]> git.ipfire.org Git - thirdparty/git.git/blob - merge-recursive.c
t1404: mark tests that muck with .git directly as REFFILES.
[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 CALLOC_ARRAY(ci, 1);
307 ci->rename_type = rename_type;
308 ci->ren1 = ren1;
309 ci->ren2 = ren2;
310
311 ci->ren1->dst_entry->processed = 0;
312 ci->ren1->dst_entry->rename_conflict_info = ci;
313 if (ren2) {
314 ci->ren2->dst_entry->rename_conflict_info = ci;
315 }
316 }
317
318 static int show(struct merge_options *opt, int v)
319 {
320 return (!opt->priv->call_depth && opt->verbosity >= v) ||
321 opt->verbosity >= 5;
322 }
323
324 __attribute__((format (printf, 3, 4)))
325 static void output(struct merge_options *opt, int v, const char *fmt, ...)
326 {
327 va_list ap;
328
329 if (!show(opt, v))
330 return;
331
332 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
333
334 va_start(ap, fmt);
335 strbuf_vaddf(&opt->obuf, fmt, ap);
336 va_end(ap);
337
338 strbuf_addch(&opt->obuf, '\n');
339 if (!opt->buffer_output)
340 flush_output(opt);
341 }
342
343 static void output_commit_title(struct merge_options *opt, struct commit *commit)
344 {
345 struct merge_remote_desc *desc;
346
347 strbuf_addchars(&opt->obuf, ' ', opt->priv->call_depth * 2);
348 desc = merge_remote_util(commit);
349 if (desc)
350 strbuf_addf(&opt->obuf, "virtual %s\n", desc->name);
351 else {
352 strbuf_add_unique_abbrev(&opt->obuf, &commit->object.oid,
353 DEFAULT_ABBREV);
354 strbuf_addch(&opt->obuf, ' ');
355 if (parse_commit(commit) != 0)
356 strbuf_addstr(&opt->obuf, _("(bad commit)\n"));
357 else {
358 const char *title;
359 const char *msg = get_commit_buffer(commit, NULL);
360 int len = find_commit_subject(msg, &title);
361 if (len)
362 strbuf_addf(&opt->obuf, "%.*s\n", len, title);
363 unuse_commit_buffer(commit, msg);
364 }
365 }
366 flush_output(opt);
367 }
368
369 static int add_cacheinfo(struct merge_options *opt,
370 const struct diff_filespec *blob,
371 const char *path, int stage, int refresh, int options)
372 {
373 struct index_state *istate = opt->repo->index;
374 struct cache_entry *ce;
375 int ret;
376
377 ce = make_cache_entry(istate, blob->mode, &blob->oid, path, stage, 0);
378 if (!ce)
379 return err(opt, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
380
381 ret = add_index_entry(istate, ce, options);
382 if (refresh) {
383 struct cache_entry *nce;
384
385 nce = refresh_cache_entry(istate, ce,
386 CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
387 if (!nce)
388 return err(opt, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
389 if (nce != ce)
390 ret = add_index_entry(istate, nce, options);
391 }
392 return ret;
393 }
394
395 static inline int merge_detect_rename(struct merge_options *opt)
396 {
397 return (opt->detect_renames >= 0) ? opt->detect_renames : 1;
398 }
399
400 static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
401 {
402 parse_tree(tree);
403 init_tree_desc(desc, tree->buffer, tree->size);
404 }
405
406 static int unpack_trees_start(struct merge_options *opt,
407 struct tree *common,
408 struct tree *head,
409 struct tree *merge)
410 {
411 int rc;
412 struct tree_desc t[3];
413 struct index_state tmp_index = { NULL };
414
415 memset(&opt->priv->unpack_opts, 0, sizeof(opt->priv->unpack_opts));
416 if (opt->priv->call_depth)
417 opt->priv->unpack_opts.index_only = 1;
418 else
419 opt->priv->unpack_opts.update = 1;
420 opt->priv->unpack_opts.merge = 1;
421 opt->priv->unpack_opts.head_idx = 2;
422 opt->priv->unpack_opts.fn = threeway_merge;
423 opt->priv->unpack_opts.src_index = opt->repo->index;
424 opt->priv->unpack_opts.dst_index = &tmp_index;
425 opt->priv->unpack_opts.aggressive = !merge_detect_rename(opt);
426 setup_unpack_trees_porcelain(&opt->priv->unpack_opts, "merge");
427
428 init_tree_desc_from_tree(t+0, common);
429 init_tree_desc_from_tree(t+1, head);
430 init_tree_desc_from_tree(t+2, merge);
431
432 rc = unpack_trees(3, t, &opt->priv->unpack_opts);
433 cache_tree_free(&opt->repo->index->cache_tree);
434
435 /*
436 * Update opt->repo->index to match the new results, AFTER saving a
437 * copy in opt->priv->orig_index. Update src_index to point to the
438 * saved copy. (verify_uptodate() checks src_index, and the original
439 * index is the one that had the necessary modification timestamps.)
440 */
441 opt->priv->orig_index = *opt->repo->index;
442 *opt->repo->index = tmp_index;
443 opt->priv->unpack_opts.src_index = &opt->priv->orig_index;
444
445 return rc;
446 }
447
448 static void unpack_trees_finish(struct merge_options *opt)
449 {
450 discard_index(&opt->priv->orig_index);
451 clear_unpack_trees_porcelain(&opt->priv->unpack_opts);
452 }
453
454 static int save_files_dirs(const struct object_id *oid,
455 struct strbuf *base, const char *path,
456 unsigned int mode, void *context)
457 {
458 struct path_hashmap_entry *entry;
459 int baselen = base->len;
460 struct merge_options *opt = context;
461
462 strbuf_addstr(base, path);
463
464 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
465 hashmap_entry_init(&entry->e, path_hash(entry->path));
466 hashmap_add(&opt->priv->current_file_dir_set, &entry->e);
467
468 strbuf_setlen(base, baselen);
469 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
470 }
471
472 static void get_files_dirs(struct merge_options *opt, struct tree *tree)
473 {
474 struct pathspec match_all;
475 memset(&match_all, 0, sizeof(match_all));
476 read_tree(opt->repo, tree,
477 &match_all, save_files_dirs, opt);
478 }
479
480 static int get_tree_entry_if_blob(struct repository *r,
481 const struct object_id *tree,
482 const char *path,
483 struct diff_filespec *dfs)
484 {
485 int ret;
486
487 ret = get_tree_entry(r, tree, path, &dfs->oid, &dfs->mode);
488 if (S_ISDIR(dfs->mode)) {
489 oidcpy(&dfs->oid, &null_oid);
490 dfs->mode = 0;
491 }
492 return ret;
493 }
494
495 /*
496 * Returns an index_entry instance which doesn't have to correspond to
497 * a real cache entry in Git's index.
498 */
499 static struct stage_data *insert_stage_data(struct repository *r,
500 const char *path,
501 struct tree *o, struct tree *a, struct tree *b,
502 struct string_list *entries)
503 {
504 struct string_list_item *item;
505 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
506 get_tree_entry_if_blob(r, &o->object.oid, path, &e->stages[1]);
507 get_tree_entry_if_blob(r, &a->object.oid, path, &e->stages[2]);
508 get_tree_entry_if_blob(r, &b->object.oid, path, &e->stages[3]);
509 item = string_list_insert(entries, path);
510 item->util = e;
511 return e;
512 }
513
514 /*
515 * Create a dictionary mapping file names to stage_data objects. The
516 * dictionary contains one entry for every path with a non-zero stage entry.
517 */
518 static struct string_list *get_unmerged(struct index_state *istate)
519 {
520 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
521 int i;
522
523 unmerged->strdup_strings = 1;
524
525 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, NULL)) {
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 int refresh = (!opt->priv->call_depth &&
1003 contents->mode != S_IFGITLINK);
1004 if (add_cacheinfo(opt, contents, path, 0, refresh,
1005 ADD_CACHE_OK_TO_ADD))
1006 return -1;
1007 }
1008 return ret;
1009 }
1010
1011 static int update_file(struct merge_options *opt,
1012 int clean,
1013 const struct diff_filespec *contents,
1014 const char *path)
1015 {
1016 return update_file_flags(opt, contents, path,
1017 opt->priv->call_depth || clean, !opt->priv->call_depth);
1018 }
1019
1020 /* Low level file merging, update and removal */
1021
1022 struct merge_file_info {
1023 struct diff_filespec blob; /* mostly use oid & mode; sometimes path */
1024 unsigned clean:1,
1025 merge:1;
1026 };
1027
1028 static int merge_3way(struct merge_options *opt,
1029 mmbuffer_t *result_buf,
1030 const struct diff_filespec *o,
1031 const struct diff_filespec *a,
1032 const struct diff_filespec *b,
1033 const char *branch1,
1034 const char *branch2,
1035 const int extra_marker_size)
1036 {
1037 mmfile_t orig, src1, src2;
1038 struct ll_merge_options ll_opts = {0};
1039 char *base, *name1, *name2;
1040 int merge_status;
1041
1042 ll_opts.renormalize = opt->renormalize;
1043 ll_opts.extra_marker_size = extra_marker_size;
1044 ll_opts.xdl_opts = opt->xdl_opts;
1045
1046 if (opt->priv->call_depth) {
1047 ll_opts.virtual_ancestor = 1;
1048 ll_opts.variant = 0;
1049 } else {
1050 switch (opt->recursive_variant) {
1051 case MERGE_VARIANT_OURS:
1052 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1053 break;
1054 case MERGE_VARIANT_THEIRS:
1055 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1056 break;
1057 default:
1058 ll_opts.variant = 0;
1059 break;
1060 }
1061 }
1062
1063 assert(a->path && b->path && o->path && opt->ancestor);
1064 if (strcmp(a->path, b->path) || strcmp(a->path, o->path) != 0) {
1065 base = mkpathdup("%s:%s", opt->ancestor, o->path);
1066 name1 = mkpathdup("%s:%s", branch1, a->path);
1067 name2 = mkpathdup("%s:%s", branch2, b->path);
1068 } else {
1069 base = mkpathdup("%s", opt->ancestor);
1070 name1 = mkpathdup("%s", branch1);
1071 name2 = mkpathdup("%s", branch2);
1072 }
1073
1074 read_mmblob(&orig, &o->oid);
1075 read_mmblob(&src1, &a->oid);
1076 read_mmblob(&src2, &b->oid);
1077
1078 merge_status = ll_merge(result_buf, a->path, &orig, base,
1079 &src1, name1, &src2, name2,
1080 opt->repo->index, &ll_opts);
1081
1082 free(base);
1083 free(name1);
1084 free(name2);
1085 free(orig.ptr);
1086 free(src1.ptr);
1087 free(src2.ptr);
1088 return merge_status;
1089 }
1090
1091 static int find_first_merges(struct repository *repo,
1092 struct object_array *result, const char *path,
1093 struct commit *a, struct commit *b)
1094 {
1095 int i, j;
1096 struct object_array merges = OBJECT_ARRAY_INIT;
1097 struct commit *commit;
1098 int contains_another;
1099
1100 char merged_revision[GIT_MAX_HEXSZ + 2];
1101 const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1102 "--all", merged_revision, NULL };
1103 struct rev_info revs;
1104 struct setup_revision_opt rev_opts;
1105
1106 memset(result, 0, sizeof(struct object_array));
1107 memset(&rev_opts, 0, sizeof(rev_opts));
1108
1109 /* get all revisions that merge commit a */
1110 xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1111 oid_to_hex(&a->object.oid));
1112 repo_init_revisions(repo, &revs, NULL);
1113 rev_opts.submodule = path;
1114 /* FIXME: can't handle linked worktrees in submodules yet */
1115 revs.single_worktree = path != NULL;
1116 setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1117
1118 /* save all revisions from the above list that contain b */
1119 if (prepare_revision_walk(&revs))
1120 die("revision walk setup failed");
1121 while ((commit = get_revision(&revs)) != NULL) {
1122 struct object *o = &(commit->object);
1123 if (in_merge_bases(b, commit))
1124 add_object_array(o, NULL, &merges);
1125 }
1126 reset_revision_walk();
1127
1128 /* Now we've got all merges that contain a and b. Prune all
1129 * merges that contain another found merge and save them in
1130 * result.
1131 */
1132 for (i = 0; i < merges.nr; i++) {
1133 struct commit *m1 = (struct commit *) merges.objects[i].item;
1134
1135 contains_another = 0;
1136 for (j = 0; j < merges.nr; j++) {
1137 struct commit *m2 = (struct commit *) merges.objects[j].item;
1138 if (i != j && in_merge_bases(m2, m1)) {
1139 contains_another = 1;
1140 break;
1141 }
1142 }
1143
1144 if (!contains_another)
1145 add_object_array(merges.objects[i].item, NULL, result);
1146 }
1147
1148 object_array_clear(&merges);
1149 return result->nr;
1150 }
1151
1152 static void print_commit(struct commit *commit)
1153 {
1154 struct strbuf sb = STRBUF_INIT;
1155 struct pretty_print_context ctx = {0};
1156 ctx.date_mode.type = DATE_NORMAL;
1157 format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1158 fprintf(stderr, "%s\n", sb.buf);
1159 strbuf_release(&sb);
1160 }
1161
1162 static int is_valid(const struct diff_filespec *dfs)
1163 {
1164 return dfs->mode != 0 && !is_null_oid(&dfs->oid);
1165 }
1166
1167 static int merge_submodule(struct merge_options *opt,
1168 struct object_id *result, const char *path,
1169 const struct object_id *base, const struct object_id *a,
1170 const struct object_id *b)
1171 {
1172 struct commit *commit_base, *commit_a, *commit_b;
1173 int parent_count;
1174 struct object_array merges;
1175
1176 int i;
1177 int search = !opt->priv->call_depth;
1178
1179 /* store a in result in case we fail */
1180 oidcpy(result, a);
1181
1182 /* we can not handle deletion conflicts */
1183 if (is_null_oid(base))
1184 return 0;
1185 if (is_null_oid(a))
1186 return 0;
1187 if (is_null_oid(b))
1188 return 0;
1189
1190 if (add_submodule_odb(path)) {
1191 output(opt, 1, _("Failed to merge submodule %s (not checked out)"), path);
1192 return 0;
1193 }
1194
1195 if (!(commit_base = lookup_commit_reference(opt->repo, base)) ||
1196 !(commit_a = lookup_commit_reference(opt->repo, a)) ||
1197 !(commit_b = lookup_commit_reference(opt->repo, b))) {
1198 output(opt, 1, _("Failed to merge submodule %s (commits not present)"), path);
1199 return 0;
1200 }
1201
1202 /* check whether both changes are forward */
1203 if (!in_merge_bases(commit_base, commit_a) ||
1204 !in_merge_bases(commit_base, commit_b)) {
1205 output(opt, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1206 return 0;
1207 }
1208
1209 /* Case #1: a is contained in b or vice versa */
1210 if (in_merge_bases(commit_a, commit_b)) {
1211 oidcpy(result, b);
1212 if (show(opt, 3)) {
1213 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1214 output_commit_title(opt, commit_b);
1215 } else if (show(opt, 2))
1216 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1217 else
1218 ; /* no output */
1219
1220 return 1;
1221 }
1222 if (in_merge_bases(commit_b, commit_a)) {
1223 oidcpy(result, a);
1224 if (show(opt, 3)) {
1225 output(opt, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1226 output_commit_title(opt, commit_a);
1227 } else if (show(opt, 2))
1228 output(opt, 2, _("Fast-forwarding submodule %s"), path);
1229 else
1230 ; /* no output */
1231
1232 return 1;
1233 }
1234
1235 /*
1236 * Case #2: There are one or more merges that contain a and b in
1237 * the submodule. If there is only one, then present it as a
1238 * suggestion to the user, but leave it marked unmerged so the
1239 * user needs to confirm the resolution.
