merge-recursive.con commit t6036, t6042: testcases for rename collision of already conflicting files (b28eeb3)
   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 "object-store.h"
  12#include "repository.h"
  13#include "commit.h"
  14#include "blob.h"
  15#include "builtin.h"
  16#include "tree-walk.h"
  17#include "diff.h"
  18#include "diffcore.h"
  19#include "tag.h"
  20#include "alloc.h"
  21#include "unpack-trees.h"
  22#include "string-list.h"
  23#include "xdiff-interface.h"
  24#include "ll-merge.h"
  25#include "attr.h"
  26#include "merge-recursive.h"
  27#include "dir.h"
  28#include "submodule.h"
  29#include "revision.h"
  30#include "commit-reach.h"
  31
  32struct path_hashmap_entry {
  33        struct hashmap_entry e;
  34        char path[FLEX_ARRAY];
  35};
  36
  37static int path_hashmap_cmp(const void *cmp_data,
  38                            const void *entry,
  39                            const void *entry_or_key,
  40                            const void *keydata)
  41{
  42        const struct path_hashmap_entry *a = entry;
  43        const struct path_hashmap_entry *b = entry_or_key;
  44        const char *key = keydata;
  45
  46        if (ignore_case)
  47                return strcasecmp(a->path, key ? key : b->path);
  48        else
  49                return strcmp(a->path, key ? key : b->path);
  50}
  51
  52static unsigned int path_hash(const char *path)
  53{
  54        return ignore_case ? strihash(path) : strhash(path);
  55}
  56
  57static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
  58                                                      char *dir)
  59{
  60        struct dir_rename_entry key;
  61
  62        if (dir == NULL)
  63                return NULL;
  64        hashmap_entry_init(&key, strhash(dir));
  65        key.dir = dir;
  66        return hashmap_get(hashmap, &key, NULL);
  67}
  68
  69static int dir_rename_cmp(const void *unused_cmp_data,
  70                          const void *entry,
  71                          const void *entry_or_key,
  72                          const void *unused_keydata)
  73{
  74        const struct dir_rename_entry *e1 = entry;
  75        const struct dir_rename_entry *e2 = entry_or_key;
  76
  77        return strcmp(e1->dir, e2->dir);
  78}
  79
  80static void dir_rename_init(struct hashmap *map)
  81{
  82        hashmap_init(map, dir_rename_cmp, NULL, 0);
  83}
  84
  85static void dir_rename_entry_init(struct dir_rename_entry *entry,
  86                                  char *directory)
  87{
  88        hashmap_entry_init(entry, strhash(directory));
  89        entry->dir = directory;
  90        entry->non_unique_new_dir = 0;
  91        strbuf_init(&entry->new_dir, 0);
  92        string_list_init(&entry->possible_new_dirs, 0);
  93}
  94
  95static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
  96                                                    char *target_file)
  97{
  98        struct collision_entry key;
  99
 100        hashmap_entry_init(&key, strhash(target_file));
 101        key.target_file = target_file;
 102        return hashmap_get(hashmap, &key, NULL);
 103}
 104
 105static int collision_cmp(void *unused_cmp_data,
 106                         const struct collision_entry *e1,
 107                         const struct collision_entry *e2,
 108                         const void *unused_keydata)
 109{
 110        return strcmp(e1->target_file, e2->target_file);
 111}
 112
 113static void collision_init(struct hashmap *map)
 114{
 115        hashmap_init(map, (hashmap_cmp_fn) collision_cmp, NULL, 0);
 116}
 117
 118static void flush_output(struct merge_options *o)
 119{
 120        if (o->buffer_output < 2 && o->obuf.len) {
 121                fputs(o->obuf.buf, stdout);
 122                strbuf_reset(&o->obuf);
 123        }
 124}
 125
 126static int err(struct merge_options *o, const char *err, ...)
 127{
 128        va_list params;
 129
 130        if (o->buffer_output < 2)
 131                flush_output(o);
 132        else {
 133                strbuf_complete(&o->obuf, '\n');
 134                strbuf_addstr(&o->obuf, "error: ");
 135        }
 136        va_start(params, err);
 137        strbuf_vaddf(&o->obuf, err, params);
 138        va_end(params);
 139        if (o->buffer_output > 1)
 140                strbuf_addch(&o->obuf, '\n');
 141        else {
 142                error("%s", o->obuf.buf);
 143                strbuf_reset(&o->obuf);
 144        }
 145
 146        return -1;
 147}
 148
 149static struct tree *shift_tree_object(struct tree *one, struct tree *two,
 150                                      const char *subtree_shift)
 151{
 152        struct object_id shifted;
 153
 154        if (!*subtree_shift) {
 155                shift_tree(&one->object.oid, &two->object.oid, &shifted, 0);
 156        } else {
 157                shift_tree_by(&one->object.oid, &two->object.oid, &shifted,
 158                              subtree_shift);
 159        }
 160        if (oideq(&two->object.oid, &shifted))
 161                return two;
 162        return lookup_tree(the_repository, &shifted);
 163}
 164
 165static struct commit *make_virtual_commit(struct tree *tree, const char *comment)
 166{
 167        struct commit *commit = alloc_commit_node(the_repository);
 168
 169        set_merge_remote_desc(commit, comment, (struct object *)commit);
 170        commit->maybe_tree = tree;
 171        commit->object.parsed = 1;
 172        return commit;
 173}
 174
 175/*
 176 * Since we use get_tree_entry(), which does not put the read object into
 177 * the object pool, we cannot rely on a == b.
 178 */
 179static int oid_eq(const struct object_id *a, const struct object_id *b)
 180{
 181        if (!a && !b)
 182                return 2;
 183        return a && b && oideq(a, b);
 184}
 185
 186enum rename_type {
 187        RENAME_NORMAL = 0,
 188        RENAME_VIA_DIR,
 189        RENAME_DELETE,
 190        RENAME_ONE_FILE_TO_ONE,
 191        RENAME_ONE_FILE_TO_TWO,
 192        RENAME_TWO_FILES_TO_ONE
 193};
 194
 195struct rename_conflict_info {
 196        enum rename_type rename_type;
 197        struct diff_filepair *pair1;
 198        struct diff_filepair *pair2;
 199        const char *branch1;
 200        const char *branch2;
 201        struct stage_data *dst_entry1;
 202        struct stage_data *dst_entry2;
 203        struct diff_filespec ren1_other;
 204        struct diff_filespec ren2_other;
 205};
 206
 207/*
 208 * Since we want to write the index eventually, we cannot reuse the index
 209 * for these (temporary) data.
 210 */
 211struct stage_data {
 212        struct {
 213                unsigned mode;
 214                struct object_id oid;
 215        } stages[4];
 216        struct rename_conflict_info *rename_conflict_info;
 217        unsigned processed:1;
 218};
 219
 220static inline void setup_rename_conflict_info(enum rename_type rename_type,
 221                                              struct diff_filepair *pair1,
 222                                              struct diff_filepair *pair2,
 223                                              const char *branch1,
 224                                              const char *branch2,
 225                                              struct stage_data *dst_entry1,
 226                                              struct stage_data *dst_entry2,
 227                                              struct merge_options *o,
 228                                              struct stage_data *src_entry1,
 229                                              struct stage_data *src_entry2)
 230{
 231        struct rename_conflict_info *ci;
 232
 233        /*
 234         * When we have two renames involved, it's easiest to get the
 235         * correct things into stage 2 and 3, and to make sure that the
 236         * content merge puts HEAD before the other branch if we just
 237         * ensure that branch1 == o->branch1.  So, simply flip arguments
 238         * around if we don't have that.
 239         */
 240        if (dst_entry2 && branch1 != o->branch1) {
 241                setup_rename_conflict_info(rename_type,
 242                                           pair2,      pair1,
 243                                           branch2,    branch1,
 244                                           dst_entry2, dst_entry1,
 245                                           o,
 246                                           src_entry2, src_entry1);
 247                return;
 248        }
 249
 250        ci = xcalloc(1, sizeof(struct rename_conflict_info));
 251        ci->rename_type = rename_type;
 252        ci->pair1 = pair1;
 253        ci->branch1 = branch1;
 254        ci->branch2 = branch2;
 255
 256        ci->dst_entry1 = dst_entry1;
 257        dst_entry1->rename_conflict_info = ci;
 258        dst_entry1->processed = 0;
 259
 260        assert(!pair2 == !dst_entry2);
 261        if (dst_entry2) {
 262                ci->dst_entry2 = dst_entry2;
 263                ci->pair2 = pair2;
 264                dst_entry2->rename_conflict_info = ci;
 265        }
 266
 267        if (rename_type == RENAME_TWO_FILES_TO_ONE) {
 268                /*
 269                 * For each rename, there could have been
 270                 * modifications on the side of history where that
 271                 * file was not renamed.
 272                 */
 273                int ostage1 = o->branch1 == branch1 ? 3 : 2;
 274                int ostage2 = ostage1 ^ 1;
 275
 276                ci->ren1_other.path = pair1->one->path;
 277                oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid);
 278                ci->ren1_other.mode = src_entry1->stages[ostage1].mode;
 279
 280                ci->ren2_other.path = pair2->one->path;
 281                oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid);
 282                ci->ren2_other.mode = src_entry2->stages[ostage2].mode;
 283        }
 284}
 285
 286static int show(struct merge_options *o, int v)
 287{
 288        return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5;
 289}
 290
 291__attribute__((format (printf, 3, 4)))
 292static void output(struct merge_options *o, int v, const char *fmt, ...)
 293{
 294        va_list ap;
 295
 296        if (!show(o, v))
 297                return;
 298
 299        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 300
 301        va_start(ap, fmt);
 302        strbuf_vaddf(&o->obuf, fmt, ap);
 303        va_end(ap);
 304
 305        strbuf_addch(&o->obuf, '\n');
 306        if (!o->buffer_output)
 307                flush_output(o);
 308}
 309
 310static void output_commit_title(struct merge_options *o, struct commit *commit)
 311{
 312        struct merge_remote_desc *desc;
 313
 314        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 315        desc = merge_remote_util(commit);
 316        if (desc)
 317                strbuf_addf(&o->obuf, "virtual %s\n", desc->name);
 318        else {
 319                strbuf_add_unique_abbrev(&o->obuf, &commit->object.oid,
 320                                         DEFAULT_ABBREV);
 321                strbuf_addch(&o->obuf, ' ');
 322                if (parse_commit(commit) != 0)
 323                        strbuf_addstr(&o->obuf, _("(bad commit)\n"));
 324                else {
 325                        const char *title;
 326                        const char *msg = get_commit_buffer(commit, NULL);
 327                        int len = find_commit_subject(msg, &title);
 328                        if (len)
 329                                strbuf_addf(&o->obuf, "%.*s\n", len, title);
 330                        unuse_commit_buffer(commit, msg);
 331                }
 332        }
 333        flush_output(o);
 334}
 335
 336static int add_cacheinfo(struct merge_options *o,
 337                         unsigned int mode, const struct object_id *oid,
 338                         const char *path, int stage, int refresh, int options)
 339{
 340        struct cache_entry *ce;
 341        int ret;
 342
 343        ce = make_cache_entry(&the_index, mode, oid ? oid : &null_oid, path, stage, 0);
 344        if (!ce)
 345                return err(o, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
 346
 347        ret = add_cache_entry(ce, options);
 348        if (refresh) {
 349                struct cache_entry *nce;
 350
 351                nce = refresh_cache_entry(&the_index, ce, CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
 352                if (!nce)
 353                        return err(o, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
 354                if (nce != ce)
 355                        ret = add_cache_entry(nce, options);
 356        }
 357        return ret;
 358}
 359
 360static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
 361{
 362        parse_tree(tree);
 363        init_tree_desc(desc, tree->buffer, tree->size);
 364}
 365
 366static int unpack_trees_start(struct merge_options *o,
 367                              struct tree *common,
 368                              struct tree *head,
 369                              struct tree *merge)
 370{
 371        int rc;
 372        struct tree_desc t[3];
 373        struct index_state tmp_index = { NULL };
 374
 375        memset(&o->unpack_opts, 0, sizeof(o->unpack_opts));
 376        if (o->call_depth)
 377                o->unpack_opts.index_only = 1;
 378        else
 379                o->unpack_opts.update = 1;
 380        o->unpack_opts.merge = 1;
 381        o->unpack_opts.head_idx = 2;
 382        o->unpack_opts.fn = threeway_merge;
 383        o->unpack_opts.src_index = &the_index;
 384        o->unpack_opts.dst_index = &tmp_index;
 385        o->unpack_opts.aggressive = !merge_detect_rename(o);
 386        setup_unpack_trees_porcelain(&o->unpack_opts, "merge");
 387
 388        init_tree_desc_from_tree(t+0, common);
 389        init_tree_desc_from_tree(t+1, head);
 390        init_tree_desc_from_tree(t+2, merge);
 391
 392        rc = unpack_trees(3, t, &o->unpack_opts);
 393        cache_tree_free(&active_cache_tree);
 394
 395        /*
 396         * Update the_index to match the new results, AFTER saving a copy
 397         * in o->orig_index.  Update src_index to point to the saved copy.
