2608 /*
2609 * Finally, record the new location.
2610 */
2611 pair->two->path = new_path;
2612 }
2613
2614 /*
2615 * Get information of all renames which occurred in 'pairs', making use of
2616 * any implicit directory renames inferred from the other side of history.
2617 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2618 * to be able to associate the correct cache entries with the rename
2619 * information; tree is always equal to either a_tree or b_tree.
2620 */
2621 static struct string_list *get_renames(struct merge_options *opt,
2622 const char *branch,
2623 struct diff_queue_struct *pairs,
2624 struct hashmap *dir_renames,
2625 struct hashmap *dir_rename_exclusions,
2626 struct tree *tree,
2627 struct tree *o_tree,
2628 struct tree *a_tree,
2629 struct tree *b_tree,
2630 struct string_list *entries,
2631 int *clean_merge)
2632 {
2633 int i;
2634 struct hashmap collisions;
2635 struct hashmap_iter iter;
2636 struct collision_entry *e;
2637 struct string_list *renames;
2638
2639 compute_collisions(&collisions, dir_renames, pairs);
2640 renames = xcalloc(1, sizeof(struct string_list));
2641
2642 for (i = 0; i < pairs->nr; ++i) {
2643 struct string_list_item *item;
2644 struct rename *re;
2645 struct diff_filepair *pair = pairs->queue[i];
2646 char *new_path; /* non-NULL only with directory renames */
2647
2648 if (pair->status != 'A' && pair->status != 'R') {
2649 diff_free_filepair(pair);
2650 continue;
2651 }
2652 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2653 dir_renames,
2654 dir_rename_exclusions,
2655 &collisions,
2656 clean_merge);
2657 if (pair->status != 'R' && !new_path) {
2658 diff_free_filepair(pair);
2659 continue;
2660 }
2661
2662 re = xmalloc(sizeof(*re));
2663 re->processed = 0;
2664 re->pair = pair;
2665 re->branch = branch;
2666 re->dir_rename_original_type = '\0';
2667 re->dir_rename_original_dest = NULL;
2668 item = string_list_lookup(entries, re->pair->one->path);
2669 if (!item)
2670 re->src_entry = insert_stage_data(opt->repo,
2671 re->pair->one->path,
2672 o_tree, a_tree, b_tree, entries);
2673 else
2674 re->src_entry = item->util;
2675
2676 item = string_list_lookup(entries, re->pair->two->path);
2677 if (!item)
2678 re->dst_entry = insert_stage_data(opt->repo,
2679 re->pair->two->path,
2680 o_tree, a_tree, b_tree, entries);
2681 else
2682 re->dst_entry = item->util;
2683 item = string_list_insert(renames, pair->one->path);
2684 item->util = re;
2685 if (new_path)
2686 apply_directory_rename_modifications(opt, pair, new_path,
2687 re, tree, o_tree,
2688 a_tree, b_tree,
2689 entries);
2690 }
2691
2692 hashmap_for_each_entry(&collisions, &iter, e,
2693 ent /* member name */) {
2694 free(e->target_file);
2695 string_list_clear(&e->source_files, 0);
2696 }
2697 hashmap_free_entries(&collisions, struct collision_entry, ent);
2698 return renames;
2699 }
2700
2701 static int process_renames(struct merge_options *opt,
2702 struct string_list *a_renames,
2703 struct string_list *b_renames)
2704 {
2705 int clean_merge = 1, i, j;
2706 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2707 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2708 const struct rename *sre;
2709
2710 for (i = 0; i < a_renames->nr; i++) {
2711 sre = a_renames->items[i].util;
2712 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2713 = (void *)sre;
2714 }
2715 for (i = 0; i < b_renames->nr; i++) {
2716 sre = b_renames->items[i].util;
2717 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2718 = (void *)sre;
2719 }
2720
2721 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2722 struct string_list *renames1, *renames2Dst;
2723 struct rename *ren1 = NULL, *ren2 = NULL;
2724 const char *ren1_src, *ren1_dst;
2725 struct string_list_item *lookup;
2726
2727 if (i >= a_renames->nr) {
2728 ren2 = b_renames->items[j++].util;
2729 } else if (j >= b_renames->nr) {
2730 ren1 = a_renames->items[i++].util;
2731 } else {
2732 int compare = strcmp(a_renames->items[i].string,
2733 b_renames->items[j].string);
2734 if (compare <= 0)
2735 ren1 = a_renames->items[i++].util;
2736 if (compare >= 0)
2737 ren2 = b_renames->items[j++].util;
2738 }
2739
2740 /* TODO: refactor, so that 1/2 are not needed */
2741 if (ren1) {
2742 renames1 = a_renames;
2743 renames2Dst = &b_by_dst;
2744 } else {
2745 renames1 = b_renames;
2746 renames2Dst = &a_by_dst;
2747 SWAP(ren2, ren1);
2748 }
2749
2750 if (ren1->processed)
2751 continue;
2752 ren1->processed = 1;
2753 ren1->dst_entry->processed = 1;
2754 /* BUG: We should only mark src_entry as processed if we
2755 * are not dealing with a rename + add-source case.
2756 */
2757 ren1->src_entry->processed = 1;
2758
2759 ren1_src = ren1->pair->one->path;
2760 ren1_dst = ren1->pair->two->path;
2761
2762 if (ren2) {
2763 /* One file renamed on both sides */
2764 const char *ren2_src = ren2->pair->one->path;
2765 const char *ren2_dst = ren2->pair->two->path;
2766 enum rename_type rename_type;
2767 if (strcmp(ren1_src, ren2_src) != 0)
2768 BUG("ren1_src != ren2_src");
2769 ren2->dst_entry->processed = 1;
2770 ren2->processed = 1;
2771 if (strcmp(ren1_dst, ren2_dst) != 0) {
2772 rename_type = RENAME_ONE_FILE_TO_TWO;
2773 clean_merge = 0;
2774 } else {
2775 rename_type = RENAME_ONE_FILE_TO_ONE;
2776 /* BUG: We should only remove ren1_src in
2777 * the base stage (think of rename +
2778 * add-source cases).
