1/*
2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
3 * Fredrik Kuivinen.
4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
5 */
6#include "cache.h"
7#include "config.h"
8#include "advice.h"
9#include "lockfile.h"
10#include "cache-tree.h"
11#include "commit.h"
12#include "blob.h"
13#include "builtin.h"
14#include "tree-walk.h"
15#include "diff.h"
16#include "diffcore.h"
17#include "tag.h"
18#include "unpack-trees.h"
19#include "string-list.h"
20#include "xdiff-interface.h"
21#include "ll-merge.h"
22#include "attr.h"
23#include "merge-recursive.h"
24#include "dir.h"
25#include "submodule.h"
26
27struct path_hashmap_entry {
28 struct hashmap_entry e;
29 char path[FLEX_ARRAY];
30};
31
32static int path_hashmap_cmp(const void *cmp_data,
33 const void *entry,
34 const void *entry_or_key,
35 const void *keydata)
36{
37 const struct path_hashmap_entry *a = entry;
38 const struct path_hashmap_entry *b = entry_or_key;
39 const char *key = keydata;
40
41 if (ignore_case)
42 return strcasecmp(a->path, key ? key : b->path);
43 else
44 return strcmp(a->path, key ? key : b->path);
45}
46
47static unsigned int path_hash(const char *path)
48{
49 return ignore_case ? strihash(path) : strhash(path);
50}
51
52static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
53 char *dir)
54{
55 struct dir_rename_entry key;
56
57 if (dir == NULL)
58 return NULL;
59 hashmap_entry_init(&key, strhash(dir));
60 key.dir = dir;
61 return hashmap_get(hashmap, &key, NULL);
62}
63
64static int dir_rename_cmp(const void *unused_cmp_data,
65 const void *entry,
66 const void *entry_or_key,
67 const void *unused_keydata)
68{
69 const struct dir_rename_entry *e1 = entry;
70 const struct dir_rename_entry *e2 = entry_or_key;
71
72 return strcmp(e1->dir, e2->dir);
73}
74
75static void dir_rename_init(struct hashmap *map)
76{
77 hashmap_init(map, dir_rename_cmp, NULL, 0);
78}
79
80static void dir_rename_entry_init(struct dir_rename_entry *entry,
81 char *directory)
82{
83 hashmap_entry_init(entry, strhash(directory));
84 entry->dir = directory;
85 entry->non_unique_new_dir = 0;
86 strbuf_init(&entry->new_dir, 0);
87 string_list_init(&entry->possible_new_dirs, 0);
88}
89
90static void flush_output(struct merge_options *o)
91{
92 if (o->buffer_output < 2 && o->obuf.len) {
93 fputs(o->obuf.buf, stdout);
94 strbuf_reset(&o->obuf);
95 }
96}
97
98static int err(struct merge_options *o, const char *err, ...)
99{
100 va_list params;
101
102 if (o->buffer_output < 2)
103 flush_output(o);
104 else {
105 strbuf_complete(&o->obuf, '\n');
106 strbuf_addstr(&o->obuf, "error: ");
107 }
108 va_start(params, err);
109 strbuf_vaddf(&o->obuf, err, params);
110 va_end(params);
111 if (o->buffer_output > 1)
112 strbuf_addch(&o->obuf, '\n');
113 else {
114 error("%s", o->obuf.buf);
115 strbuf_reset(&o->obuf);
116 }
117
118 return -1;
119}
120
121static struct tree *shift_tree_object(struct tree *one, struct tree *two,
122 const char *subtree_shift)
123{
124 struct object_id shifted;
125
126 if (!*subtree_shift) {
127 shift_tree(&one->object.oid, &two->object.oid, &shifted, 0);
128 } else {
129 shift_tree_by(&one->object.oid, &two->object.oid, &shifted,
130 subtree_shift);
131 }
132 if (!oidcmp(&two->object.oid, &shifted))
133 return two;
134 return lookup_tree(&shifted);
135}
136
137static struct commit *make_virtual_commit(struct tree *tree, const char *comment)
138{
139 struct commit *commit = alloc_commit_node();
140
141 set_merge_remote_desc(commit, comment, (struct object *)commit);
142 commit->tree = tree;
143 commit->object.parsed = 1;
144 return commit;
145}
146
147/*
148 * Since we use get_tree_entry(), which does not put the read object into
149 * the object pool, we cannot rely on a == b.
150 */
151static int oid_eq(const struct object_id *a, const struct object_id *b)
152{
153 if (!a && !b)
154 return 2;
155 return a && b && oidcmp(a, b) == 0;
156}
157
158enum rename_type {
159 RENAME_NORMAL = 0,
160 RENAME_DELETE,
161 RENAME_ONE_FILE_TO_ONE,
162 RENAME_ONE_FILE_TO_TWO,
163 RENAME_TWO_FILES_TO_ONE
164};
165
166struct rename_conflict_info {
167 enum rename_type rename_type;
168 struct diff_filepair *pair1;
169 struct diff_filepair *pair2;
170 const char *branch1;
171 const char *branch2;
172 struct stage_data *dst_entry1;
173 struct stage_data *dst_entry2;
174 struct diff_filespec ren1_other;
175 struct diff_filespec ren2_other;
176};
177
178/*
179 * Since we want to write the index eventually, we cannot reuse the index
180 * for these (temporary) data.
181 */
182struct stage_data {
183 struct {
184 unsigned mode;
185 struct object_id oid;
186 } stages[4];
187 struct rename_conflict_info *rename_conflict_info;
188 unsigned processed:1;
189};
190
191static inline void setup_rename_conflict_info(enum rename_type rename_type,
192 struct diff_filepair *pair1,
193 struct diff_filepair *pair2,
194 const char *branch1,
195 const char *branch2,
196 struct stage_data *dst_entry1,
197 struct stage_data *dst_entry2,
198 struct merge_options *o,
199 struct stage_data *src_entry1,
200 struct stage_data *src_entry2)
201{
202 struct rename_conflict_info *ci = xcalloc(1, sizeof(struct rename_conflict_info));
203 ci->rename_type = rename_type;
204 ci->pair1 = pair1;
205 ci->branch1 = branch1;
206 ci->branch2 = branch2;
207
208 ci->dst_entry1 = dst_entry1;
209 dst_entry1->rename_conflict_info = ci;
210 dst_entry1->processed = 0;
211
212 assert(!pair2 == !dst_entry2);
213 if (dst_entry2) {
214 ci->dst_entry2 = dst_entry2;
215 ci->pair2 = pair2;
216 dst_entry2->rename_conflict_info = ci;
217 }
218
219 if (rename_type == RENAME_TWO_FILES_TO_ONE) {
220 /*
221 * For each rename, there could have been
222 * modifications on the side of history where that
223 * file was not renamed.
224 */
225 int ostage1 = o->branch1 == branch1 ? 3 : 2;
226 int ostage2 = ostage1 ^ 1;
227
228 ci->ren1_other.path = pair1->one->path;
229 oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid);
230 ci->ren1_other.mode = src_entry1->stages[ostage1].mode;
231
232 ci->ren2_other.path = pair2->one->path;
233 oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid);
234 ci->ren2_other.mode = src_entry2->stages[ostage2].mode;
235 }
236}
237
238static int show(struct merge_options *o, int v)
239{
240 return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5;
241}
242
243__attribute__((format (printf, 3, 4)))
244static void output(struct merge_options *o, int v, const char *fmt, ...)
245{
246 va_list ap;
247
248 if (!show(o, v))
249 return;
250
251 strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
252
253 va_start(ap, fmt);
254 strbuf_vaddf(&o->obuf, fmt, ap);
255 va_end(ap);
256
257 strbuf_addch(&o->obuf, '\n');
258 if (!o->buffer_output)
259 flush_output(o);
260}
261
262static void output_commit_title(struct merge_options *o, struct commit *commit)
263{
264 strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
265 if (commit->util)
266 strbuf_addf(&o->obuf, "virtual %s\n",
267 merge_remote_util(commit)->name);
268 else {
269 strbuf_add_unique_abbrev(&o->obuf, commit->object.oid.hash,
270 DEFAULT_ABBREV);
271 strbuf_addch(&o->obuf, ' ');
272 if (parse_commit(commit) != 0)
273 strbuf_addstr(&o->obuf, _("(bad commit)\n"));
274 else {
275 const char *title;
276 const char *msg = get_commit_buffer(commit, NULL);
277 int len = find_commit_subject(msg, &title);
278 if (len)
279 strbuf_addf(&o->obuf, "%.*s\n", len, title);
280 unuse_commit_buffer(commit, msg);
281 }
282 }
283 flush_output(o);
284}
285
286static int add_cacheinfo(struct merge_options *o,
287 unsigned int mode, const struct object_id *oid,
288 const char *path, int stage, int refresh, int options)
289{
290 struct cache_entry *ce;
291 int ret;
292
293 ce = make_cache_entry(mode, oid ? oid->hash : null_sha1, path, stage, 0);
294 if (!ce)
295 return err(o, _("addinfo_cache failed for path '%s'"), path);
296
297 ret = add_cache_entry(ce, options);
298 if (refresh) {
299 struct cache_entry *nce;
300
301 nce = refresh_cache_entry(ce, CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
302 if (!nce)
303 return err(o, _("addinfo_cache failed for path '%s'"), path);
304 if (nce != ce)
305 ret = add_cache_entry(nce, options);
306 }
307 return ret;
308}
309
310static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
311{
312 parse_tree(tree);
313 init_tree_desc(desc, tree->buffer, tree->size);
314}
315
316static int git_merge_trees(int index_only,
317 struct tree *common,
318 struct tree *head,
319 struct tree *merge)
320{
321 int rc;
322 struct tree_desc t[3];
323 struct unpack_trees_options opts;
324
325 memset(&opts, 0, sizeof(opts));
326 if (index_only)
327 opts.index_only = 1;
328 else
329 opts.update = 1;
330 opts.merge = 1;
331 opts.head_idx = 2;
332 opts.fn = threeway_merge;
333 opts.src_index = &the_index;
334 opts.dst_index = &the_index;
335 setup_unpack_trees_porcelain(&opts, "merge");
336
337 init_tree_desc_from_tree(t+0, common);
338 init_tree_desc_from_tree(t+1, head);
339 init_tree_desc_from_tree(t+2, merge);
340
341 rc = unpack_trees(3, t, &opts);
342 cache_tree_free(&active_cache_tree);
343 return rc;
344}
345
346struct tree *write_tree_from_memory(struct merge_options *o)
347{
348 struct tree *result = NULL;
349
350 if (unmerged_cache()) {
351 int i;
352 fprintf(stderr, "BUG: There are unmerged index entries:\n");
353 for (i = 0; i < active_nr; i++) {
354 const struct cache_entry *ce = active_cache[i];
355 if (ce_stage(ce))
356 fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
357 (int)ce_namelen(ce), ce->name);
358 }
359 die("BUG: unmerged index entries in merge-recursive.c");
360 }
361
362 if (!active_cache_tree)
363 active_cache_tree = cache_tree();
364
365 if (!cache_tree_fully_valid(active_cache_tree) &&
366 cache_tree_update(&the_index, 0) < 0) {
367 err(o, _("error building trees"));
368 return NULL;
369 }
370
371 result = lookup_tree(&active_cache_tree->oid);
372
373 return result;
374}
375
376static int save_files_dirs(const unsigned char *sha1,
377 struct strbuf *base, const char *path,
378 unsigned int mode, int stage, void *context)
379{
380 struct path_hashmap_entry *entry;
381 int baselen = base->len;
382 struct merge_options *o = context;
383
384 strbuf_addstr(base, path);
385
386 FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
387 hashmap_entry_init(entry, path_hash(entry->path));
388 hashmap_add(&o->current_file_dir_set, entry);
389
390 strbuf_setlen(base, baselen);
391 return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
392}
393
394static void get_files_dirs(struct merge_options *o, struct tree *tree)
395{
396 struct pathspec match_all;
397 memset(&match_all, 0, sizeof(match_all));
398 read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
399}
400
401/*
402 * Returns an index_entry instance which doesn't have to correspond to
403 * a real cache entry in Git's index.