1240 */
1241
1242 /* Skip the search if makes no sense to the calling context. */
1243 if (!search)
1244 return 0;
1245
1246 /* find commit which merges them */
1247 parent_count = find_first_merges(opt->repo, &merges, path,
1248 commit_a, commit_b);
1249 switch (parent_count) {
1250 case 0:
1251 output(opt, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1252 break;
1253
1254 case 1:
1255 output(opt, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1256 output(opt, 2, _("Found a possible merge resolution for the submodule:\n"));
1257 print_commit((struct commit *) merges.objects[0].item);
1258 output(opt, 2, _(
1259 "If this is correct simply add it to the index "
1260 "for example\n"
1261 "by using:\n\n"
1262 " git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1263 "which will accept this suggestion.\n"),
1264 oid_to_hex(&merges.objects[0].item->oid), path);
1265 break;
1266
1267 default:
1268 output(opt, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1269 for (i = 0; i < merges.nr; i++)
1270 print_commit((struct commit *) merges.objects[i].item);
1271 }
1272
1273 object_array_clear(&merges);
1274 return 0;
1275 }
1276
1277 static int merge_mode_and_contents(struct merge_options *opt,
1278 const struct diff_filespec *o,
1279 const struct diff_filespec *a,
1280 const struct diff_filespec *b,
1281 const char *filename,
1282 const char *branch1,
1283 const char *branch2,
1284 const int extra_marker_size,
1285 struct merge_file_info *result)
1286 {
1287 if (opt->branch1 != branch1) {
1288 /*
1289 * It's weird getting a reverse merge with HEAD on the bottom
1290 * side of the conflict markers and the other branch on the
1291 * top. Fix that.
1292 */
1293 return merge_mode_and_contents(opt, o, b, a,
1294 filename,
1295 branch2, branch1,
1296 extra_marker_size, result);
1297 }
1298
1299 result->merge = 0;
1300 result->clean = 1;
1301
1302 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1303 result->clean = 0;
1304 if (S_ISREG(a->mode)) {
1305 result->blob.mode = a->mode;
1306 oidcpy(&result->blob.oid, &a->oid);
1307 } else {
1308 result->blob.mode = b->mode;
1309 oidcpy(&result->blob.oid, &b->oid);
1310 }
1311 } else {
1312 if (!oideq(&a->oid, &o->oid) && !oideq(&b->oid, &o->oid))
1313 result->merge = 1;
1314
1315 /*
1316 * Merge modes
1317 */
1318 if (a->mode == b->mode || a->mode == o->mode)
1319 result->blob.mode = b->mode;
1320 else {
1321 result->blob.mode = a->mode;
1322 if (b->mode != o->mode) {
1323 result->clean = 0;
1324 result->merge = 1;
1325 }
1326 }
1327
1328 if (oideq(&a->oid, &b->oid) || oideq(&a->oid, &o->oid))
1329 oidcpy(&result->blob.oid, &b->oid);
1330 else if (oideq(&b->oid, &o->oid))
1331 oidcpy(&result->blob.oid, &a->oid);
1332 else if (S_ISREG(a->mode)) {
1333 mmbuffer_t result_buf;
1334 int ret = 0, merge_status;
1335
1336 merge_status = merge_3way(opt, &result_buf, o, a, b,
1337 branch1, branch2,
1338 extra_marker_size);
1339
1340 if ((merge_status < 0) || !result_buf.ptr)
1341 ret = err(opt, _("Failed to execute internal merge"));
1342
1343 if (!ret &&
1344 write_object_file(result_buf.ptr, result_buf.size,
1345 blob_type, &result->blob.oid))
1346 ret = err(opt, _("Unable to add %s to database"),
1347 a->path);
1348
1349 free(result_buf.ptr);
1350 if (ret)
1351 return ret;
1352 result->clean = (merge_status == 0);
1353 } else if (S_ISGITLINK(a->mode)) {
1354 result->clean = merge_submodule(opt, &result->blob.oid,
1355 o->path,
1356 &o->oid,
1357 &a->oid,
1358 &b->oid);
1359 } else if (S_ISLNK(a->mode)) {
1360 switch (opt->recursive_variant) {
1361 case MERGE_VARIANT_NORMAL:
1362 oidcpy(&result->blob.oid, &a->oid);
1363 if (!oideq(&a->oid, &b->oid))
1364 result->clean = 0;
1365 break;
1366 case MERGE_VARIANT_OURS:
1367 oidcpy(&result->blob.oid, &a->oid);
1368 break;
1369 case MERGE_VARIANT_THEIRS:
1370 oidcpy(&result->blob.oid, &b->oid);
1371 break;
1372 }
1373 } else
1374 BUG("unsupported object type in the tree");
1375 }
1376
1377 if (result->merge)
1378 output(opt, 2, _("Auto-merging %s"), filename);
1379
1380 return 0;
1381 }
1382
1383 static int handle_rename_via_dir(struct merge_options *opt,
1384 struct rename_conflict_info *ci)
1385 {
1386 /*
1387 * Handle file adds that need to be renamed due to directory rename
1388 * detection. This differs from handle_rename_normal, because
1389 * there is no content merge to do; just move the file into the
1390 * desired final location.
1391 */
1392 const struct rename *ren = ci->ren1;
1393 const struct diff_filespec *dest = ren->pair->two;
1394 char *file_path = dest->path;
1395 int mark_conflicted = (opt->detect_directory_renames ==
1396 MERGE_DIRECTORY_RENAMES_CONFLICT);
1397 assert(ren->dir_rename_original_dest);
1398
1399 if (!opt->priv->call_depth && would_lose_untracked(opt, dest->path)) {
1400 mark_conflicted = 1;
1401 file_path = unique_path(opt, dest->path, ren->branch);
1402 output(opt, 1, _("Error: Refusing to lose untracked file at %s; "
1403 "writing to %s instead."),
1404 dest->path, file_path);
1405 }
1406
1407 if (mark_conflicted) {
1408 /*
1409 * Write the file in worktree at file_path. In the index,
1410 * only record the file at dest->path in the appropriate
1411 * higher stage.
1412 */
1413 if (update_file(opt, 0, dest, file_path))
1414 return -1;
1415 if (file_path != dest->path)
1416 free(file_path);
1417 if (update_stages(opt, dest->path, NULL,
1418 ren->branch == opt->branch1 ? dest : NULL,
1419 ren->branch == opt->branch1 ? NULL : dest))
1420 return -1;
1421 return 0; /* not clean, but conflicted */
1422 } else {
1423 /* Update dest->path both in index and in worktree */
1424 if (update_file(opt, 1, dest, dest->path))
1425 return -1;
1426 return 1; /* clean */
1427 }
1428 }
1429
1430 static int handle_change_delete(struct merge_options *opt,
1431 const char *path, const char *old_path,
1432 const struct diff_filespec *o,
1433 const struct diff_filespec *changed,
1434 const char *change_branch,
1435 const char *delete_branch,
1436 const char *change, const char *change_past)
1437 {
1438 char *alt_path = NULL;
1439 const char *update_path = path;
1440 int ret = 0;
1441
1442 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0) ||
1443 (!opt->priv->call_depth && would_lose_untracked(opt, path))) {
1444 update_path = alt_path = unique_path(opt, path, change_branch);
1445 }
1446
1447 if (opt->priv->call_depth) {
1448 /*
1449 * We cannot arbitrarily accept either a_sha or b_sha as
1450 * correct; since there is no true "middle point" between
1451 * them, simply reuse the base version for virtual merge base.
1452 */
1453 ret = remove_file_from_index(opt->repo->index, path);
1454 if (!ret)
1455 ret = update_file(opt, 0, o, update_path);
1456 } else {
1457 /*
1458 * Despite the four nearly duplicate messages and argument
1459 * lists below and the ugliness of the nested if-statements,
1460 * having complete messages makes the job easier for
1461 * translators.
1462 *
1463 * The slight variance among the cases is due to the fact
1464 * that:
1465 * 1) directory/file conflicts (in effect if
1466 * !alt_path) could cause us to need to write the
1467 * file to a different path.
1468 * 2) renames (in effect if !old_path) could mean that
1469 * there are two names for the path that the user
1470 * may know the file by.
1471 */
1472 if (!alt_path) {
1473 if (!old_path) {
1474 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1475 "and %s in %s. Version %s of %s left in tree."),
1476 change, path, delete_branch, change_past,
1477 change_branch, change_branch, path);
1478 } else {
1479 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1480 "and %s to %s in %s. Version %s of %s left in tree."),
1481 change, old_path, delete_branch, change_past, path,
1482 change_branch, change_branch, path);
1483 }
1484 } else {
1485 if (!old_path) {
1486 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1487 "and %s in %s. Version %s of %s left in tree at %s."),
1488 change, path, delete_branch, change_past,
1489 change_branch, change_branch, path, alt_path);
1490 } else {
1491 output(opt, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1492 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1493 change, old_path, delete_branch, change_past, path,
1494 change_branch, change_branch, path, alt_path);
1495 }
1496 }
1497 /*
1498 * No need to call update_file() on path when change_branch ==
1499 * opt->branch1 && !alt_path, since that would needlessly touch
1500 * path. We could call update_file_flags() with update_cache=0
1501 * and update_wd=0, but that's a no-op.
1502 */
1503 if (change_branch != opt->branch1 || alt_path)
1504 ret = update_file(opt, 0, changed, update_path);
1505 }
1506 free(alt_path);
1507
1508 return ret;
1509 }
1510
1511 static int handle_rename_delete(struct merge_options *opt,
1512 struct rename_conflict_info *ci)
1513 {
1514 const struct rename *ren = ci->ren1;
1515 const struct diff_filespec *orig = ren->pair->one;
1516 const struct diff_filespec *dest = ren->pair->two;
1517 const char *rename_branch = ren->branch;
1518 const char *delete_branch = (opt->branch1 == ren->branch ?
1519 opt->branch2 : opt->branch1);
1520
1521 if (handle_change_delete(opt,
1522 opt->priv->call_depth ? orig->path : dest->path,
1523 opt->priv->call_depth ? NULL : orig->path,
1524 orig, dest,
1525 rename_branch, delete_branch,
1526 _("rename"), _("renamed")))
1527 return -1;
1528
1529 if (opt->priv->call_depth)
1530 return remove_file_from_index(opt->repo->index, dest->path);
1531 else
1532 return update_stages(opt, dest->path, NULL,
1533 rename_branch == opt->branch1 ? dest : NULL,
1534 rename_branch == opt->branch1 ? NULL : dest);
1535 }
1536
1537 static int handle_file_collision(struct merge_options *opt,
1538 const char *collide_path,
1539 const char *prev_path1,
1540 const char *prev_path2,
1541 const char *branch1, const char *branch2,
1542 struct diff_filespec *a,
1543 struct diff_filespec *b)
1544 {
1545 struct merge_file_info mfi;
1546 struct diff_filespec null;
1547 char *alt_path = NULL;
1548 const char *update_path = collide_path;
1549
1550 /*
1551 * It's easiest to get the correct things into stage 2 and 3, and
1552 * to make sure that the content merge puts HEAD before the other
1553 * branch if we just ensure that branch1 == opt->branch1. So, simply
1554 * flip arguments around if we don't have that.
1555 */
1556 if (branch1 != opt->branch1) {
1557 return handle_file_collision(opt, collide_path,
1558 prev_path2, prev_path1,
1559 branch2, branch1,
1560 b, a);
1561 }
1562
1563 /* Remove rename sources if rename/add or rename/rename(2to1) */
1564 if (prev_path1)
1565 remove_file(opt, 1, prev_path1,
1566 opt->priv->call_depth || would_lose_untracked(opt, prev_path1));
1567 if (prev_path2)
1568 remove_file(opt, 1, prev_path2,
1569 opt->priv->call_depth || would_lose_untracked(opt, prev_path2));
1570
1571 /*
1572 * Remove the collision path, if it wouldn't cause dirty contents
1573 * or an untracked file to get lost. We'll either overwrite with
1574 * merged contents, or just write out to differently named files.
1575 */
1576 if (was_dirty(opt, collide_path)) {
1577 output(opt, 1, _("Refusing to lose dirty file at %s"),
1578 collide_path);
1579 update_path = alt_path = unique_path(opt, collide_path, "merged");
1580 } else if (would_lose_untracked(opt, collide_path)) {
1581 /*
1582 * Only way we get here is if both renames were from
1583 * a directory rename AND user had an untracked file
1584 * at the location where both files end up after the
1585 * two directory renames. See testcase 10d of t6043.
1586 */
1587 output(opt, 1, _("Refusing to lose untracked file at "
1588 "%s, even though it's in the way."),
1589 collide_path);
1590 update_path = alt_path = unique_path(opt, collide_path, "merged");
1591 } else {
1592 /*
1593 * FIXME: It's possible that the two files are identical
1594 * and that the current working copy happens to match, in
1595 * which case we are unnecessarily touching the working
1596 * tree file. It's not a likely enough scenario that I
1597 * want to code up the checks for it and a better fix is
1598 * available if we restructure how unpack_trees() and
1599 * merge-recursive interoperate anyway, so punting for
1600 * now...