2779 */
2780 remove_file(opt, 1, ren1_src, 1);
2781 update_entry(ren1->dst_entry,
2782 ren1->pair->one,
2783 ren1->pair->two,
2784 ren2->pair->two);
2785 }
2786 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2787 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2788 /* Two different files renamed to the same thing */
2789 char *ren2_dst;
2790 ren2 = lookup->util;
2791 ren2_dst = ren2->pair->two->path;
2792 if (strcmp(ren1_dst, ren2_dst) != 0)
2793 BUG("ren1_dst != ren2_dst");
2794
2795 clean_merge = 0;
2796 ren2->processed = 1;
2797 /*
2798 * BUG: We should only mark src_entry as processed
2799 * if we are not dealing with a rename + add-source
2800 * case.
2801 */
2802 ren2->src_entry->processed = 1;
2803
2804 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2805 opt, ren1, ren2);
2806 } else {
2807 /* Renamed in 1, maybe changed in 2 */
2808 /* we only use sha1 and mode of these */
2809 struct diff_filespec src_other, dst_other;
2810 int try_merge;
2811
2812 /*
2813 * unpack_trees loads entries from common-commit
2814 * into stage 1, from head-commit into stage 2, and
2815 * from merge-commit into stage 3. We keep track
2816 * of which side corresponds to the rename.
2817 */
2818 int renamed_stage = a_renames == renames1 ? 2 : 3;
2819 int other_stage = a_renames == renames1 ? 3 : 2;
2820
2821 /* BUG: We should only remove ren1_src in the base
2822 * stage and in other_stage (think of rename +
2823 * add-source case).
2824 */
2825 remove_file(opt, 1, ren1_src,
2826 renamed_stage == 2 || !was_tracked(opt, ren1_src));
2827
2828 oidcpy(&src_other.oid,
2829 &ren1->src_entry->stages[other_stage].oid);
2830 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2831 oidcpy(&dst_other.oid,
2832 &ren1->dst_entry->stages[other_stage].oid);
2833 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2834 try_merge = 0;
2835
2836 if (oideq(&src_other.oid, &null_oid) &&
2837 ren1->dir_rename_original_type == 'A') {
2838 setup_rename_conflict_info(RENAME_VIA_DIR,
2839 opt, ren1, NULL);
2840 } else if (oideq(&src_other.oid, &null_oid)) {
2841 setup_rename_conflict_info(RENAME_DELETE,
2842 opt, ren1, NULL);
2843 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2844 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2845 /*
2846 * Added file on the other side identical to
2847 * the file being renamed: clean merge.
2848 * Also, there is no need to overwrite the
2849 * file already in the working copy, so call
2850 * update_file_flags() instead of
2851 * update_file().
2852 */
2853 if (update_file_flags(opt,
2854 ren1->pair->two,
2855 ren1_dst,
2856 1, /* update_cache */
2857 0 /* update_wd */))
2858 clean_merge = -1;
2859 } else if (!oideq(&dst_other.oid, &null_oid)) {
2860 /*
2861 * Probably not a clean merge, but it's
2862 * premature to set clean_merge to 0 here,
2863 * because if the rename merges cleanly and
2864 * the merge exactly matches the newly added
2865 * file, then the merge will be clean.
2866 */
2867 setup_rename_conflict_info(RENAME_ADD,
2868 opt, ren1, NULL);
2869 } else
2870 try_merge = 1;
2871
2872 if (clean_merge < 0)
2873 goto cleanup_and_return;
2874 if (try_merge) {
2875 struct diff_filespec *o, *a, *b;
2876 src_other.path = (char *)ren1_src;
2877
2878 o = ren1->pair->one;
2879 if (a_renames == renames1) {
2880 a = ren1->pair->two;
2881 b = &src_other;
2882 } else {
2883 b = ren1->pair->two;
2884 a = &src_other;
2885 }
2886 update_entry(ren1->dst_entry, o, a, b);
2887 setup_rename_conflict_info(RENAME_NORMAL,
2888 opt, ren1, NULL);
2889 }
2890 }
2891 }
2892 cleanup_and_return:
2893 string_list_clear(&a_by_dst, 0);
2894 string_list_clear(&b_by_dst, 0);
2895
2896 return clean_merge;
2897 }
2898
2899 struct rename_info {
2900 struct string_list *head_renames;
2901 struct string_list *merge_renames;
2902 };
2903
2904 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2905 struct hashmap *dir_renames)
2906 {
2907 struct hashmap_iter iter;
2908 struct dir_rename_entry *e;
2909
2910 hashmap_for_each_entry(dir_renames, &iter, e,
2911 ent /* member name */) {
2912 free(e->dir);
2913 strbuf_release(&e->new_dir);
2914 /* possible_new_dirs already cleared in get_directory_renames */
2915 }
2916 hashmap_free_entries(dir_renames, struct dir_rename_entry, ent);
2917 free(dir_renames);
2918
2919 free(pairs->queue);
2920 free(pairs);
2921 }
2922
2923 static int detect_and_process_renames(struct merge_options *opt,
2924 struct tree *common,
2925 struct tree *head,
2926 struct tree *merge,
2927 struct string_list *entries,
2928 struct rename_info *ri)
2929 {
2930 struct diff_queue_struct *head_pairs, *merge_pairs;
2931 struct hashmap *dir_re_head, *dir_re_merge;
2932 int clean = 1;
2933
2934 ri->head_renames = NULL;
2935 ri->merge_renames = NULL;
2936
2937 if (!merge_detect_rename(opt))
2938 return 1;
2939
2940 head_pairs = get_diffpairs(opt, common, head);
2941 merge_pairs = get_diffpairs(opt, common, merge);
2942
2943 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2944 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2945 !opt->priv->call_depth)) {
2946 dir_re_head = get_directory_renames(head_pairs);
2947 dir_re_merge = get_directory_renames(merge_pairs);
2948
2949 handle_directory_level_conflicts(opt,
2950 dir_re_head, head,
2951 dir_re_merge, merge);
2952 } else {
2953 dir_re_head = xmalloc(sizeof(*dir_re_head));
2954 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2955 dir_rename_init(dir_re_head);
2956 dir_rename_init(dir_re_merge);
2957 }
2958
2959 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2960 dir_re_merge, dir_re_head, head,
2961 common, head, merge, entries,
2962 &clean);
2963 if (clean < 0)
2964 goto cleanup;
2965 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2966 dir_re_head, dir_re_merge, merge,
2967 common, head, merge, entries,
2968 &clean);
2969 if (clean < 0)
2970 goto cleanup;
2971 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2972
2973 cleanup:
2974 /*
2975 * Some cleanup is deferred until cleanup_renames() because the
2976 * data structures are still needed and referenced in
2977 * process_entry(). But there are a few things we can free now.