404 */
405static struct stage_data *insert_stage_data(const char *path,
406 struct tree *o, struct tree *a, struct tree *b,
407 struct string_list *entries)
408{
409 struct string_list_item *item;
410 struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
411 get_tree_entry(o->object.oid.hash, path,
412 e->stages[1].oid.hash, &e->stages[1].mode);
413 get_tree_entry(a->object.oid.hash, path,
414 e->stages[2].oid.hash, &e->stages[2].mode);
415 get_tree_entry(b->object.oid.hash, path,
416 e->stages[3].oid.hash, &e->stages[3].mode);
417 item = string_list_insert(entries, path);
418 item->util = e;
419 return e;
420}
421
422/*
423 * Create a dictionary mapping file names to stage_data objects. The
424 * dictionary contains one entry for every path with a non-zero stage entry.
425 */
426static struct string_list *get_unmerged(void)
427{
428 struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
429 int i;
430
431 unmerged->strdup_strings = 1;
432
433 for (i = 0; i < active_nr; i++) {
434 struct string_list_item *item;
435 struct stage_data *e;
436 const struct cache_entry *ce = active_cache[i];
437 if (!ce_stage(ce))
438 continue;
439
440 item = string_list_lookup(unmerged, ce->name);
441 if (!item) {
442 item = string_list_insert(unmerged, ce->name);
443 item->util = xcalloc(1, sizeof(struct stage_data));
444 }
445 e = item->util;
446 e->stages[ce_stage(ce)].mode = ce->ce_mode;
447 oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
448 }
449
450 return unmerged;
451}
452
453static int string_list_df_name_compare(const char *one, const char *two)
454{
455 int onelen = strlen(one);
456 int twolen = strlen(two);
457 /*
458 * Here we only care that entries for D/F conflicts are
459 * adjacent, in particular with the file of the D/F conflict
460 * appearing before files below the corresponding directory.
461 * The order of the rest of the list is irrelevant for us.
462 *
463 * To achieve this, we sort with df_name_compare and provide
464 * the mode S_IFDIR so that D/F conflicts will sort correctly.
465 * We use the mode S_IFDIR for everything else for simplicity,
466 * since in other cases any changes in their order due to
467 * sorting cause no problems for us.
468 */
469 int cmp = df_name_compare(one, onelen, S_IFDIR,
470 two, twolen, S_IFDIR);
471 /*
472 * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
473 * that 'foo' comes before 'foo/bar'.
474 */
475 if (cmp)
476 return cmp;
477 return onelen - twolen;
478}
479
480static void record_df_conflict_files(struct merge_options *o,
481 struct string_list *entries)
482{
483 /* If there is a D/F conflict and the file for such a conflict
484 * currently exist in the working tree, we want to allow it to be
485 * removed to make room for the corresponding directory if needed.
486 * The files underneath the directories of such D/F conflicts will
487 * be processed before the corresponding file involved in the D/F
488 * conflict. If the D/F directory ends up being removed by the
489 * merge, then we won't have to touch the D/F file. If the D/F
490 * directory needs to be written to the working copy, then the D/F
491 * file will simply be removed (in make_room_for_path()) to make
492 * room for the necessary paths. Note that if both the directory
493 * and the file need to be present, then the D/F file will be
494 * reinstated with a new unique name at the time it is processed.
495 */
496 struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
497 const char *last_file = NULL;
498 int last_len = 0;
499 int i;
500
501 /*
502 * If we're merging merge-bases, we don't want to bother with
503 * any working directory changes.
504 */
505 if (o->call_depth)
506 return;
507
508 /* Ensure D/F conflicts are adjacent in the entries list. */
509 for (i = 0; i < entries->nr; i++) {
510 struct string_list_item *next = &entries->items[i];
511 string_list_append(&df_sorted_entries, next->string)->util =
512 next->util;
513 }
514 df_sorted_entries.cmp = string_list_df_name_compare;
515 string_list_sort(&df_sorted_entries);
516
517 string_list_clear(&o->df_conflict_file_set, 1);
518 for (i = 0; i < df_sorted_entries.nr; i++) {
519 const char *path = df_sorted_entries.items[i].string;
520 int len = strlen(path);
521 struct stage_data *e = df_sorted_entries.items[i].util;
522
523 /*
524 * Check if last_file & path correspond to a D/F conflict;
525 * i.e. whether path is last_file+'/'+<something>.
526 * If so, record that it's okay to remove last_file to make
527 * room for path and friends if needed.
528 */
529 if (last_file &&
530 len > last_len &&
531 memcmp(path, last_file, last_len) == 0 &&
532 path[last_len] == '/') {
533 string_list_insert(&o->df_conflict_file_set, last_file);
534 }
535
536 /*
537 * Determine whether path could exist as a file in the
538 * working directory as a possible D/F conflict. This
539 * will only occur when it exists in stage 2 as a
540 * file.
541 */
542 if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
543 last_file = path;
544 last_len = len;
545 } else {
546 last_file = NULL;
547 }
548 }
549 string_list_clear(&df_sorted_entries, 0);
550}
551
552struct rename {
553 struct diff_filepair *pair;
554 /*
555 * Purpose of src_entry and dst_entry:
556 *
557 * If 'before' is renamed to 'after' then src_entry will contain
558 * the versions of 'before' from the merge_base, HEAD, and MERGE in
559 * stages 1, 2, and 3; dst_entry will contain the respective
560 * versions of 'after' in corresponding locations. Thus, we have a
561 * total of six modes and oids, though some will be null. (Stage 0
562 * is ignored; we're interested in handling conflicts.)
563 *
564 * Since we don't turn on break-rewrites by default, neither
565 * src_entry nor dst_entry can have all three of their stages have
566 * non-null oids, meaning at most four of the six will be non-null.
567 * Also, since this is a rename, both src_entry and dst_entry will
568 * have at least one non-null oid, meaning at least two will be
569 * non-null. Of the six oids, a typical rename will have three be
570 * non-null. Only two implies a rename/delete, and four implies a
571 * rename/add.
572 */
573 struct stage_data *src_entry;
574 struct stage_data *dst_entry;
575 unsigned processed:1;
576};
577
578static int update_stages(struct merge_options *opt, const char *path,
579 const struct diff_filespec *o,
580 const struct diff_filespec *a,
581 const struct diff_filespec *b)
582{
583
584 /*
585 * NOTE: It is usually a bad idea to call update_stages on a path
586 * before calling update_file on that same path, since it can
587 * sometimes lead to spurious "refusing to lose untracked file..."
588 * messages from update_file (via make_room_for path via
589 * would_lose_untracked). Instead, reverse the order of the calls
590 * (executing update_file first and then update_stages).
591 */
592 int clear = 1;
593 int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
594 if (clear)
595 if (remove_file_from_cache(path))
596 return -1;
597 if (o)
598 if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
599 return -1;
600 if (a)
601 if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
602 return -1;
603 if (b)
604 if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
605 return -1;
606 return 0;
607}
608
609static void update_entry(struct stage_data *entry,
610 struct diff_filespec *o,
611 struct diff_filespec *a,
612 struct diff_filespec *b)
613{
614 entry->processed = 0;
615 entry->stages[1].mode = o->mode;
616 entry->stages[2].mode = a->mode;
617 entry->stages[3].mode = b->mode;
618 oidcpy(&entry->stages[1].oid, &o->oid);
619 oidcpy(&entry->stages[2].oid, &a->oid);
620 oidcpy(&entry->stages[3].oid, &b->oid);
621}
622
623static int remove_file(struct merge_options *o, int clean,
624 const char *path, int no_wd)
625{
626 int update_cache = o->call_depth || clean;
627 int update_working_directory = !o->call_depth && !no_wd;
628
629 if (update_cache) {
630 if (remove_file_from_cache(path))
631 return -1;
632 }
633 if (update_working_directory) {
634 if (ignore_case) {
635 struct cache_entry *ce;
636 ce = cache_file_exists(path, strlen(path), ignore_case);
637 if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
638 return 0;
639 }
640 if (remove_path(path))
641 return -1;
642 }
643 return 0;
644}
645
646/* add a string to a strbuf, but converting "/" to "_" */
647static void add_flattened_path(struct strbuf *out, const char *s)
648{
649 size_t i = out->len;
650 strbuf_addstr(out, s);
651 for (; i < out->len; i++)
652 if (out->buf[i] == '/')
653 out->buf[i] = '_';
654}
655
656static char *unique_path(struct merge_options *o, const char *path, const char *branch)
657{
658 struct path_hashmap_entry *entry;
659 struct strbuf newpath = STRBUF_INIT;
660 int suffix = 0;
661 size_t base_len;
662
663 strbuf_addf(&newpath, "%s~", path);
664 add_flattened_path(&newpath, branch);
665
666 base_len = newpath.len;
667 while (hashmap_get_from_hash(&o->current_file_dir_set,
668 path_hash(newpath.buf), newpath.buf) ||
669 (!o->call_depth && file_exists(newpath.buf))) {
670 strbuf_setlen(&newpath, base_len);
671 strbuf_addf(&newpath, "_%d", suffix++);
672 }
673
674 FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
675 hashmap_entry_init(entry, path_hash(entry->path));
676 hashmap_add(&o->current_file_dir_set, entry);
677 return strbuf_detach(&newpath, NULL);
678}
679
680/**
681 * Check whether a directory in the index is in the way of an incoming
682 * file. Return 1 if so. If check_working_copy is non-zero, also
683 * check the working directory. If empty_ok is non-zero, also return
684 * 0 in the case where the working-tree dir exists but is empty.
685 */
686static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
687{
688 int pos;
689 struct strbuf dirpath = STRBUF_INIT;
690 struct stat st;
691
692 strbuf_addstr(&dirpath, path);
693 strbuf_addch(&dirpath, '/');
694
695 pos = cache_name_pos(dirpath.buf, dirpath.len);
696
697 if (pos < 0)
698 pos = -1 - pos;
699 if (pos < active_nr &&
700 !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
701 strbuf_release(&dirpath);
702 return 1;
703 }
704
705 strbuf_release(&dirpath);
706 return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
707 !(empty_ok && is_empty_dir(path));
708}
709
710static int was_tracked(const char *path)
711{
712 int pos = cache_name_pos(path, strlen(path));
713
714 if (0 <= pos)
715 /* we have been tracking this path */
716 return 1;
717
718 /*
719 * Look for an unmerged entry for the path,
720 * specifically stage #2, which would indicate
721 * that "our" side before the merge started
722 * had the path tracked (and resulted in a conflict).