1601 */
1602 remove_file(opt, 0, collide_path, 0);
1603 }
1604
1605 /* Store things in diff_filespecs for functions that need it */
1606 null.path = (char *)collide_path;
1607 oidcpy(&null.oid, &null_oid);
1608 null.mode = 0;
1609
1610 if (merge_mode_and_contents(opt, &null, a, b, collide_path,
1611 branch1, branch2, opt->priv->call_depth * 2, &mfi))
1612 return -1;
1613 mfi.clean &= !alt_path;
1614 if (update_file(opt, mfi.clean, &mfi.blob, update_path))
1615 return -1;
1616 if (!mfi.clean && !opt->priv->call_depth &&
1617 update_stages(opt, collide_path, NULL, a, b))
1618 return -1;
1619 free(alt_path);
1620 /*
1621 * FIXME: If both a & b both started with conflicts (only possible
1622 * if they came from a rename/rename(2to1)), but had IDENTICAL
1623 * contents including those conflicts, then in the next line we claim
1624 * it was clean. If someone cares about this case, we should have the
1625 * caller notify us if we started with conflicts.
1626 */
1627 return mfi.clean;
1628 }
1629
1630 static int handle_rename_add(struct merge_options *opt,
1631 struct rename_conflict_info *ci)
1632 {
1633 /* a was renamed to c, and a separate c was added. */
1634 struct diff_filespec *a = ci->ren1->pair->one;
1635 struct diff_filespec *c = ci->ren1->pair->two;
1636 char *path = c->path;
1637 char *prev_path_desc;
1638 struct merge_file_info mfi;
1639
1640 const char *rename_branch = ci->ren1->branch;
1641 const char *add_branch = (opt->branch1 == rename_branch ?
1642 opt->branch2 : opt->branch1);
1643 int other_stage = (ci->ren1->branch == opt->branch1 ? 3 : 2);
1644
1645 output(opt, 1, _("CONFLICT (rename/add): "
1646 "Rename %s->%s in %s. Added %s in %s"),
1647 a->path, c->path, rename_branch,
1648 c->path, add_branch);
1649
1650 prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1651 ci->ren1->src_entry->stages[other_stage].path = a->path;
1652 if (merge_mode_and_contents(opt, a, c,
1653 &ci->ren1->src_entry->stages[other_stage],
1654 prev_path_desc,
1655 opt->branch1, opt->branch2,
1656 1 + opt->priv->call_depth * 2, &mfi))
1657 return -1;
1658 free(prev_path_desc);
1659
1660 ci->ren1->dst_entry->stages[other_stage].path = mfi.blob.path = c->path;
1661 return handle_file_collision(opt,
1662 c->path, a->path, NULL,
1663 rename_branch, add_branch,
1664 &mfi.blob,
1665 &ci->ren1->dst_entry->stages[other_stage]);
1666 }
1667
1668 static char *find_path_for_conflict(struct merge_options *opt,
1669 const char *path,
1670 const char *branch1,
1671 const char *branch2)
1672 {
1673 char *new_path = NULL;
1674 if (dir_in_way(opt->repo->index, path, !opt->priv->call_depth, 0)) {
1675 new_path = unique_path(opt, path, branch1);
1676 output(opt, 1, _("%s is a directory in %s adding "
1677 "as %s instead"),
1678 path, branch2, new_path);
1679 } else if (would_lose_untracked(opt, path)) {
1680 new_path = unique_path(opt, path, branch1);
1681 output(opt, 1, _("Refusing to lose untracked file"
1682 " at %s; adding as %s instead"),
1683 path, new_path);
1684 }
1685
1686 return new_path;
1687 }
1688
1689 /*
1690 * Toggle the stage number between "ours" and "theirs" (2 and 3).
1691 */
1692 static inline int flip_stage(int stage)
1693 {
1694 return (2 + 3) - stage;
1695 }
1696
1697 static int handle_rename_rename_1to2(struct merge_options *opt,
1698 struct rename_conflict_info *ci)
1699 {
1700 /* One file was renamed in both branches, but to different names. */
1701 struct merge_file_info mfi;
1702 struct diff_filespec *add;
1703 struct diff_filespec *o = ci->ren1->pair->one;
1704 struct diff_filespec *a = ci->ren1->pair->two;
1705 struct diff_filespec *b = ci->ren2->pair->two;
1706 char *path_desc;
1707
1708 output(opt, 1, _("CONFLICT (rename/rename): "
1709 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1710 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1711 o->path, a->path, ci->ren1->branch,
1712 o->path, b->path, ci->ren2->branch,
1713 opt->priv->call_depth ? _(" (left unresolved)") : "");
1714
1715 path_desc = xstrfmt("%s and %s, both renamed from %s",
1716 a->path, b->path, o->path);
1717 if (merge_mode_and_contents(opt, o, a, b, path_desc,
1718 ci->ren1->branch, ci->ren2->branch,
1719 opt->priv->call_depth * 2, &mfi))
1720 return -1;
1721 free(path_desc);
1722
1723 if (opt->priv->call_depth)
1724 remove_file_from_index(opt->repo->index, o->path);
1725
1726 /*
1727 * For each destination path, we need to see if there is a
1728 * rename/add collision. If not, we can write the file out
1729 * to the specified location.
1730 */
1731 add = &ci->ren1->dst_entry->stages[flip_stage(2)];
1732 if (is_valid(add)) {
1733 add->path = mfi.blob.path = a->path;
1734 if (handle_file_collision(opt, a->path,
1735 NULL, NULL,
1736 ci->ren1->branch,
1737 ci->ren2->branch,
1738 &mfi.blob, add) < 0)
1739 return -1;
1740 } else {
1741 char *new_path = find_path_for_conflict(opt, a->path,
1742 ci->ren1->branch,
1743 ci->ren2->branch);
1744 if (update_file(opt, 0, &mfi.blob,
1745 new_path ? new_path : a->path))
1746 return -1;
1747 free(new_path);
1748 if (!opt->priv->call_depth &&
1749 update_stages(opt, a->path, NULL, a, NULL))
1750 return -1;
1751 }
1752
1753 if (!mfi.clean && mfi.blob.mode == a->mode &&
1754 oideq(&mfi.blob.oid, &a->oid)) {
1755 /*
1756 * Getting here means we were attempting to merge a binary
1757 * blob. Since we can't merge binaries, the merge algorithm
1758 * just takes one side. But we don't want to copy the
1759 * contents of one side to both paths; we'd rather use the
1760 * original content at the given path for each path.
1761 */
1762 oidcpy(&mfi.blob.oid, &b->oid);
1763 mfi.blob.mode = b->mode;
1764 }
1765 add = &ci->ren2->dst_entry->stages[flip_stage(3)];
1766 if (is_valid(add)) {
1767 add->path = mfi.blob.path = b->path;
1768 if (handle_file_collision(opt, b->path,
1769 NULL, NULL,
1770 ci->ren1->branch,
1771 ci->ren2->branch,
1772 add, &mfi.blob) < 0)
1773 return -1;
1774 } else {
1775 char *new_path = find_path_for_conflict(opt, b->path,
1776 ci->ren2->branch,
1777 ci->ren1->branch);
1778 if (update_file(opt, 0, &mfi.blob,
1779 new_path ? new_path : b->path))
1780 return -1;
1781 free(new_path);
1782 if (!opt->priv->call_depth &&
1783 update_stages(opt, b->path, NULL, NULL, b))
1784 return -1;
1785 }
1786
1787 return 0;
1788 }
1789
1790 static int handle_rename_rename_2to1(struct merge_options *opt,
1791 struct rename_conflict_info *ci)
1792 {
1793 /* Two files, a & b, were renamed to the same thing, c. */
1794 struct diff_filespec *a = ci->ren1->pair->one;
1795 struct diff_filespec *b = ci->ren2->pair->one;
1796 struct diff_filespec *c1 = ci->ren1->pair->two;
1797 struct diff_filespec *c2 = ci->ren2->pair->two;
1798 char *path = c1->path; /* == c2->path */
1799 char *path_side_1_desc;
1800 char *path_side_2_desc;
1801 struct merge_file_info mfi_c1;
1802 struct merge_file_info mfi_c2;
1803 int ostage1, ostage2;
1804
1805 output(opt, 1, _("CONFLICT (rename/rename): "
1806 "Rename %s->%s in %s. "
1807 "Rename %s->%s in %s"),
1808 a->path, c1->path, ci->ren1->branch,
1809 b->path, c2->path, ci->ren2->branch);
1810
1811 path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1812 path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1813 ostage1 = ci->ren1->branch == opt->branch1 ? 3 : 2;
1814 ostage2 = flip_stage(ostage1);
1815 ci->ren1->src_entry->stages[ostage1].path = a->path;
1816 ci->ren2->src_entry->stages[ostage2].path = b->path;
1817 if (merge_mode_and_contents(opt, a, c1,
1818 &ci->ren1->src_entry->stages[ostage1],
1819 path_side_1_desc,
1820 opt->branch1, opt->branch2,
1821 1 + opt->priv->call_depth * 2, &mfi_c1) ||
1822 merge_mode_and_contents(opt, b,
1823 &ci->ren2->src_entry->stages[ostage2],
1824 c2, path_side_2_desc,
1825 opt->branch1, opt->branch2,
1826 1 + opt->priv->call_depth * 2, &mfi_c2))
1827 return -1;
1828 free(path_side_1_desc);
1829 free(path_side_2_desc);
1830 mfi_c1.blob.path = path;
1831 mfi_c2.blob.path = path;
1832
1833 return handle_file_collision(opt, path, a->path, b->path,
1834 ci->ren1->branch, ci->ren2->branch,
1835 &mfi_c1.blob, &mfi_c2.blob);
1836 }
1837
1838 /*
1839 * Get the diff_filepairs changed between o_tree and tree.
1840 */
1841 static struct diff_queue_struct *get_diffpairs(struct merge_options *opt,
1842 struct tree *o_tree,
1843 struct tree *tree)
1844 {
1845 struct diff_queue_struct *ret;
1846 struct diff_options opts;
1847
1848 repo_diff_setup(opt->repo, &opts);
1849 opts.flags.recursive = 1;
1850 opts.flags.rename_empty = 0;
1851 opts.detect_rename = merge_detect_rename(opt);
1852 /*
1853 * We do not have logic to handle the detection of copies. In
1854 * fact, it may not even make sense to add such logic: would we
1855 * really want a change to a base file to be propagated through
1856 * multiple other files by a merge?
1857 */
1858 if (opts.detect_rename > DIFF_DETECT_RENAME)
1859 opts.detect_rename = DIFF_DETECT_RENAME;
1860 opts.rename_limit = (opt->rename_limit >= 0) ? opt->rename_limit : 1000;
1861 opts.rename_score = opt->rename_score;
1862 opts.show_rename_progress = opt->show_rename_progress;
1863 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1864 diff_setup_done(&opts);
1865 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1866 diffcore_std(&opts);
1867 if (opts.needed_rename_limit > opt->priv->needed_rename_limit)
1868 opt->priv->needed_rename_limit = opts.needed_rename_limit;
1869
1870 ret = xmalloc(sizeof(*ret));
1871 *ret = diff_queued_diff;
1872
1873 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1874 diff_queued_diff.nr = 0;
1875 diff_queued_diff.queue = NULL;
1876 diff_flush(&opts);
1877 return ret;
1878 }
1879
1880 static int tree_has_path(struct repository *r, struct tree *tree,
1881 const char *path)
1882 {
1883 struct object_id hashy;
1884 unsigned short mode_o;
1885
1886 return !get_tree_entry(r,
1887 &tree->object.oid, path,
1888 &hashy, &mode_o);
1889 }
1890
1891 /*
1892 * Return a new string that replaces the beginning portion (which matches
1893 * entry->dir), with entry->new_dir. In perl-speak:
1894 * new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1895 * NOTE:
1896 * Caller must ensure that old_path starts with entry->dir + '/'.
1897 */
1898 static char *apply_dir_rename(struct dir_rename_entry *entry,
1899 const char *old_path)
1900 {
1901 struct strbuf new_path = STRBUF_INIT;
1902 int oldlen, newlen;
1903
1904 if (entry->non_unique_new_dir)
1905 return NULL;
1906
1907 oldlen = strlen(entry->dir);
1908 if (entry->new_dir.len == 0)
1909 /*
1910 * If someone renamed/merged a subdirectory into the root
1911 * directory (e.g. 'some/subdir' -> ''), then we want to
1912 * avoid returning
1913 * '' + '/filename'
1914 * as the rename; we need to make old_path + oldlen advance
1915 * past the '/' character.
1916 */
1917 oldlen++;
1918 newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1919 strbuf_grow(&new_path, newlen);
1920 strbuf_addbuf(&new_path, &entry->new_dir);
1921 strbuf_addstr(&new_path, &old_path[oldlen]);
1922
1923 return strbuf_detach(&new_path, NULL);
1924 }
1925
1926 static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1927 char **old_dir, char **new_dir)
1928 {
1929 char *end_of_old, *end_of_new;
1930
1931 /* Default return values: NULL, meaning no rename */
1932 *old_dir = NULL;
1933 *new_dir = NULL;
1934
1935 /*
1936 * For
1937 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1938 * the "e/foo.c" part is the same, we just want to know that
1939 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1940 * so, for this example, this function returns "a/b/c/d" in
1941 * *old_dir and "a/b/some/thing/else" in *new_dir.
1942 */
1943
1944 /*
1945 * If the basename of the file changed, we don't care. We want
1946 * to know which portion of the directory, if any, changed.
1947 */
1948 end_of_old = strrchr(old_path, '/');
1949 end_of_new = strrchr(new_path, '/');
1950
1951 /*
1952 * If end_of_old is NULL, old_path wasn't in a directory, so there
1953 * could not be a directory rename (our rule elsewhere that a
1954 * directory which still exists is not considered to have been
1955 * renamed means the root directory can never be renamed -- because
1956 * the root directory always exists).
1957 */
1958 if (end_of_old == NULL)
1959 return; /* Note: *old_dir and *new_dir are still NULL */
1960
1961 /*
1962 * If new_path contains no directory (end_of_new is NULL), then we
1963 * have a rename of old_path's directory to the root directory.
1964 */
1965 if (end_of_new == NULL) {
1966 *old_dir = xstrndup(old_path, end_of_old - old_path);
1967 *new_dir = xstrdup("");
1968 return;
1969 }
1970
1971 /* Find the first non-matching character traversing backwards */
1972 while (*--end_of_new == *--end_of_old &&
1973 end_of_old != old_path &&
1974 end_of_new != new_path)
1975 ; /* Do nothing; all in the while loop */
1976
1977 /*
1978 * If both got back to the beginning of their strings, then the
1979 * directory didn't change at all, only the basename did.