 398         * (verify_uptodate() checks src_index, and the original index is
 399         * the one that had the necessary modification timestamps.)
 400         */
 401        o->orig_index = the_index;
 402        the_index = tmp_index;
 403        o->unpack_opts.src_index = &o->orig_index;
 404
 405        return rc;
 406}
 407
 408static void unpack_trees_finish(struct merge_options *o)
 409{
 410        discard_index(&o->orig_index);
 411        clear_unpack_trees_porcelain(&o->unpack_opts);
 412}
 413
 414struct tree *write_tree_from_memory(struct merge_options *o)
 415{
 416        struct tree *result = NULL;
 417
 418        if (unmerged_cache()) {
 419                int i;
 420                fprintf(stderr, "BUG: There are unmerged index entries:\n");
 421                for (i = 0; i < active_nr; i++) {
 422                        const struct cache_entry *ce = active_cache[i];
 423                        if (ce_stage(ce))
 424                                fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
 425                                        (int)ce_namelen(ce), ce->name);
 426                }
 427                BUG("unmerged index entries in merge-recursive.c");
 428        }
 429
 430        if (!active_cache_tree)
 431                active_cache_tree = cache_tree();
 432
 433        if (!cache_tree_fully_valid(active_cache_tree) &&
 434            cache_tree_update(&the_index, 0) < 0) {
 435                err(o, _("error building trees"));
 436                return NULL;
 437        }
 438
 439        result = lookup_tree(the_repository, &active_cache_tree->oid);
 440
 441        return result;
 442}
 443
 444static int save_files_dirs(const struct object_id *oid,
 445                           struct strbuf *base, const char *path,
 446                           unsigned int mode, int stage, void *context)
 447{
 448        struct path_hashmap_entry *entry;
 449        int baselen = base->len;
 450        struct merge_options *o = context;
 451
 452        strbuf_addstr(base, path);
 453
 454        FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
 455        hashmap_entry_init(entry, path_hash(entry->path));
 456        hashmap_add(&o->current_file_dir_set, entry);
 457
 458        strbuf_setlen(base, baselen);
 459        return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
 460}
 461
 462static void get_files_dirs(struct merge_options *o, struct tree *tree)
 463{
 464        struct pathspec match_all;
 465        memset(&match_all, 0, sizeof(match_all));
 466        read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
 467}
 468
 469static int get_tree_entry_if_blob(const struct object_id *tree,
 470                                  const char *path,
 471                                  struct object_id *hashy,
 472                                  unsigned int *mode_o)
 473{
 474        int ret;
 475
 476        ret = get_tree_entry(tree, path, hashy, mode_o);
 477        if (S_ISDIR(*mode_o)) {
 478                oidcpy(hashy, &null_oid);
 479                *mode_o = 0;
 480        }
 481        return ret;
 482}
 483
 484/*
 485 * Returns an index_entry instance which doesn't have to correspond to
 486 * a real cache entry in Git's index.
 487 */
 488static struct stage_data *insert_stage_data(const char *path,
 489                struct tree *o, struct tree *a, struct tree *b,
 490                struct string_list *entries)
 491{
 492        struct string_list_item *item;
 493        struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
 494        get_tree_entry_if_blob(&o->object.oid, path,
 495                               &e->stages[1].oid, &e->stages[1].mode);
 496        get_tree_entry_if_blob(&a->object.oid, path,
 497                               &e->stages[2].oid, &e->stages[2].mode);
 498        get_tree_entry_if_blob(&b->object.oid, path,
 499                               &e->stages[3].oid, &e->stages[3].mode);
 500        item = string_list_insert(entries, path);
 501        item->util = e;
 502        return e;
 503}
 504
 505/*
 506 * Create a dictionary mapping file names to stage_data objects. The
 507 * dictionary contains one entry for every path with a non-zero stage entry.
 508 */
 509static struct string_list *get_unmerged(void)
 510{
 511        struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
 512        int i;
 513
 514        unmerged->strdup_strings = 1;
 515
 516        for (i = 0; i < active_nr; i++) {
 517                struct string_list_item *item;
 518                struct stage_data *e;
 519                const struct cache_entry *ce = active_cache[i];
 520                if (!ce_stage(ce))
 521                        continue;
 522
 523                item = string_list_lookup(unmerged, ce->name);
 524                if (!item) {
 525                        item = string_list_insert(unmerged, ce->name);
 526                        item->util = xcalloc(1, sizeof(struct stage_data));
 527                }
 528                e = item->util;
 529                e->stages[ce_stage(ce)].mode = ce->ce_mode;
 530                oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
 531        }
 532
 533        return unmerged;
 534}
 535
 536static int string_list_df_name_compare(const char *one, const char *two)
 537{
 538        int onelen = strlen(one);
 539        int twolen = strlen(two);
 540        /*
 541         * Here we only care that entries for D/F conflicts are
 542         * adjacent, in particular with the file of the D/F conflict
 543         * appearing before files below the corresponding directory.
 544         * The order of the rest of the list is irrelevant for us.
 545         *
 546         * To achieve this, we sort with df_name_compare and provide
 547         * the mode S_IFDIR so that D/F conflicts will sort correctly.
 548         * We use the mode S_IFDIR for everything else for simplicity,
 549         * since in other cases any changes in their order due to
 550         * sorting cause no problems for us.
 551         */
 552        int cmp = df_name_compare(one, onelen, S_IFDIR,
 553                                  two, twolen, S_IFDIR);
 554        /*
 555         * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
 556         * that 'foo' comes before 'foo/bar'.
 557         */
 558        if (cmp)
 559                return cmp;
 560        return onelen - twolen;
 561}
 562
 563static void record_df_conflict_files(struct merge_options *o,
 564                                     struct string_list *entries)
 565{
 566        /* If there is a D/F conflict and the file for such a conflict
 567         * currently exists in the working tree, we want to allow it to be
 568         * removed to make room for the corresponding directory if needed.
 569         * The files underneath the directories of such D/F conflicts will
 570         * be processed before the corresponding file involved in the D/F
 571         * conflict.  If the D/F directory ends up being removed by the
 572         * merge, then we won't have to touch the D/F file.  If the D/F
 573         * directory needs to be written to the working copy, then the D/F
 574         * file will simply be removed (in make_room_for_path()) to make
 575         * room for the necessary paths.  Note that if both the directory
 576         * and the file need to be present, then the D/F file will be
 577         * reinstated with a new unique name at the time it is processed.
 578         */
 579        struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
 580        const char *last_file = NULL;
 581        int last_len = 0;
 582        int i;
 583
 584        /*
 585         * If we're merging merge-bases, we don't want to bother with
 586         * any working directory changes.
 587         */
 588        if (o->call_depth)
 589                return;
 590
 591        /* Ensure D/F conflicts are adjacent in the entries list. */
 592        for (i = 0; i < entries->nr; i++) {
 593                struct string_list_item *next = &entries->items[i];
 594                string_list_append(&df_sorted_entries, next->string)->util =
 595                                   next->util;
 596        }
 597        df_sorted_entries.cmp = string_list_df_name_compare;
 598        string_list_sort(&df_sorted_entries);
 599
 600        string_list_clear(&o->df_conflict_file_set, 1);
 601        for (i = 0; i < df_sorted_entries.nr; i++) {
 602                const char *path = df_sorted_entries.items[i].string;
 603                int len = strlen(path);
 604                struct stage_data *e = df_sorted_entries.items[i].util;
 605
 606                /*
 607                 * Check if last_file & path correspond to a D/F conflict;
 608                 * i.e. whether path is last_file+'/'+<something>.
 609                 * If so, record that it's okay to remove last_file to make
 610                 * room for path and friends if needed.
 611                 */
 612                if (last_file &&
 613                    len > last_len &&
 614                    memcmp(path, last_file, last_len) == 0 &&
 615                    path[last_len] == '/') {
 616                        string_list_insert(&o->df_conflict_file_set, last_file);
 617                }
 618
 619                /*
 620                 * Determine whether path could exist as a file in the
 621                 * working directory as a possible D/F conflict.  This
 622                 * will only occur when it exists in stage 2 as a
 623                 * file.
 624                 */
 625                if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
 626                        last_file = path;
 627                        last_len = len;
 628                } else {
 629                        last_file = NULL;
 630                }
 631        }
 632        string_list_clear(&df_sorted_entries, 0);
 633}
 634
 635struct rename {
 636        struct diff_filepair *pair;
 637        /*
 638         * Purpose of src_entry and dst_entry:
 639         *
 640         * If 'before' is renamed to 'after' then src_entry will contain
 641         * the versions of 'before' from the merge_base, HEAD, and MERGE in
 642         * stages 1, 2, and 3; dst_entry will contain the respective
 643         * versions of 'after' in corresponding locations.  Thus, we have a
 644         * total of six modes and oids, though some will be null.  (Stage 0
 645         * is ignored; we're interested in handling conflicts.)
 646         *
 647         * Since we don't turn on break-rewrites by default, neither
 648         * src_entry nor dst_entry can have all three of their stages have
 649         * non-null oids, meaning at most four of the six will be non-null.
 650         * Also, since this is a rename, both src_entry and dst_entry will
 651         * have at least one non-null oid, meaning at least two will be
 652         * non-null.  Of the six oids, a typical rename will have three be
 653         * non-null.  Only two implies a rename/delete, and four implies a
 654         * rename/add.
 655         */
 656        struct stage_data *src_entry;
 657        struct stage_data *dst_entry;
 658        unsigned add_turned_into_rename:1;
 659        unsigned processed:1;
 660};
 661
 662static int update_stages(struct merge_options *opt, const char *path,
 663                         const struct diff_filespec *o,
 664                         const struct diff_filespec *a,
 665                         const struct diff_filespec *b)
 666{
 667
 668        /*
 669         * NOTE: It is usually a bad idea to call update_stages on a path
 670         * before calling update_file on that same path, since it can
 671         * sometimes lead to spurious "refusing to lose untracked file..."
 672         * messages from update_file (via make_room_for path via
 673         * would_lose_untracked).  Instead, reverse the order of the calls
 674         * (executing update_file first and then update_stages).