723 */
724 for (pos = -1 - pos;
725 pos < active_nr && !strcmp(path, active_cache[pos]->name);
726 pos++)
727 if (ce_stage(active_cache[pos]) == 2)
728 return 1;
729 return 0;
730}
731
732static int would_lose_untracked(const char *path)
733{
734 return !was_tracked(path) && file_exists(path);
735}
736
737static int make_room_for_path(struct merge_options *o, const char *path)
738{
739 int status, i;
740 const char *msg = _("failed to create path '%s'%s");
741
742 /* Unlink any D/F conflict files that are in the way */
743 for (i = 0; i < o->df_conflict_file_set.nr; i++) {
744 const char *df_path = o->df_conflict_file_set.items[i].string;
745 size_t pathlen = strlen(path);
746 size_t df_pathlen = strlen(df_path);
747 if (df_pathlen < pathlen &&
748 path[df_pathlen] == '/' &&
749 strncmp(path, df_path, df_pathlen) == 0) {
750 output(o, 3,
751 _("Removing %s to make room for subdirectory\n"),
752 df_path);
753 unlink(df_path);
754 unsorted_string_list_delete_item(&o->df_conflict_file_set,
755 i, 0);
756 break;
757 }
758 }
759
760 /* Make sure leading directories are created */
761 status = safe_create_leading_directories_const(path);
762 if (status) {
763 if (status == SCLD_EXISTS)
764 /* something else exists */
765 return err(o, msg, path, _(": perhaps a D/F conflict?"));
766 return err(o, msg, path, "");
767 }
768
769 /*
770 * Do not unlink a file in the work tree if we are not
771 * tracking it.
772 */
773 if (would_lose_untracked(path))
774 return err(o, _("refusing to lose untracked file at '%s'"),
775 path);
776
777 /* Successful unlink is good.. */
778 if (!unlink(path))
779 return 0;
780 /* .. and so is no existing file */
781 if (errno == ENOENT)
782 return 0;
783 /* .. but not some other error (who really cares what?) */
784 return err(o, msg, path, _(": perhaps a D/F conflict?"));
785}
786
787static int update_file_flags(struct merge_options *o,
788 const struct object_id *oid,
789 unsigned mode,
790 const char *path,
791 int update_cache,
792 int update_wd)
793{
794 int ret = 0;
795
796 if (o->call_depth)
797 update_wd = 0;
798
799 if (update_wd) {
800 enum object_type type;
801 void *buf;
802 unsigned long size;
803
804 if (S_ISGITLINK(mode)) {
805 /*
806 * We may later decide to recursively descend into
807 * the submodule directory and update its index
808 * and/or work tree, but we do not do that now.
809 */
810 update_wd = 0;
811 goto update_index;
812 }
813
814 buf = read_sha1_file(oid->hash, &type, &size);
815 if (!buf)
816 return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
817 if (type != OBJ_BLOB) {
818 ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
819 goto free_buf;
820 }
821 if (S_ISREG(mode)) {
822 struct strbuf strbuf = STRBUF_INIT;
823 if (convert_to_working_tree(path, buf, size, &strbuf)) {
824 free(buf);
825 size = strbuf.len;
826 buf = strbuf_detach(&strbuf, NULL);
827 }
828 }
829
830 if (make_room_for_path(o, path) < 0) {
831 update_wd = 0;
832 goto free_buf;
833 }
834 if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
835 int fd;
836 if (mode & 0100)
837 mode = 0777;
838 else
839 mode = 0666;
840 fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
841 if (fd < 0) {
842 ret = err(o, _("failed to open '%s': %s"),
843 path, strerror(errno));
844 goto free_buf;
845 }
846 write_in_full(fd, buf, size);
847 close(fd);
848 } else if (S_ISLNK(mode)) {
849 char *lnk = xmemdupz(buf, size);
850 safe_create_leading_directories_const(path);
851 unlink(path);
852 if (symlink(lnk, path))
853 ret = err(o, _("failed to symlink '%s': %s"),
854 path, strerror(errno));
855 free(lnk);
856 } else
857 ret = err(o,
858 _("do not know what to do with %06o %s '%s'"),
859 mode, oid_to_hex(oid), path);
860 free_buf:
861 free(buf);
862 }
863 update_index:
864 if (!ret && update_cache)
865 add_cacheinfo(o, mode, oid, path, 0, update_wd, ADD_CACHE_OK_TO_ADD);
866 return ret;
867}
868
869static int update_file(struct merge_options *o,
870 int clean,
871 const struct object_id *oid,
872 unsigned mode,
873 const char *path)
874{
875 return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
876}
877
878/* Low level file merging, update and removal */
879
880struct merge_file_info {
881 struct object_id oid;
882 unsigned mode;
883 unsigned clean:1,
884 merge:1;
885};
886
887static int merge_3way(struct merge_options *o,
888 mmbuffer_t *result_buf,
889 const struct diff_filespec *one,
890 const struct diff_filespec *a,
891 const struct diff_filespec *b,
892 const char *branch1,
893 const char *branch2)
894{
895 mmfile_t orig, src1, src2;
896 struct ll_merge_options ll_opts = {0};
897 char *base_name, *name1, *name2;
898 int merge_status;
899
900 ll_opts.renormalize = o->renormalize;
901 ll_opts.xdl_opts = o->xdl_opts;
902
903 if (o->call_depth) {
904 ll_opts.virtual_ancestor = 1;
905 ll_opts.variant = 0;
906 } else {
907 switch (o->recursive_variant) {
908 case MERGE_RECURSIVE_OURS:
909 ll_opts.variant = XDL_MERGE_FAVOR_OURS;
910 break;
911 case MERGE_RECURSIVE_THEIRS:
912 ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
913 break;
914 default:
915 ll_opts.variant = 0;
916 break;
917 }
918 }
919
920 if (strcmp(a->path, b->path) ||
921 (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
922 base_name = o->ancestor == NULL ? NULL :
923 mkpathdup("%s:%s", o->ancestor, one->path);
924 name1 = mkpathdup("%s:%s", branch1, a->path);
925 name2 = mkpathdup("%s:%s", branch2, b->path);
926 } else {
927 base_name = o->ancestor == NULL ? NULL :
928 mkpathdup("%s", o->ancestor);
929 name1 = mkpathdup("%s", branch1);
930 name2 = mkpathdup("%s", branch2);
931 }
932
933 read_mmblob(&orig, &one->oid);
934 read_mmblob(&src1, &a->oid);
935 read_mmblob(&src2, &b->oid);
936
937 merge_status = ll_merge(result_buf, a->path, &orig, base_name,
938 &src1, name1, &src2, name2, &ll_opts);
939
940 free(base_name);
941 free(name1);
942 free(name2);
943 free(orig.ptr);
944 free(src1.ptr);
945 free(src2.ptr);
946 return merge_status;
947}
948
949static int merge_file_1(struct merge_options *o,
950 const struct diff_filespec *one,
951 const struct diff_filespec *a,
952 const struct diff_filespec *b,
953 const char *branch1,
954 const char *branch2,
955 struct merge_file_info *result)
956{
957 result->merge = 0;
958 result->clean = 1;
959
960 if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
961 result->clean = 0;
962 if (S_ISREG(a->mode)) {
963 result->mode = a->mode;
964 oidcpy(&result->oid, &a->oid);
965 } else {
966 result->mode = b->mode;
967 oidcpy(&result->oid, &b->oid);
968 }
969 } else {
970 if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
971 result->merge = 1;
972
973 /*
974 * Merge modes
975 */
976 if (a->mode == b->mode || a->mode == one->mode)
977 result->mode = b->mode;
978 else {
979 result->mode = a->mode;
980 if (b->mode != one->mode) {
981 result->clean = 0;
982 result->merge = 1;
983 }
984 }
985
986 if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
987 oidcpy(&result->oid, &b->oid);
988 else if (oid_eq(&b->oid, &one->oid))
989 oidcpy(&result->oid, &a->oid);
990 else if (S_ISREG(a->mode)) {
991 mmbuffer_t result_buf;
992 int ret = 0, merge_status;
993
994 merge_status = merge_3way(o, &result_buf, one, a, b,
995 branch1, branch2);
996
997 if ((merge_status < 0) || !result_buf.ptr)
998 ret = err(o, _("Failed to execute internal merge"));
999
1000 if (!ret && write_sha1_file(result_buf.ptr, result_buf.size,
1001 blob_type, result->oid.hash))
1002 ret = err(o, _("Unable to add %s to database"),
1003 a->path);
1004
1005 free(result_buf.ptr);
1006 if (ret)
1007 return ret;
1008 result->clean = (merge_status == 0);
1009 } else if (S_ISGITLINK(a->mode)) {
1010 result->clean = merge_submodule(&result->oid,
1011 one->path,
1012 &one->oid,
1013 &a->oid,
1014 &b->oid,
1015 !o->call_depth);
1016 } else if (S_ISLNK(a->mode)) {
1017 oidcpy(&result->oid, &a->oid);
1018
1019 if (!oid_eq(&a->oid, &b->oid))
1020 result->clean = 0;
1021 } else
1022 die("BUG: unsupported object type in the tree");
1023 }
1024
1025 return 0;
1026}
1027
1028static int merge_file_special_markers(struct merge_options *o,
1029 const struct diff_filespec *one,
1030 const struct diff_filespec *a,
1031 const struct diff_filespec *b,
1032 const char *branch1,
1033 const char *filename1,
1034 const char *branch2,
1035 const char *filename2,
1036 struct merge_file_info *mfi)
1037{
1038 char *side1 = NULL;
1039 char *side2 = NULL;
1040 int ret;
1041
1042 if (filename1)
1043 side1 = xstrfmt("%s:%s", branch1, filename1);
1044 if (filename2)
1045 side2 = xstrfmt("%s:%s", branch2, filename2);
1046
1047 ret = merge_file_1(o, one, a, b,
1048 side1 ? side1 : branch1,
1049 side2 ? side2 : branch2, mfi);
1050 free(side1);
1051 free(side2);
1052 return ret;
1053}
1054
1055static int merge_file_one(struct merge_options *o,
1056 const char *path,
1057 const struct object_id *o_oid, int o_mode,
1058 const struct object_id *a_oid, int a_mode,
1059 const struct object_id *b_oid, int b_mode,
1060 const char *branch1,
1061 const char *branch2,
1062 struct merge_file_info *mfi)
1063{
1064 struct diff_filespec one, a, b;
1065
1066 one.path = a.path = b.path = (char *)path;
1067 oidcpy(&one.oid, o_oid);
1068 one.mode = o_mode;
1069 oidcpy(&a.oid, a_oid);
1070 a.mode = a_mode;
1071 oidcpy(&b.oid, b_oid);
1072 b.mode = b_mode;
1073 return merge_file_1(o, &one, &a, &b, branch1, branch2, mfi);
1074}
1075
1076static int handle_change_delete(struct merge_options *o,
1077 const char *path, const char *old_path,
1078 const struct object_id *o_oid, int o_mode,
1079 const struct object_id *changed_oid,
1080 int changed_mode,
1081 const char *change_branch,
1082 const char *delete_branch,
1083 const char *change, const char *change_past)
1084{
1085 char *alt_path = NULL;
1086 const char *update_path = path;
1087 int ret = 0;
1088
1089 if (dir_in_way(path, !o->call_depth, 0)) {
1090 update_path = alt_path = unique_path(o, path, change_branch);
1091 }
1092
1093 if (o->call_depth) {
1094 /*
1095 * We cannot arbitrarily accept either a_sha or b_sha as
1096 * correct; since there is no true "middle point" between
1097 * them, simply reuse the base version for virtual merge base.