1980 */
1981 if (end_of_old == old_path && end_of_new == new_path &&
1982 *end_of_old == *end_of_new)
1983 return; /* Note: *old_dir and *new_dir are still NULL */
1984
1985 /*
1986 * If end_of_new got back to the beginning of its string, and
1987 * end_of_old got back to the beginning of some subdirectory, then
1988 * we have a rename/merge of a subdirectory into the root, which
1989 * needs slightly special handling.
1990 *
1991 * Note: There is no need to consider the opposite case, with a
1992 * rename/merge of the root directory into some subdirectory
1993 * because as noted above the root directory always exists so it
1994 * cannot be considered to be renamed.
1995 */
1996 if (end_of_new == new_path &&
1997 end_of_old != old_path && end_of_old[-1] == '/') {
1998 *old_dir = xstrndup(old_path, --end_of_old - old_path);
1999 *new_dir = xstrdup("");
2000 return;
2001 }
2002
2003 /*
2004 * We've found the first non-matching character in the directory
2005 * paths. That means the current characters we were looking at
2006 * were part of the first non-matching subdir name going back from
2007 * the end of the strings. Get the whole name by advancing both
2008 * end_of_old and end_of_new to the NEXT '/' character. That will
2009 * represent the entire directory rename.
2010 *
2011 * The reason for the increment is cases like
2012 * a/b/star/foo/whatever.c -> a/b/tar/foo/random.c
2013 * After dropping the basename and going back to the first
2014 * non-matching character, we're now comparing:
2015 * a/b/s and a/b/
2016 * and we want to be comparing:
2017 * a/b/star/ and a/b/tar/
2018 * but without the pre-increment, the one on the right would stay
2019 * a/b/.
2020 */
2021 end_of_old = strchr(++end_of_old, '/');
2022 end_of_new = strchr(++end_of_new, '/');
2023
2024 /* Copy the old and new directories into *old_dir and *new_dir. */
2025 *old_dir = xstrndup(old_path, end_of_old - old_path);
2026 *new_dir = xstrndup(new_path, end_of_new - new_path);
2027 }
2028
2029 static void remove_hashmap_entries(struct hashmap *dir_renames,
2030 struct string_list *items_to_remove)
2031 {
2032 int i;
2033 struct dir_rename_entry *entry;
2034
2035 for (i = 0; i < items_to_remove->nr; i++) {
2036 entry = items_to_remove->items[i].util;
2037 hashmap_remove(dir_renames, &entry->ent, NULL);
2038 }
2039 string_list_clear(items_to_remove, 0);
2040 }
2041
2042 /*
2043 * See if there is a directory rename for path, and if there are any file
2044 * level conflicts for the renamed location. If there is a rename and
2045 * there are no conflicts, return the new name. Otherwise, return NULL.
2046 */
2047 static char *handle_path_level_conflicts(struct merge_options *opt,
2048 const char *path,
2049 struct dir_rename_entry *entry,
2050 struct hashmap *collisions,
2051 struct tree *tree)
2052 {
2053 char *new_path = NULL;
2054 struct collision_entry *collision_ent;
2055 int clean = 1;
2056 struct strbuf collision_paths = STRBUF_INIT;
2057
2058 /*
2059 * entry has the mapping of old directory name to new directory name
2060 * that we want to apply to path.
2061 */
2062 new_path = apply_dir_rename(entry, path);
2063
2064 if (!new_path) {
2065 /* This should only happen when entry->non_unique_new_dir set */
2066 if (!entry->non_unique_new_dir)
2067 BUG("entry->non_unqiue_dir not set and !new_path");
2068 output(opt, 1, _("CONFLICT (directory rename split): "
2069 "Unclear where to place %s because directory "
2070 "%s was renamed to multiple other directories, "
2071 "with no destination getting a majority of the "
2072 "files."),
2073 path, entry->dir);
2074 clean = 0;
2075 return NULL;
2076 }
2077
2078 /*
2079 * The caller needs to have ensured that it has pre-populated
2080 * collisions with all paths that map to new_path. Do a quick check
2081 * to ensure that's the case.
2082 */
2083 collision_ent = collision_find_entry(collisions, new_path);
2084 if (collision_ent == NULL)
2085 BUG("collision_ent is NULL");
2086
2087 /*
2088 * Check for one-sided add/add/.../add conflicts, i.e.
2089 * where implicit renames from the other side doing
2090 * directory rename(s) can affect this side of history
2091 * to put multiple paths into the same location. Warn
2092 * and bail on directory renames for such paths.
2093 */
2094 if (collision_ent->reported_already) {
2095 clean = 0;
2096 } else if (tree_has_path(opt->repo, tree, new_path)) {
2097 collision_ent->reported_already = 1;
2098 strbuf_add_separated_string_list(&collision_paths, ", ",
2099 &collision_ent->source_files);
2100 output(opt, 1, _("CONFLICT (implicit dir rename): Existing "
2101 "file/dir at %s in the way of implicit "
2102 "directory rename(s) putting the following "
2103 "path(s) there: %s."),
2104 new_path, collision_paths.buf);
2105 clean = 0;
2106 } else if (collision_ent->source_files.nr > 1) {
2107 collision_ent->reported_already = 1;
2108 strbuf_add_separated_string_list(&collision_paths, ", ",
2109 &collision_ent->source_files);
2110 output(opt, 1, _("CONFLICT (implicit dir rename): Cannot map "
2111 "more than one path to %s; implicit directory "
2112 "renames tried to put these paths there: %s"),
2113 new_path, collision_paths.buf);
2114 clean = 0;
2115 }
2116
2117 /* Free memory we no longer need */
2118 strbuf_release(&collision_paths);
2119 if (!clean && new_path) {
2120 free(new_path);
2121 return NULL;
2122 }
2123
2124 return new_path;
2125 }
2126
2127 /*
2128 * There are a couple things we want to do at the directory level:
2129 * 1. Check for both sides renaming to the same thing, in order to avoid
2130 * implicit renaming of files that should be left in place. (See
2131 * testcase 6b in t6043 for details.)
2132 * 2. Prune directory renames if there are still files left in the
2133 * the original directory. These represent a partial directory rename,
2134 * i.e. a rename where only some of the files within the directory
2135 * were renamed elsewhere. (Technically, this could be done earlier
2136 * in get_directory_renames(), except that would prevent us from
2137 * doing the previous check and thus failing testcase 6b.)
2138 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
2139 * In the future, we could potentially record this info as well and
2140 * omit reporting rename/rename(1to2) conflicts for each path within
2141 * the affected directories, thus cleaning up the merge output.
2142 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2143 * directory level, because merging directories is fine. If it
2144 * causes conflicts for files within those merged directories, then
2145 * that should be detected at the individual path level.
2146 */
2147 static void handle_directory_level_conflicts(struct merge_options *opt,
2148 struct hashmap *dir_re_head,
2149 struct tree *head,
2150 struct hashmap *dir_re_merge,
2151 struct tree *merge)
2152 {
2153 struct hashmap_iter iter;
2154 struct dir_rename_entry *head_ent;
2155 struct dir_rename_entry *merge_ent;
2156
2157 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2158 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2159
2160 hashmap_for_each_entry(dir_re_head, &iter, head_ent,
2161 ent /* member name */) {
2162 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2163 if (merge_ent &&
2164 !head_ent->non_unique_new_dir &&
2165 !merge_ent->non_unique_new_dir &&
2166 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2167 /* 1. Renamed identically; remove it from both sides */
2168 string_list_append(&remove_from_head,
2169 head_ent->dir)->util = head_ent;
2170 strbuf_release(&head_ent->new_dir);
2171 string_list_append(&remove_from_merge,
2172 merge_ent->dir)->util = merge_ent;
2173 strbuf_release(&merge_ent->new_dir);
2174 } else if (tree_has_path(opt->repo, head, head_ent->dir)) {
2175 /* 2. This wasn't a directory rename after all */
2176 string_list_append(&remove_from_head,
2177 head_ent->dir)->util = head_ent;
2178 strbuf_release(&head_ent->new_dir);
2179 }
2180 }
2181
2182 remove_hashmap_entries(dir_re_head, &remove_from_head);
2183 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2184
2185 hashmap_for_each_entry(dir_re_merge, &iter, merge_ent,
2186 ent /* member name */) {
2187 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2188 if (tree_has_path(opt->repo, merge, merge_ent->dir)) {
2189 /* 2. This wasn't a directory rename after all */
2190 string_list_append(&remove_from_merge,
2191 merge_ent->dir)->util = merge_ent;
2192 } else if (head_ent &&
2193 !head_ent->non_unique_new_dir &&
2194 !merge_ent->non_unique_new_dir) {
2195 /* 3. rename/rename(1to2) */
2196 /*
2197 * We can assume it's not rename/rename(1to1) because
2198 * that was case (1), already checked above. So we
2199 * know that head_ent->new_dir and merge_ent->new_dir
2200 * are different strings.
2201 */
2202 output(opt, 1, _("CONFLICT (rename/rename): "
2203 "Rename directory %s->%s in %s. "
2204 "Rename directory %s->%s in %s"),
2205 head_ent->dir, head_ent->new_dir.buf, opt->branch1,
2206 head_ent->dir, merge_ent->new_dir.buf, opt->branch2);
2207 string_list_append(&remove_from_head,
2208 head_ent->dir)->util = head_ent;
2209 strbuf_release(&head_ent->new_dir);
2210 string_list_append(&remove_from_merge,
2211 merge_ent->dir)->util = merge_ent;
2212 strbuf_release(&merge_ent->new_dir);
2213 }
2214 }
2215
2216 remove_hashmap_entries(dir_re_head, &remove_from_head);
2217 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2218 }
2219
2220 static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs)
2221 {
2222 struct hashmap *dir_renames;
2223 struct hashmap_iter iter;
2224 struct dir_rename_entry *entry;
2225 int i;
2226
2227 /*
2228 * Typically, we think of a directory rename as all files from a
2229 * certain directory being moved to a target directory. However,
2230 * what if someone first moved two files from the original
2231 * directory in one commit, and then renamed the directory
2232 * somewhere else in a later commit? At merge time, we just know
2233 * that files from the original directory went to two different
2234 * places, and that the bulk of them ended up in the same place.
2235 * We want each directory rename to represent where the bulk of the
2236 * files from that directory end up; this function exists to find
2237 * where the bulk of the files went.
2238 *
2239 * The first loop below simply iterates through the list of file
2240 * renames, finding out how often each directory rename pair
2241 * possibility occurs.
2242 */
2243 dir_renames = xmalloc(sizeof(*dir_renames));
2244 dir_rename_init(dir_renames);
2245 for (i = 0; i < pairs->nr; ++i) {
2246 struct string_list_item *item;
2247 int *count;
2248 struct diff_filepair *pair = pairs->queue[i];
2249 char *old_dir, *new_dir;
2250
2251 /* File not part of directory rename if it wasn't renamed */
2252 if (pair->status != 'R')
2253 continue;
2254
2255 get_renamed_dir_portion(pair->one->path, pair->two->path,
2256 &old_dir, &new_dir);
2257 if (!old_dir)
2258 /* Directory didn't change at all; ignore this one. */
2259 continue;
2260
2261 entry = dir_rename_find_entry(dir_renames, old_dir);
2262 if (!entry) {
2263 entry = xmalloc(sizeof(*entry));
2264 dir_rename_entry_init(entry, old_dir);
2265 hashmap_put(dir_renames, &entry->ent);
2266 } else {
2267 free(old_dir);
2268 }
2269 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2270 if (!item) {
2271 item = string_list_insert(&entry->possible_new_dirs,
2272 new_dir);
2273 item->util = xcalloc(1, sizeof(int));
2274 } else {
2275 free(new_dir);
2276 }
2277 count = item->util;
2278 *count += 1;
2279 }
2280
2281 /*
2282 * For each directory with files moved out of it, we find out which
2283 * target directory received the most files so we can declare it to
2284 * be the "winning" target location for the directory rename. This
2285 * winner gets recorded in new_dir. If there is no winner
2286 * (multiple target directories received the same number of files),
2287 * we set non_unique_new_dir. Once we've determined the winner (or
2288 * that there is no winner), we no longer need possible_new_dirs.
2289 */
2290 hashmap_for_each_entry(dir_renames, &iter, entry,
2291 ent /* member name */) {
2292 int max = 0;
2293 int bad_max = 0;
2294 char *best = NULL;
2295
2296 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2297 int *count = entry->possible_new_dirs.items[i].util;
2298
2299 if (*count == max)
2300 bad_max = max;
2301 else if (*count > max) {
2302 max = *count;
2303 best = entry->possible_new_dirs.items[i].string;
2304 }
2305 }
2306 if (bad_max == max)
2307 entry->non_unique_new_dir = 1;
2308 else {
2309 assert(entry->new_dir.len == 0);
2310 strbuf_addstr(&entry->new_dir, best);
2311 }
2312 /*
2313 * The relevant directory sub-portion of the original full
2314 * filepaths were xstrndup'ed before inserting into
2315 * possible_new_dirs, and instead of manually iterating the
2316 * list and free'ing each, just lie and tell
2317 * possible_new_dirs that it did the strdup'ing so that it
2318 * will free them for us.
2319 */
2320 entry->possible_new_dirs.strdup_strings = 1;
2321 string_list_clear(&entry->possible_new_dirs, 1);
2322 }
2323
2324 return dir_renames;
2325 }
2326
2327 static struct dir_rename_entry *check_dir_renamed(const char *path,
2328 struct hashmap *dir_renames)
2329 {
2330 char *temp = xstrdup(path);
2331 char *end;
2332 struct dir_rename_entry *entry = NULL;
2333
2334 while ((end = strrchr(temp, '/'))) {
2335 *end = '\0';
2336 entry = dir_rename_find_entry(dir_renames, temp);
2337 if (entry)
2338 break;
2339 }
2340 free(temp);
2341 return entry;
2342 }
2343
2344 static void compute_collisions(struct hashmap *collisions,
2345 struct hashmap *dir_renames,
2346 struct diff_queue_struct *pairs)
2347 {
2348 int i;
2349
2350 /*
2351 * Multiple files can be mapped to the same path due to directory
2352 * renames done by the other side of history. Since that other
2353 * side of history could have merged multiple directories into one,
2354 * if our side of history added the same file basename to each of
2355 * those directories, then all N of them would get implicitly
2356 * renamed by the directory rename detection into the same path,
2357 * and we'd get an add/add/.../add conflict, and all those adds
2358 * from *this* side of history. This is not representable in the
2359 * index, and users aren't going to easily be able to make sense of
2360 * it. So we need to provide a good warning about what's
2361 * happening, and fall back to no-directory-rename detection
2362 * behavior for those paths.