 675         */
 676        int clear = 1;
 677        int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
 678        if (clear)
 679                if (remove_file_from_cache(path))
 680                        return -1;
 681        if (o)
 682                if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
 683                        return -1;
 684        if (a)
 685                if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
 686                        return -1;
 687        if (b)
 688                if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
 689                        return -1;
 690        return 0;
 691}
 692
 693static int update_stages_for_stage_data(struct merge_options *opt,
 694                                        const char *path,
 695                                        const struct stage_data *stage_data)
 696{
 697        struct diff_filespec o, a, b;
 698
 699        o.mode = stage_data->stages[1].mode;
 700        oidcpy(&o.oid, &stage_data->stages[1].oid);
 701
 702        a.mode = stage_data->stages[2].mode;
 703        oidcpy(&a.oid, &stage_data->stages[2].oid);
 704
 705        b.mode = stage_data->stages[3].mode;
 706        oidcpy(&b.oid, &stage_data->stages[3].oid);
 707
 708        return update_stages(opt, path,
 709                             is_null_oid(&o.oid) ? NULL : &o,
 710                             is_null_oid(&a.oid) ? NULL : &a,
 711                             is_null_oid(&b.oid) ? NULL : &b);
 712}
 713
 714static void update_entry(struct stage_data *entry,
 715                         struct diff_filespec *o,
 716                         struct diff_filespec *a,
 717                         struct diff_filespec *b)
 718{
 719        entry->processed = 0;
 720        entry->stages[1].mode = o->mode;
 721        entry->stages[2].mode = a->mode;
 722        entry->stages[3].mode = b->mode;
 723        oidcpy(&entry->stages[1].oid, &o->oid);
 724        oidcpy(&entry->stages[2].oid, &a->oid);
 725        oidcpy(&entry->stages[3].oid, &b->oid);
 726}
 727
 728static int remove_file(struct merge_options *o, int clean,
 729                       const char *path, int no_wd)
 730{
 731        int update_cache = o->call_depth || clean;
 732        int update_working_directory = !o->call_depth && !no_wd;
 733
 734        if (update_cache) {
 735                if (remove_file_from_cache(path))
 736                        return -1;
 737        }
 738        if (update_working_directory) {
 739                if (ignore_case) {
 740                        struct cache_entry *ce;
 741                        ce = cache_file_exists(path, strlen(path), ignore_case);
 742                        if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
 743                                return 0;
 744                }
 745                if (remove_path(path))
 746                        return -1;
 747        }
 748        return 0;
 749}
 750
 751/* add a string to a strbuf, but converting "/" to "_" */
 752static void add_flattened_path(struct strbuf *out, const char *s)
 753{
 754        size_t i = out->len;
 755        strbuf_addstr(out, s);
 756        for (; i < out->len; i++)
 757                if (out->buf[i] == '/')
 758                        out->buf[i] = '_';
 759}
 760
 761static char *unique_path(struct merge_options *o, const char *path, const char *branch)
 762{
 763        struct path_hashmap_entry *entry;
 764        struct strbuf newpath = STRBUF_INIT;
 765        int suffix = 0;
 766        size_t base_len;
 767
 768        strbuf_addf(&newpath, "%s~", path);
 769        add_flattened_path(&newpath, branch);
 770
 771        base_len = newpath.len;
 772        while (hashmap_get_from_hash(&o->current_file_dir_set,
 773                                     path_hash(newpath.buf), newpath.buf) ||
 774               (!o->call_depth && file_exists(newpath.buf))) {
 775                strbuf_setlen(&newpath, base_len);
 776                strbuf_addf(&newpath, "_%d", suffix++);
 777        }
 778
 779        FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
 780        hashmap_entry_init(entry, path_hash(entry->path));
 781        hashmap_add(&o->current_file_dir_set, entry);
 782        return strbuf_detach(&newpath, NULL);
 783}
 784
 785/**
 786 * Check whether a directory in the index is in the way of an incoming
 787 * file.  Return 1 if so.  If check_working_copy is non-zero, also
 788 * check the working directory.  If empty_ok is non-zero, also return
 789 * 0 in the case where the working-tree dir exists but is empty.
 790 */
 791static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
 792{
 793        int pos;
 794        struct strbuf dirpath = STRBUF_INIT;
 795        struct stat st;
 796
 797        strbuf_addstr(&dirpath, path);
 798        strbuf_addch(&dirpath, '/');
 799
 800        pos = cache_name_pos(dirpath.buf, dirpath.len);
 801
 802        if (pos < 0)
 803                pos = -1 - pos;
 804        if (pos < active_nr &&
 805            !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
 806                strbuf_release(&dirpath);
 807                return 1;
 808        }
 809
 810        strbuf_release(&dirpath);
 811        return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
 812                !(empty_ok && is_empty_dir(path));
 813}
 814
 815/*
 816 * Returns whether path was tracked in the index before the merge started,
 817 * and its oid and mode match the specified values
 818 */
 819static int was_tracked_and_matches(struct merge_options *o, const char *path,
 820                                   const struct object_id *oid, unsigned mode)
 821{
 822        int pos = index_name_pos(&o->orig_index, path, strlen(path));
 823        struct cache_entry *ce;
 824
 825        if (0 > pos)
 826                /* we were not tracking this path before the merge */
 827                return 0;
 828
 829        /* See if the file we were tracking before matches */
 830        ce = o->orig_index.cache[pos];
 831        return (oid_eq(&ce->oid, oid) && ce->ce_mode == mode);
 832}
 833
 834/*
 835 * Returns whether path was tracked in the index before the merge started
 836 */
 837static int was_tracked(struct merge_options *o, const char *path)
 838{
 839        int pos = index_name_pos(&o->orig_index, path, strlen(path));
 840
 841        if (0 <= pos)
 842                /* we were tracking this path before the merge */
 843                return 1;
 844
 845        return 0;
 846}
 847
 848static int would_lose_untracked(const char *path)
 849{
 850        /*
 851         * This may look like it can be simplified to:
 852         *   return !was_tracked(o, path) && file_exists(path)
 853         * but it can't.  This function needs to know whether path was in
 854         * the working tree due to EITHER having been tracked in the index
 855         * before the merge OR having been put into the working copy and
 856         * index by unpack_trees().  Due to that either-or requirement, we
 857         * check the current index instead of the original one.
 858         *
 859         * Note that we do not need to worry about merge-recursive itself
 860         * updating the index after unpack_trees() and before calling this
 861         * function, because we strictly require all code paths in
 862         * merge-recursive to update the working tree first and the index
 863         * second.  Doing otherwise would break
 864         * update_file()/would_lose_untracked(); see every comment in this
 865         * file which mentions "update_stages".
 866         */
 867        int pos = cache_name_pos(path, strlen(path));
 868
 869        if (pos < 0)
 870                pos = -1 - pos;
 871        while (pos < active_nr &&
 872               !strcmp(path, active_cache[pos]->name)) {
 873                /*
 874                 * If stage #0, it is definitely tracked.
 875                 * If it has stage #2 then it was tracked
 876                 * before this merge started.  All other
 877                 * cases the path was not tracked.
 878                 */
 879                switch (ce_stage(active_cache[pos])) {
 880                case 0:
 881                case 2:
 882                        return 0;
 883                }
 884                pos++;
 885        }
 886        return file_exists(path);
 887}
 888
 889static int was_dirty(struct merge_options *o, const char *path)
 890{
 891        struct cache_entry *ce;
 892        int dirty = 1;
 893
 894        if (o->call_depth || !was_tracked(o, path))
 895                return !dirty;
 896
 897        ce = index_file_exists(o->unpack_opts.src_index,
 898                               path, strlen(path), ignore_case);
 899        dirty = verify_uptodate(ce, &o->unpack_opts) != 0;
 900        return dirty;
 901}
 902
 903static int make_room_for_path(struct merge_options *o, const char *path)
 904{
 905        int status, i;
 906        const char *msg = _("failed to create path '%s'%s");
 907
 908        /* Unlink any D/F conflict files that are in the way */
 909        for (i = 0; i < o->df_conflict_file_set.nr; i++) {
 910                const char *df_path = o->df_conflict_file_set.items[i].string;
 911                size_t pathlen = strlen(path);
 912                size_t df_pathlen = strlen(df_path);
 913                if (df_pathlen < pathlen &&
 914                    path[df_pathlen] == '/' &&
 915                    strncmp(path, df_path, df_pathlen) == 0) {
 916                        output(o, 3,
 917                               _("Removing %s to make room for subdirectory\n"),
 918                               df_path);
 919                        unlink(df_path);
 920                        unsorted_string_list_delete_item(&o->df_conflict_file_set,
 921                                                         i, 0);
 922                        break;
 923                }
 924        }
 925
 926        /* Make sure leading directories are created */
 927        status = safe_create_leading_directories_const(path);
 928        if (status) {
 929                if (status == SCLD_EXISTS)
 930                        /* something else exists */
 931                        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 932                return err(o, msg, path, "");
 933        }
 934
 935        /*
 936         * Do not unlink a file in the work tree if we are not
 937         * tracking it.
 938         */
 939        if (would_lose_untracked(path))
 940                return err(o, _("refusing to lose untracked file at '%s'"),
 941                           path);
 942
 943        /* Successful unlink is good.. */
 944        if (!unlink(path))
 945                return 0;
 946        /* .. and so is no existing file */
 947        if (errno == ENOENT)
 948                return 0;
 949        /* .. but not some other error (who really cares what?) */
 950        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 951}
 952
 953static int update_file_flags(struct merge_options *o,
 954                             const struct object_id *oid,
 955                             unsigned mode,
 956                             const char *path,
 957                             int update_cache,
 958                             int update_wd)
 959{
 960        int ret = 0;
 961
 962        if (o->call_depth)
 963                update_wd = 0;
 964
 965        if (update_wd) {
 966                enum object_type type;
 967                void *buf;
 968                unsigned long size;
 969
 970                if (S_ISGITLINK(mode)) {
 971                        /*
 972                         * We may later decide to recursively descend into
 973                         * the submodule directory and update its index
 974                         * and/or work tree, but we do not do that now.
 975                         */
 976                        update_wd = 0;
 977                        goto update_index;
 978                }
 979
 980                buf = read_object_file(oid, &type, &size);
 981                if (!buf)
 982                        return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
 983                if (type != OBJ_BLOB) {
 984                        ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
 985                        goto free_buf;
 986                }
 987                if (S_ISREG(mode)) {
 988                        struct strbuf strbuf = STRBUF_INIT;
 989                        if (convert_to_working_tree(&the_index, path, buf, size, &strbuf)) {
 990                                free(buf);
 991                                size = strbuf.len;
 992                                buf = strbuf_detach(&strbuf, NULL);
 993                        }
 994                }
 995
 996                if (make_room_for_path(o, path) < 0) {
 997                        update_wd = 0;
 998                        goto free_buf;
 999                }
1000                if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
1001                        int fd;
1002                        if (mode & 0100)
1003                                mode = 0777;
1004                        else
1005                                mode = 0666;
1006                        fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
1007                        if (fd < 0) {
1008                                ret = err(o, _("failed to open '%s': %s"),
1009                                          path, strerror(errno));
1010                                goto free_buf;
1011                        }
1012                        write_in_full(fd, buf, size);
1013                        close(fd);
1014                } else if (S_ISLNK(mode)) {
1015                        char *lnk = xmemdupz(buf, size);
1016                        safe_create_leading_directories_const(path);
1017                        unlink(path);
1018                        if (symlink(lnk, path))
1019                                ret = err(o, _("failed to symlink '%s': %s"),
1020                                          path, strerror(errno));
1021                        free(lnk);
1022                } else
1023                        ret = err(o,
1024                                  _("do not know what to do with %06o %s '%s'"),
1025                                  mode, oid_to_hex(oid), path);
1026        free_buf:
1027                free(buf);
1028        }
1029update_index:
1030        if (!ret && update_cache)
1031                if (add_cacheinfo(o, mode, oid, path, 0, update_wd,
1032                                  ADD_CACHE_OK_TO_ADD))
1033                        return -1;
1034        return ret;
1035}
1036
1037static int update_file(struct merge_options *o,
1038                       int clean,
1039                       const struct object_id *oid,
1040                       unsigned mode,
1041                       const char *path)
1042{
1043        return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
1044}
1045
1046/* Low level file merging, update and removal */
1047
1048struct merge_file_info {
1049        struct object_id oid;
1050        unsigned mode;
1051        unsigned clean:1,
1052                 merge:1;
1053};
1054
1055static int merge_3way(struct merge_options *o,
1056                      mmbuffer_t *result_buf,
1057                      const struct diff_filespec *one,
1058                      const struct diff_filespec *a,
1059                      const struct diff_filespec *b,
1060                      const char *branch1,
1061                      const char *branch2)
1062{
1063        mmfile_t orig, src1, src2;
1064        struct ll_merge_options ll_opts = {0};
1065        char *base_name, *name1, *name2;
1066        int merge_status;
1067
1068        ll_opts.renormalize = o->renormalize;
1069        ll_opts.xdl_opts = o->xdl_opts;
1070
1071        if (o->call_depth) {
1072                ll_opts.virtual_ancestor = 1;
1073                ll_opts.variant = 0;
1074        } else {
1075                switch (o->recursive_variant) {
1076                case MERGE_RECURSIVE_OURS:
1077                        ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1078                        break;
1079                case MERGE_RECURSIVE_THEIRS:
1080                        ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1081                        break;
1082                default:
1083                        ll_opts.variant = 0;
1084                        break;
1085                }
1086        }
1087
1088        if (strcmp(a->path, b->path) ||
1089            (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1090                base_name = o->ancestor == NULL ? NULL :
1091                        mkpathdup("%s:%s", o->ancestor, one->path);
1092                name1 = mkpathdup("%s:%s", branch1, a->path);
1093                name2 = mkpathdup("%s:%s", branch2, b->path);
1094        } else {
1095                base_name = o->ancestor == NULL ? NULL :
1096                        mkpathdup("%s", o->ancestor);
1097                name1 = mkpathdup("%s", branch1);
1098                name2 = mkpathdup("%s", branch2);
1099        }
1100
1101        read_mmblob(&orig, &one->oid);
1102        read_mmblob(&src1, &a->oid);
1103        read_mmblob(&src2, &b->oid);
1104
1105        merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1106                                &src1, name1, &src2, name2, &ll_opts);
1107
1108        free(base_name);
1109        free(name1);
1110        free(name2);
1111        free(orig.ptr);
1112        free(src1.ptr);
1113        free(src2.ptr);
1114        return merge_status;
1115}
1116
1117static int find_first_merges(struct object_array *result, const char *path,
1118                             struct commit *a, struct commit *b)
1119{
1120        int i, j;
1121        struct object_array merges = OBJECT_ARRAY_INIT;
1122        struct commit *commit;
1123        int contains_another;
1124
1125        char merged_revision[42];
1126        const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1127                                   "--all", merged_revision, NULL };
1128        struct rev_info revs;
1129        struct setup_revision_opt rev_opts;
1130
1131        memset(result, 0, sizeof(struct object_array));
1132        memset(&rev_opts, 0, sizeof(rev_opts));
1133
1134        /* get all revisions that merge commit a */
1135        xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1136                  oid_to_hex(&a->object.oid));
1137        init_revisions(&revs, NULL);
1138        rev_opts.submodule = path;
1139        /* FIXME: can't handle linked worktrees in submodules yet */
1140        revs.single_worktree = path != NULL;
1141        setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1142
1143        /* save all revisions from the above list that contain b */
1144        if (prepare_revision_walk(&revs))
1145                die("revision walk setup failed");
1146        while ((commit = get_revision(&revs)) != NULL) {
1147                struct object *o = &(commit->object);
1148                if (in_merge_bases(b, commit))
1149                        add_object_array(o, NULL, &merges);
1150        }
1151        reset_revision_walk();
1152
1153        /* Now we've got all merges that contain a and b. Prune all
1154         * merges that contain another found merge and save them in
1155         * result.