2978 */
2979 initial_cleanup_rename(head_pairs, dir_re_head);
2980 initial_cleanup_rename(merge_pairs, dir_re_merge);
2981
2982 return clean;
2983 }
2984
2985 static void final_cleanup_rename(struct string_list *rename)
2986 {
2987 const struct rename *re;
2988 int i;
2989
2990 if (rename == NULL)
2991 return;
2992
2993 for (i = 0; i < rename->nr; i++) {
2994 re = rename->items[i].util;
2995 diff_free_filepair(re->pair);
2996 }
2997 string_list_clear(rename, 1);
2998 free(rename);
2999 }
3000
3001 static void final_cleanup_renames(struct rename_info *re_info)
3002 {
3003 final_cleanup_rename(re_info->head_renames);
3004 final_cleanup_rename(re_info->merge_renames);
3005 }
3006
3007 static int read_oid_strbuf(struct merge_options *opt,
3008 const struct object_id *oid,
3009 struct strbuf *dst)
3010 {
3011 void *buf;
3012 enum object_type type;
3013 unsigned long size;
3014 buf = read_object_file(oid, &type, &size);
3015 if (!buf)
3016 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
3017 if (type != OBJ_BLOB) {
3018 free(buf);
3019 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
3020 }
3021 strbuf_attach(dst, buf, size, size + 1);
3022 return 0;
3023 }
3024
3025 static int blob_unchanged(struct merge_options *opt,
3026 const struct diff_filespec *o,
3027 const struct diff_filespec *a,
3028 int renormalize, const char *path)
3029 {
3030 struct strbuf obuf = STRBUF_INIT;
3031 struct strbuf abuf = STRBUF_INIT;
3032 int ret = 0; /* assume changed for safety */
3033 const struct index_state *idx = opt->repo->index;
3034
3035 if (a->mode != o->mode)
3036 return 0;
3037 if (oideq(&o->oid, &a->oid))
3038 return 1;
3039 if (!renormalize)
3040 return 0;
3041
3042 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
3043 read_oid_strbuf(opt, &a->oid, &abuf))
3044 goto error_return;
3045 /*
3046 * Note: binary | is used so that both renormalizations are
3047 * performed. Comparison can be skipped if both files are
3048 * unchanged since their sha1s have already been compared.
3049 */
3050 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3051 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3052 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3053
3054 error_return:
3055 strbuf_release(&obuf);
3056 strbuf_release(&abuf);
3057 return ret;
3058 }
3059
3060 static int handle_modify_delete(struct merge_options *opt,
3061 const char *path,
3062 const struct diff_filespec *o,
3063 const struct diff_filespec *a,
3064 const struct diff_filespec *b)
3065 {
3066 const char *modify_branch, *delete_branch;
3067 const struct diff_filespec *changed;
3068
3069 if (is_valid(a)) {
3070 modify_branch = opt->branch1;
3071 delete_branch = opt->branch2;
3072 changed = a;
3073 } else {
3074 modify_branch = opt->branch2;
3075 delete_branch = opt->branch1;
3076 changed = b;
3077 }
3078
3079 return handle_change_delete(opt,
3080 path, NULL,
3081 o, changed,
3082 modify_branch, delete_branch,
3083 _("modify"), _("modified"));
3084 }
3085
3086 static int handle_content_merge(struct merge_file_info *mfi,
3087 struct merge_options *opt,
3088 const char *path,
3089 int is_dirty,
3090 const struct diff_filespec *o,
3091 const struct diff_filespec *a,
3092 const struct diff_filespec *b,
3093 struct rename_conflict_info *ci)
3094 {
3095 const char *reason = _("content");
3096 unsigned df_conflict_remains = 0;
3097
3098 if (!is_valid(o))
3099 reason = _("add/add");
3100
3101 assert(o->path && a->path && b->path);
3102 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3103 S_ISGITLINK(ci->ren1->pair->two->mode)))
3104 df_conflict_remains = 1;
3105
3106 if (merge_mode_and_contents(opt, o, a, b, path,
3107 opt->branch1, opt->branch2,
3108 opt->priv->call_depth * 2, mfi))
3109 return -1;
3110
3111 /*
3112 * We can skip updating the working tree file iff:
3113 * a) The merge is clean
3114 * b) The merge matches what was in HEAD (content, mode, pathname)
3115 * c) The target path is usable (i.e. not involved in D/F conflict)
3116 */
3117 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3118 !df_conflict_remains) {
3119 int pos;
3120 struct cache_entry *ce;
3121
3122 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3123 if (add_cacheinfo(opt, &mfi->blob, path,
3124 0, (!opt->priv->call_depth && !is_dirty), 0))
3125 return -1;
3126 /*
3127 * However, add_cacheinfo() will delete the old cache entry
3128 * and add a new one. We need to copy over any skip_worktree
3129 * flag to avoid making the file appear as if it were
3130 * deleted by the user.