1098 */
1099 ret = remove_file_from_cache(path);
1100 if (!ret)
1101 ret = update_file(o, 0, o_oid, o_mode, update_path);
1102 } else {
1103 if (!alt_path) {
1104 if (!old_path) {
1105 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1106 "and %s in %s. Version %s of %s left in tree."),
1107 change, path, delete_branch, change_past,
1108 change_branch, change_branch, path);
1109 } else {
1110 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1111 "and %s to %s in %s. Version %s of %s left in tree."),
1112 change, old_path, delete_branch, change_past, path,
1113 change_branch, change_branch, path);
1114 }
1115 } else {
1116 if (!old_path) {
1117 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1118 "and %s in %s. Version %s of %s left in tree at %s."),
1119 change, path, delete_branch, change_past,
1120 change_branch, change_branch, path, alt_path);
1121 } else {
1122 output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1123 "and %s to %s in %s. Version %s of %s left in tree at %s."),
1124 change, old_path, delete_branch, change_past, path,
1125 change_branch, change_branch, path, alt_path);
1126 }
1127 }
1128 /*
1129 * No need to call update_file() on path when change_branch ==
1130 * o->branch1 && !alt_path, since that would needlessly touch
1131 * path. We could call update_file_flags() with update_cache=0
1132 * and update_wd=0, but that's a no-op.
1133 */
1134 if (change_branch != o->branch1 || alt_path)
1135 ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1136 }
1137 free(alt_path);
1138
1139 return ret;
1140}
1141
1142static int conflict_rename_delete(struct merge_options *o,
1143 struct diff_filepair *pair,
1144 const char *rename_branch,
1145 const char *delete_branch)
1146{
1147 const struct diff_filespec *orig = pair->one;
1148 const struct diff_filespec *dest = pair->two;
1149
1150 if (handle_change_delete(o,
1151 o->call_depth ? orig->path : dest->path,
1152 o->call_depth ? NULL : orig->path,
1153 &orig->oid, orig->mode,
1154 &dest->oid, dest->mode,
1155 rename_branch, delete_branch,
1156 _("rename"), _("renamed")))
1157 return -1;
1158
1159 if (o->call_depth)
1160 return remove_file_from_cache(dest->path);
1161 else
1162 return update_stages(o, dest->path, NULL,
1163 rename_branch == o->branch1 ? dest : NULL,
1164 rename_branch == o->branch1 ? NULL : dest);
1165}
1166
1167static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1168 struct stage_data *entry,
1169 int stage)
1170{
1171 struct object_id *oid = &entry->stages[stage].oid;
1172 unsigned mode = entry->stages[stage].mode;
1173 if (mode == 0 || is_null_oid(oid))
1174 return NULL;
1175 oidcpy(&target->oid, oid);
1176 target->mode = mode;
1177 return target;
1178}
1179
1180static int handle_file(struct merge_options *o,
1181 struct diff_filespec *rename,
1182 int stage,
1183 struct rename_conflict_info *ci)
1184{
1185 char *dst_name = rename->path;
1186 struct stage_data *dst_entry;
1187 const char *cur_branch, *other_branch;
1188 struct diff_filespec other;
1189 struct diff_filespec *add;
1190 int ret;
1191
1192 if (stage == 2) {
1193 dst_entry = ci->dst_entry1;
1194 cur_branch = ci->branch1;
1195 other_branch = ci->branch2;
1196 } else {
1197 dst_entry = ci->dst_entry2;
1198 cur_branch = ci->branch2;
1199 other_branch = ci->branch1;
1200 }
1201
1202 add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1203 if (add) {
1204 char *add_name = unique_path(o, rename->path, other_branch);
1205 if (update_file(o, 0, &add->oid, add->mode, add_name))
1206 return -1;
1207
1208 remove_file(o, 0, rename->path, 0);
1209 dst_name = unique_path(o, rename->path, cur_branch);
1210 } else {
1211 if (dir_in_way(rename->path, !o->call_depth, 0)) {
1212 dst_name = unique_path(o, rename->path, cur_branch);
1213 output(o, 1, _("%s is a directory in %s adding as %s instead"),
1214 rename->path, other_branch, dst_name);
1215 }
1216 }
1217 if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1218 ; /* fall through, do allow dst_name to be released */
1219 else if (stage == 2)
1220 ret = update_stages(o, rename->path, NULL, rename, add);
1221 else
1222 ret = update_stages(o, rename->path, NULL, add, rename);
1223
1224 if (dst_name != rename->path)
1225 free(dst_name);
1226
1227 return ret;
1228}
1229
1230static int conflict_rename_rename_1to2(struct merge_options *o,
1231 struct rename_conflict_info *ci)
1232{
1233 /* One file was renamed in both branches, but to different names. */
1234 struct diff_filespec *one = ci->pair1->one;
1235 struct diff_filespec *a = ci->pair1->two;
1236 struct diff_filespec *b = ci->pair2->two;
1237
1238 output(o, 1, _("CONFLICT (rename/rename): "
1239 "Rename \"%s\"->\"%s\" in branch \"%s\" "
1240 "rename \"%s\"->\"%s\" in \"%s\"%s"),
1241 one->path, a->path, ci->branch1,
1242 one->path, b->path, ci->branch2,
1243 o->call_depth ? _(" (left unresolved)") : "");
1244 if (o->call_depth) {
1245 struct merge_file_info mfi;
1246 struct diff_filespec other;
1247 struct diff_filespec *add;
1248 if (merge_file_one(o, one->path,
1249 &one->oid, one->mode,
1250 &a->oid, a->mode,
1251 &b->oid, b->mode,
1252 ci->branch1, ci->branch2, &mfi))
1253 return -1;
1254
1255 /*
1256 * FIXME: For rename/add-source conflicts (if we could detect
1257 * such), this is wrong. We should instead find a unique
1258 * pathname and then either rename the add-source file to that
1259 * unique path, or use that unique path instead of src here.
1260 */
1261 if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1262 return -1;
1263
1264 /*
1265 * Above, we put the merged content at the merge-base's
1266 * path. Now we usually need to delete both a->path and
1267 * b->path. However, the rename on each side of the merge
1268 * could also be involved in a rename/add conflict. In
1269 * such cases, we should keep the added file around,
1270 * resolving the conflict at that path in its favor.
1271 */
1272 add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1273 if (add) {
1274 if (update_file(o, 0, &add->oid, add->mode, a->path))
1275 return -1;
1276 }
1277 else
1278 remove_file_from_cache(a->path);
1279 add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1280 if (add) {
1281 if (update_file(o, 0, &add->oid, add->mode, b->path))
1282 return -1;
1283 }
1284 else
1285 remove_file_from_cache(b->path);
1286 } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1287 return -1;
1288
1289 return 0;
1290}
1291
1292static int conflict_rename_rename_2to1(struct merge_options *o,
1293 struct rename_conflict_info *ci)
1294{
1295 /* Two files, a & b, were renamed to the same thing, c. */
1296 struct diff_filespec *a = ci->pair1->one;
1297 struct diff_filespec *b = ci->pair2->one;
1298 struct diff_filespec *c1 = ci->pair1->two;
1299 struct diff_filespec *c2 = ci->pair2->two;
1300 char *path = c1->path; /* == c2->path */
1301 struct merge_file_info mfi_c1;
1302 struct merge_file_info mfi_c2;
1303 int ret;
1304
1305 output(o, 1, _("CONFLICT (rename/rename): "
1306 "Rename %s->%s in %s. "
1307 "Rename %s->%s in %s"),
1308 a->path, c1->path, ci->branch1,
1309 b->path, c2->path, ci->branch2);
1310
1311 remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1312 remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1313
1314 if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1315 o->branch1, c1->path,
1316 o->branch2, ci->ren1_other.path, &mfi_c1) ||
1317 merge_file_special_markers(o, b, &ci->ren2_other, c2,
1318 o->branch1, ci->ren2_other.path,
1319 o->branch2, c2->path, &mfi_c2))
1320 return -1;
1321
1322 if (o->call_depth) {
1323 /*
1324 * If mfi_c1.clean && mfi_c2.clean, then it might make
1325 * sense to do a two-way merge of those results. But, I
1326 * think in all cases, it makes sense to have the virtual
1327 * merge base just undo the renames; they can be detected
1328 * again later for the non-recursive merge.
1329 */
1330 remove_file(o, 0, path, 0);
1331 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1332 if (!ret)
1333 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1334 b->path);
1335 } else {
1336 char *new_path1 = unique_path(o, path, ci->branch1);
1337 char *new_path2 = unique_path(o, path, ci->branch2);
1338 output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1339 a->path, new_path1, b->path, new_path2);
1340 remove_file(o, 0, path, 0);
1341 ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1342 if (!ret)
1343 ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1344 new_path2);
1345 free(new_path2);
1346 free(new_path1);
1347 }
1348
1349 return ret;
1350}
1351
1352/*
1353 * Get the diff_filepairs changed between o_tree and tree.
1354 */
1355static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1356 struct tree *o_tree,
1357 struct tree *tree)
1358{
1359 struct diff_queue_struct *ret;
1360 struct diff_options opts;
1361
1362 diff_setup(&opts);
1363 opts.flags.recursive = 1;
1364 opts.flags.rename_empty = 0;
1365 opts.detect_rename = DIFF_DETECT_RENAME;
1366 opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1367 o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1368 1000;
1369 opts.rename_score = o->rename_score;
1370 opts.show_rename_progress = o->show_rename_progress;
1371 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1372 diff_setup_done(&opts);
1373 diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1374 diffcore_std(&opts);
1375 if (opts.needed_rename_limit > o->needed_rename_limit)
1376 o->needed_rename_limit = opts.needed_rename_limit;
1377
1378 ret = xmalloc(sizeof(*ret));
1379 *ret = diff_queued_diff;
1380
1381 opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1382 diff_queued_diff.nr = 0;
1383 diff_queued_diff.queue = NULL;
1384 diff_flush(&opts);
1385 return ret;
1386}
1387
1388static int tree_has_path(struct tree *tree, const char *path)
1389{
1390 unsigned char hashy[GIT_MAX_RAWSZ];
1391 unsigned int mode_o;
1392
1393 return !get_tree_entry(tree->object.oid.hash, path,
1394 hashy, &mode_o);
1395}
1396
1397static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1398 char **old_dir, char **new_dir)
1399{
1400 char *end_of_old, *end_of_new;
1401 int old_len, new_len;
1402
1403 *old_dir = NULL;
1404 *new_dir = NULL;
1405
1406 /*
1407 * For
1408 * "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1409 * the "e/foo.c" part is the same, we just want to know that
1410 * "a/b/c/d" was renamed to "a/b/some/thing/else"
1411 * so, for this example, this function returns "a/b/c/d" in
1412 * *old_dir and "a/b/some/thing/else" in *new_dir.
1413 *
1414 * Also, if the basename of the file changed, we don't care. We
1415 * want to know which portion of the directory, if any, changed.
1416 */
1417 end_of_old = strrchr(old_path, '/');
1418 end_of_new = strrchr(new_path, '/');
1419
1420 if (end_of_old == NULL || end_of_new == NULL)
1421 return;
1422 while (*--end_of_new == *--end_of_old &&
1423 end_of_old != old_path &&
1424 end_of_new != new_path)
1425 ; /* Do nothing; all in the while loop */
1426 /*
1427 * We've found the first non-matching character in the directory
1428 * paths. That means the current directory we were comparing
1429 * represents the rename. Move end_of_old and end_of_new back
1430 * to the full directory name.