1156         */
1157        for (i = 0; i < merges.nr; i++) {
1158                struct commit *m1 = (struct commit *) merges.objects[i].item;
1159
1160                contains_another = 0;
1161                for (j = 0; j < merges.nr; j++) {
1162                        struct commit *m2 = (struct commit *) merges.objects[j].item;
1163                        if (i != j && in_merge_bases(m2, m1)) {
1164                                contains_another = 1;
1165                                break;
1166                        }
1167                }
1168
1169                if (!contains_another)
1170                        add_object_array(merges.objects[i].item, NULL, result);
1171        }
1172
1173        object_array_clear(&merges);
1174        return result->nr;
1175}
1176
1177static void print_commit(struct commit *commit)
1178{
1179        struct strbuf sb = STRBUF_INIT;
1180        struct pretty_print_context ctx = {0};
1181        ctx.date_mode.type = DATE_NORMAL;
1182        format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1183        fprintf(stderr, "%s\n", sb.buf);
1184        strbuf_release(&sb);
1185}
1186
1187static int merge_submodule(struct merge_options *o,
1188                           struct object_id *result, const char *path,
1189                           const struct object_id *base, const struct object_id *a,
1190                           const struct object_id *b)
1191{
1192        struct commit *commit_base, *commit_a, *commit_b;
1193        int parent_count;
1194        struct object_array merges;
1195
1196        int i;
1197        int search = !o->call_depth;
1198
1199        /* store a in result in case we fail */
1200        oidcpy(result, a);
1201
1202        /* we can not handle deletion conflicts */
1203        if (is_null_oid(base))
1204                return 0;
1205        if (is_null_oid(a))
1206                return 0;
1207        if (is_null_oid(b))
1208                return 0;
1209
1210        if (add_submodule_odb(path)) {
1211                output(o, 1, _("Failed to merge submodule %s (not checked out)"), path);
1212                return 0;
1213        }
1214
1215        if (!(commit_base = lookup_commit_reference(the_repository, base)) ||
1216            !(commit_a = lookup_commit_reference(the_repository, a)) ||
1217            !(commit_b = lookup_commit_reference(the_repository, b))) {
1218                output(o, 1, _("Failed to merge submodule %s (commits not present)"), path);
1219                return 0;
1220        }
1221
1222        /* check whether both changes are forward */
1223        if (!in_merge_bases(commit_base, commit_a) ||
1224            !in_merge_bases(commit_base, commit_b)) {
1225                output(o, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1226                return 0;
1227        }
1228
1229        /* Case #1: a is contained in b or vice versa */
1230        if (in_merge_bases(commit_a, commit_b)) {
1231                oidcpy(result, b);
1232                if (show(o, 3)) {
1233                        output(o, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1234                        output_commit_title(o, commit_b);
1235                } else if (show(o, 2))
1236                        output(o, 2, _("Fast-forwarding submodule %s"), path);
1237                else
1238                        ; /* no output */
1239
1240                return 1;
1241        }
1242        if (in_merge_bases(commit_b, commit_a)) {
1243                oidcpy(result, a);
1244                if (show(o, 3)) {
1245                        output(o, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1246                        output_commit_title(o, commit_a);
1247                } else if (show(o, 2))
1248                        output(o, 2, _("Fast-forwarding submodule %s"), path);
1249                else
1250                        ; /* no output */
1251
1252                return 1;
1253        }
1254
1255        /*
1256         * Case #2: There are one or more merges that contain a and b in
1257         * the submodule. If there is only one, then present it as a
1258         * suggestion to the user, but leave it marked unmerged so the
1259         * user needs to confirm the resolution.
1260         */
1261
1262        /* Skip the search if makes no sense to the calling context.  */
1263        if (!search)
1264                return 0;
1265
1266        /* find commit which merges them */
1267        parent_count = find_first_merges(&merges, path, commit_a, commit_b);
1268        switch (parent_count) {
1269        case 0:
1270                output(o, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1271                break;
1272
1273        case 1:
1274                output(o, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1275                output(o, 2, _("Found a possible merge resolution for the submodule:\n"));
1276                print_commit((struct commit *) merges.objects[0].item);
1277                output(o, 2, _(
1278                       "If this is correct simply add it to the index "
1279                       "for example\n"
1280                       "by using:\n\n"
1281                       "  git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1282                       "which will accept this suggestion.\n"),
1283                       oid_to_hex(&merges.objects[0].item->oid), path);
1284                break;
1285
1286        default:
1287                output(o, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1288                for (i = 0; i < merges.nr; i++)
1289                        print_commit((struct commit *) merges.objects[i].item);
1290        }
1291
1292        object_array_clear(&merges);
1293        return 0;
1294}
1295
1296static int merge_mode_and_contents(struct merge_options *o,
1297                                   const struct diff_filespec *one,
1298                                   const struct diff_filespec *a,
1299                                   const struct diff_filespec *b,
1300                                   const char *filename,
1301                                   const char *branch1,
1302                                   const char *branch2,
1303                                   struct merge_file_info *result)
1304{
1305        if (o->branch1 != branch1) {
1306                /*
1307                 * It's weird getting a reverse merge with HEAD on the bottom
1308                 * side of the conflict markers and the other branch on the
1309                 * top.  Fix that.
1310                 */
1311                return merge_mode_and_contents(o, one, b, a,
1312                                               filename,
1313                                               branch2, branch1, result);
1314        }
1315
1316        result->merge = 0;
1317        result->clean = 1;
1318
1319        if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1320                result->clean = 0;
1321                if (S_ISREG(a->mode)) {
1322                        result->mode = a->mode;
1323                        oidcpy(&result->oid, &a->oid);
1324                } else {
1325                        result->mode = b->mode;
1326                        oidcpy(&result->oid, &b->oid);
1327                }
1328        } else {
1329                if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1330                        result->merge = 1;
1331
1332                /*
1333                 * Merge modes
1334                 */
1335                if (a->mode == b->mode || a->mode == one->mode)
1336                        result->mode = b->mode;
1337                else {
1338                        result->mode = a->mode;
1339                        if (b->mode != one->mode) {
1340                                result->clean = 0;
1341                                result->merge = 1;
1342                        }
1343                }
1344
1345                if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1346                        oidcpy(&result->oid, &b->oid);
1347                else if (oid_eq(&b->oid, &one->oid))
1348                        oidcpy(&result->oid, &a->oid);
1349                else if (S_ISREG(a->mode)) {
1350                        mmbuffer_t result_buf;
1351                        int ret = 0, merge_status;
1352
1353                        merge_status = merge_3way(o, &result_buf, one, a, b,
1354                                                  branch1, branch2);
1355
1356                        if ((merge_status < 0) || !result_buf.ptr)
1357                                ret = err(o, _("Failed to execute internal merge"));
1358
1359                        if (!ret &&
1360                            write_object_file(result_buf.ptr, result_buf.size,
1361                                              blob_type, &result->oid))
1362                                ret = err(o, _("Unable to add %s to database"),
1363                                          a->path);
1364
1365                        free(result_buf.ptr);
1366                        if (ret)
1367                                return ret;
1368                        result->clean = (merge_status == 0);
1369                } else if (S_ISGITLINK(a->mode)) {
1370                        result->clean = merge_submodule(o, &result->oid,
1371                                                        one->path,
1372                                                        &one->oid,
1373                                                        &a->oid,
1374                                                        &b->oid);
1375                } else if (S_ISLNK(a->mode)) {
1376                        switch (o->recursive_variant) {
1377                        case MERGE_RECURSIVE_NORMAL:
1378                                oidcpy(&result->oid, &a->oid);
1379                                if (!oid_eq(&a->oid, &b->oid))
1380                                        result->clean = 0;
1381                                break;
1382                        case MERGE_RECURSIVE_OURS:
1383                                oidcpy(&result->oid, &a->oid);
1384                                break;
1385                        case MERGE_RECURSIVE_THEIRS:
1386                                oidcpy(&result->oid, &b->oid);
1387                                break;
1388                        }
1389                } else
1390                        BUG("unsupported object type in the tree");
1391        }
1392
1393        if (result->merge)
1394                output(o, 2, _("Auto-merging %s"), filename);
1395
1396        return 0;
1397}
1398
1399static int handle_rename_via_dir(struct merge_options *o,
1400                                 struct diff_filepair *pair,
1401                                 const char *rename_branch,
1402                                 const char *other_branch)
1403{
1404        /*
1405         * Handle file adds that need to be renamed due to directory rename
1406         * detection.  This differs from handle_rename_normal, because
1407         * there is no content merge to do; just move the file into the
1408         * desired final location.
1409         */
1410        const struct diff_filespec *dest = pair->two;
1411
1412        if (!o->call_depth && would_lose_untracked(dest->path)) {
1413                char *alt_path = unique_path(o, dest->path, rename_branch);
1414
1415                output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1416                               "writing to %s instead."),
1417                       dest->path, alt_path);
1418                /*
1419                 * Write the file in worktree at alt_path, but not in the
1420                 * index.  Instead, write to dest->path for the index but
1421                 * only at the higher appropriate stage.
1422                 */
1423                if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1424                        return -1;
1425                free(alt_path);
1426                return update_stages(o, dest->path, NULL,
1427                                     rename_branch == o->branch1 ? dest : NULL,
1428                                     rename_branch == o->branch1 ? NULL : dest);
1429        }
1430
1431        /* Update dest->path both in index and in worktree */
1432        if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1433                return -1;
1434        return 0;
1435}
1436
1437static int handle_change_delete(struct merge_options *o,
1438                                const char *path, const char *old_path,
1439                                const struct object_id *o_oid, int o_mode,
1440                                const struct object_id *changed_oid,
1441                                int changed_mode,
1442                                const char *change_branch,
1443                                const char *delete_branch,
1444                                const char *change, const char *change_past)
1445{
1446        char *alt_path = NULL;
1447        const char *update_path = path;
1448        int ret = 0;
1449
1450        if (dir_in_way(path, !o->call_depth, 0) ||
1451            (!o->call_depth && would_lose_untracked(path))) {
1452                update_path = alt_path = unique_path(o, path, change_branch);
1453        }
1454
1455        if (o->call_depth) {
1456                /*
1457                 * We cannot arbitrarily accept either a_sha or b_sha as
1458                 * correct; since there is no true "middle point" between
1459                 * them, simply reuse the base version for virtual merge base.