3131 */
3132 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3133 ce = opt->priv->orig_index.cache[pos];
3134 if (ce_skip_worktree(ce)) {
3135 pos = index_name_pos(opt->repo->index, path, strlen(path));
3136 ce = opt->repo->index->cache[pos];
3137 ce->ce_flags |= CE_SKIP_WORKTREE;
3138 }
3139 return mfi->clean;
3140 }
3141
3142 if (!mfi->clean) {
3143 if (S_ISGITLINK(mfi->blob.mode))
3144 reason = _("submodule");
3145 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3146 reason, path);
3147 if (ci && !df_conflict_remains)
3148 if (update_stages(opt, path, o, a, b))
3149 return -1;
3150 }
3151
3152 if (df_conflict_remains || is_dirty) {
3153 char *new_path;
3154 if (opt->priv->call_depth) {
3155 remove_file_from_index(opt->repo->index, path);
3156 } else {
3157 if (!mfi->clean) {
3158 if (update_stages(opt, path, o, a, b))
3159 return -1;
3160 } else {
3161 int file_from_stage2 = was_tracked(opt, path);
3162
3163 if (update_stages(opt, path, NULL,
3164 file_from_stage2 ? &mfi->blob : NULL,
3165 file_from_stage2 ? NULL : &mfi->blob))
3166 return -1;
3167 }
3168
3169 }
3170 new_path = unique_path(opt, path, ci->ren1->branch);
3171 if (is_dirty) {
3172 output(opt, 1, _("Refusing to lose dirty file at %s"),
3173 path);
3174 }
3175 output(opt, 1, _("Adding as %s instead"), new_path);
3176 if (update_file(opt, 0, &mfi->blob, new_path)) {
3177 free(new_path);
3178 return -1;
3179 }
3180 free(new_path);
3181 mfi->clean = 0;
3182 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3183 return -1;
3184 return !is_dirty && mfi->clean;
3185 }
3186
3187 static int handle_rename_normal(struct merge_options *opt,
3188 const char *path,
3189 const struct diff_filespec *o,
3190 const struct diff_filespec *a,
3191 const struct diff_filespec *b,
3192 struct rename_conflict_info *ci)
3193 {
3194 struct rename *ren = ci->ren1;
3195 struct merge_file_info mfi;
3196 int clean;
3197 int side = (ren->branch == opt->branch1 ? 2 : 3);
3198
3199 /* Merge the content and write it out */
3200 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3201 o, a, b, ci);
3202
3203 if (clean &&
3204 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3205 ren->dir_rename_original_dest) {
3206 if (update_stages(opt, path,
3207 NULL,
3208 side == 2 ? &mfi.blob : NULL,
3209 side == 2 ? NULL : &mfi.blob))
3210 return -1;
3211 clean = 0; /* not clean, but conflicted */
3212 }
3213 return clean;
3214 }
3215
3216 static void dir_rename_warning(const char *msg,
3217 int is_add,
3218 int clean,
3219 struct merge_options *opt,
3220 struct rename *ren)
3221 {
3222 const char *other_branch;
3223 other_branch = (ren->branch == opt->branch1 ?
3224 opt->branch2 : opt->branch1);
3225 if (is_add) {
3226 output(opt, clean ? 2 : 1, msg,
3227 ren->pair->one->path, ren->branch,
3228 other_branch, ren->pair->two->path);
3229 return;
3230 }
3231 output(opt, clean ? 2 : 1, msg,
3232 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3233 other_branch, ren->pair->two->path);
3234 }
3235 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3236 struct rename *ren)
3237 {
3238 const char *msg;
3239 int clean = 1, is_add;
3240
3241 if (!ren)
3242 return clean;
3243
3244 /* Return early if ren was not affected/created by a directory rename */
3245 if (!ren->dir_rename_original_dest)
3246 return clean;
3247
3248 /* Sanity checks */
3249 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3250 assert(ren->dir_rename_original_type == 'A' ||
3251 ren->dir_rename_original_type == 'R');
3252
3253 /* Check whether to treat directory renames as a conflict */
3254 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3255
3256 is_add = (ren->dir_rename_original_type == 'A');
3257 if (ren->dir_rename_original_type == 'A' && clean) {
3258 msg = _("Path updated: %s added in %s inside a "
3259 "directory that was renamed in %s; moving it to %s.");
3260 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3261 msg = _("CONFLICT (file location): %s added in %s "
3262 "inside a directory that was renamed in %s, "
3263 "suggesting it should perhaps be moved to %s.");
3264 } else if (ren->dir_rename_original_type == 'R' && clean) {
3265 msg = _("Path updated: %s renamed to %s in %s, inside a "
3266 "directory that was renamed in %s; moving it to %s.");
3267 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3268 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3269 "inside a directory that was renamed in %s, "
3270 "suggesting it should perhaps be moved to %s.");
3271 } else {
3272 BUG("Impossible dir_rename_original_type/clean combination");
3273 }
3274 dir_rename_warning(msg, is_add, clean, opt, ren);
3275
3276 return clean;
3277 }
3278
3279 /* Per entry merge function */
3280 static int process_entry(struct merge_options *opt,
3281 const char *path, struct stage_data *entry)
3282 {
3283 int clean_merge = 1;
3284 int normalize = opt->renormalize;
3285
3286 struct diff_filespec *o = &entry->stages[1];
3287 struct diff_filespec *a = &entry->stages[2];
3288 struct diff_filespec *b = &entry->stages[3];
3289 int o_valid = is_valid(o);
3290 int a_valid = is_valid(a);
3291 int b_valid = is_valid(b);
3292 o->path = a->path = b->path = (char*)path;
3293
3294 entry->processed = 1;
3295 if (entry->rename_conflict_info) {
3296 struct rename_conflict_info *ci = entry->rename_conflict_info;
3297 struct diff_filespec *temp;
3298 int path_clean;
3299
3300 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3301 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3302
3303 /*
3304 * For cases with a single rename, {o,a,b}->path have all been
3305 * set to the rename target path; we need to set two of these
3306 * back to the rename source.