1431 */
1432 if (*end_of_old == '/')
1433 end_of_old++;
1434 if (*end_of_old != '/')
1435 end_of_new++;
1436 end_of_old = strchr(end_of_old, '/');
1437 end_of_new = strchr(end_of_new, '/');
1438
1439 /*
1440 * It may have been the case that old_path and new_path were the same
1441 * directory all along. Don't claim a rename if they're the same.
1442 */
1443 old_len = end_of_old - old_path;
1444 new_len = end_of_new - new_path;
1445
1446 if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1447 *old_dir = xstrndup(old_path, old_len);
1448 *new_dir = xstrndup(new_path, new_len);
1449 }
1450}
1451
1452static void remove_hashmap_entries(struct hashmap *dir_renames,
1453 struct string_list *items_to_remove)
1454{
1455 int i;
1456 struct dir_rename_entry *entry;
1457
1458 for (i = 0; i < items_to_remove->nr; i++) {
1459 entry = items_to_remove->items[i].util;
1460 hashmap_remove(dir_renames, entry, NULL);
1461 }
1462 string_list_clear(items_to_remove, 0);
1463}
1464
1465/*
1466 * There are a couple things we want to do at the directory level:
1467 * 1. Check for both sides renaming to the same thing, in order to avoid
1468 * implicit renaming of files that should be left in place. (See
1469 * testcase 6b in t6043 for details.)
1470 * 2. Prune directory renames if there are still files left in the
1471 * the original directory. These represent a partial directory rename,
1472 * i.e. a rename where only some of the files within the directory
1473 * were renamed elsewhere. (Technically, this could be done earlier
1474 * in get_directory_renames(), except that would prevent us from
1475 * doing the previous check and thus failing testcase 6b.)
1476 * 3. Check for rename/rename(1to2) conflicts (at the directory level).
1477 * In the future, we could potentially record this info as well and
1478 * omit reporting rename/rename(1to2) conflicts for each path within
1479 * the affected directories, thus cleaning up the merge output.
1480 * NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1481 * directory level, because merging directories is fine. If it
1482 * causes conflicts for files within those merged directories, then
1483 * that should be detected at the individual path level.
1484 */
1485static void handle_directory_level_conflicts(struct merge_options *o,
1486 struct hashmap *dir_re_head,
1487 struct tree *head,
1488 struct hashmap *dir_re_merge,
1489 struct tree *merge)
1490{
1491 struct hashmap_iter iter;
1492 struct dir_rename_entry *head_ent;
1493 struct dir_rename_entry *merge_ent;
1494
1495 struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1496 struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1497
1498 hashmap_iter_init(dir_re_head, &iter);
1499 while ((head_ent = hashmap_iter_next(&iter))) {
1500 merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1501 if (merge_ent &&
1502 !head_ent->non_unique_new_dir &&
1503 !merge_ent->non_unique_new_dir &&
1504 !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1505 /* 1. Renamed identically; remove it from both sides */
1506 string_list_append(&remove_from_head,
1507 head_ent->dir)->util = head_ent;
1508 strbuf_release(&head_ent->new_dir);
1509 string_list_append(&remove_from_merge,
1510 merge_ent->dir)->util = merge_ent;
1511 strbuf_release(&merge_ent->new_dir);
1512 } else if (tree_has_path(head, head_ent->dir)) {
1513 /* 2. This wasn't a directory rename after all */
1514 string_list_append(&remove_from_head,
1515 head_ent->dir)->util = head_ent;
1516 strbuf_release(&head_ent->new_dir);
1517 }
1518 }
1519
1520 remove_hashmap_entries(dir_re_head, &remove_from_head);
1521 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1522
1523 hashmap_iter_init(dir_re_merge, &iter);
1524 while ((merge_ent = hashmap_iter_next(&iter))) {
1525 head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1526 if (tree_has_path(merge, merge_ent->dir)) {
1527 /* 2. This wasn't a directory rename after all */
1528 string_list_append(&remove_from_merge,
1529 merge_ent->dir)->util = merge_ent;
1530 } else if (head_ent &&
1531 !head_ent->non_unique_new_dir &&
1532 !merge_ent->non_unique_new_dir) {
1533 /* 3. rename/rename(1to2) */
1534 /*
1535 * We can assume it's not rename/rename(1to1) because
1536 * that was case (1), already checked above. So we
1537 * know that head_ent->new_dir and merge_ent->new_dir
1538 * are different strings.
1539 */
1540 output(o, 1, _("CONFLICT (rename/rename): "
1541 "Rename directory %s->%s in %s. "
1542 "Rename directory %s->%s in %s"),
1543 head_ent->dir, head_ent->new_dir.buf, o->branch1,
1544 head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1545 string_list_append(&remove_from_head,
1546 head_ent->dir)->util = head_ent;
1547 strbuf_release(&head_ent->new_dir);
1548 string_list_append(&remove_from_merge,
1549 merge_ent->dir)->util = merge_ent;
1550 strbuf_release(&merge_ent->new_dir);
1551 }
1552 }
1553
1554 remove_hashmap_entries(dir_re_head, &remove_from_head);
1555 remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1556}
1557
1558static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1559 struct tree *tree)
1560{
1561 struct hashmap *dir_renames;
1562 struct hashmap_iter iter;
1563 struct dir_rename_entry *entry;
1564 int i;
1565
1566 /*
1567 * Typically, we think of a directory rename as all files from a
1568 * certain directory being moved to a target directory. However,
1569 * what if someone first moved two files from the original
1570 * directory in one commit, and then renamed the directory
1571 * somewhere else in a later commit? At merge time, we just know
1572 * that files from the original directory went to two different
1573 * places, and that the bulk of them ended up in the same place.
1574 * We want each directory rename to represent where the bulk of the
1575 * files from that directory end up; this function exists to find
1576 * where the bulk of the files went.
1577 *
1578 * The first loop below simply iterates through the list of file
1579 * renames, finding out how often each directory rename pair
1580 * possibility occurs.
1581 */
1582 dir_renames = xmalloc(sizeof(struct hashmap));
1583 dir_rename_init(dir_renames);
1584 for (i = 0; i < pairs->nr; ++i) {
1585 struct string_list_item *item;
1586 int *count;
1587 struct diff_filepair *pair = pairs->queue[i];
1588 char *old_dir, *new_dir;
1589
1590 /* File not part of directory rename if it wasn't renamed */
1591 if (pair->status != 'R')
1592 continue;
1593
1594 get_renamed_dir_portion(pair->one->path, pair->two->path,
1595 &old_dir, &new_dir);
1596 if (!old_dir)
1597 /* Directory didn't change at all; ignore this one. */
1598 continue;
1599
1600 entry = dir_rename_find_entry(dir_renames, old_dir);
1601 if (!entry) {
1602 entry = xmalloc(sizeof(struct dir_rename_entry));
1603 dir_rename_entry_init(entry, old_dir);
1604 hashmap_put(dir_renames, entry);
1605 } else {
1606 free(old_dir);
1607 }
1608 item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1609 if (!item) {
1610 item = string_list_insert(&entry->possible_new_dirs,
1611 new_dir);
1612 item->util = xcalloc(1, sizeof(int));
1613 } else {
1614 free(new_dir);
1615 }
1616 count = item->util;
1617 *count += 1;
1618 }
1619
1620 /*
1621 * For each directory with files moved out of it, we find out which
1622 * target directory received the most files so we can declare it to
1623 * be the "winning" target location for the directory rename. This
1624 * winner gets recorded in new_dir. If there is no winner
1625 * (multiple target directories received the same number of files),
1626 * we set non_unique_new_dir. Once we've determined the winner (or
1627 * that there is no winner), we no longer need possible_new_dirs.
1628 */
1629 hashmap_iter_init(dir_renames, &iter);
1630 while ((entry = hashmap_iter_next(&iter))) {
1631 int max = 0;
1632 int bad_max = 0;
1633 char *best = NULL;
1634
1635 for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1636 int *count = entry->possible_new_dirs.items[i].util;
1637
1638 if (*count == max)
1639 bad_max = max;
1640 else if (*count > max) {
1641 max = *count;
1642 best = entry->possible_new_dirs.items[i].string;
1643 }
1644 }
1645 if (bad_max == max)
1646 entry->non_unique_new_dir = 1;
1647 else {
1648 assert(entry->new_dir.len == 0);
1649 strbuf_addstr(&entry->new_dir, best);
1650 }
1651 /*
1652 * The relevant directory sub-portion of the original full
1653 * filepaths were xstrndup'ed before inserting into
1654 * possible_new_dirs, and instead of manually iterating the
1655 * list and free'ing each, just lie and tell
1656 * possible_new_dirs that it did the strdup'ing so that it
1657 * will free them for us.
1658 */
1659 entry->possible_new_dirs.strdup_strings = 1;
1660 string_list_clear(&entry->possible_new_dirs, 1);
1661 }
1662
1663 return dir_renames;
1664}
1665
1666/*
1667 * Get information of all renames which occurred in 'pairs', making use of
1668 * any implicit directory renames inferred from the other side of history.
1669 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
1670 * to be able to associate the correct cache entries with the rename
1671 * information; tree is always equal to either a_tree or b_tree.
1672 */
1673static struct string_list *get_renames(struct merge_options *o,
1674 struct diff_queue_struct *pairs,
1675 struct tree *tree,
1676 struct tree *o_tree,
1677 struct tree *a_tree,
1678 struct tree *b_tree,
1679 struct string_list *entries)
1680{
1681 int i;
1682 struct string_list *renames;
1683
1684 renames = xcalloc(1, sizeof(struct string_list));
1685
1686 for (i = 0; i < pairs->nr; ++i) {
1687 struct string_list_item *item;
1688 struct rename *re;
1689 struct diff_filepair *pair = pairs->queue[i];
1690
1691 if (pair->status != 'R') {
1692 diff_free_filepair(pair);
1693 continue;
1694 }
1695 re = xmalloc(sizeof(*re));
1696 re->processed = 0;
1697 re->pair = pair;
1698 item = string_list_lookup(entries, re->pair->one->path);
1699 if (!item)
1700 re->src_entry = insert_stage_data(re->pair->one->path,
1701 o_tree, a_tree, b_tree, entries);
1702 else
1703 re->src_entry = item->util;
1704
1705 item = string_list_lookup(entries, re->pair->two->path);
1706 if (!item)
1707 re->dst_entry = insert_stage_data(re->pair->two->path,
1708 o_tree, a_tree, b_tree, entries);
1709 else
1710 re->dst_entry = item->util;
1711 item = string_list_insert(renames, pair->one->path);
1712 item->util = re;
1713 }
1714 return renames;
1715}
1716
1717static int process_renames(struct merge_options *o,
1718 struct string_list *a_renames,
1719 struct string_list *b_renames)
1720{
1721 int clean_merge = 1, i, j;
1722 struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
1723 struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
1724 const struct rename *sre;
1725
1726 for (i = 0; i < a_renames->nr; i++) {
1727 sre = a_renames->items[i].util;
1728 string_list_insert(&a_by_dst, sre->pair->two->path)->util
1729 = (void *)sre;
1730 }
1731 for (i = 0; i < b_renames->nr; i++) {
1732 sre = b_renames->items[i].util;
1733 string_list_insert(&b_by_dst, sre->pair->two->path)->util
1734 = (void *)sre;
1735 }
1736
1737 for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
1738 struct string_list *renames1, *renames2Dst;
1739 struct rename *ren1 = NULL, *ren2 = NULL;
1740 const char *branch1, *branch2;
1741 const char *ren1_src, *ren1_dst;
1742 struct string_list_item *lookup;
1743
1744 if (i >= a_renames->nr) {
1745 ren2 = b_renames->items[j++].util;
1746 } else if (j >= b_renames->nr) {
1747 ren1 = a_renames->items[i++].util;
1748 } else {
1749 int compare = strcmp(a_renames->items[i].string,
1750 b_renames->items[j].string);
1751 if (compare <= 0)
1752 ren1 = a_renames->items[i++].util;
1753 if (compare >= 0)
1754 ren2 = b_renames->items[j++].util;
1755 }
1756
1757 /* TODO: refactor, so that 1/2 are not needed */
1758 if (ren1) {
1759 renames1 = a_renames;
1760 renames2Dst = &b_by_dst;
1761 branch1 = o->branch1;
1762 branch2 = o->branch2;
1763 } else {
1764 renames1 = b_renames;
1765 renames2Dst = &a_by_dst;
1766 branch1 = o->branch2;
1767 branch2 = o->branch1;
1768 SWAP(ren2, ren1);
1769 }
1770
1771 if (ren1->processed)
1772 continue;
1773 ren1->processed = 1;
1774 ren1->dst_entry->processed = 1;
1775 /* BUG: We should only mark src_entry as processed if we
1776 * are not dealing with a rename + add-source case.