1460                 */
1461                ret = remove_file_from_cache(path);
1462                if (!ret)
1463                        ret = update_file(o, 0, o_oid, o_mode, update_path);
1464        } else {
1465                /*
1466                 * Despite the four nearly duplicate messages and argument
1467                 * lists below and the ugliness of the nested if-statements,
1468                 * having complete messages makes the job easier for
1469                 * translators.
1470                 *
1471                 * The slight variance among the cases is due to the fact
1472                 * that:
1473                 *   1) directory/file conflicts (in effect if
1474                 *      !alt_path) could cause us to need to write the
1475                 *      file to a different path.
1476                 *   2) renames (in effect if !old_path) could mean that
1477                 *      there are two names for the path that the user
1478                 *      may know the file by.
1479                 */
1480                if (!alt_path) {
1481                        if (!old_path) {
1482                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1483                                       "and %s in %s. Version %s of %s left in tree."),
1484                                       change, path, delete_branch, change_past,
1485                                       change_branch, change_branch, path);
1486                        } else {
1487                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1488                                       "and %s to %s in %s. Version %s of %s left in tree."),
1489                                       change, old_path, delete_branch, change_past, path,
1490                                       change_branch, change_branch, path);
1491                        }
1492                } else {
1493                        if (!old_path) {
1494                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1495                                       "and %s in %s. Version %s of %s left in tree at %s."),
1496                                       change, path, delete_branch, change_past,
1497                                       change_branch, change_branch, path, alt_path);
1498                        } else {
1499                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1500                                       "and %s to %s in %s. Version %s of %s left in tree at %s."),
1501                                       change, old_path, delete_branch, change_past, path,
1502                                       change_branch, change_branch, path, alt_path);
1503                        }
1504                }
1505                /*
1506                 * No need to call update_file() on path when change_branch ==
1507                 * o->branch1 && !alt_path, since that would needlessly touch
1508                 * path.  We could call update_file_flags() with update_cache=0
1509                 * and update_wd=0, but that's a no-op.
1510                 */
1511                if (change_branch != o->branch1 || alt_path)
1512                        ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1513        }
1514        free(alt_path);
1515
1516        return ret;
1517}
1518
1519static int handle_rename_delete(struct merge_options *o,
1520                                struct diff_filepair *pair,
1521                                const char *rename_branch,
1522                                const char *delete_branch)
1523{
1524        const struct diff_filespec *orig = pair->one;
1525        const struct diff_filespec *dest = pair->two;
1526
1527        if (handle_change_delete(o,
1528                                 o->call_depth ? orig->path : dest->path,
1529                                 o->call_depth ? NULL : orig->path,
1530                                 &orig->oid, orig->mode,
1531                                 &dest->oid, dest->mode,
1532                                 rename_branch, delete_branch,
1533                                 _("rename"), _("renamed")))
1534                return -1;
1535
1536        if (o->call_depth)
1537                return remove_file_from_cache(dest->path);
1538        else
1539                return update_stages(o, dest->path, NULL,
1540                                     rename_branch == o->branch1 ? dest : NULL,
1541                                     rename_branch == o->branch1 ? NULL : dest);
1542}
1543
1544static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1545                                                 struct stage_data *entry,
1546                                                 int stage)
1547{
1548        struct object_id *oid = &entry->stages[stage].oid;
1549        unsigned mode = entry->stages[stage].mode;
1550        if (mode == 0 || is_null_oid(oid))
1551                return NULL;
1552        oidcpy(&target->oid, oid);
1553        target->mode = mode;
1554        return target;
1555}
1556
1557static int handle_file(struct merge_options *o,
1558                        struct diff_filespec *rename,
1559                        int stage,
1560                        struct rename_conflict_info *ci)
1561{
1562        char *dst_name = rename->path;
1563        struct stage_data *dst_entry;
1564        const char *cur_branch, *other_branch;
1565        struct diff_filespec other;
1566        struct diff_filespec *add;
1567        int ret;
1568
1569        if (stage == 2) {
1570                dst_entry = ci->dst_entry1;
1571                cur_branch = ci->branch1;
1572                other_branch = ci->branch2;
1573        } else {
1574                dst_entry = ci->dst_entry2;
1575                cur_branch = ci->branch2;
1576                other_branch = ci->branch1;
1577        }
1578
1579        add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1580        if (add) {
1581                int ren_src_was_dirty = was_dirty(o, rename->path);
1582                char *add_name = unique_path(o, rename->path, other_branch);
1583                if (update_file(o, 0, &add->oid, add->mode, add_name))
1584                        return -1;
1585
1586                if (ren_src_was_dirty) {
1587                        output(o, 1, _("Refusing to lose dirty file at %s"),
1588                               rename->path);
1589                }
1590                /*
1591                 * Because the double negatives somehow keep confusing me...
1592                 *    1) update_wd iff !ren_src_was_dirty.
1593                 *    2) no_wd iff !update_wd
1594                 *    3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty
1595                 */
1596                remove_file(o, 0, rename->path, ren_src_was_dirty);
1597                dst_name = unique_path(o, rename->path, cur_branch);
1598        } else {
1599                if (dir_in_way(rename->path, !o->call_depth, 0)) {
1600                        dst_name = unique_path(o, rename->path, cur_branch);
1601                        output(o, 1, _("%s is a directory in %s adding as %s instead"),
1602                               rename->path, other_branch, dst_name);
1603                } else if (!o->call_depth &&
1604                           would_lose_untracked(rename->path)) {
1605                        dst_name = unique_path(o, rename->path, cur_branch);
1606                        output(o, 1, _("Refusing to lose untracked file at %s; "
1607                                       "adding as %s instead"),
1608                               rename->path, dst_name);
1609                }
1610        }
1611        if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1612                ; /* fall through, do allow dst_name to be released */
1613        else if (stage == 2)
1614                ret = update_stages(o, rename->path, NULL, rename, add);
1615        else
1616                ret = update_stages(o, rename->path, NULL, add, rename);
1617
1618        if (dst_name != rename->path)
1619                free(dst_name);
1620
1621        return ret;
1622}
1623
1624static int handle_rename_rename_1to2(struct merge_options *o,
1625                                     struct rename_conflict_info *ci)
1626{
1627        /* One file was renamed in both branches, but to different names. */
1628        struct diff_filespec *one = ci->pair1->one;
1629        struct diff_filespec *a = ci->pair1->two;
1630        struct diff_filespec *b = ci->pair2->two;
1631
1632        output(o, 1, _("CONFLICT (rename/rename): "
1633               "Rename \"%s\"->\"%s\" in branch \"%s\" "
1634               "rename \"%s\"->\"%s\" in \"%s\"%s"),
1635               one->path, a->path, ci->branch1,
1636               one->path, b->path, ci->branch2,
1637               o->call_depth ? _(" (left unresolved)") : "");
1638        if (o->call_depth) {
1639                struct merge_file_info mfi;
1640                struct diff_filespec other;
1641                struct diff_filespec *add;
1642                if (merge_mode_and_contents(o, one, a, b, one->path,
1643                                            ci->branch1, ci->branch2, &mfi))
1644                        return -1;
1645
1646                /*
1647                 * FIXME: For rename/add-source conflicts (if we could detect
1648                 * such), this is wrong.  We should instead find a unique
1649                 * pathname and then either rename the add-source file to that
1650                 * unique path, or use that unique path instead of src here.
1651                 */
1652                if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1653                        return -1;
1654
1655                /*
1656                 * Above, we put the merged content at the merge-base's
1657                 * path.  Now we usually need to delete both a->path and
1658                 * b->path.  However, the rename on each side of the merge
1659                 * could also be involved in a rename/add conflict.  In
1660                 * such cases, we should keep the added file around,
1661                 * resolving the conflict at that path in its favor.
1662                 */
1663                add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1664                if (add) {
1665                        if (update_file(o, 0, &add->oid, add->mode, a->path))
1666                                return -1;
1667                }
1668                else
1669                        remove_file_from_cache(a->path);
1670                add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1671                if (add) {
1672                        if (update_file(o, 0, &add->oid, add->mode, b->path))
1673                                return -1;
1674                }
1675                else
1676                        remove_file_from_cache(b->path);
1677        } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1678                return -1;
1679
1680        return 0;
1681}
1682
1683static int handle_rename_rename_2to1(struct merge_options *o,
1684                                     struct rename_conflict_info *ci)
1685{
1686        /* Two files, a & b, were renamed to the same thing, c. */
1687        struct diff_filespec *a = ci->pair1->one;
1688        struct diff_filespec *b = ci->pair2->one;
1689        struct diff_filespec *c1 = ci->pair1->two;
1690        struct diff_filespec *c2 = ci->pair2->two;
1691        char *path = c1->path; /* == c2->path */
1692        char *path_side_1_desc;
1693        char *path_side_2_desc;
1694        struct merge_file_info mfi_c1;
1695        struct merge_file_info mfi_c2;
1696        int ret;
1697
1698        output(o, 1, _("CONFLICT (rename/rename): "
1699               "Rename %s->%s in %s. "
1700               "Rename %s->%s in %s"),
1701               a->path, c1->path, ci->branch1,
1702               b->path, c2->path, ci->branch2);
1703
1704        remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1705        remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1706
1707        path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1708        path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1709        if (merge_mode_and_contents(o, a, c1, &ci->ren1_other, path_side_1_desc,
1710                                    o->branch1, o->branch2, &mfi_c1) ||
1711            merge_mode_and_contents(o, b, &ci->ren2_other, c2, path_side_2_desc,
1712                                    o->branch1, o->branch2, &mfi_c2))
1713                return -1;
1714        free(path_side_1_desc);
1715        free(path_side_2_desc);
1716
1717        if (o->call_depth) {
1718                /*
1719                 * If mfi_c1.clean && mfi_c2.clean, then it might make
1720                 * sense to do a two-way merge of those results.  But, I
1721                 * think in all cases, it makes sense to have the virtual
1722                 * merge base just undo the renames; they can be detected
1723                 * again later for the non-recursive merge.
1724                 */
1725                remove_file(o, 0, path, 0);
1726                ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1727                if (!ret)
1728                        ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1729                                          b->path);
1730        } else {
1731                char *new_path1 = unique_path(o, path, ci->branch1);
1732                char *new_path2 = unique_path(o, path, ci->branch2);
1733                output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1734                       a->path, new_path1, b->path, new_path2);
1735                if (was_dirty(o, path))
1736                        output(o, 1, _("Refusing to lose dirty file at %s"),
1737                               path);
1738                else if (would_lose_untracked(path))
1739                        /*
1740                         * Only way we get here is if both renames were from
1741                         * a directory rename AND user had an untracked file
1742                         * at the location where both files end up after the
1743                         * two directory renames.  See testcase 10d of t6043.
1744                         */
1745                        output(o, 1, _("Refusing to lose untracked file at "
1746                                       "%s, even though it's in the way."),
1747                               path);
1748                else
1749                        remove_file(o, 0, path, 0);
1750                ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1751                if (!ret)
1752                        ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1753                                          new_path2);
1754                /*
1755                 * unpack_trees() actually populates the index for us for
1756                 * "normal" rename/rename(2to1) situtations so that the
1757                 * correct entries are at the higher stages, which would
1758                 * make the call below to update_stages_for_stage_data
1759                 * unnecessary.  However, if either of the renames came
1760                 * from a directory rename, then unpack_trees() will not
1761                 * have gotten the right data loaded into the index, so we
1762                 * need to do so now.  (While it'd be tempting to move this
1763                 * call to update_stages_for_stage_data() to
1764                 * apply_directory_rename_modifications(), that would break
1765                 * our intermediate calls to would_lose_untracked() since
1766                 * those rely on the current in-memory index.  See also the
1767                 * big "NOTE" in update_stages()).
1768                 */
1769                if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1770                        ret = -1;
1771
1772                free(new_path2);
1773                free(new_path1);
1774        }
1775
1776        return ret;
1777}
1778
1779/*
1780 * Get the diff_filepairs changed between o_tree and tree.
1781 */
1782static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1783                                               struct tree *o_tree,
1784                                               struct tree *tree)
1785{
1786        struct diff_queue_struct *ret;
1787        struct diff_options opts;
1788
1789        diff_setup(&opts);
1790        opts.flags.recursive = 1;
1791        opts.flags.rename_empty = 0;
1792        opts.detect_rename = merge_detect_rename(o);
1793        /*
1794         * We do not have logic to handle the detection of copies.  In
1795         * fact, it may not even make sense to add such logic: would we
1796         * really want a change to a base file to be propagated through
1797         * multiple other files by a merge?