2363 *
2364 * See testcases 9e and all of section 5 from t6043 for examples.
2365 */
2366 collision_init(collisions);
2367
2368 for (i = 0; i < pairs->nr; ++i) {
2369 struct dir_rename_entry *dir_rename_ent;
2370 struct collision_entry *collision_ent;
2371 char *new_path;
2372 struct diff_filepair *pair = pairs->queue[i];
2373
2374 if (pair->status != 'A' && pair->status != 'R')
2375 continue;
2376 dir_rename_ent = check_dir_renamed(pair->two->path,
2377 dir_renames);
2378 if (!dir_rename_ent)
2379 continue;
2380
2381 new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2382 if (!new_path)
2383 /*
2384 * dir_rename_ent->non_unique_new_path is true, which
2385 * means there is no directory rename for us to use,
2386 * which means it won't cause us any additional
2387 * collisions.
2388 */
2389 continue;
2390 collision_ent = collision_find_entry(collisions, new_path);
2391 if (!collision_ent) {
2392 CALLOC_ARRAY(collision_ent, 1);
2393 hashmap_entry_init(&collision_ent->ent,
2394 strhash(new_path));
2395 hashmap_put(collisions, &collision_ent->ent);
2396 collision_ent->target_file = new_path;
2397 } else {
2398 free(new_path);
2399 }
2400 string_list_insert(&collision_ent->source_files,
2401 pair->two->path);
2402 }
2403 }
2404
2405 static char *check_for_directory_rename(struct merge_options *opt,
2406 const char *path,
2407 struct tree *tree,
2408 struct hashmap *dir_renames,
2409 struct hashmap *dir_rename_exclusions,
2410 struct hashmap *collisions,
2411 int *clean_merge)
2412 {
2413 char *new_path = NULL;
2414 struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2415 struct dir_rename_entry *oentry = NULL;
2416
2417 if (!entry)
2418 return new_path;
2419
2420 /*
2421 * This next part is a little weird. We do not want to do an
2422 * implicit rename into a directory we renamed on our side, because
2423 * that will result in a spurious rename/rename(1to2) conflict. An
2424 * example:
2425 * Base commit: dumbdir/afile, otherdir/bfile
2426 * Side 1: smrtdir/afile, otherdir/bfile
2427 * Side 2: dumbdir/afile, dumbdir/bfile
2428 * Here, while working on Side 1, we could notice that otherdir was
2429 * renamed/merged to dumbdir, and change the diff_filepair for
2430 * otherdir/bfile into a rename into dumbdir/bfile. However, Side
2431 * 2 will notice the rename from dumbdir to smrtdir, and do the
2432 * transitive rename to move it from dumbdir/bfile to
2433 * smrtdir/bfile. That gives us bfile in dumbdir vs being in
2434 * smrtdir, a rename/rename(1to2) conflict. We really just want
2435 * the file to end up in smrtdir. And the way to achieve that is
2436 * to not let Side1 do the rename to dumbdir, since we know that is
2437 * the source of one of our directory renames.
2438 *
2439 * That's why oentry and dir_rename_exclusions is here.
2440 *
2441 * As it turns out, this also prevents N-way transient rename
2442 * confusion; See testcases 9c and 9d of t6043.
2443 */
2444 oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2445 if (oentry) {
2446 output(opt, 1, _("WARNING: Avoiding applying %s -> %s rename "
2447 "to %s, because %s itself was renamed."),
2448 entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2449 } else {
2450 new_path = handle_path_level_conflicts(opt, path, entry,
2451 collisions, tree);
2452 *clean_merge &= (new_path != NULL);
2453 }
2454
2455 return new_path;
2456 }
2457
2458 static void apply_directory_rename_modifications(struct merge_options *opt,
2459 struct diff_filepair *pair,
2460 char *new_path,
2461 struct rename *re,
2462 struct tree *tree,
2463 struct tree *o_tree,
2464 struct tree *a_tree,
2465 struct tree *b_tree,
2466 struct string_list *entries)
2467 {
2468 struct string_list_item *item;
2469 int stage = (tree == a_tree ? 2 : 3);
2470 int update_wd;
2471
2472 /*
2473 * In all cases where we can do directory rename detection,
2474 * unpack_trees() will have read pair->two->path into the
2475 * index and the working copy. We need to remove it so that
2476 * we can instead place it at new_path. It is guaranteed to
2477 * not be untracked (unpack_trees() would have errored out
2478 * saying the file would have been overwritten), but it might
2479 * be dirty, though.
2480 */
2481 update_wd = !was_dirty(opt, pair->two->path);
2482 if (!update_wd)
2483 output(opt, 1, _("Refusing to lose dirty file at %s"),
2484 pair->two->path);
2485 remove_file(opt, 1, pair->two->path, !update_wd);
2486
2487 /* Find or create a new re->dst_entry */
2488 item = string_list_lookup(entries, new_path);
2489 if (item) {
2490 /*
2491 * Since we're renaming on this side of history, and it's
2492 * due to a directory rename on the other side of history
2493 * (which we only allow when the directory in question no
2494 * longer exists on the other side of history), the
2495 * original entry for re->dst_entry is no longer
2496 * necessary...
2497 */
2498 re->dst_entry->processed = 1;
2499
2500 /*
2501 * ...because we'll be using this new one.
2502 */
2503 re->dst_entry = item->util;
2504 } else {
2505 /*
2506 * re->dst_entry is for the before-dir-rename path, and we
2507 * need it to hold information for the after-dir-rename
2508 * path. Before creating a new entry, we need to mark the
2509 * old one as unnecessary (...unless it is shared by
2510 * src_entry, i.e. this didn't use to be a rename, in which
2511 * case we can just allow the normal processing to happen
2512 * for it).
2513 */
2514 if (pair->status == 'R')
2515 re->dst_entry->processed = 1;
2516
2517 re->dst_entry = insert_stage_data(opt->repo, new_path,
2518 o_tree, a_tree, b_tree,
2519 entries);
2520 item = string_list_insert(entries, new_path);
2521 item->util = re->dst_entry;
2522 }
2523
2524 /*
2525 * Update the stage_data with the information about the path we are
2526 * moving into place. That slot will be empty and available for us
2527 * to write to because of the collision checks in
2528 * handle_path_level_conflicts(). In other words,
2529 * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2530 * open for us to write to.
2531 *
2532 * It may be tempting to actually update the index at this point as
2533 * well, using update_stages_for_stage_data(), but as per the big
2534 * "NOTE" in update_stages(), doing so will modify the current
2535 * in-memory index which will break calls to would_lose_untracked()
2536 * that we need to make. Instead, we need to just make sure that
2537 * the various handle_rename_*() functions update the index
2538 * explicitly rather than relying on unpack_trees() to have done it.
2539 */
2540 get_tree_entry(opt->repo,
2541 &tree->object.oid,
2542 pair->two->path,
2543 &re->dst_entry->stages[stage].oid,
2544 &re->dst_entry->stages[stage].mode);
2545
2546 /*
2547 * Record the original change status (or 'type' of change). If it
2548 * was originally an add ('A'), this lets us differentiate later
2549 * between a RENAME_DELETE conflict and RENAME_VIA_DIR (they
2550 * otherwise look the same). If it was originally a rename ('R'),
2551 * this lets us remember and report accurately about the transitive
2552 * renaming that occurred via the directory rename detection. Also,
2553 * record the original destination name.
2554 */
2555 re->dir_rename_original_type = pair->status;
2556 re->dir_rename_original_dest = pair->two->path;
2557
2558 /*
2559 * We don't actually look at pair->status again, but it seems
2560 * pedagogically correct to adjust it.
2561 */
2562 pair->status = 'R';
2563
2564 /*
2565 * Finally, record the new location.
2566 */
2567 pair->two->path = new_path;
2568 }
2569
2570 /*
2571 * Get information of all renames which occurred in 'pairs', making use of
2572 * any implicit directory renames inferred from the other side of history.
2573 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2574 * to be able to associate the correct cache entries with the rename
2575 * information; tree is always equal to either a_tree or b_tree.
2576 */
2577 static struct string_list *get_renames(struct merge_options *opt,
2578 const char *branch,
2579 struct diff_queue_struct *pairs,
2580 struct hashmap *dir_renames,
2581 struct hashmap *dir_rename_exclusions,
2582 struct tree *tree,
2583 struct tree *o_tree,
2584 struct tree *a_tree,
2585 struct tree *b_tree,
2586 struct string_list *entries,
2587 int *clean_merge)
2588 {
2589 int i;
2590 struct hashmap collisions;
2591 struct hashmap_iter iter;
2592 struct collision_entry *e;
2593 struct string_list *renames;
2594
2595 compute_collisions(&collisions, dir_renames, pairs);
2596 CALLOC_ARRAY(renames, 1);
2597
2598 for (i = 0; i < pairs->nr; ++i) {
2599 struct string_list_item *item;
2600 struct rename *re;
2601 struct diff_filepair *pair = pairs->queue[i];
2602 char *new_path; /* non-NULL only with directory renames */
2603
2604 if (pair->status != 'A' && pair->status != 'R') {
2605 diff_free_filepair(pair);
2606 continue;
2607 }
2608 new_path = check_for_directory_rename(opt, pair->two->path, tree,
2609 dir_renames,
2610 dir_rename_exclusions,
2611 &collisions,
2612 clean_merge);
2613 if (pair->status != 'R' && !new_path) {
2614 diff_free_filepair(pair);
2615 continue;
2616 }
2617
2618 re = xmalloc(sizeof(*re));
2619 re->processed = 0;
2620 re->pair = pair;
2621 re->branch = branch;
2622 re->dir_rename_original_type = '\0';
2623 re->dir_rename_original_dest = NULL;
2624 item = string_list_lookup(entries, re->pair->one->path);
2625 if (!item)
2626 re->src_entry = insert_stage_data(opt->repo,
2627 re->pair->one->path,
2628 o_tree, a_tree, b_tree, entries);
2629 else
2630 re->src_entry = item->util;
2631
2632 item = string_list_lookup(entries, re->pair->two->path);
2633 if (!item)
2634 re->dst_entry = insert_stage_data(opt->repo,
2635 re->pair->two->path,
2636 o_tree, a_tree, b_tree, entries);
2637 else
2638 re->dst_entry = item->util;
2639 item = string_list_insert(renames, pair->one->path);
2640 item->util = re;
2641 if (new_path)
2642 apply_directory_rename_modifications(opt, pair, new_path,
2643 re, tree, o_tree,
2644 a_tree, b_tree,
2645 entries);
2646 }
2647
2648 hashmap_for_each_entry(&collisions, &iter, e,
2649 ent /* member name */) {
2650 free(e->target_file);
2651 string_list_clear(&e->source_files, 0);
2652 }
2653 hashmap_clear_and_free(&collisions, struct collision_entry, ent);
2654 return renames;
2655 }
2656
2657 static int process_renames(struct merge_options *opt,
2658 struct string_list *a_renames,
2659 struct string_list *b_renames)
2660 {
2661 int clean_merge = 1, i, j;
2662 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2663 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2664 const struct rename *sre;
2665
2666 for (i = 0; i < a_renames->nr; i++) {
2667 sre = a_renames->items[i].util;
2668 string_list_insert(&a_by_dst, sre->pair->two->path)->util
2669 = (void *)sre;
2670 }
2671 for (i = 0; i < b_renames->nr; i++) {
2672 sre = b_renames->items[i].util;
2673 string_list_insert(&b_by_dst, sre->pair->two->path)->util
2674 = (void *)sre;
2675 }
2676
2677 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2678 struct string_list *renames1, *renames2Dst;
2679 struct rename *ren1 = NULL, *ren2 = NULL;
2680 const char *ren1_src, *ren1_dst;
2681 struct string_list_item *lookup;
2682
2683 if (i >= a_renames->nr) {
2684 ren2 = b_renames->items[j++].util;
2685 } else if (j >= b_renames->nr) {
2686 ren1 = a_renames->items[i++].util;
2687 } else {
2688 int compare = strcmp(a_renames->items[i].string,
2689 b_renames->items[j].string);
2690 if (compare <= 0)
2691 ren1 = a_renames->items[i++].util;
2692 if (compare >= 0)
2693 ren2 = b_renames->items[j++].util;
2694 }
2695
2696 /* TODO: refactor, so that 1/2 are not needed */
2697 if (ren1) {
2698 renames1 = a_renames;
2699 renames2Dst = &b_by_dst;
2700 } else {
2701 renames1 = b_renames;
2702 renames2Dst = &a_by_dst;
2703 SWAP(ren2, ren1);
2704 }
2705
2706 if (ren1->processed)
2707 continue;
2708 ren1->processed = 1;
2709 ren1->dst_entry->processed = 1;
2710 /* BUG: We should only mark src_entry as processed if we
2711 * are not dealing with a rename + add-source case.
2712 */
2713 ren1->src_entry->processed = 1;
2714
2715 ren1_src = ren1->pair->one->path;
2716 ren1_dst = ren1->pair->two->path;
2717
2718 if (ren2) {
2719 /* One file renamed on both sides */
2720 const char *ren2_src = ren2->pair->one->path;
2721 const char *ren2_dst = ren2->pair->two->path;
2722 enum rename_type rename_type;
2723 if (strcmp(ren1_src, ren2_src) != 0)
2724 BUG("ren1_src != ren2_src");
2725 ren2->dst_entry->processed = 1;
2726 ren2->processed = 1;
2727 if (strcmp(ren1_dst, ren2_dst) != 0) {
2728 rename_type = RENAME_ONE_FILE_TO_TWO;
2729 clean_merge = 0;
2730 } else {
2731 rename_type = RENAME_ONE_FILE_TO_ONE;
2732 /* BUG: We should only remove ren1_src in
2733 * the base stage (think of rename +
2734 * add-source cases).