1777 */
1778 ren1->src_entry->processed = 1;
1779
1780 ren1_src = ren1->pair->one->path;
1781 ren1_dst = ren1->pair->two->path;
1782
1783 if (ren2) {
1784 /* One file renamed on both sides */
1785 const char *ren2_src = ren2->pair->one->path;
1786 const char *ren2_dst = ren2->pair->two->path;
1787 enum rename_type rename_type;
1788 if (strcmp(ren1_src, ren2_src) != 0)
1789 die("BUG: ren1_src != ren2_src");
1790 ren2->dst_entry->processed = 1;
1791 ren2->processed = 1;
1792 if (strcmp(ren1_dst, ren2_dst) != 0) {
1793 rename_type = RENAME_ONE_FILE_TO_TWO;
1794 clean_merge = 0;
1795 } else {
1796 rename_type = RENAME_ONE_FILE_TO_ONE;
1797 /* BUG: We should only remove ren1_src in
1798 * the base stage (think of rename +
1799 * add-source cases).
1800 */
1801 remove_file(o, 1, ren1_src, 1);
1802 update_entry(ren1->dst_entry,
1803 ren1->pair->one,
1804 ren1->pair->two,
1805 ren2->pair->two);
1806 }
1807 setup_rename_conflict_info(rename_type,
1808 ren1->pair,
1809 ren2->pair,
1810 branch1,
1811 branch2,
1812 ren1->dst_entry,
1813 ren2->dst_entry,
1814 o,
1815 NULL,
1816 NULL);
1817 } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
1818 /* Two different files renamed to the same thing */
1819 char *ren2_dst;
1820 ren2 = lookup->util;
1821 ren2_dst = ren2->pair->two->path;
1822 if (strcmp(ren1_dst, ren2_dst) != 0)
1823 die("BUG: ren1_dst != ren2_dst");
1824
1825 clean_merge = 0;
1826 ren2->processed = 1;
1827 /*
1828 * BUG: We should only mark src_entry as processed
1829 * if we are not dealing with a rename + add-source
1830 * case.
1831 */
1832 ren2->src_entry->processed = 1;
1833
1834 setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
1835 ren1->pair,
1836 ren2->pair,
1837 branch1,
1838 branch2,
1839 ren1->dst_entry,
1840 ren2->dst_entry,
1841 o,
1842 ren1->src_entry,
1843 ren2->src_entry);
1844
1845 } else {
1846 /* Renamed in 1, maybe changed in 2 */
1847 /* we only use sha1 and mode of these */
1848 struct diff_filespec src_other, dst_other;
1849 int try_merge;
1850
1851 /*
1852 * unpack_trees loads entries from common-commit
1853 * into stage 1, from head-commit into stage 2, and
1854 * from merge-commit into stage 3. We keep track
1855 * of which side corresponds to the rename.
1856 */
1857 int renamed_stage = a_renames == renames1 ? 2 : 3;
1858 int other_stage = a_renames == renames1 ? 3 : 2;
1859
1860 /* BUG: We should only remove ren1_src in the base
1861 * stage and in other_stage (think of rename +
1862 * add-source case).
1863 */
1864 remove_file(o, 1, ren1_src,
1865 renamed_stage == 2 || !was_tracked(ren1_src));
1866
1867 oidcpy(&src_other.oid,
1868 &ren1->src_entry->stages[other_stage].oid);
1869 src_other.mode = ren1->src_entry->stages[other_stage].mode;
1870 oidcpy(&dst_other.oid,
1871 &ren1->dst_entry->stages[other_stage].oid);
1872 dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
1873 try_merge = 0;
1874
1875 if (oid_eq(&src_other.oid, &null_oid)) {
1876 setup_rename_conflict_info(RENAME_DELETE,
1877 ren1->pair,
1878 NULL,
1879 branch1,
1880 branch2,
1881 ren1->dst_entry,
1882 NULL,
1883 o,
1884 NULL,
1885 NULL);
1886 } else if ((dst_other.mode == ren1->pair->two->mode) &&
1887 oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
1888 /*
1889 * Added file on the other side identical to
1890 * the file being renamed: clean merge.
1891 * Also, there is no need to overwrite the
1892 * file already in the working copy, so call
1893 * update_file_flags() instead of
1894 * update_file().
1895 */
1896 if (update_file_flags(o,
1897 &ren1->pair->two->oid,
1898 ren1->pair->two->mode,
1899 ren1_dst,
1900 1, /* update_cache */
1901 0 /* update_wd */))
1902 clean_merge = -1;
1903 } else if (!oid_eq(&dst_other.oid, &null_oid)) {
1904 clean_merge = 0;
1905 try_merge = 1;
1906 output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
1907 "%s added in %s"),
1908 ren1_src, ren1_dst, branch1,
1909 ren1_dst, branch2);
1910 if (o->call_depth) {
1911 struct merge_file_info mfi;
1912 if (merge_file_one(o, ren1_dst, &null_oid, 0,
1913 &ren1->pair->two->oid,
1914 ren1->pair->two->mode,
1915 &dst_other.oid,
1916 dst_other.mode,
1917 branch1, branch2, &mfi)) {
1918 clean_merge = -1;
1919 goto cleanup_and_return;
1920 }
1921 output(o, 1, _("Adding merged %s"), ren1_dst);
1922 if (update_file(o, 0, &mfi.oid,
1923 mfi.mode, ren1_dst))
1924 clean_merge = -1;
1925 try_merge = 0;
1926 } else {
1927 char *new_path = unique_path(o, ren1_dst, branch2);
1928 output(o, 1, _("Adding as %s instead"), new_path);
1929 if (update_file(o, 0, &dst_other.oid,
1930 dst_other.mode, new_path))
1931 clean_merge = -1;
1932 free(new_path);
1933 }
1934 } else
1935 try_merge = 1;
1936
1937 if (clean_merge < 0)
1938 goto cleanup_and_return;
1939 if (try_merge) {
1940 struct diff_filespec *one, *a, *b;
1941 src_other.path = (char *)ren1_src;
1942
1943 one = ren1->pair->one;
1944 if (a_renames == renames1) {
1945 a = ren1->pair->two;
1946 b = &src_other;
1947 } else {
1948 b = ren1->pair->two;
1949 a = &src_other;
1950 }
1951 update_entry(ren1->dst_entry, one, a, b);
1952 setup_rename_conflict_info(RENAME_NORMAL,
1953 ren1->pair,
1954 NULL,
1955 branch1,
1956 NULL,
1957 ren1->dst_entry,
1958 NULL,
1959 o,
1960 NULL,
1961 NULL);
1962 }
1963 }
1964 }
1965cleanup_and_return:
1966 string_list_clear(&a_by_dst, 0);
1967 string_list_clear(&b_by_dst, 0);
1968
1969 return clean_merge;
1970}
1971
1972struct rename_info {
1973 struct string_list *head_renames;
1974 struct string_list *merge_renames;
1975};
1976
1977static void initial_cleanup_rename(struct diff_queue_struct *pairs,
1978 struct hashmap *dir_renames)
1979{
1980 struct hashmap_iter iter;
1981 struct dir_rename_entry *e;
1982
1983 hashmap_iter_init(dir_renames, &iter);
1984 while ((e = hashmap_iter_next(&iter))) {
1985 free(e->dir);
1986 strbuf_release(&e->new_dir);
1987 /* possible_new_dirs already cleared in get_directory_renames */
1988 }
1989 hashmap_free(dir_renames, 1);
1990 free(dir_renames);
1991
1992 free(pairs->queue);
1993 free(pairs);
1994}
1995
1996static int handle_renames(struct merge_options *o,
1997 struct tree *common,
1998 struct tree *head,
1999 struct tree *merge,
2000 struct string_list *entries,
2001 struct rename_info *ri)
2002{
2003 struct diff_queue_struct *head_pairs, *merge_pairs;
2004 struct hashmap *dir_re_head, *dir_re_merge;
2005 int clean;
2006
2007 ri->head_renames = NULL;
2008 ri->merge_renames = NULL;
2009
2010 if (!o->detect_rename)
2011 return 1;
2012
2013 head_pairs = get_diffpairs(o, common, head);
2014 merge_pairs = get_diffpairs(o, common, merge);
2015
2016 dir_re_head = get_directory_renames(head_pairs, head);
2017 dir_re_merge = get_directory_renames(merge_pairs, merge);
2018
2019 handle_directory_level_conflicts(o,
2020 dir_re_head, head,
2021 dir_re_merge, merge);
2022
2023 ri->head_renames = get_renames(o, head_pairs, head,
2024 common, head, merge, entries);
2025 ri->merge_renames = get_renames(o, merge_pairs, merge,
2026 common, head, merge, entries);
2027 clean = process_renames(o, ri->head_renames, ri->merge_renames);
2028
2029 /*
2030 * Some cleanup is deferred until cleanup_renames() because the
2031 * data structures are still needed and referenced in
2032 * process_entry(). But there are a few things we can free now.