1798         */
1799        if (opts.detect_rename > DIFF_DETECT_RENAME)
1800                opts.detect_rename = DIFF_DETECT_RENAME;
1801        opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1802                            o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1803                            1000;
1804        opts.rename_score = o->rename_score;
1805        opts.show_rename_progress = o->show_rename_progress;
1806        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1807        diff_setup_done(&opts);
1808        diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1809        diffcore_std(&opts);
1810        if (opts.needed_rename_limit > o->needed_rename_limit)
1811                o->needed_rename_limit = opts.needed_rename_limit;
1812
1813        ret = xmalloc(sizeof(*ret));
1814        *ret = diff_queued_diff;
1815
1816        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1817        diff_queued_diff.nr = 0;
1818        diff_queued_diff.queue = NULL;
1819        diff_flush(&opts);
1820        return ret;
1821}
1822
1823static int tree_has_path(struct tree *tree, const char *path)
1824{
1825        struct object_id hashy;
1826        unsigned int mode_o;
1827
1828        return !get_tree_entry(&tree->object.oid, path,
1829                               &hashy, &mode_o);
1830}
1831
1832/*
1833 * Return a new string that replaces the beginning portion (which matches
1834 * entry->dir), with entry->new_dir.  In perl-speak:
1835 *   new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1836 * NOTE:
1837 *   Caller must ensure that old_path starts with entry->dir + '/'.
1838 */
1839static char *apply_dir_rename(struct dir_rename_entry *entry,
1840                              const char *old_path)
1841{
1842        struct strbuf new_path = STRBUF_INIT;
1843        int oldlen, newlen;
1844
1845        if (entry->non_unique_new_dir)
1846                return NULL;
1847
1848        oldlen = strlen(entry->dir);
1849        newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1850        strbuf_grow(&new_path, newlen);
1851        strbuf_addbuf(&new_path, &entry->new_dir);
1852        strbuf_addstr(&new_path, &old_path[oldlen]);
1853
1854        return strbuf_detach(&new_path, NULL);
1855}
1856
1857static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1858                                    char **old_dir, char **new_dir)
1859{
1860        char *end_of_old, *end_of_new;
1861        int old_len, new_len;
1862
1863        *old_dir = NULL;
1864        *new_dir = NULL;
1865
1866        /*
1867         * For
1868         *    "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1869         * the "e/foo.c" part is the same, we just want to know that
1870         *    "a/b/c/d" was renamed to "a/b/some/thing/else"
1871         * so, for this example, this function returns "a/b/c/d" in
1872         * *old_dir and "a/b/some/thing/else" in *new_dir.
1873         *
1874         * Also, if the basename of the file changed, we don't care.  We
1875         * want to know which portion of the directory, if any, changed.
1876         */
1877        end_of_old = strrchr(old_path, '/');
1878        end_of_new = strrchr(new_path, '/');
1879
1880        if (end_of_old == NULL || end_of_new == NULL)
1881                return;
1882        while (*--end_of_new == *--end_of_old &&
1883               end_of_old != old_path &&
1884               end_of_new != new_path)
1885                ; /* Do nothing; all in the while loop */
1886        /*
1887         * We've found the first non-matching character in the directory
1888         * paths.  That means the current directory we were comparing
1889         * represents the rename.  Move end_of_old and end_of_new back
1890         * to the full directory name.
1891         */
1892        if (*end_of_old == '/')
1893                end_of_old++;
1894        if (*end_of_old != '/')
1895                end_of_new++;
1896        end_of_old = strchr(end_of_old, '/');
1897        end_of_new = strchr(end_of_new, '/');
1898
1899        /*
1900         * It may have been the case that old_path and new_path were the same
1901         * directory all along.  Don't claim a rename if they're the same.
1902         */
1903        old_len = end_of_old - old_path;
1904        new_len = end_of_new - new_path;
1905
1906        if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1907                *old_dir = xstrndup(old_path, old_len);
1908                *new_dir = xstrndup(new_path, new_len);
1909        }
1910}
1911
1912static void remove_hashmap_entries(struct hashmap *dir_renames,
1913                                   struct string_list *items_to_remove)
1914{
1915        int i;
1916        struct dir_rename_entry *entry;
1917
1918        for (i = 0; i < items_to_remove->nr; i++) {
1919                entry = items_to_remove->items[i].util;
1920                hashmap_remove(dir_renames, entry, NULL);
1921        }
1922        string_list_clear(items_to_remove, 0);
1923}
1924
1925/*
1926 * See if there is a directory rename for path, and if there are any file
1927 * level conflicts for the renamed location.  If there is a rename and
1928 * there are no conflicts, return the new name.  Otherwise, return NULL.
1929 */
1930static char *handle_path_level_conflicts(struct merge_options *o,
1931                                         const char *path,
1932                                         struct dir_rename_entry *entry,
1933                                         struct hashmap *collisions,
1934                                         struct tree *tree)
1935{
1936        char *new_path = NULL;
1937        struct collision_entry *collision_ent;
1938        int clean = 1;
1939        struct strbuf collision_paths = STRBUF_INIT;
1940
1941        /*
1942         * entry has the mapping of old directory name to new directory name
1943         * that we want to apply to path.
1944         */
1945        new_path = apply_dir_rename(entry, path);
1946
1947        if (!new_path) {
1948                /* This should only happen when entry->non_unique_new_dir set */
1949                if (!entry->non_unique_new_dir)
1950                        BUG("entry->non_unqiue_dir not set and !new_path");
1951                output(o, 1, _("CONFLICT (directory rename split): "
1952                               "Unclear where to place %s because directory "
1953                               "%s was renamed to multiple other directories, "
1954                               "with no destination getting a majority of the "
1955                               "files."),
1956                       path, entry->dir);
1957                clean = 0;
1958                return NULL;
1959        }
1960
1961        /*
1962         * The caller needs to have ensured that it has pre-populated
1963         * collisions with all paths that map to new_path.  Do a quick check
1964         * to ensure that's the case.
1965         */
1966        collision_ent = collision_find_entry(collisions, new_path);
1967        if (collision_ent == NULL)
1968                BUG("collision_ent is NULL");
1969
1970        /*
1971         * Check for one-sided add/add/.../add conflicts, i.e.
1972         * where implicit renames from the other side doing
1973         * directory rename(s) can affect this side of history
1974         * to put multiple paths into the same location.  Warn
1975         * and bail on directory renames for such paths.
1976         */
1977        if (collision_ent->reported_already) {
1978                clean = 0;
1979        } else if (tree_has_path(tree, new_path)) {
1980                collision_ent->reported_already = 1;
1981                strbuf_add_separated_string_list(&collision_paths, ", ",
1982                                                 &collision_ent->source_files);
1983                output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1984                               "file/dir at %s in the way of implicit "
1985                               "directory rename(s) putting the following "
1986                               "path(s) there: %s."),
1987                       new_path, collision_paths.buf);
1988                clean = 0;
1989        } else if (collision_ent->source_files.nr > 1) {
1990                collision_ent->reported_already = 1;
1991                strbuf_add_separated_string_list(&collision_paths, ", ",
1992                                                 &collision_ent->source_files);
1993                output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1994                               "more than one path to %s; implicit directory "
1995                               "renames tried to put these paths there: %s"),
1996                       new_path, collision_paths.buf);
1997                clean = 0;
1998        }
1999
2000        /* Free memory we no longer need */
2001        strbuf_release(&collision_paths);
2002        if (!clean && new_path) {
2003                free(new_path);
2004                return NULL;
2005        }
2006
2007        return new_path;
2008}
2009
2010/*
2011 * There are a couple things we want to do at the directory level:
2012 *   1. Check for both sides renaming to the same thing, in order to avoid
2013 *      implicit renaming of files that should be left in place.  (See
2014 *      testcase 6b in t6043 for details.)
2015 *   2. Prune directory renames if there are still files left in the
2016 *      the original directory.  These represent a partial directory rename,
2017 *      i.e. a rename where only some of the files within the directory
2018 *      were renamed elsewhere.  (Technically, this could be done earlier
2019 *      in get_directory_renames(), except that would prevent us from
2020 *      doing the previous check and thus failing testcase 6b.)
2021 *   3. Check for rename/rename(1to2) conflicts (at the directory level).
2022 *      In the future, we could potentially record this info as well and
2023 *      omit reporting rename/rename(1to2) conflicts for each path within
2024 *      the affected directories, thus cleaning up the merge output.
2025 *   NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2026 *         directory level, because merging directories is fine.  If it
2027 *         causes conflicts for files within those merged directories, then
2028 *         that should be detected at the individual path level.
2029 */
2030static void handle_directory_level_conflicts(struct merge_options *o,
2031                                             struct hashmap *dir_re_head,
2032                                             struct tree *head,
2033                                             struct hashmap *dir_re_merge,
2034                                             struct tree *merge)
2035{
2036        struct hashmap_iter iter;
2037        struct dir_rename_entry *head_ent;
2038        struct dir_rename_entry *merge_ent;
2039
2040        struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2041        struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2042
2043        hashmap_iter_init(dir_re_head, &iter);
2044        while ((head_ent = hashmap_iter_next(&iter))) {
2045                merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2046                if (merge_ent &&
2047                    !head_ent->non_unique_new_dir &&
2048                    !merge_ent->non_unique_new_dir &&
2049                    !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2050                        /* 1. Renamed identically; remove it from both sides */
2051                        string_list_append(&remove_from_head,
2052                                           head_ent->dir)->util = head_ent;
2053                        strbuf_release(&head_ent->new_dir);
2054                        string_list_append(&remove_from_merge,
2055                                           merge_ent->dir)->util = merge_ent;
2056                        strbuf_release(&merge_ent->new_dir);
2057                } else if (tree_has_path(head, head_ent->dir)) {
2058                        /* 2. This wasn't a directory rename after all */
2059                        string_list_append(&remove_from_head,
2060                                           head_ent->dir)->util = head_ent;
2061                        strbuf_release(&head_ent->new_dir);
2062                }
2063        }
2064
2065        remove_hashmap_entries(dir_re_head, &remove_from_head);
2066        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2067
2068        hashmap_iter_init(dir_re_merge, &iter);
2069        while ((merge_ent = hashmap_iter_next(&iter))) {
2070                head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2071                if (tree_has_path(merge, merge_ent->dir)) {
2072                        /* 2. This wasn't a directory rename after all */
2073                        string_list_append(&remove_from_merge,
2074                                           merge_ent->dir)->util = merge_ent;
2075                } else if (head_ent &&
2076                           !head_ent->non_unique_new_dir &&
2077                           !merge_ent->non_unique_new_dir) {
2078                        /* 3. rename/rename(1to2) */
2079                        /*
2080                         * We can assume it's not rename/rename(1to1) because
2081                         * that was case (1), already checked above.  So we
2082                         * know that head_ent->new_dir and merge_ent->new_dir
2083                         * are different strings.
2084                         */
2085                        output(o, 1, _("CONFLICT (rename/rename): "
2086                                       "Rename directory %s->%s in %s. "
2087                                       "Rename directory %s->%s in %s"),
2088                               head_ent->dir, head_ent->new_dir.buf, o->branch1,
2089                               head_ent->dir, merge_ent->new_dir.buf, o->branch2);
2090                        string_list_append(&remove_from_head,
2091                                           head_ent->dir)->util = head_ent;
2092                        strbuf_release(&head_ent->new_dir);
2093                        string_list_append(&remove_from_merge,
2094                                           merge_ent->dir)->util = merge_ent;
2095                        strbuf_release(&merge_ent->new_dir);
2096                }
2097        }
2098
2099        remove_hashmap_entries(dir_re_head, &remove_from_head);
2100        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2101}
2102
2103static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
2104                                             struct tree *tree)
2105{
2106        struct hashmap *dir_renames;
2107        struct hashmap_iter iter;
2108        struct dir_rename_entry *entry;
2109        int i;
2110
2111        /*
2112         * Typically, we think of a directory rename as all files from a
2113         * certain directory being moved to a target directory.  However,
2114         * what if someone first moved two files from the original
2115         * directory in one commit, and then renamed the directory
2116         * somewhere else in a later commit?  At merge time, we just know
2117         * that files from the original directory went to two different
2118         * places, and that the bulk of them ended up in the same place.