2735 */
2736 remove_file(opt, 1, ren1_src, 1);
2737 update_entry(ren1->dst_entry,
2738 ren1->pair->one,
2739 ren1->pair->two,
2740 ren2->pair->two);
2741 }
2742 setup_rename_conflict_info(rename_type, opt, ren1, ren2);
2743 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2744 /* Two different files renamed to the same thing */
2745 char *ren2_dst;
2746 ren2 = lookup->util;
2747 ren2_dst = ren2->pair->two->path;
2748 if (strcmp(ren1_dst, ren2_dst) != 0)
2749 BUG("ren1_dst != ren2_dst");
2750
2751 clean_merge = 0;
2752 ren2->processed = 1;
2753 /*
2754 * BUG: We should only mark src_entry as processed
2755 * if we are not dealing with a rename + add-source
2756 * case.
2757 */
2758 ren2->src_entry->processed = 1;
2759
2760 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2761 opt, ren1, ren2);
2762 } else {
2763 /* Renamed in 1, maybe changed in 2 */
2764 /* we only use sha1 and mode of these */
2765 struct diff_filespec src_other, dst_other;
2766 int try_merge;
2767
2768 /*
2769 * unpack_trees loads entries from common-commit
2770 * into stage 1, from head-commit into stage 2, and
2771 * from merge-commit into stage 3. We keep track
2772 * of which side corresponds to the rename.
2773 */
2774 int renamed_stage = a_renames == renames1 ? 2 : 3;
2775 int other_stage = a_renames == renames1 ? 3 : 2;
2776
2777 /* BUG: We should only remove ren1_src in the base
2778 * stage and in other_stage (think of rename +
2779 * add-source case).
2780 */
2781 remove_file(opt, 1, ren1_src,
2782 renamed_stage == 2 || !was_tracked(opt, ren1_src));
2783
2784 oidcpy(&src_other.oid,
2785 &ren1->src_entry->stages[other_stage].oid);
2786 src_other.mode = ren1->src_entry->stages[other_stage].mode;
2787 oidcpy(&dst_other.oid,
2788 &ren1->dst_entry->stages[other_stage].oid);
2789 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2790 try_merge = 0;
2791
2792 if (oideq(&src_other.oid, &null_oid) &&
2793 ren1->dir_rename_original_type == 'A') {
2794 setup_rename_conflict_info(RENAME_VIA_DIR,
2795 opt, ren1, NULL);
2796 } else if (oideq(&src_other.oid, &null_oid)) {
2797 setup_rename_conflict_info(RENAME_DELETE,
2798 opt, ren1, NULL);
2799 } else if ((dst_other.mode == ren1->pair->two->mode) &&
2800 oideq(&dst_other.oid, &ren1->pair->two->oid)) {
2801 /*
2802 * Added file on the other side identical to
2803 * the file being renamed: clean merge.
2804 * Also, there is no need to overwrite the
2805 * file already in the working copy, so call
2806 * update_file_flags() instead of
2807 * update_file().
2808 */
2809 if (update_file_flags(opt,
2810 ren1->pair->two,
2811 ren1_dst,
2812 1, /* update_cache */
2813 0 /* update_wd */))
2814 clean_merge = -1;
2815 } else if (!oideq(&dst_other.oid, &null_oid)) {
2816 /*
2817 * Probably not a clean merge, but it's
2818 * premature to set clean_merge to 0 here,
2819 * because if the rename merges cleanly and
2820 * the merge exactly matches the newly added
2821 * file, then the merge will be clean.
2822 */
2823 setup_rename_conflict_info(RENAME_ADD,
2824 opt, ren1, NULL);
2825 } else
2826 try_merge = 1;
2827
2828 if (clean_merge < 0)
2829 goto cleanup_and_return;
2830 if (try_merge) {
2831 struct diff_filespec *o, *a, *b;
2832 src_other.path = (char *)ren1_src;
2833
2834 o = ren1->pair->one;
2835 if (a_renames == renames1) {
2836 a = ren1->pair->two;
2837 b = &src_other;
2838 } else {
2839 b = ren1->pair->two;
2840 a = &src_other;
2841 }
2842 update_entry(ren1->dst_entry, o, a, b);
2843 setup_rename_conflict_info(RENAME_NORMAL,
2844 opt, ren1, NULL);
2845 }
2846 }
2847 }
2848 cleanup_and_return:
2849 string_list_clear(&a_by_dst, 0);
2850 string_list_clear(&b_by_dst, 0);
2851
2852 return clean_merge;
2853 }
2854
2855 struct rename_info {
2856 struct string_list *head_renames;
2857 struct string_list *merge_renames;
2858 };
2859
2860 static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2861 struct hashmap *dir_renames)
2862 {
2863 struct hashmap_iter iter;
2864 struct dir_rename_entry *e;
2865
2866 hashmap_for_each_entry(dir_renames, &iter, e,
2867 ent /* member name */) {
2868 free(e->dir);
2869 strbuf_release(&e->new_dir);
2870 /* possible_new_dirs already cleared in get_directory_renames */
2871 }
2872 hashmap_clear_and_free(dir_renames, struct dir_rename_entry, ent);
2873 free(dir_renames);
2874
2875 free(pairs->queue);
2876 free(pairs);
2877 }
2878
2879 static int detect_and_process_renames(struct merge_options *opt,
2880 struct tree *common,
2881 struct tree *head,
2882 struct tree *merge,
2883 struct string_list *entries,
2884 struct rename_info *ri)
2885 {
2886 struct diff_queue_struct *head_pairs, *merge_pairs;
2887 struct hashmap *dir_re_head, *dir_re_merge;
2888 int clean = 1;
2889
2890 ri->head_renames = NULL;
2891 ri->merge_renames = NULL;
2892
2893 if (!merge_detect_rename(opt))
2894 return 1;
2895
2896 head_pairs = get_diffpairs(opt, common, head);
2897 merge_pairs = get_diffpairs(opt, common, merge);
2898
2899 if ((opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE) ||
2900 (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
2901 !opt->priv->call_depth)) {
2902 dir_re_head = get_directory_renames(head_pairs);
2903 dir_re_merge = get_directory_renames(merge_pairs);
2904
2905 handle_directory_level_conflicts(opt,
2906 dir_re_head, head,
2907 dir_re_merge, merge);
2908 } else {
2909 dir_re_head = xmalloc(sizeof(*dir_re_head));
2910 dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2911 dir_rename_init(dir_re_head);
2912 dir_rename_init(dir_re_merge);
2913 }
2914
2915 ri->head_renames = get_renames(opt, opt->branch1, head_pairs,
2916 dir_re_merge, dir_re_head, head,
2917 common, head, merge, entries,
2918 &clean);
2919 if (clean < 0)
2920 goto cleanup;
2921 ri->merge_renames = get_renames(opt, opt->branch2, merge_pairs,
2922 dir_re_head, dir_re_merge, merge,
2923 common, head, merge, entries,
2924 &clean);
2925 if (clean < 0)
2926 goto cleanup;
2927 clean &= process_renames(opt, ri->head_renames, ri->merge_renames);
2928
2929 cleanup:
2930 /*
2931 * Some cleanup is deferred until cleanup_renames() because the
2932 * data structures are still needed and referenced in
2933 * process_entry(). But there are a few things we can free now.
2934 */
2935 initial_cleanup_rename(head_pairs, dir_re_head);
2936 initial_cleanup_rename(merge_pairs, dir_re_merge);
2937
2938 return clean;
2939 }
2940
2941 static void final_cleanup_rename(struct string_list *rename)
2942 {
2943 const struct rename *re;
2944 int i;
2945
2946 if (rename == NULL)
2947 return;
2948
2949 for (i = 0; i < rename->nr; i++) {
2950 re = rename->items[i].util;
2951 diff_free_filepair(re->pair);
2952 }
2953 string_list_clear(rename, 1);
2954 free(rename);
2955 }
2956
2957 static void final_cleanup_renames(struct rename_info *re_info)
2958 {
2959 final_cleanup_rename(re_info->head_renames);
2960 final_cleanup_rename(re_info->merge_renames);
2961 }
2962
2963 static int read_oid_strbuf(struct merge_options *opt,
2964 const struct object_id *oid,
2965 struct strbuf *dst)
2966 {
2967 void *buf;
2968 enum object_type type;
2969 unsigned long size;
2970 buf = read_object_file(oid, &type, &size);
2971 if (!buf)
2972 return err(opt, _("cannot read object %s"), oid_to_hex(oid));
2973 if (type != OBJ_BLOB) {
2974 free(buf);
2975 return err(opt, _("object %s is not a blob"), oid_to_hex(oid));
2976 }
2977 strbuf_attach(dst, buf, size, size + 1);
2978 return 0;
2979 }
2980
2981 static int blob_unchanged(struct merge_options *opt,
2982 const struct diff_filespec *o,
2983 const struct diff_filespec *a,
2984 int renormalize, const char *path)
2985 {
2986 struct strbuf obuf = STRBUF_INIT;
2987 struct strbuf abuf = STRBUF_INIT;
2988 int ret = 0; /* assume changed for safety */
2989 const struct index_state *idx = opt->repo->index;
2990
2991 if (a->mode != o->mode)
2992 return 0;
2993 if (oideq(&o->oid, &a->oid))
2994 return 1;
2995 if (!renormalize)
2996 return 0;
2997
2998 if (read_oid_strbuf(opt, &o->oid, &obuf) ||
2999 read_oid_strbuf(opt, &a->oid, &abuf))
3000 goto error_return;
3001 /*
3002 * Note: binary | is used so that both renormalizations are
3003 * performed. Comparison can be skipped if both files are
3004 * unchanged since their sha1s have already been compared.
3005 */
3006 if (renormalize_buffer(idx, path, obuf.buf, obuf.len, &obuf) |
3007 renormalize_buffer(idx, path, abuf.buf, abuf.len, &abuf))
3008 ret = (obuf.len == abuf.len && !memcmp(obuf.buf, abuf.buf, obuf.len));
3009
3010 error_return:
3011 strbuf_release(&obuf);
3012 strbuf_release(&abuf);
3013 return ret;
3014 }
3015
3016 static int handle_modify_delete(struct merge_options *opt,
3017 const char *path,
3018 const struct diff_filespec *o,
3019 const struct diff_filespec *a,
3020 const struct diff_filespec *b)
3021 {
3022 const char *modify_branch, *delete_branch;
3023 const struct diff_filespec *changed;
3024
3025 if (is_valid(a)) {
3026 modify_branch = opt->branch1;
3027 delete_branch = opt->branch2;
3028 changed = a;
3029 } else {
3030 modify_branch = opt->branch2;
3031 delete_branch = opt->branch1;
3032 changed = b;
3033 }
3034
3035 return handle_change_delete(opt,
3036 path, NULL,
3037 o, changed,
3038 modify_branch, delete_branch,
3039 _("modify"), _("modified"));
3040 }
3041
3042 static int handle_content_merge(struct merge_file_info *mfi,
3043 struct merge_options *opt,
3044 const char *path,
3045 int is_dirty,
3046 const struct diff_filespec *o,
3047 const struct diff_filespec *a,
3048 const struct diff_filespec *b,
3049 struct rename_conflict_info *ci)
3050 {
3051 const char *reason = _("content");
3052 unsigned df_conflict_remains = 0;
3053
3054 if (!is_valid(o))
3055 reason = _("add/add");
3056
3057 assert(o->path && a->path && b->path);
3058 if (ci && dir_in_way(opt->repo->index, path, !opt->priv->call_depth,
3059 S_ISGITLINK(ci->ren1->pair->two->mode)))
3060 df_conflict_remains = 1;
3061
3062 if (merge_mode_and_contents(opt, o, a, b, path,
3063 opt->branch1, opt->branch2,
3064 opt->priv->call_depth * 2, mfi))
3065 return -1;
3066
3067 /*
3068 * We can skip updating the working tree file iff:
3069 * a) The merge is clean
3070 * b) The merge matches what was in HEAD (content, mode, pathname)
3071 * c) The target path is usable (i.e. not involved in D/F conflict)
3072 */
3073 if (mfi->clean && was_tracked_and_matches(opt, path, &mfi->blob) &&
3074 !df_conflict_remains) {
3075 int pos;
3076 struct cache_entry *ce;
3077
3078 output(opt, 3, _("Skipped %s (merged same as existing)"), path);
3079 if (add_cacheinfo(opt, &mfi->blob, path,
3080 0, (!opt->priv->call_depth && !is_dirty), 0))
3081 return -1;
3082 /*
3083 * However, add_cacheinfo() will delete the old cache entry
3084 * and add a new one. We need to copy over any skip_worktree
3085 * flag to avoid making the file appear as if it were
3086 * deleted by the user.
3087 */
3088 pos = index_name_pos(&opt->priv->orig_index, path, strlen(path));
3089 ce = opt->priv->orig_index.cache[pos];
3090 if (ce_skip_worktree(ce)) {
3091 pos = index_name_pos(opt->repo->index, path, strlen(path));
3092 ce = opt->repo->index->cache[pos];
3093 ce->ce_flags |= CE_SKIP_WORKTREE;
3094 }
3095 return mfi->clean;
3096 }
3097
3098 if (!mfi->clean) {
3099 if (S_ISGITLINK(mfi->blob.mode))
3100 reason = _("submodule");
3101 output(opt, 1, _("CONFLICT (%s): Merge conflict in %s"),
3102 reason, path);
3103 if (ci && !df_conflict_remains)
3104 if (update_stages(opt, path, o, a, b))
3105 return -1;
3106 }
3107
3108 if (df_conflict_remains || is_dirty) {
3109 char *new_path;
3110 if (opt->priv->call_depth) {
3111 remove_file_from_index(opt->repo->index, path);
3112 } else {
3113 if (!mfi->clean) {
3114 if (update_stages(opt, path, o, a, b))
3115 return -1;
3116 } else {
3117 int file_from_stage2 = was_tracked(opt, path);
3118
3119 if (update_stages(opt, path, NULL,
3120 file_from_stage2 ? &mfi->blob : NULL,
3121 file_from_stage2 ? NULL : &mfi->blob))
3122 return -1;
3123 }
3124
3125 }
3126 new_path = unique_path(opt, path, ci->ren1->branch);
3127 if (is_dirty) {
3128 output(opt, 1, _("Refusing to lose dirty file at %s"),
3129 path);
3130 }
3131 output(opt, 1, _("Adding as %s instead"), new_path);
3132 if (update_file(opt, 0, &mfi->blob, new_path)) {
3133 free(new_path);
3134 return -1;
3135 }
3136 free(new_path);
3137 mfi->clean = 0;
3138 } else if (update_file(opt, mfi->clean, &mfi->blob, path))
3139 return -1;
3140 return !is_dirty && mfi->clean;
3141 }
3142
3143 static int handle_rename_normal(struct merge_options *opt,
3144 const char *path,
3145 const struct diff_filespec *o,
3146 const struct diff_filespec *a,
3147 const struct diff_filespec *b,
3148 struct rename_conflict_info *ci)
3149 {
3150 struct rename *ren = ci->ren1;
3151 struct merge_file_info mfi;
3152 int clean;
3153 int side = (ren->branch == opt->branch1 ? 2 : 3);
3154
3155 /* Merge the content and write it out */
3156 clean = handle_content_merge(&mfi, opt, path, was_dirty(opt, path),
3157 o, a, b, ci);
3158
3159 if (clean &&
3160 opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_CONFLICT &&
3161 ren->dir_rename_original_dest) {
3162 if (update_stages(opt, path,
3163 NULL,
3164 side == 2 ? &mfi.blob : NULL,
3165 side == 2 ? NULL : &mfi.blob))
3166 return -1;
3167 clean = 0; /* not clean, but conflicted */
3168 }
3169 return clean;
3170 }
3171
3172 static void dir_rename_warning(const char *msg,
3173 int is_add,
3174 int clean,
3175 struct merge_options *opt,
3176 struct rename *ren)
3177 {
3178 const char *other_branch;
3179 other_branch = (ren->branch == opt->branch1 ?