3307 * For rename/rename conflicts, we'll manually fix paths below.
3308 */
3309 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3310 o->path = temp->path = ci->ren1->pair->one->path;
3311 if (ci->ren2) {
3312 assert(opt->branch1 == ci->ren1->branch);
3313 }
3314
3315 switch (ci->rename_type) {
3316 case RENAME_NORMAL:
3317 case RENAME_ONE_FILE_TO_ONE:
3318 clean_merge = handle_rename_normal(opt, path, o, a, b,
3319 ci);
3320 break;
3321 case RENAME_VIA_DIR:
3322 clean_merge = handle_rename_via_dir(opt, ci);
3323 break;
3324 case RENAME_ADD:
3325 /*
3326 * Probably unclean merge, but if the renamed file
3327 * merges cleanly and the result can then be
3328 * two-way merged cleanly with the added file, I
3329 * guess it's a clean merge?
3330 */
3331 clean_merge = handle_rename_add(opt, ci);
3332 break;
3333 case RENAME_DELETE:
3334 clean_merge = 0;
3335 if (handle_rename_delete(opt, ci))
3336 clean_merge = -1;
3337 break;
3338 case RENAME_ONE_FILE_TO_TWO:
3339 /*
3340 * Manually fix up paths; note:
3341 * ren[12]->pair->one->path are equal.
3342 */
3343 o->path = ci->ren1->pair->one->path;
3344 a->path = ci->ren1->pair->two->path;
3345 b->path = ci->ren2->pair->two->path;
3346
3347 clean_merge = 0;
3348 if (handle_rename_rename_1to2(opt, ci))
3349 clean_merge = -1;
3350 break;
3351 case RENAME_TWO_FILES_TO_ONE:
3352 /*
3353 * Manually fix up paths; note,
3354 * ren[12]->pair->two->path are actually equal.
3355 */
3356 o->path = NULL;
3357 a->path = ci->ren1->pair->two->path;
3358 b->path = ci->ren2->pair->two->path;
3359
3360 /*
3361 * Probably unclean merge, but if the two renamed
3362 * files merge cleanly and the two resulting files
3363 * can then be two-way merged cleanly, I guess it's
3364 * a clean merge?
3365 */
3366 clean_merge = handle_rename_rename_2to1(opt, ci);
3367 break;
3368 default:
3369 entry->processed = 0;
3370 break;
3371 }
3372 if (path_clean < clean_merge)
3373 clean_merge = path_clean;
3374 } else if (o_valid && (!a_valid || !b_valid)) {
3375 /* Case A: Deleted in one */
3376 if ((!a_valid && !b_valid) ||
3377 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3378 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3379 /* Deleted in both or deleted in one and
3380 * unchanged in the other */
3381 if (a_valid)
3382 output(opt, 2, _("Removing %s"), path);
3383 /* do not touch working file if it did not exist */
3384 remove_file(opt, 1, path, !a_valid);
3385 } else {
3386 /* Modify/delete; deleted side may have put a directory in the way */
3387 clean_merge = 0;
3388 if (handle_modify_delete(opt, path, o, a, b))
3389 clean_merge = -1;
3390 }
3391 } else if ((!o_valid && a_valid && !b_valid) ||
3392 (!o_valid && !a_valid && b_valid)) {
3393 /* Case B: Added in one. */
3394 /* [nothing|directory] -> ([nothing|directory], file) */
3395
3396 const char *add_branch;
3397 const char *other_branch;
3398 const char *conf;
3399 const struct diff_filespec *contents;
3400
3401 if (a_valid) {
3402 add_branch = opt->branch1;
3403 other_branch = opt->branch2;
3404 contents = a;
3405 conf = _("file/directory");
3406 } else {
3407 add_branch = opt->branch2;
3408 other_branch = opt->branch1;
3409 contents = b;
3410 conf = _("directory/file");
3411 }
3412 if (dir_in_way(opt->repo->index, path,
3413 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3414 0)) {
3415 char *new_path = unique_path(opt, path, add_branch);
3416 clean_merge = 0;
3417 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3418 "Adding %s as %s"),
3419 conf, path, other_branch, path, new_path);
3420 if (update_file(opt, 0, contents, new_path))
3421 clean_merge = -1;
3422 else if (opt->priv->call_depth)
3423 remove_file_from_index(opt->repo->index, path);
3424 free(new_path);
3425 } else {
3426 output(opt, 2, _("Adding %s"), path);
3427 /* do not overwrite file if already present */
3428 if (update_file_flags(opt, contents, path, 1, !a_valid))
3429 clean_merge = -1;
3430 }
3431 } else if (a_valid && b_valid) {
3432 if (!o_valid) {
3433 /* Case C: Added in both (check for same permissions) */
3434 output(opt, 1,
3435 _("CONFLICT (add/add): Merge conflict in %s"),
3436 path);
3437 clean_merge = handle_file_collision(opt,
3438 path, NULL, NULL,
3439 opt->branch1,
3440 opt->branch2,
3441 a, b);
3442 } else {
3443 /* case D: Modified in both, but differently. */
3444 struct merge_file_info mfi;
3445 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3446 clean_merge = handle_content_merge(&mfi, opt, path,
3447 is_dirty,
3448 o, a, b, NULL);
3449 }
3450 } else if (!o_valid && !a_valid && !b_valid) {
3451 /*
3452 * this entry was deleted altogether. a_mode == 0 means
3453 * we had that path and want to actively remove it.