2119         * We want each directory rename to represent where the bulk of the
2120         * files from that directory end up; this function exists to find
2121         * where the bulk of the files went.
2122         *
2123         * The first loop below simply iterates through the list of file
2124         * renames, finding out how often each directory rename pair
2125         * possibility occurs.
2126         */
2127        dir_renames = xmalloc(sizeof(*dir_renames));
2128        dir_rename_init(dir_renames);
2129        for (i = 0; i < pairs->nr; ++i) {
2130                struct string_list_item *item;
2131                int *count;
2132                struct diff_filepair *pair = pairs->queue[i];
2133                char *old_dir, *new_dir;
2134
2135                /* File not part of directory rename if it wasn't renamed */
2136                if (pair->status != 'R')
2137                        continue;
2138
2139                get_renamed_dir_portion(pair->one->path, pair->two->path,
2140                                        &old_dir,        &new_dir);
2141                if (!old_dir)
2142                        /* Directory didn't change at all; ignore this one. */
2143                        continue;
2144
2145                entry = dir_rename_find_entry(dir_renames, old_dir);
2146                if (!entry) {
2147                        entry = xmalloc(sizeof(*entry));
2148                        dir_rename_entry_init(entry, old_dir);
2149                        hashmap_put(dir_renames, entry);
2150                } else {
2151                        free(old_dir);
2152                }
2153                item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2154                if (!item) {
2155                        item = string_list_insert(&entry->possible_new_dirs,
2156                                                  new_dir);
2157                        item->util = xcalloc(1, sizeof(int));
2158                } else {
2159                        free(new_dir);
2160                }
2161                count = item->util;
2162                *count += 1;
2163        }
2164
2165        /*
2166         * For each directory with files moved out of it, we find out which
2167         * target directory received the most files so we can declare it to
2168         * be the "winning" target location for the directory rename.  This
2169         * winner gets recorded in new_dir.  If there is no winner
2170         * (multiple target directories received the same number of files),
2171         * we set non_unique_new_dir.  Once we've determined the winner (or
2172         * that there is no winner), we no longer need possible_new_dirs.
2173         */
2174        hashmap_iter_init(dir_renames, &iter);
2175        while ((entry = hashmap_iter_next(&iter))) {
2176                int max = 0;
2177                int bad_max = 0;
2178                char *best = NULL;
2179
2180                for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2181                        int *count = entry->possible_new_dirs.items[i].util;
2182
2183                        if (*count == max)
2184                                bad_max = max;
2185                        else if (*count > max) {
2186                                max = *count;
2187                                best = entry->possible_new_dirs.items[i].string;
2188                        }
2189                }
2190                if (bad_max == max)
2191                        entry->non_unique_new_dir = 1;
2192                else {
2193                        assert(entry->new_dir.len == 0);
2194                        strbuf_addstr(&entry->new_dir, best);
2195                }
2196                /*
2197                 * The relevant directory sub-portion of the original full
2198                 * filepaths were xstrndup'ed before inserting into
2199                 * possible_new_dirs, and instead of manually iterating the
2200                 * list and free'ing each, just lie and tell
2201                 * possible_new_dirs that it did the strdup'ing so that it
2202                 * will free them for us.
2203                 */
2204                entry->possible_new_dirs.strdup_strings = 1;
2205                string_list_clear(&entry->possible_new_dirs, 1);
2206        }
2207
2208        return dir_renames;
2209}
2210
2211static struct dir_rename_entry *check_dir_renamed(const char *path,
2212                                                  struct hashmap *dir_renames)
2213{
2214        char *temp = xstrdup(path);
2215        char *end;
2216        struct dir_rename_entry *entry = NULL;;
2217
2218        while ((end = strrchr(temp, '/'))) {
2219                *end = '\0';
2220                entry = dir_rename_find_entry(dir_renames, temp);
2221                if (entry)
2222                        break;
2223        }
2224        free(temp);
2225        return entry;
2226}
2227
2228static void compute_collisions(struct hashmap *collisions,
2229                               struct hashmap *dir_renames,
2230                               struct diff_queue_struct *pairs)
2231{
2232        int i;
2233
2234        /*
2235         * Multiple files can be mapped to the same path due to directory
2236         * renames done by the other side of history.  Since that other
2237         * side of history could have merged multiple directories into one,
2238         * if our side of history added the same file basename to each of
2239         * those directories, then all N of them would get implicitly
2240         * renamed by the directory rename detection into the same path,
2241         * and we'd get an add/add/.../add conflict, and all those adds
2242         * from *this* side of history.  This is not representable in the
2243         * index, and users aren't going to easily be able to make sense of
2244         * it.  So we need to provide a good warning about what's
2245         * happening, and fall back to no-directory-rename detection
2246         * behavior for those paths.
2247         *
2248         * See testcases 9e and all of section 5 from t6043 for examples.
2249         */
2250        collision_init(collisions);
2251
2252        for (i = 0; i < pairs->nr; ++i) {
2253                struct dir_rename_entry *dir_rename_ent;
2254                struct collision_entry *collision_ent;
2255                char *new_path;
2256                struct diff_filepair *pair = pairs->queue[i];
2257
2258                if (pair->status != 'A' && pair->status != 'R')
2259                        continue;
2260                dir_rename_ent = check_dir_renamed(pair->two->path,
2261                                                   dir_renames);
2262                if (!dir_rename_ent)
2263                        continue;
2264
2265                new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2266                if (!new_path)
2267                        /*
2268                         * dir_rename_ent->non_unique_new_path is true, which
2269                         * means there is no directory rename for us to use,
2270                         * which means it won't cause us any additional
2271                         * collisions.
2272                         */
2273                        continue;
2274                collision_ent = collision_find_entry(collisions, new_path);
2275                if (!collision_ent) {
2276                        collision_ent = xcalloc(1,
2277                                                sizeof(struct collision_entry));
2278                        hashmap_entry_init(collision_ent, strhash(new_path));
2279                        hashmap_put(collisions, collision_ent);
2280                        collision_ent->target_file = new_path;
2281                } else {
2282                        free(new_path);
2283                }
2284                string_list_insert(&collision_ent->source_files,
2285                                   pair->two->path);
2286        }
2287}
2288
2289static char *check_for_directory_rename(struct merge_options *o,
2290                                        const char *path,
2291                                        struct tree *tree,
2292                                        struct hashmap *dir_renames,
2293                                        struct hashmap *dir_rename_exclusions,
2294                                        struct hashmap *collisions,
2295                                        int *clean_merge)
2296{
2297        char *new_path = NULL;
2298        struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2299        struct dir_rename_entry *oentry = NULL;
2300
2301        if (!entry)
2302                return new_path;
2303
2304        /*
2305         * This next part is a little weird.  We do not want to do an
2306         * implicit rename into a directory we renamed on our side, because
2307         * that will result in a spurious rename/rename(1to2) conflict.  An
2308         * example:
2309         *   Base commit: dumbdir/afile, otherdir/bfile
2310         *   Side 1:      smrtdir/afile, otherdir/bfile
2311         *   Side 2:      dumbdir/afile, dumbdir/bfile
2312         * Here, while working on Side 1, we could notice that otherdir was
2313         * renamed/merged to dumbdir, and change the diff_filepair for
2314         * otherdir/bfile into a rename into dumbdir/bfile.  However, Side
2315         * 2 will notice the rename from dumbdir to smrtdir, and do the
2316         * transitive rename to move it from dumbdir/bfile to
2317         * smrtdir/bfile.  That gives us bfile in dumbdir vs being in
2318         * smrtdir, a rename/rename(1to2) conflict.  We really just want
2319         * the file to end up in smrtdir.  And the way to achieve that is
2320         * to not let Side1 do the rename to dumbdir, since we know that is
2321         * the source of one of our directory renames.
2322         *
2323         * That's why oentry and dir_rename_exclusions is here.
2324         *
2325         * As it turns out, this also prevents N-way transient rename
2326         * confusion; See testcases 9c and 9d of t6043.
2327         */
2328        oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2329        if (oentry) {
2330                output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2331                               "to %s, because %s itself was renamed."),
2332                       entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2333        } else {
2334                new_path = handle_path_level_conflicts(o, path, entry,
2335                                                       collisions, tree);
2336                *clean_merge &= (new_path != NULL);
2337        }
2338
2339        return new_path;
2340}
2341
2342static void apply_directory_rename_modifications(struct merge_options *o,
2343                                                 struct diff_filepair *pair,
2344                                                 char *new_path,
2345                                                 struct rename *re,
2346                                                 struct tree *tree,
2347                                                 struct tree *o_tree,
2348                                                 struct tree *a_tree,
2349                                                 struct tree *b_tree,
2350                                                 struct string_list *entries,
2351                                                 int *clean)
2352{
2353        struct string_list_item *item;
2354        int stage = (tree == a_tree ? 2 : 3);
2355        int update_wd;
2356
2357        /*
2358         * In all cases where we can do directory rename detection,
2359         * unpack_trees() will have read pair->two->path into the
2360         * index and the working copy.  We need to remove it so that
2361         * we can instead place it at new_path.  It is guaranteed to
2362         * not be untracked (unpack_trees() would have errored out
2363         * saying the file would have been overwritten), but it might
2364         * be dirty, though.
2365         */
2366        update_wd = !was_dirty(o, pair->two->path);
2367        if (!update_wd)
2368                output(o, 1, _("Refusing to lose dirty file at %s"),
2369                       pair->two->path);
2370        remove_file(o, 1, pair->two->path, !update_wd);
2371
2372        /* Find or create a new re->dst_entry */
2373        item = string_list_lookup(entries, new_path);
2374        if (item) {
2375                /*
2376                 * Since we're renaming on this side of history, and it's
2377                 * due to a directory rename on the other side of history
2378                 * (which we only allow when the directory in question no
2379                 * longer exists on the other side of history), the
2380                 * original entry for re->dst_entry is no longer
2381                 * necessary...
2382                 */
2383                re->dst_entry->processed = 1;
2384
2385                /*
2386                 * ...because we'll be using this new one.
2387                 */
2388                re->dst_entry = item->util;
2389        } else {
2390                /*
2391                 * re->dst_entry is for the before-dir-rename path, and we
2392                 * need it to hold information for the after-dir-rename
2393                 * path.  Before creating a new entry, we need to mark the
2394                 * old one as unnecessary (...unless it is shared by
2395                 * src_entry, i.e. this didn't use to be a rename, in which
2396                 * case we can just allow the normal processing to happen
2397                 * for it).
2398                 */
2399                if (pair->status == 'R')
2400                        re->dst_entry->processed = 1;
2401
2402                re->dst_entry = insert_stage_data(new_path,
2403                                                  o_tree, a_tree, b_tree,
2404                                                  entries);
2405                item = string_list_insert(entries, new_path);
2406                item->util = re->dst_entry;
2407        }
2408
2409        /*
2410         * Update the stage_data with the information about the path we are
2411         * moving into place.  That slot will be empty and available for us
2412         * to write to because of the collision checks in
2413         * handle_path_level_conflicts().  In other words,
2414         * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2415         * open for us to write to.
2416         *
2417         * It may be tempting to actually update the index at this point as
2418         * well, using update_stages_for_stage_data(), but as per the big
2419         * "NOTE" in update_stages(), doing so will modify the current
2420         * in-memory index which will break calls to would_lose_untracked()
2421         * that we need to make.  Instead, we need to just make sure that
2422         * the various handle_rename_*() functions update the index
2423         * explicitly rather than relying on unpack_trees() to have done it.
2424         */
2425        get_tree_entry(&tree->object.oid,
2426                       pair->two->path,
2427                       &re->dst_entry->stages[stage].oid,
2428                       &re->dst_entry->stages[stage].mode);
2429
2430        /* Update pair status */
2431        if (pair->status == 'A') {
2432                /*
2433                 * Recording rename information for this add makes it look
2434                 * like a rename/delete conflict.  Make sure we can
2435                 * correctly handle this as an add that was moved to a new
2436                 * directory instead of reporting a rename/delete conflict.
2437                 */
2438                re->add_turned_into_rename = 1;
2439        }
2440        /*
2441         * We don't actually look at pair->status again, but it seems
2442         * pedagogically correct to adjust it.
2443         */
2444        pair->status = 'R';
2445
2446        /*
2447         * Finally, record the new location.