2033 */
2034 initial_cleanup_rename(head_pairs, dir_re_head);
2035 initial_cleanup_rename(merge_pairs, dir_re_merge);
2036
2037 return clean;
2038}
2039
2040static void final_cleanup_rename(struct string_list *rename)
2041{
2042 const struct rename *re;
2043 int i;
2044
2045 if (rename == NULL)
2046 return;
2047
2048 for (i = 0; i < rename->nr; i++) {
2049 re = rename->items[i].util;
2050 diff_free_filepair(re->pair);
2051 }
2052 string_list_clear(rename, 1);
2053 free(rename);
2054}
2055
2056static void final_cleanup_renames(struct rename_info *re_info)
2057{
2058 final_cleanup_rename(re_info->head_renames);
2059 final_cleanup_rename(re_info->merge_renames);
2060}
2061
2062static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2063{
2064 return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2065}
2066
2067static int read_oid_strbuf(struct merge_options *o,
2068 const struct object_id *oid, struct strbuf *dst)
2069{
2070 void *buf;
2071 enum object_type type;
2072 unsigned long size;
2073 buf = read_sha1_file(oid->hash, &type, &size);
2074 if (!buf)
2075 return err(o, _("cannot read object %s"), oid_to_hex(oid));
2076 if (type != OBJ_BLOB) {
2077 free(buf);
2078 return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2079 }
2080 strbuf_attach(dst, buf, size, size + 1);
2081 return 0;
2082}
2083
2084static int blob_unchanged(struct merge_options *opt,
2085 const struct object_id *o_oid,
2086 unsigned o_mode,
2087 const struct object_id *a_oid,
2088 unsigned a_mode,
2089 int renormalize, const char *path)
2090{
2091 struct strbuf o = STRBUF_INIT;
2092 struct strbuf a = STRBUF_INIT;
2093 int ret = 0; /* assume changed for safety */
2094
2095 if (a_mode != o_mode)
2096 return 0;
2097 if (oid_eq(o_oid, a_oid))
2098 return 1;
2099 if (!renormalize)
2100 return 0;
2101
2102 assert(o_oid && a_oid);
2103 if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2104 goto error_return;
2105 /*
2106 * Note: binary | is used so that both renormalizations are
2107 * performed. Comparison can be skipped if both files are
2108 * unchanged since their sha1s have already been compared.
2109 */
2110 if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2111 renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2112 ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2113
2114error_return:
2115 strbuf_release(&o);
2116 strbuf_release(&a);
2117 return ret;
2118}
2119
2120static int handle_modify_delete(struct merge_options *o,
2121 const char *path,
2122 struct object_id *o_oid, int o_mode,
2123 struct object_id *a_oid, int a_mode,
2124 struct object_id *b_oid, int b_mode)
2125{
2126 const char *modify_branch, *delete_branch;
2127 struct object_id *changed_oid;
2128 int changed_mode;
2129
2130 if (a_oid) {
2131 modify_branch = o->branch1;
2132 delete_branch = o->branch2;
2133 changed_oid = a_oid;
2134 changed_mode = a_mode;
2135 } else {
2136 modify_branch = o->branch2;
2137 delete_branch = o->branch1;
2138 changed_oid = b_oid;
2139 changed_mode = b_mode;
2140 }
2141
2142 return handle_change_delete(o,
2143 path, NULL,
2144 o_oid, o_mode,
2145 changed_oid, changed_mode,
2146 modify_branch, delete_branch,
2147 _("modify"), _("modified"));
2148}
2149
2150static int merge_content(struct merge_options *o,
2151 const char *path,
2152 struct object_id *o_oid, int o_mode,
2153 struct object_id *a_oid, int a_mode,
2154 struct object_id *b_oid, int b_mode,
2155 struct rename_conflict_info *rename_conflict_info)
2156{
2157 const char *reason = _("content");
2158 const char *path1 = NULL, *path2 = NULL;
2159 struct merge_file_info mfi;
2160 struct diff_filespec one, a, b;
2161 unsigned df_conflict_remains = 0;
2162
2163 if (!o_oid) {
2164 reason = _("add/add");
2165 o_oid = (struct object_id *)&null_oid;
2166 }
2167 one.path = a.path = b.path = (char *)path;
2168 oidcpy(&one.oid, o_oid);
2169 one.mode = o_mode;
2170 oidcpy(&a.oid, a_oid);
2171 a.mode = a_mode;
2172 oidcpy(&b.oid, b_oid);
2173 b.mode = b_mode;
2174
2175 if (rename_conflict_info) {
2176 struct diff_filepair *pair1 = rename_conflict_info->pair1;
2177
2178 path1 = (o->branch1 == rename_conflict_info->branch1) ?
2179 pair1->two->path : pair1->one->path;
2180 /* If rename_conflict_info->pair2 != NULL, we are in
2181 * RENAME_ONE_FILE_TO_ONE case. Otherwise, we have a
2182 * normal rename.
2183 */
2184 path2 = (rename_conflict_info->pair2 ||
2185 o->branch2 == rename_conflict_info->branch1) ?
2186 pair1->two->path : pair1->one->path;
2187
2188 if (dir_in_way(path, !o->call_depth,
2189 S_ISGITLINK(pair1->two->mode)))
2190 df_conflict_remains = 1;
2191 }
2192 if (merge_file_special_markers(o, &one, &a, &b,
2193 o->branch1, path1,
2194 o->branch2, path2, &mfi))
2195 return -1;
2196
2197 if (mfi.clean && !df_conflict_remains &&
2198 oid_eq(&mfi.oid, a_oid) && mfi.mode == a_mode) {
2199 int path_renamed_outside_HEAD;
2200 output(o, 3, _("Skipped %s (merged same as existing)"), path);
2201 /*
2202 * The content merge resulted in the same file contents we
2203 * already had. We can return early if those file contents
2204 * are recorded at the correct path (which may not be true
2205 * if the merge involves a rename).
2206 */
2207 path_renamed_outside_HEAD = !path2 || !strcmp(path, path2);
2208 if (!path_renamed_outside_HEAD) {
2209 add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2210 0, (!o->call_depth), 0);
2211 return mfi.clean;
2212 }
2213 } else
2214 output(o, 2, _("Auto-merging %s"), path);
2215
2216 if (!mfi.clean) {
2217 if (S_ISGITLINK(mfi.mode))
2218 reason = _("submodule");
2219 output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2220 reason, path);
2221 if (rename_conflict_info && !df_conflict_remains)
2222 if (update_stages(o, path, &one, &a, &b))
2223 return -1;
2224 }
2225
2226 if (df_conflict_remains) {
2227 char *new_path;
2228 if (o->call_depth) {
2229 remove_file_from_cache(path);
2230 } else {
2231 if (!mfi.clean) {
2232 if (update_stages(o, path, &one, &a, &b))
2233 return -1;
2234 } else {
2235 int file_from_stage2 = was_tracked(path);
2236 struct diff_filespec merged;
2237 oidcpy(&merged.oid, &mfi.oid);
2238 merged.mode = mfi.mode;
2239
2240 if (update_stages(o, path, NULL,
2241 file_from_stage2 ? &merged : NULL,
2242 file_from_stage2 ? NULL : &merged))
2243 return -1;
2244 }
2245
2246 }
2247 new_path = unique_path(o, path, rename_conflict_info->branch1);
2248 output(o, 1, _("Adding as %s instead"), new_path);
2249 if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2250 free(new_path);
2251 return -1;
2252 }
2253 free(new_path);
2254 mfi.clean = 0;
2255 } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2256 return -1;
2257 return mfi.clean;
2258}
2259
2260/* Per entry merge function */
2261static int process_entry(struct merge_options *o,
2262 const char *path, struct stage_data *entry)
2263{
2264 int clean_merge = 1;
2265 int normalize = o->renormalize;
2266 unsigned o_mode = entry->stages[1].mode;
2267 unsigned a_mode = entry->stages[2].mode;
2268 unsigned b_mode = entry->stages[3].mode;
2269 struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2270 struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2271 struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2272
2273 entry->processed = 1;
2274 if (entry->rename_conflict_info) {
2275 struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2276 switch (conflict_info->rename_type) {
2277 case RENAME_NORMAL:
2278 case RENAME_ONE_FILE_TO_ONE:
2279 clean_merge = merge_content(o, path,
2280 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2281 conflict_info);
2282 break;
2283 case RENAME_DELETE:
2284 clean_merge = 0;
2285 if (conflict_rename_delete(o,
2286 conflict_info->pair1,
2287 conflict_info->branch1,
2288 conflict_info->branch2))
2289 clean_merge = -1;
2290 break;
2291 case RENAME_ONE_FILE_TO_TWO:
2292 clean_merge = 0;
2293 if (conflict_rename_rename_1to2(o, conflict_info))
2294 clean_merge = -1;
2295 break;
2296 case RENAME_TWO_FILES_TO_ONE:
2297 clean_merge = 0;
2298 if (conflict_rename_rename_2to1(o, conflict_info))
2299 clean_merge = -1;
2300 break;
2301 default:
2302 entry->processed = 0;
2303 break;
2304 }
2305 } else if (o_oid && (!a_oid || !b_oid)) {
2306 /* Case A: Deleted in one */
2307 if ((!a_oid && !b_oid) ||
2308 (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2309 (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2310 /* Deleted in both or deleted in one and
2311 * unchanged in the other */
2312 if (a_oid)
2313 output(o, 2, _("Removing %s"), path);
2314 /* do not touch working file if it did not exist */
2315 remove_file(o, 1, path, !a_oid);
2316 } else {
2317 /* Modify/delete; deleted side may have put a directory in the way */
2318 clean_merge = 0;
2319 if (handle_modify_delete(o, path, o_oid, o_mode,
2320 a_oid, a_mode, b_oid, b_mode))
2321 clean_merge = -1;
2322 }
2323 } else if ((!o_oid && a_oid && !b_oid) ||
2324 (!o_oid && !a_oid && b_oid)) {
2325 /* Case B: Added in one. */
2326 /* [nothing|directory] -> ([nothing|directory], file) */
2327
2328 const char *add_branch;
2329 const char *other_branch;
2330 unsigned mode;
2331 const struct object_id *oid;
2332 const char *conf;
2333
2334 if (a_oid) {
2335 add_branch = o->branch1;
2336 other_branch = o->branch2;
2337 mode = a_mode;
2338 oid = a_oid;
2339 conf = _("file/directory");
2340 } else {
2341 add_branch = o->branch2;
2342 other_branch = o->branch1;
2343 mode = b_mode;
2344 oid = b_oid;
2345 conf = _("directory/file");
2346 }
2347 if (dir_in_way(path,
2348 !o->call_depth && !S_ISGITLINK(a_mode),
2349 0)) {
2350 char *new_path = unique_path(o, path, add_branch);
2351 clean_merge = 0;
2352 output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
2353 "Adding %s as %s"),
2354 conf, path, other_branch, path, new_path);
2355 if (update_file(o, 0, oid, mode, new_path))
2356 clean_merge = -1;
2357 else if (o->call_depth)
2358 remove_file_from_cache(path);
2359 free(new_path);
2360 } else {
2361 output(o, 2, _("Adding %s"), path);
2362 /* do not overwrite file if already present */
2363 if (update_file_flags(o, oid, mode, path, 1, !a_oid))
2364 clean_merge = -1;
2365 }
2366 } else if (a_oid && b_oid) {
2367 /* Case C: Added in both (check for same permissions) and */
2368 /* case D: Modified in both, but differently. */
2369 clean_merge = merge_content(o, path,
2370 o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2371 NULL);
2372 } else if (!o_oid && !a_oid && !b_oid) {
2373 /*
2374 * this entry was deleted altogether. a_mode == 0 means
2375 * we had that path and want to actively remove it.