3454 */
3455 remove_file(opt, 1, path, !a->mode);
3456 } else
3457 BUG("fatal merge failure, shouldn't happen.");
3458
3459 return clean_merge;
3460 }
3461
3462 static int merge_trees_internal(struct merge_options *opt,
3463 struct tree *head,
3464 struct tree *merge,
3465 struct tree *merge_base,
3466 struct tree **result)
3467 {
3468 struct index_state *istate = opt->repo->index;
3469 int code, clean;
3470
3471 if (opt->subtree_shift) {
3472 merge = shift_tree_object(opt->repo, head, merge,
3473 opt->subtree_shift);
3474 merge_base = shift_tree_object(opt->repo, head, merge_base,
3475 opt->subtree_shift);
3476 }
3477
3478 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3479 output(opt, 0, _("Already up to date!"));
3480 *result = head;
3481 return 1;
3482 }
3483
3484 code = unpack_trees_start(opt, merge_base, head, merge);
3485
3486 if (code != 0) {
3487 if (show(opt, 4) || opt->priv->call_depth)
3488 err(opt, _("merging of trees %s and %s failed"),
3489 oid_to_hex(&head->object.oid),
3490 oid_to_hex(&merge->object.oid));
3491 unpack_trees_finish(opt);
3492 return -1;
3493 }
3494
3495 if (unmerged_index(istate)) {
3496 struct string_list *entries;
3497 struct rename_info re_info;
3498 int i;
3499 /*
3500 * Only need the hashmap while processing entries, so
3501 * initialize it here and free it when we are done running
3502 * through the entries. Keeping it in the merge_options as
3503 * opposed to decaring a local hashmap is for convenience
3504 * so that we don't have to pass it to around.
3505 */
3506 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3507 NULL, 512);
3508 get_files_dirs(opt, head);
3509 get_files_dirs(opt, merge);
3510
3511 entries = get_unmerged(opt->repo->index);
3512 clean = detect_and_process_renames(opt, merge_base, head, merge,
3513 entries, &re_info);
3514 record_df_conflict_files(opt, entries);
3515 if (clean < 0)
3516 goto cleanup;
3517 for (i = entries->nr-1; 0 <= i; i--) {
3518 const char *path = entries->items[i].string;
3519 struct stage_data *e = entries->items[i].util;
3520 if (!e->processed) {
3521 int ret = process_entry(opt, path, e);
3522 if (!ret)
3523 clean = 0;
3524 else if (ret < 0) {
3525 clean = ret;
3526 goto cleanup;
3527 }
3528 }
3529 }
3530 for (i = 0; i < entries->nr; i++) {
3531 struct stage_data *e = entries->items[i].util;
3532 if (!e->processed)
3533 BUG("unprocessed path??? %s",
3534 entries->items[i].string);
3535 }
3536
3537 cleanup:
3538 final_cleanup_renames(&re_info);
3539
3540 string_list_clear(entries, 1);
3541 free(entries);
3542
3543 hashmap_free_entries(&opt->priv->current_file_dir_set,
3544 struct path_hashmap_entry, e);
3545
3546 if (clean < 0) {
3547 unpack_trees_finish(opt);
3548 return clean;
3549 }
3550 }
3551 else
3552 clean = 1;
3553
3554 unpack_trees_finish(opt);
3555
3556 if (opt->priv->call_depth &&
3557 !(*result = write_in_core_index_as_tree(opt->repo)))
3558 return -1;
3559
3560 return clean;
3561 }
3562
3563 static struct commit_list *reverse_commit_list(struct commit_list *list)
3564 {
3565 struct commit_list *next = NULL, *current, *backup;
3566 for (current = list; current; current = backup) {
3567 backup = current->next;
3568 current->next = next;
3569 next = current;
3570 }
3571 return next;
3572 }
3573
3574 /*
3575 * Merge the commits h1 and h2, return the resulting virtual
3576 * commit object and a flag indicating the cleanness of the merge.