3180 opt->branch2 : opt->branch1);
3181 if (is_add) {
3182 output(opt, clean ? 2 : 1, msg,
3183 ren->pair->one->path, ren->branch,
3184 other_branch, ren->pair->two->path);
3185 return;
3186 }
3187 output(opt, clean ? 2 : 1, msg,
3188 ren->pair->one->path, ren->dir_rename_original_dest, ren->branch,
3189 other_branch, ren->pair->two->path);
3190 }
3191 static int warn_about_dir_renamed_entries(struct merge_options *opt,
3192 struct rename *ren)
3193 {
3194 const char *msg;
3195 int clean = 1, is_add;
3196
3197 if (!ren)
3198 return clean;
3199
3200 /* Return early if ren was not affected/created by a directory rename */
3201 if (!ren->dir_rename_original_dest)
3202 return clean;
3203
3204 /* Sanity checks */
3205 assert(opt->detect_directory_renames > MERGE_DIRECTORY_RENAMES_NONE);
3206 assert(ren->dir_rename_original_type == 'A' ||
3207 ren->dir_rename_original_type == 'R');
3208
3209 /* Check whether to treat directory renames as a conflict */
3210 clean = (opt->detect_directory_renames == MERGE_DIRECTORY_RENAMES_TRUE);
3211
3212 is_add = (ren->dir_rename_original_type == 'A');
3213 if (ren->dir_rename_original_type == 'A' && clean) {
3214 msg = _("Path updated: %s added in %s inside a "
3215 "directory that was renamed in %s; moving it to %s.");
3216 } else if (ren->dir_rename_original_type == 'A' && !clean) {
3217 msg = _("CONFLICT (file location): %s added in %s "
3218 "inside a directory that was renamed in %s, "
3219 "suggesting it should perhaps be moved to %s.");
3220 } else if (ren->dir_rename_original_type == 'R' && clean) {
3221 msg = _("Path updated: %s renamed to %s in %s, inside a "
3222 "directory that was renamed in %s; moving it to %s.");
3223 } else if (ren->dir_rename_original_type == 'R' && !clean) {
3224 msg = _("CONFLICT (file location): %s renamed to %s in %s, "
3225 "inside a directory that was renamed in %s, "
3226 "suggesting it should perhaps be moved to %s.");
3227 } else {
3228 BUG("Impossible dir_rename_original_type/clean combination");
3229 }
3230 dir_rename_warning(msg, is_add, clean, opt, ren);
3231
3232 return clean;
3233 }
3234
3235 /* Per entry merge function */
3236 static int process_entry(struct merge_options *opt,
3237 const char *path, struct stage_data *entry)
3238 {
3239 int clean_merge = 1;
3240 int normalize = opt->renormalize;
3241
3242 struct diff_filespec *o = &entry->stages[1];
3243 struct diff_filespec *a = &entry->stages[2];
3244 struct diff_filespec *b = &entry->stages[3];
3245 int o_valid = is_valid(o);
3246 int a_valid = is_valid(a);
3247 int b_valid = is_valid(b);
3248 o->path = a->path = b->path = (char*)path;
3249
3250 entry->processed = 1;
3251 if (entry->rename_conflict_info) {
3252 struct rename_conflict_info *ci = entry->rename_conflict_info;
3253 struct diff_filespec *temp;
3254 int path_clean;
3255
3256 path_clean = warn_about_dir_renamed_entries(opt, ci->ren1);
3257 path_clean &= warn_about_dir_renamed_entries(opt, ci->ren2);
3258
3259 /*
3260 * For cases with a single rename, {o,a,b}->path have all been
3261 * set to the rename target path; we need to set two of these
3262 * back to the rename source.
3263 * For rename/rename conflicts, we'll manually fix paths below.
3264 */
3265 temp = (opt->branch1 == ci->ren1->branch) ? b : a;
3266 o->path = temp->path = ci->ren1->pair->one->path;
3267 if (ci->ren2) {
3268 assert(opt->branch1 == ci->ren1->branch);
3269 }
3270
3271 switch (ci->rename_type) {
3272 case RENAME_NORMAL:
3273 case RENAME_ONE_FILE_TO_ONE:
3274 clean_merge = handle_rename_normal(opt, path, o, a, b,
3275 ci);
3276 break;
3277 case RENAME_VIA_DIR:
3278 clean_merge = handle_rename_via_dir(opt, ci);
3279 break;
3280 case RENAME_ADD:
3281 /*
3282 * Probably unclean merge, but if the renamed file
3283 * merges cleanly and the result can then be
3284 * two-way merged cleanly with the added file, I
3285 * guess it's a clean merge?
3286 */
3287 clean_merge = handle_rename_add(opt, ci);
3288 break;
3289 case RENAME_DELETE:
3290 clean_merge = 0;
3291 if (handle_rename_delete(opt, ci))
3292 clean_merge = -1;
3293 break;
3294 case RENAME_ONE_FILE_TO_TWO:
3295 /*
3296 * Manually fix up paths; note:
3297 * ren[12]->pair->one->path are equal.
3298 */
3299 o->path = ci->ren1->pair->one->path;
3300 a->path = ci->ren1->pair->two->path;
3301 b->path = ci->ren2->pair->two->path;
3302
3303 clean_merge = 0;
3304 if (handle_rename_rename_1to2(opt, ci))
3305 clean_merge = -1;
3306 break;
3307 case RENAME_TWO_FILES_TO_ONE:
3308 /*
3309 * Manually fix up paths; note,
3310 * ren[12]->pair->two->path are actually equal.
3311 */
3312 o->path = NULL;
3313 a->path = ci->ren1->pair->two->path;
3314 b->path = ci->ren2->pair->two->path;
3315
3316 /*
3317 * Probably unclean merge, but if the two renamed
3318 * files merge cleanly and the two resulting files
3319 * can then be two-way merged cleanly, I guess it's
3320 * a clean merge?
3321 */
3322 clean_merge = handle_rename_rename_2to1(opt, ci);
3323 break;
3324 default:
3325 entry->processed = 0;
3326 break;
3327 }
3328 if (path_clean < clean_merge)
3329 clean_merge = path_clean;
3330 } else if (o_valid && (!a_valid || !b_valid)) {
3331 /* Case A: Deleted in one */
3332 if ((!a_valid && !b_valid) ||
3333 (!b_valid && blob_unchanged(opt, o, a, normalize, path)) ||
3334 (!a_valid && blob_unchanged(opt, o, b, normalize, path))) {
3335 /* Deleted in both or deleted in one and
3336 * unchanged in the other */
3337 if (a_valid)
3338 output(opt, 2, _("Removing %s"), path);
3339 /* do not touch working file if it did not exist */
3340 remove_file(opt, 1, path, !a_valid);
3341 } else {
3342 /* Modify/delete; deleted side may have put a directory in the way */
3343 clean_merge = 0;
3344 if (handle_modify_delete(opt, path, o, a, b))
3345 clean_merge = -1;
3346 }
3347 } else if ((!o_valid && a_valid && !b_valid) ||
3348 (!o_valid && !a_valid && b_valid)) {
3349 /* Case B: Added in one. */
3350 /* [nothing|directory] -> ([nothing|directory], file) */
3351
3352 const char *add_branch;
3353 const char *other_branch;
3354 const char *conf;
3355 const struct diff_filespec *contents;
3356
3357 if (a_valid) {
3358 add_branch = opt->branch1;
3359 other_branch = opt->branch2;
3360 contents = a;
3361 conf = _("file/directory");
3362 } else {
3363 add_branch = opt->branch2;
3364 other_branch = opt->branch1;
3365 contents = b;
3366 conf = _("directory/file");
3367 }
3368 if (dir_in_way(opt->repo->index, path,
3369 !opt->priv->call_depth && !S_ISGITLINK(a->mode),
3370 0)) {
3371 char *new_path = unique_path(opt, path, add_branch);
3372 clean_merge = 0;
3373 output(opt, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3374 "Adding %s as %s"),
3375 conf, path, other_branch, path, new_path);
3376 if (update_file(opt, 0, contents, new_path))
3377 clean_merge = -1;
3378 else if (opt->priv->call_depth)
3379 remove_file_from_index(opt->repo->index, path);
3380 free(new_path);
3381 } else {
3382 output(opt, 2, _("Adding %s"), path);
3383 /* do not overwrite file if already present */
3384 if (update_file_flags(opt, contents, path, 1, !a_valid))
3385 clean_merge = -1;
3386 }
3387 } else if (a_valid && b_valid) {
3388 if (!o_valid) {
3389 /* Case C: Added in both (check for same permissions) */
3390 output(opt, 1,
3391 _("CONFLICT (add/add): Merge conflict in %s"),
3392 path);
3393 clean_merge = handle_file_collision(opt,
3394 path, NULL, NULL,
3395 opt->branch1,
3396 opt->branch2,
3397 a, b);
3398 } else {
3399 /* case D: Modified in both, but differently. */
3400 struct merge_file_info mfi;
3401 int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3402 clean_merge = handle_content_merge(&mfi, opt, path,
3403 is_dirty,
3404 o, a, b, NULL);
3405 }
3406 } else if (!o_valid && !a_valid && !b_valid) {
3407 /*
3408 * this entry was deleted altogether. a_mode == 0 means
3409 * we had that path and want to actively remove it.
3410 */
3411 remove_file(opt, 1, path, !a->mode);
3412 } else
3413 BUG("fatal merge failure, shouldn't happen.");
3414
3415 return clean_merge;
3416 }
3417
3418 static int merge_trees_internal(struct merge_options *opt,
3419 struct tree *head,
3420 struct tree *merge,
3421 struct tree *merge_base,
3422 struct tree **result)
3423 {
3424 struct index_state *istate = opt->repo->index;
3425 int code, clean;
3426
3427 if (opt->subtree_shift) {
3428 merge = shift_tree_object(opt->repo, head, merge,
3429 opt->subtree_shift);
3430 merge_base = shift_tree_object(opt->repo, head, merge_base,
3431 opt->subtree_shift);
3432 }
3433
3434 if (oideq(&merge_base->object.oid, &merge->object.oid)) {
3435 output(opt, 0, _("Already up to date!"));
3436 *result = head;
3437 return 1;
3438 }
3439
3440 code = unpack_trees_start(opt, merge_base, head, merge);
3441
3442 if (code != 0) {
3443 if (show(opt, 4) || opt->priv->call_depth)
3444 err(opt, _("merging of trees %s and %s failed"),
3445 oid_to_hex(&head->object.oid),
3446 oid_to_hex(&merge->object.oid));
3447 unpack_trees_finish(opt);
3448 return -1;
3449 }
3450
3451 if (unmerged_index(istate)) {
3452 struct string_list *entries;
3453 struct rename_info re_info;
3454 int i;
3455 /*
3456 * Only need the hashmap while processing entries, so
3457 * initialize it here and free it when we are done running
3458 * through the entries. Keeping it in the merge_options as
3459 * opposed to decaring a local hashmap is for convenience
3460 * so that we don't have to pass it to around.
3461 */
3462 hashmap_init(&opt->priv->current_file_dir_set, path_hashmap_cmp,
3463 NULL, 512);
3464 get_files_dirs(opt, head);
3465 get_files_dirs(opt, merge);
3466
3467 entries = get_unmerged(opt->repo->index);
3468 clean = detect_and_process_renames(opt, merge_base, head, merge,
3469 entries, &re_info);
3470 record_df_conflict_files(opt, entries);
3471 if (clean < 0)
3472 goto cleanup;
3473 for (i = entries->nr-1; 0 <= i; i--) {
3474 const char *path = entries->items[i].string;
3475 struct stage_data *e = entries->items[i].util;
3476 if (!e->processed) {
3477 int ret = process_entry(opt, path, e);
3478 if (!ret)
3479 clean = 0;
3480 else if (ret < 0) {
3481 clean = ret;
3482 goto cleanup;
3483 }
3484 }
3485 }
3486 for (i = 0; i < entries->nr; i++) {
3487 struct stage_data *e = entries->items[i].util;
3488 if (!e->processed)
3489 BUG("unprocessed path??? %s",
3490 entries->items[i].string);
3491 }
3492
3493 cleanup:
3494 final_cleanup_renames(&re_info);
3495
3496 string_list_clear(entries, 1);
3497 free(entries);
3498
3499 hashmap_clear_and_free(&opt->priv->current_file_dir_set,
3500 struct path_hashmap_entry, e);
3501
3502 if (clean < 0) {
3503 unpack_trees_finish(opt);
3504 return clean;
3505 }
3506 }
3507 else
3508 clean = 1;
3509
3510 unpack_trees_finish(opt);
3511
3512 if (opt->priv->call_depth &&
3513 !(*result = write_in_core_index_as_tree(opt->repo)))
3514 return -1;
3515
3516 return clean;
3517 }
3518
3519 /*
3520 * Merge the commits h1 and h2, returning a flag (int) indicating the
3521 * cleanness of the merge. Also, if opt->priv->call_depth, create a
3522 * virtual commit and write its location to *result.