2376 */
2377 remove_file(o, 1, path, !a_mode);
2378 } else
2379 die("BUG: fatal merge failure, shouldn't happen.");
2380
2381 return clean_merge;
2382}
2383
2384int merge_trees(struct merge_options *o,
2385 struct tree *head,
2386 struct tree *merge,
2387 struct tree *common,
2388 struct tree **result)
2389{
2390 int code, clean;
2391
2392 if (o->subtree_shift) {
2393 merge = shift_tree_object(head, merge, o->subtree_shift);
2394 common = shift_tree_object(head, common, o->subtree_shift);
2395 }
2396
2397 if (oid_eq(&common->object.oid, &merge->object.oid)) {
2398 struct strbuf sb = STRBUF_INIT;
2399
2400 if (!o->call_depth && index_has_changes(&sb)) {
2401 err(o, _("Dirty index: cannot merge (dirty: %s)"),
2402 sb.buf);
2403 return 0;
2404 }
2405 output(o, 0, _("Already up to date!"));
2406 *result = head;
2407 return 1;
2408 }
2409
2410 code = git_merge_trees(o->call_depth, common, head, merge);
2411
2412 if (code != 0) {
2413 if (show(o, 4) || o->call_depth)
2414 err(o, _("merging of trees %s and %s failed"),
2415 oid_to_hex(&head->object.oid),
2416 oid_to_hex(&merge->object.oid));
2417 return -1;
2418 }
2419
2420 if (unmerged_cache()) {
2421 struct string_list *entries;
2422 struct rename_info re_info;
2423 int i;
2424 /*
2425 * Only need the hashmap while processing entries, so
2426 * initialize it here and free it when we are done running
2427 * through the entries. Keeping it in the merge_options as
2428 * opposed to decaring a local hashmap is for convenience
2429 * so that we don't have to pass it to around.
2430 */
2431 hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
2432 get_files_dirs(o, head);
2433 get_files_dirs(o, merge);
2434
2435 entries = get_unmerged();
2436 clean = handle_renames(o, common, head, merge, entries,
2437 &re_info);
2438 record_df_conflict_files(o, entries);
2439 if (clean < 0)
2440 goto cleanup;
2441 for (i = entries->nr-1; 0 <= i; i--) {
2442 const char *path = entries->items[i].string;
2443 struct stage_data *e = entries->items[i].util;
2444 if (!e->processed) {
2445 int ret = process_entry(o, path, e);
2446 if (!ret)
2447 clean = 0;
2448 else if (ret < 0) {
2449 clean = ret;
2450 goto cleanup;
2451 }
2452 }
2453 }
2454 for (i = 0; i < entries->nr; i++) {
2455 struct stage_data *e = entries->items[i].util;
2456 if (!e->processed)
2457 die("BUG: unprocessed path??? %s",
2458 entries->items[i].string);
2459 }
2460
2461cleanup:
2462 final_cleanup_renames(&re_info);
2463
2464 string_list_clear(entries, 1);
2465 free(entries);
2466
2467 hashmap_free(&o->current_file_dir_set, 1);
2468
2469 if (clean < 0)
2470 return clean;
2471 }
2472 else
2473 clean = 1;
2474
2475 if (o->call_depth && !(*result = write_tree_from_memory(o)))
2476 return -1;
2477
2478 return clean;
2479}
2480
2481static struct commit_list *reverse_commit_list(struct commit_list *list)
2482{
2483 struct commit_list *next = NULL, *current, *backup;
2484 for (current = list; current; current = backup) {
2485 backup = current->next;
2486 current->next = next;
2487 next = current;
2488 }
2489 return next;
2490}
2491
2492/*
2493 * Merge the commits h1 and h2, return the resulting virtual
2494 * commit object and a flag indicating the cleanness of the merge.
2495 */
2496int merge_recursive(struct merge_options *o,
2497 struct commit *h1,
2498 struct commit *h2,
2499 struct commit_list *ca,
2500 struct commit **result)
2501{
2502 struct commit_list *iter;
2503 struct commit *merged_common_ancestors;
2504 struct tree *mrtree = mrtree;
2505 int clean;
2506
2507 if (show(o, 4)) {
2508 output(o, 4, _("Merging:"));
2509 output_commit_title(o, h1);
2510 output_commit_title(o, h2);
2511 }
2512
2513 if (!ca) {
2514 ca = get_merge_bases(h1, h2);
2515 ca = reverse_commit_list(ca);
2516 }
2517
2518 if (show(o, 5)) {
2519 unsigned cnt = commit_list_count(ca);
2520
2521 output(o, 5, Q_("found %u common ancestor:",
2522 "found %u common ancestors:", cnt), cnt);
2523 for (iter = ca; iter; iter = iter->next)
2524 output_commit_title(o, iter->item);
2525 }
2526
2527 merged_common_ancestors = pop_commit(&ca);
2528 if (merged_common_ancestors == NULL) {
2529 /* if there is no common ancestor, use an empty tree */
2530 struct tree *tree;
2531
2532 tree = lookup_tree(the_hash_algo->empty_tree);
2533 merged_common_ancestors = make_virtual_commit(tree, "ancestor");
2534 }
2535
2536 for (iter = ca; iter; iter = iter->next) {
2537 const char *saved_b1, *saved_b2;
2538 o->call_depth++;
2539 /*
2540 * When the merge fails, the result contains files
2541 * with conflict markers. The cleanness flag is
2542 * ignored (unless indicating an error), it was never
2543 * actually used, as result of merge_trees has always
2544 * overwritten it: the committed "conflicts" were
2545 * already resolved.
2546 */
2547 discard_cache();
2548 saved_b1 = o->branch1;
2549 saved_b2 = o->branch2;
2550 o->branch1 = "Temporary merge branch 1";
2551 o->branch2 = "Temporary merge branch 2";
2552 if (merge_recursive(o, merged_common_ancestors, iter->item,
2553 NULL, &merged_common_ancestors) < 0)
2554 return -1;
2555 o->branch1 = saved_b1;
2556 o->branch2 = saved_b2;
2557 o->call_depth--;
2558
2559 if (!merged_common_ancestors)
2560 return err(o, _("merge returned no commit"));
2561 }
2562
2563 discard_cache();
2564 if (!o->call_depth)
2565 read_cache();
2566
2567 o->ancestor = "merged common ancestors";
2568 clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,
2569 &mrtree);
2570 if (clean < 0) {
2571 flush_output(o);
2572 return clean;
2573 }
2574
2575 if (o->call_depth) {
2576 *result = make_virtual_commit(mrtree, "merged tree");
2577 commit_list_insert(h1, &(*result)->parents);
2578 commit_list_insert(h2, &(*result)->parents->next);
2579 }
2580 flush_output(o);
2581 if (!o->call_depth && o->buffer_output < 2)
2582 strbuf_release(&o->obuf);
2583 if (show(o, 2))
2584 diff_warn_rename_limit("merge.renamelimit",
2585 o->needed_rename_limit, 0);
2586 return clean;
2587}
2588
2589static struct commit *get_ref(const struct object_id *oid, const char *name)
2590{
2591 struct object *object;
2592
2593 object = deref_tag(parse_object(oid), name, strlen(name));
2594 if (!object)
2595 return NULL;
2596 if (object->type == OBJ_TREE)
2597 return make_virtual_commit((struct tree*)object, name);
2598 if (object->type != OBJ_COMMIT)
2599 return NULL;
2600 if (parse_commit((struct commit *)object))
2601 return NULL;
2602 return (struct commit *)object;
2603}
2604
2605int merge_recursive_generic(struct merge_options *o,
2606 const struct object_id *head,
2607 const struct object_id *merge,
2608 int num_base_list,
2609 const struct object_id **base_list,
2610 struct commit **result)
2611{
2612 int clean;
2613 struct lock_file lock = LOCK_INIT;
2614 struct commit *head_commit = get_ref(head, o->branch1);
2615 struct commit *next_commit = get_ref(merge, o->branch2);
2616 struct commit_list *ca = NULL;
2617
2618 if (base_list) {
2619 int i;
2620 for (i = 0; i < num_base_list; ++i) {
2621 struct commit *base;
2622 if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
2623 return err(o, _("Could not parse object '%s'"),
2624 oid_to_hex(base_list[i]));
2625 commit_list_insert(base, &ca);
2626 }
2627 }
2628
2629 hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
2630 clean = merge_recursive(o, head_commit, next_commit, ca,
2631 result);
2632 if (clean < 0)
2633 return clean;
2634
2635 if (active_cache_changed &&
2636 write_locked_index(&the_index, &lock, COMMIT_LOCK))
2637 return err(o, _("Unable to write index."));
2638
2639 return clean ? 0 : 1;
2640}
2641
2642static void merge_recursive_config(struct merge_options *o)
2643{
2644 git_config_get_int("merge.verbosity", &o->verbosity);
2645 git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
2646 git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
2647 git_config(git_xmerge_config, NULL);
2648}
2649
2650void init_merge_options(struct merge_options *o)
2651{
2652 const char *merge_verbosity;
2653 memset(o, 0, sizeof(struct merge_options));
2654 o->verbosity = 2;
2655 o->buffer_output = 1;
2656 o->diff_rename_limit = -1;
2657 o->merge_rename_limit = -1;
2658 o->renormalize = 0;
2659 o->detect_rename = 1;
2660 merge_recursive_config(o);
2661 merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
2662 if (merge_verbosity)
2663 o->verbosity = strtol(merge_verbosity, NULL, 10);
2664 if (o->verbosity >= 5)
2665 o->buffer_output = 0;
2666 strbuf_init(&o->obuf, 0);
2667 string_list_init(&o->df_conflict_file_set, 1);
2668}
2669
2670int parse_merge_opt(struct merge_options *o, const char *s)
2671{
2672 const char *arg;
2673
2674 if (!s || !*s)
2675 return -1;
2676 if (!strcmp(s, "ours"))
2677 o->recursive_variant = MERGE_RECURSIVE_OURS;
2678 else if (!strcmp(s, "theirs"))
2679 o->recursive_variant = MERGE_RECURSIVE_THEIRS;
2680 else if (!strcmp(s, "subtree"))
2681 o->subtree_shift = "";
2682 else if (skip_prefix(s, "subtree=", &arg))
2683 o->subtree_shift = arg;
2684 else if (!strcmp(s, "patience"))
2685 o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
2686 else if (!strcmp(s, "histogram"))
2687 o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
2688 else if (skip_prefix(s, "diff-algorithm=", &arg)) {
2689 long value = parse_algorithm_value(arg);
2690 if (value < 0)
2691 return -1;
2692 /* clear out previous settings */
2693 DIFF_XDL_CLR(o, NEED_MINIMAL);
2694 o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
2695 o->xdl_opts |= value;
2696 }
2697 else if (!strcmp(s, "ignore-space-change"))
2698 DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
2699 else if (!strcmp(s, "ignore-all-space"))
2700 DIFF_XDL_SET(o, IGNORE_WHITESPACE);
2701 else if (!strcmp(s, "ignore-space-at-eol"))
2702 DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
2703 else if (!strcmp(s, "ignore-cr-at-eol"))
2704 DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
2705 else if (!strcmp(s, "renormalize"))
2706 o->renormalize = 1;
2707 else if (!strcmp(s, "no-renormalize"))
2708 o->renormalize = 0;
2709 else if (!strcmp(s, "no-renames"))
2710 o->detect_rename = 0;
2711 else if (!strcmp(s, "find-renames")) {
2712 o->detect_rename = 1;
2713 o->rename_score = 0;
2714 }
2715 else if (skip_prefix(s, "find-renames=", &arg) ||
2716 skip_prefix(s, "rename-threshold=", &arg)) {
2717 if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
2718 return -1;
2719 o->detect_rename = 1;
2720 }
2721 else
2722 return -1;
2723 return 0;
2724}