3577 */
3578 static int merge_recursive_internal(struct merge_options *opt,
3579 struct commit *h1,
3580 struct commit *h2,
3581 struct commit_list *merge_bases,
3582 struct commit **result)
3583 {
3584 struct commit_list *iter;
3585 struct commit *merged_merge_bases;
3586 struct tree *result_tree;
3587 int clean;
3588 const char *ancestor_name;
3589 struct strbuf merge_base_abbrev = STRBUF_INIT;
3590
3591 if (show(opt, 4)) {
3592 output(opt, 4, _("Merging:"));
3593 output_commit_title(opt, h1);
3594 output_commit_title(opt, h2);
3595 }
3596
3597 if (!merge_bases) {
3598 merge_bases = get_merge_bases(h1, h2);
3599 merge_bases = reverse_commit_list(merge_bases);
3600 }
3601
3602 if (show(opt, 5)) {
3603 unsigned cnt = commit_list_count(merge_bases);
3604
3605 output(opt, 5, Q_("found %u common ancestor:",
3606 "found %u common ancestors:", cnt), cnt);
3607 for (iter = merge_bases; iter; iter = iter->next)
3608 output_commit_title(opt, iter->item);
3609 }
3610
3611 merged_merge_bases = pop_commit(&merge_bases);
3612 if (merged_merge_bases == NULL) {
3613 /* if there is no common ancestor, use an empty tree */
3614 struct tree *tree;
3615
3616 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3617 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3618 "ancestor");
3619 ancestor_name = "empty tree";
3620 } else if (opt->ancestor && !opt->priv->call_depth) {
3621 ancestor_name = opt->ancestor;
3622 } else if (merge_bases) {
3623 ancestor_name = "merged common ancestors";
3624 } else {
3625 strbuf_add_unique_abbrev(&merge_base_abbrev,
3626 &merged_merge_bases->object.oid,
3627 DEFAULT_ABBREV);
3628 ancestor_name = merge_base_abbrev.buf;
3629 }
3630
3631 for (iter = merge_bases; iter; iter = iter->next) {
3632 const char *saved_b1, *saved_b2;
3633 opt->priv->call_depth++;
3634 /*
3635 * When the merge fails, the result contains files
3636 * with conflict markers. The cleanness flag is
3637 * ignored (unless indicating an error), it was never
3638 * actually used, as result of merge_trees has always
3639 * overwritten it: the committed "conflicts" were
3640 * already resolved.
3641 */
3642 discard_index(opt->repo->index);
3643 saved_b1 = opt->branch1;
3644 saved_b2 = opt->branch2;
3645 opt->branch1 = "Temporary merge branch 1";
3646 opt->branch2 = "Temporary merge branch 2";
3647 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3648 NULL, &merged_merge_bases) < 0)
3649 return -1;
3650 opt->branch1 = saved_b1;
3651 opt->branch2 = saved_b2;
3652 opt->priv->call_depth--;
3653
3654 if (!merged_merge_bases)
3655 return err(opt, _("merge returned no commit"));
3656 }
3657
3658 discard_index(opt->repo->index);
3659 if (!opt->priv->call_depth)
3660 repo_read_index(opt->repo);
3661
3662 opt->ancestor = ancestor_name;
3663 clean = merge_trees_internal(opt,
3664 repo_get_commit_tree(opt->repo, h1),
3665 repo_get_commit_tree(opt->repo, h2),
3666 repo_get_commit_tree(opt->repo,
3667 merged_merge_bases),
3668 &result_tree);
3669 strbuf_release(&merge_base_abbrev);
3670 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3671 if (clean < 0) {
3672 flush_output(opt);
3673 return clean;
3674 }
3675
3676 if (opt->priv->call_depth) {
3677 *result = make_virtual_commit(opt->repo, result_tree,
3678 "merged tree");
3679 commit_list_insert(h1, &(*result)->parents);
3680 commit_list_insert(h2, &(*result)->parents->next);
3681 }
3682 return clean;
3683 }
3684
3685 static int merge_start(struct merge_options *opt, struct tree *head)
3686 {
3687 struct strbuf sb = STRBUF_INIT;
3688
3689 /* Sanity checks on opt */
3690 assert(opt->repo);
3691
3692 assert(opt->branch1 && opt->branch2);
3693
3694 assert(opt->detect_renames >= -1 &&
3695 opt->detect_renames <= DIFF_DETECT_COPY);
3696 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3697 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3698 assert(opt->rename_limit >= -1);
3699 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3700 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3701
3702 assert(opt->xdl_opts >= 0);
3703 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3704 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3705
3706 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3707 assert(opt->buffer_output <= 2);
3708 assert(opt->obuf.len == 0);
3709
3710 assert(opt->priv == NULL);
3711
3712 /* Sanity check on repo state; index must match head */
3713 if (repo_index_has_changes(opt->repo, head, &sb)) {
3714 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3715 sb.buf);
3716 strbuf_release(&sb);
3717 return -1;
3718 }
3719
3720 opt->priv = xcalloc(1, sizeof(*opt->priv));
3721 string_list_init(&opt->priv->df_conflict_file_set, 1);
3722 return 0;
3723 }
3724
3725 static void merge_finalize(struct merge_options *opt)
3726 {
3727 flush_output(opt);
3728 if (!opt->priv->call_depth && opt->buffer_output < 2)
3729 strbuf_release(&opt->obuf);
3730 if (show(opt, 2))
3731 diff_warn_rename_limit("merge.renamelimit",
3732 opt->priv->needed_rename_limit, 0);
3733 FREE_AND_NULL(opt->priv);
3734 }
3735
3736 int merge_trees(struct merge_options *opt,
3737 struct tree *head,
3738 struct tree *merge,
3739 struct tree *merge_base)
3740 {
3741 int clean;
3742 struct tree *ignored;
3743
3744 assert(opt->ancestor != NULL);
3745
3746 if (merge_start(opt, head))
3747 return -1;
3748 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3749 merge_finalize(opt);
3750
3751 return clean;
3752 }
3753
3754 int merge_recursive(struct merge_options *opt,
3755 struct commit *h1,
3756 struct commit *h2,
3757 struct commit_list *merge_bases,
3758 struct commit **result)
3759 {
3760 int clean;
3761
3762 assert(opt->ancestor == NULL ||
3763 !strcmp(opt->ancestor, "constructed merge base"));
3764
3765 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3766 return -1;
3767 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3768 merge_finalize(opt);