3523 */
3524 static int merge_recursive_internal(struct merge_options *opt,
3525 struct commit *h1,
3526 struct commit *h2,
3527 struct commit_list *merge_bases,
3528 struct commit **result)
3529 {
3530 struct commit_list *iter;
3531 struct commit *merged_merge_bases;
3532 struct tree *result_tree;
3533 int clean;
3534 const char *ancestor_name;
3535 struct strbuf merge_base_abbrev = STRBUF_INIT;
3536
3537 if (show(opt, 4)) {
3538 output(opt, 4, _("Merging:"));
3539 output_commit_title(opt, h1);
3540 output_commit_title(opt, h2);
3541 }
3542
3543 if (!merge_bases) {
3544 merge_bases = get_merge_bases(h1, h2);
3545 merge_bases = reverse_commit_list(merge_bases);
3546 }
3547
3548 if (show(opt, 5)) {
3549 unsigned cnt = commit_list_count(merge_bases);
3550
3551 output(opt, 5, Q_("found %u common ancestor:",
3552 "found %u common ancestors:", cnt), cnt);
3553 for (iter = merge_bases; iter; iter = iter->next)
3554 output_commit_title(opt, iter->item);
3555 }
3556
3557 merged_merge_bases = pop_commit(&merge_bases);
3558 if (merged_merge_bases == NULL) {
3559 /* if there is no common ancestor, use an empty tree */
3560 struct tree *tree;
3561
3562 tree = lookup_tree(opt->repo, opt->repo->hash_algo->empty_tree);
3563 merged_merge_bases = make_virtual_commit(opt->repo, tree,
3564 "ancestor");
3565 ancestor_name = "empty tree";
3566 } else if (opt->ancestor && !opt->priv->call_depth) {
3567 ancestor_name = opt->ancestor;
3568 } else if (merge_bases) {
3569 ancestor_name = "merged common ancestors";
3570 } else {
3571 strbuf_add_unique_abbrev(&merge_base_abbrev,
3572 &merged_merge_bases->object.oid,
3573 DEFAULT_ABBREV);
3574 ancestor_name = merge_base_abbrev.buf;
3575 }
3576
3577 for (iter = merge_bases; iter; iter = iter->next) {
3578 const char *saved_b1, *saved_b2;
3579 opt->priv->call_depth++;
3580 /*
3581 * When the merge fails, the result contains files
3582 * with conflict markers. The cleanness flag is
3583 * ignored (unless indicating an error), it was never
3584 * actually used, as result of merge_trees has always
3585 * overwritten it: the committed "conflicts" were
3586 * already resolved.
3587 */
3588 discard_index(opt->repo->index);
3589 saved_b1 = opt->branch1;
3590 saved_b2 = opt->branch2;
3591 opt->branch1 = "Temporary merge branch 1";
3592 opt->branch2 = "Temporary merge branch 2";
3593 if (merge_recursive_internal(opt, merged_merge_bases, iter->item,
3594 NULL, &merged_merge_bases) < 0)
3595 return -1;
3596 opt->branch1 = saved_b1;
3597 opt->branch2 = saved_b2;
3598 opt->priv->call_depth--;
3599
3600 if (!merged_merge_bases)
3601 return err(opt, _("merge returned no commit"));
3602 }
3603
3604 discard_index(opt->repo->index);
3605 if (!opt->priv->call_depth)
3606 repo_read_index(opt->repo);
3607
3608 opt->ancestor = ancestor_name;
3609 clean = merge_trees_internal(opt,
3610 repo_get_commit_tree(opt->repo, h1),
3611 repo_get_commit_tree(opt->repo, h2),
3612 repo_get_commit_tree(opt->repo,
3613 merged_merge_bases),
3614 &result_tree);
3615 strbuf_release(&merge_base_abbrev);
3616 opt->ancestor = NULL; /* avoid accidental re-use of opt->ancestor */
3617 if (clean < 0) {
3618 flush_output(opt);
3619 return clean;
3620 }
3621
3622 if (opt->priv->call_depth) {
3623 *result = make_virtual_commit(opt->repo, result_tree,
3624 "merged tree");
3625 commit_list_insert(h1, &(*result)->parents);
3626 commit_list_insert(h2, &(*result)->parents->next);
3627 }
3628 return clean;
3629 }
3630
3631 static int merge_start(struct merge_options *opt, struct tree *head)
3632 {
3633 struct strbuf sb = STRBUF_INIT;
3634
3635 /* Sanity checks on opt */
3636 assert(opt->repo);
3637
3638 assert(opt->branch1 && opt->branch2);
3639
3640 assert(opt->detect_renames >= -1 &&
3641 opt->detect_renames <= DIFF_DETECT_COPY);
3642 assert(opt->detect_directory_renames >= MERGE_DIRECTORY_RENAMES_NONE &&
3643 opt->detect_directory_renames <= MERGE_DIRECTORY_RENAMES_TRUE);
3644 assert(opt->rename_limit >= -1);
3645 assert(opt->rename_score >= 0 && opt->rename_score <= MAX_SCORE);
3646 assert(opt->show_rename_progress >= 0 && opt->show_rename_progress <= 1);
3647
3648 assert(opt->xdl_opts >= 0);
3649 assert(opt->recursive_variant >= MERGE_VARIANT_NORMAL &&
3650 opt->recursive_variant <= MERGE_VARIANT_THEIRS);
3651
3652 assert(opt->verbosity >= 0 && opt->verbosity <= 5);
3653 assert(opt->buffer_output <= 2);
3654 assert(opt->obuf.len == 0);
3655
3656 assert(opt->priv == NULL);
3657
3658 /* Sanity check on repo state; index must match head */
3659 if (repo_index_has_changes(opt->repo, head, &sb)) {
3660 err(opt, _("Your local changes to the following files would be overwritten by merge:\n %s"),
3661 sb.buf);
3662 strbuf_release(&sb);
3663 return -1;
3664 }
3665
3666 CALLOC_ARRAY(opt->priv, 1);
3667 string_list_init(&opt->priv->df_conflict_file_set, 1);
3668 return 0;
3669 }
3670
3671 static void merge_finalize(struct merge_options *opt)
3672 {
3673 flush_output(opt);
3674 if (!opt->priv->call_depth && opt->buffer_output < 2)
3675 strbuf_release(&opt->obuf);
3676 if (show(opt, 2))
3677 diff_warn_rename_limit("merge.renamelimit",
3678 opt->priv->needed_rename_limit, 0);
3679 FREE_AND_NULL(opt->priv);
3680 }
3681
3682 int merge_trees(struct merge_options *opt,
3683 struct tree *head,
3684 struct tree *merge,
3685 struct tree *merge_base)
3686 {
3687 int clean;
3688 struct tree *ignored;
3689
3690 assert(opt->ancestor != NULL);
3691
3692 if (merge_start(opt, head))
3693 return -1;
3694 clean = merge_trees_internal(opt, head, merge, merge_base, &ignored);
3695 merge_finalize(opt);
3696
3697 return clean;
3698 }
3699
3700 int merge_recursive(struct merge_options *opt,
3701 struct commit *h1,
3702 struct commit *h2,
3703 struct commit_list *merge_bases,
3704 struct commit **result)
3705 {
3706 int clean;
3707
3708 assert(opt->ancestor == NULL ||
3709 !strcmp(opt->ancestor, "constructed merge base"));
3710
3711 if (merge_start(opt, repo_get_commit_tree(opt->repo, h1)))
3712 return -1;
3713 clean = merge_recursive_internal(opt, h1, h2, merge_bases, result);
3714 merge_finalize(opt);
3715
3716 return clean;
3717 }
3718
3719 static struct commit *get_ref(struct repository *repo,
3720 const struct object_id *oid,
3721 const char *name)
3722 {
3723 struct object *object;
3724
3725 object = deref_tag(repo, parse_object(repo, oid),
3726 name, strlen(name));
3727 if (!object)
3728 return NULL;
3729 if (object->type == OBJ_TREE)
3730 return make_virtual_commit(repo, (struct tree*)object, name);
3731 if (object->type != OBJ_COMMIT)
3732 return NULL;
3733 if (parse_commit((struct commit *)object))
3734 return NULL;
3735 return (struct commit *)object;
3736 }
3737
3738 int merge_recursive_generic(struct merge_options *opt,
3739 const struct object_id *head,
3740 const struct object_id *merge,
3741 int num_merge_bases,
3742 const struct object_id **merge_bases,
3743 struct commit **result)
3744 {
3745 int clean;
3746 struct lock_file lock = LOCK_INIT;
3747 struct commit *head_commit = get_ref(opt->repo, head, opt->branch1);
3748 struct commit *next_commit = get_ref(opt->repo, merge, opt->branch2);
3749 struct commit_list *ca = NULL;
3750
3751 if (merge_bases) {
3752 int i;
3753 for (i = 0; i < num_merge_bases; ++i) {
3754 struct commit *base;
3755 if (!(base = get_ref(opt->repo, merge_bases[i],
3756 oid_to_hex(merge_bases[i]))))
3757 return err(opt, _("Could not parse object '%s'"),
3758 oid_to_hex(merge_bases[i]));
3759 commit_list_insert(base, &ca);
3760 }
3761 if (num_merge_bases == 1)
3762 opt->ancestor = "constructed merge base";
3763 }
3764
3765 repo_hold_locked_index(opt->repo, &lock, LOCK_DIE_ON_ERROR);
3766 clean = merge_recursive(opt, head_commit, next_commit, ca,
3767 result);
3768 if (clean < 0) {
3769 rollback_lock_file(&lock);
3770 return clean;
3771 }
3772
3773 if (write_locked_index(opt->repo->index, &lock,
3774 COMMIT_LOCK | SKIP_IF_UNCHANGED))
3775 return err(opt, _("Unable to write index."));
3776
3777 return clean ? 0 : 1;
3778 }
3779
3780 static void merge_recursive_config(struct merge_options *opt)
3781 {
3782 char *value = NULL;
3783 int renormalize = 0;
3784 git_config_get_int("merge.verbosity", &opt->verbosity);
3785 git_config_get_int("diff.renamelimit", &opt->rename_limit);
3786 git_config_get_int("merge.renamelimit", &opt->rename_limit);
3787 git_config_get_bool("merge.renormalize", &renormalize);
3788 opt->renormalize = renormalize;
3789 if (!git_config_get_string("diff.renames", &value)) {
3790 opt->detect_renames = git_config_rename("diff.renames", value);
3791 free(value);
3792 }
3793 if (!git_config_get_string("merge.renames", &value)) {
3794 opt->detect_renames = git_config_rename("merge.renames", value);
3795 free(value);
3796 }
3797 if (!git_config_get_string("merge.directoryrenames", &value)) {
3798 int boolval = git_parse_maybe_bool(value);
3799 if (0 <= boolval) {
3800 opt->detect_directory_renames = boolval ?
3801 MERGE_DIRECTORY_RENAMES_TRUE :
3802 MERGE_DIRECTORY_RENAMES_NONE;
3803 } else if (!strcasecmp(value, "conflict")) {
3804 opt->detect_directory_renames =
3805 MERGE_DIRECTORY_RENAMES_CONFLICT;
3806 } /* avoid erroring on values from future versions of git */
3807 free(value);
3808 }
3809 git_config(git_xmerge_config, NULL);
3810 }
3811
3812 void init_merge_options(struct merge_options *opt,
3813 struct repository *repo)
3814 {
3815 const char *merge_verbosity;
3816 memset(opt, 0, sizeof(struct merge_options));
3817
3818 opt->repo = repo;
3819
3820 opt->detect_renames = -1;
3821 opt->detect_directory_renames = MERGE_DIRECTORY_RENAMES_CONFLICT;
3822 opt->rename_limit = -1;
3823
3824 opt->verbosity = 2;
3825 opt->buffer_output = 1;
3826 strbuf_init(&opt->obuf, 0);
3827
3828 opt->renormalize = 0;
3829
3830 merge_recursive_config(opt);
3831 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3832 if (merge_verbosity)
3833 opt->verbosity = strtol(merge_verbosity, NULL, 10);
3834 if (opt->verbosity >= 5)
3835 opt->buffer_output = 0;
3836 }
3837
3838 int parse_merge_opt(struct merge_options *opt, const char *s)
3839 {
3840 const char *arg;
3841
3842 if (!s || !*s)
3843 return -1;
3844 if (!strcmp(s, "ours"))
3845 opt->recursive_variant = MERGE_VARIANT_OURS;
3846 else if (!strcmp(s, "theirs"))
3847 opt->recursive_variant = MERGE_VARIANT_THEIRS;
3848 else if (!strcmp(s, "subtree"))
3849 opt->subtree_shift = "";
3850 else if (skip_prefix(s, "subtree=", &arg))
3851 opt->subtree_shift = arg;
3852 else if (!strcmp(s, "patience"))
3853 opt->xdl_opts = DIFF_WITH_ALG(opt, PATIENCE_DIFF);
3854 else if (!strcmp(s, "histogram"))
3855 opt->xdl_opts = DIFF_WITH_ALG(opt, HISTOGRAM_DIFF);
3856 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3857 long value = parse_algorithm_value(arg);
3858 if (value < 0)
3859 return -1;
3860 /* clear out previous settings */
3861 DIFF_XDL_CLR(opt, NEED_MINIMAL);
3862 opt->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3863 opt->xdl_opts |= value;
3864 }
3865 else if (!strcmp(s, "ignore-space-change"))
3866 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_CHANGE);
3867 else if (!strcmp(s, "ignore-all-space"))
3868 DIFF_XDL_SET(opt, IGNORE_WHITESPACE);
3869 else if (!strcmp(s, "ignore-space-at-eol"))
3870 DIFF_XDL_SET(opt, IGNORE_WHITESPACE_AT_EOL);
3871 else if (!strcmp(s, "ignore-cr-at-eol"))
3872 DIFF_XDL_SET(opt, IGNORE_CR_AT_EOL);
3873 else if (!strcmp(s, "renormalize"))
3874 opt->renormalize = 1;
3875 else if (!strcmp(s, "no-renormalize"))
3876 opt->renormalize = 0;
3877 else if (!strcmp(s, "no-renames"))
3878 opt->detect_renames = 0;
3879 else if (!strcmp(s, "find-renames")) {
3880 opt->detect_renames = 1;
3881 opt->rename_score = 0;
3882 }
3883 else if (skip_prefix(s, "find-renames=", &arg) ||
3884 skip_prefix(s, "rename-threshold=", &arg)) {
3885 if ((opt->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3886 return -1;
3887 opt->detect_renames = 1;
3888 }
3889 /*
3890 * Please update $__git_merge_strategy_options in
3891 * git-completion.bash when you add new options
3892 */
3893 else
3894 return -1;
3895 return 0;
3896 }