3769
3770 return clean;
3771 }
3772
3773 static struct commit *get_ref(struct repository *repo,
3774 const struct object_id *oid,
3775 const char *name)
3776 {
3777 struct object *object;
3778
3779 object = deref_tag(repo, parse_object(repo, oid),
3780 name, strlen(name));
3781 if (!object)
3782 return NULL;
3783 if (object->type == OBJ_TREE)
3784 return make_virtual_commit(repo, (struct tree*)object, name);
3785 if (object->type != OBJ_COMMIT)
3786 return NULL;
3787 if (parse_commit((struct commit *)object))
3788 return NULL;
3789 return (struct commit *)object;
3790 }
3791
3792 int merge_recursive_generic(struct merge_options *opt,
3793 const struct object_id *head,
3794 const struct object_id *merge,
3795 int num_merge_bases,
3796 const struct object_id **merge_bases,
3797 struct commit **result)
3798 {
3799 int clean;
3800 struct lock_file lock = LOCK_INIT;
3801 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3802 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3803 struct commit_list *ca = NULL;
3804
3805 if (merge_bases) {
3806 int i;
3807 for (i = 0; i < num_merge_bases; ++i) {
3808 struct commit *base;
3809 if (!(base = get_ref(opt->repo, merge_bases[i],
3810 oid_to_hex(merge_bases[i]))))
3811 return err(opt, _("Could not parse object '%s'"),
3812 oid_to_hex(merge_bases[i]));
3813 commit_list_insert(base, &ca);
3814 }
3815 if (num_merge_bases == 1)
3816 opt->ancestor = "constructed merge base";
3817 }
3818
3819 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3820 clean = merge_recursive(opt, head_commit, next_commit, ca,
3821 result);
3822 if (clean < 0) {
3823 rollback_lock_file(&lock);
3824 return clean;
3825 }
3826
3827 if (write_locked_index(opt->repo->index, &lock,
3828 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3829 return err(opt, _("Unable to write index."));
3830
3831 return clean ? 0 : 1;
3832 }
3833
3834 static void merge_recursive_config(struct merge_options *opt)
3835 {
3836 char *value = NULL;
3837 git_config_get_int("merge.verbosity", &opt->verbosity);
3838 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3839 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3840 if (!git_config_get_string("diff.renames", &value)) {
3841 opt->detect_renames = git_config_rename("diff.renames", value);
3842 free(value);
3843 }
3844 if (!git_config_get_string("merge.renames", &value)) {
3845 opt->detect_renames = git_config_rename("merge.renames", value);
3846 free(value);
3847 }
3848 if (!git_config_get_string("merge.directoryrenames", &value)) {
3849 int boolval = git_parse_maybe_bool(value);
3850 if (0 <= boolval) {
3851 opt->detect_directory_renames = boolval ?
3852 MERGE_DIRECTORY_RENAMES_TRUE :
3853 MERGE_DIRECTORY_RENAMES_NONE;
3854 } else if (!strcasecmp(value, "conflict")) {
3855 opt->detect_directory_renames =
3856 MERGE_DIRECTORY_RENAMES_CONFLICT;
3857 } /* avoid erroring on values from future versions of git */
3858 free(value);
3859 }
3860 git_config(git_xmerge_config, NULL);
3861 }
3862
3863 void init_merge_options(struct merge_options *opt,
3864 struct repository *repo)
3865 {
3866 const char *merge_verbosity;
3867 memset(opt, 0, sizeof(struct merge_options));
3868
3869 opt->repo = repo;
3870
3871 opt->detect_renames = -1;
3872 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3873 opt->rename_limit = -1;
3874
3875 opt->verbosity = 2;
3876 opt->buffer_output = 1;
3877 strbuf_init(&opt->obuf, 0);
3878
3879 opt->renormalize = 0;
3880
3881 merge_recursive_config(opt);
3882 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3883 if (merge_verbosity)
3884 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3885 if (opt->verbosity >= 5)
3886 opt->buffer_output = 0;
3887 }
3888
3889 int parse_merge_opt(struct merge_options *opt, const char *s)
3890 {
3891 const char *arg;
3892
3893 if (!s || !*s)
3894 return -1;
3895 if (!strcmp(s, "ours"))
3896 opt->recursive_variant = MERGE_VARIANT_OURS;
3897 else if (!strcmp(s, "theirs"))
3898 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3899 else if (!strcmp(s, "subtree"))
3900 opt->subtree_shift = "";
3901 else if (skip_prefix(s, "subtree=", &arg))
3902 opt->subtree_shift = arg;
3903 else if (!strcmp(s, "patience"))
3904 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3905 else if (!strcmp(s, "histogram"))
3906 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3907 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3908 long value = parse_algorithm_value(arg);
3909 if (value < 0)
3910 return -1;
3911 /* clear out previous settings */
3912 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3913 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3914 opt->xdl_opts |= value;
3915 }
3916 else if (!strcmp(s, "ignore-space-change"))
3917 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3918 else if (!strcmp(s, "ignore-all-space"))
3919 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3920 else if (!strcmp(s, "ignore-space-at-eol"))
3921 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3922 else if (!strcmp(s, "ignore-cr-at-eol"))
3923 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3924 else if (!strcmp(s, "renormalize"))
3925 opt->renormalize = 1;
3926 else if (!strcmp(s, "no-renormalize"))
3927 opt->renormalize = 0;
3928 else if (!strcmp(s, "no-renames"))
3929 opt->detect_renames = 0;
3930 else if (!strcmp(s, "find-renames")) {
3931 opt->detect_renames = 1;
3932 opt->rename_score = 0;
3933 }
3934 else if (skip_prefix(s, "find-renames=", &arg) ||
3935 skip_prefix(s, "rename-threshold=", &arg)) {
3936 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3937 return -1;
3938 opt->detect_renames = 1;
3939 }
3940 /*
3941 * Please update $__git_merge_strategy_options in
3942 * git-completion.bash when you add new options
3943 */
3944 else
3945 return -1;
3946 return 0;
3947 }