2448         */
2449        pair->two->path = new_path;
2450}
2451
2452/*
2453 * Get information of all renames which occurred in 'pairs', making use of
2454 * any implicit directory renames inferred from the other side of history.
2455 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2456 * to be able to associate the correct cache entries with the rename
2457 * information; tree is always equal to either a_tree or b_tree.
2458 */
2459static struct string_list *get_renames(struct merge_options *o,
2460                                       struct diff_queue_struct *pairs,
2461                                       struct hashmap *dir_renames,
2462                                       struct hashmap *dir_rename_exclusions,
2463                                       struct tree *tree,
2464                                       struct tree *o_tree,
2465                                       struct tree *a_tree,
2466                                       struct tree *b_tree,
2467                                       struct string_list *entries,
2468                                       int *clean_merge)
2469{
2470        int i;
2471        struct hashmap collisions;
2472        struct hashmap_iter iter;
2473        struct collision_entry *e;
2474        struct string_list *renames;
2475
2476        compute_collisions(&collisions, dir_renames, pairs);
2477        renames = xcalloc(1, sizeof(struct string_list));
2478
2479        for (i = 0; i < pairs->nr; ++i) {
2480                struct string_list_item *item;
2481                struct rename *re;
2482                struct diff_filepair *pair = pairs->queue[i];
2483                char *new_path; /* non-NULL only with directory renames */
2484
2485                if (pair->status != 'A' && pair->status != 'R') {
2486                        diff_free_filepair(pair);
2487                        continue;
2488                }
2489                new_path = check_for_directory_rename(o, pair->two->path, tree,
2490                                                      dir_renames,
2491                                                      dir_rename_exclusions,
2492                                                      &collisions,
2493                                                      clean_merge);
2494                if (pair->status != 'R' && !new_path) {
2495                        diff_free_filepair(pair);
2496                        continue;
2497                }
2498
2499                re = xmalloc(sizeof(*re));
2500                re->processed = 0;
2501                re->add_turned_into_rename = 0;
2502                re->pair = pair;
2503                item = string_list_lookup(entries, re->pair->one->path);
2504                if (!item)
2505                        re->src_entry = insert_stage_data(re->pair->one->path,
2506                                        o_tree, a_tree, b_tree, entries);
2507                else
2508                        re->src_entry = item->util;
2509
2510                item = string_list_lookup(entries, re->pair->two->path);
2511                if (!item)
2512                        re->dst_entry = insert_stage_data(re->pair->two->path,
2513                                        o_tree, a_tree, b_tree, entries);
2514                else
2515                        re->dst_entry = item->util;
2516                item = string_list_insert(renames, pair->one->path);
2517                item->util = re;
2518                if (new_path)
2519                        apply_directory_rename_modifications(o, pair, new_path,
2520                                                             re, tree, o_tree,
2521                                                             a_tree, b_tree,
2522                                                             entries,
2523                                                             clean_merge);
2524        }
2525
2526        hashmap_iter_init(&collisions, &iter);
2527        while ((e = hashmap_iter_next(&iter))) {
2528                free(e->target_file);
2529                string_list_clear(&e->source_files, 0);
2530        }
2531        hashmap_free(&collisions, 1);
2532        return renames;
2533}
2534
2535static int process_renames(struct merge_options *o,
2536                           struct string_list *a_renames,
2537                           struct string_list *b_renames)
2538{
2539        int clean_merge = 1, i, j;
2540        struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2541        struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2542        const struct rename *sre;
2543
2544        for (i = 0; i < a_renames->nr; i++) {
2545                sre = a_renames->items[i].util;
2546                string_list_insert(&a_by_dst, sre->pair->two->path)->util
2547                        = (void *)sre;
2548        }
2549        for (i = 0; i < b_renames->nr; i++) {
2550                sre = b_renames->items[i].util;
2551                string_list_insert(&b_by_dst, sre->pair->two->path)->util
2552                        = (void *)sre;
2553        }
2554
2555        for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2556                struct string_list *renames1, *renames2Dst;
2557                struct rename *ren1 = NULL, *ren2 = NULL;
2558                const char *branch1, *branch2;
2559                const char *ren1_src, *ren1_dst;
2560                struct string_list_item *lookup;
2561
2562                if (i >= a_renames->nr) {
2563                        ren2 = b_renames->items[j++].util;
2564                } else if (j >= b_renames->nr) {
2565                        ren1 = a_renames->items[i++].util;
2566                } else {
2567                        int compare = strcmp(a_renames->items[i].string,
2568                                             b_renames->items[j].string);
2569                        if (compare <= 0)
2570                                ren1 = a_renames->items[i++].util;
2571                        if (compare >= 0)
2572                                ren2 = b_renames->items[j++].util;
2573                }
2574
2575                /* TODO: refactor, so that 1/2 are not needed */
2576                if (ren1) {
2577                        renames1 = a_renames;
2578                        renames2Dst = &b_by_dst;
2579                        branch1 = o->branch1;
2580                        branch2 = o->branch2;
2581                } else {
2582                        renames1 = b_renames;
2583                        renames2Dst = &a_by_dst;
2584                        branch1 = o->branch2;
2585                        branch2 = o->branch1;
2586                        SWAP(ren2, ren1);
2587                }
2588
2589                if (ren1->processed)
2590                        continue;
2591                ren1->processed = 1;
2592                ren1->dst_entry->processed = 1;
2593                /* BUG: We should only mark src_entry as processed if we
2594                 * are not dealing with a rename + add-source case.
2595                 */
2596                ren1->src_entry->processed = 1;
2597
2598                ren1_src = ren1->pair->one->path;
2599                ren1_dst = ren1->pair->two->path;
2600
2601                if (ren2) {
2602                        /* One file renamed on both sides */
2603                        const char *ren2_src = ren2->pair->one->path;
2604                        const char *ren2_dst = ren2->pair->two->path;
2605                        enum rename_type rename_type;
2606                        if (strcmp(ren1_src, ren2_src) != 0)
2607                                BUG("ren1_src != ren2_src");
2608                        ren2->dst_entry->processed = 1;
2609                        ren2->processed = 1;
2610                        if (strcmp(ren1_dst, ren2_dst) != 0) {
2611                                rename_type = RENAME_ONE_FILE_TO_TWO;
2612                                clean_merge = 0;
2613                        } else {
2614                                rename_type = RENAME_ONE_FILE_TO_ONE;
2615                                /* BUG: We should only remove ren1_src in
2616                                 * the base stage (think of rename +
2617                                 * add-source cases).
2618                                 */
2619                                remove_file(o, 1, ren1_src, 1);
2620                                update_entry(ren1->dst_entry,
2621                                             ren1->pair->one,
2622                                             ren1->pair->two,
2623                                             ren2->pair->two);
2624                        }
2625                        setup_rename_conflict_info(rename_type,
2626                                                   ren1->pair,
2627                                                   ren2->pair,
2628                                                   branch1,
2629                                                   branch2,
2630                                                   ren1->dst_entry,
2631                                                   ren2->dst_entry,
2632                                                   o,
2633                                                   NULL,
2634                                                   NULL);
2635                } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2636                        /* Two different files renamed to the same thing */
2637                        char *ren2_dst;
2638                        ren2 = lookup->util;
2639                        ren2_dst = ren2->pair->two->path;
2640                        if (strcmp(ren1_dst, ren2_dst) != 0)
2641                                BUG("ren1_dst != ren2_dst");
2642
2643                        clean_merge = 0;
2644                        ren2->processed = 1;
2645                        /*
2646                         * BUG: We should only mark src_entry as processed
2647                         * if we are not dealing with a rename + add-source
2648                         * case.
2649                         */
2650                        ren2->src_entry->processed = 1;
2651
2652                        setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2653                                                   ren1->pair,
2654                                                   ren2->pair,
2655                                                   branch1,
2656                                                   branch2,
2657                                                   ren1->dst_entry,
2658                                                   ren2->dst_entry,
2659                                                   o,
2660                                                   ren1->src_entry,
2661                                                   ren2->src_entry);
2662
2663                } else {
2664                        /* Renamed in 1, maybe changed in 2 */
2665                        /* we only use sha1 and mode of these */
2666                        struct diff_filespec src_other, dst_other;
2667                        int try_merge;
2668
2669                        /*
2670                         * unpack_trees loads entries from common-commit
2671                         * into stage 1, from head-commit into stage 2, and
2672                         * from merge-commit into stage 3.  We keep track
2673                         * of which side corresponds to the rename.
2674                         */
2675                        int renamed_stage = a_renames == renames1 ? 2 : 3;
2676                        int other_stage =   a_renames == renames1 ? 3 : 2;
2677
2678                        /* BUG: We should only remove ren1_src in the base
2679                         * stage and in other_stage (think of rename +
2680                         * add-source case).
2681                         */
2682                        remove_file(o, 1, ren1_src,
2683                                    renamed_stage == 2 || !was_tracked(o, ren1_src));
2684
2685                        oidcpy(&src_other.oid,
2686                               &ren1->src_entry->stages[other_stage].oid);
2687                        src_other.mode = ren1->src_entry->stages[other_stage].mode;
2688                        oidcpy(&dst_other.oid,
2689                               &ren1->dst_entry->stages[other_stage].oid);
2690                        dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2691                        try_merge = 0;
2692
2693                        if (oid_eq(&src_other.oid, &null_oid) &&
2694                            ren1->add_turned_into_rename) {
2695                                setup_rename_conflict_info(RENAME_VIA_DIR,
2696                                                           ren1->pair,
2697                                                           NULL,
2698                                                           branch1,
2699                                                           branch2,
2700                                                           ren1->dst_entry,
2701                                                           NULL,
2702                                                           o,
2703                                                           NULL,
2704                                                           NULL);
2705                        } else if (oid_eq(&src_other.oid, &null_oid)) {
2706                                setup_rename_conflict_info(RENAME_DELETE,
2707                                                           ren1->pair,
2708                                                           NULL,
2709                                                           branch1,
2710                                                           branch2,
2711                                                           ren1->dst_entry,
2712                                                           NULL,
2713                                                           o,
2714                                                           NULL,
2715                                                           NULL);
2716                        } else if ((dst_other.mode == ren1->pair->two->mode) &&
2717                                   oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2718                                /*
2719                                 * Added file on the other side identical to
2720                                 * the file being renamed: clean merge.
2721                                 * Also, there is no need to overwrite the
2722                                 * file already in the working copy, so call
2723                                 * update_file_flags() instead of
2724                                 * update_file().
2725                                 */
2726                                if (update_file_flags(o,
2727                                                      &ren1->pair->two->oid,
2728                                                      ren1->pair->two->mode,
2729                                                      ren1_dst,
2730                                                      1, /* update_cache */
2731                                                      0  /* update_wd    */))
2732                                        clean_merge = -1;
2733                        } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2734                                clean_merge = 0;
2735                                try_merge = 1;
2736                                output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2737                                       "%s added in %s"),
2738                                       ren1_src, ren1_dst, branch1,
2739                                       ren1_dst, branch2);
2740                                if (o->call_depth) {
2741                                        struct merge_file_info mfi;
2742                                        struct diff_filespec one, a, b;
2743
2744                                        oidcpy(&one.oid, &null_oid);
2745                                        one.mode = 0;
2746                                        one.path = ren1->pair->two->path;
2747
2748                                        oidcpy(&a.oid, &ren1->pair->two->oid);
2749                                        a.mode = ren1->pair->two->mode;
2750                                        a.path = one.path;
2751
2752                                        oidcpy(&b.oid, &dst_other.oid);
2753                                        b.mode = dst_other.mode;
2754                                        b.path = one.path;
2755
2756                                        if (merge_mode_and_contents(o, &one, &a, &b, ren1_dst,
2757                                                                    branch1, branch2,
2758                                                                    &mfi)) {
2759                                                clean_merge = -1;
2760                                                goto cleanup_and_return;
2761                                        }
2762                                        output(o, 1, _("Adding merged %s"), ren1_dst);
2763                                        if (update_file(o, 0, &mfi.oid,
2764                                                        mfi.mode, ren1_dst))
2765                                                clean_merge = -1;
2766                                        try_merge = 0;
2767                                } else {
2768                                        char *new_path = unique_path(o, ren1_dst, branch2);
2769                                        output(o, 1, _("Adding as %s instead"), new_path);
2770                                        if (update_file(o, 0, &dst_other.oid,
2771                                                        dst_other.mode, new_path))
2772                                                clean_merge = -1;
2773                                        free(new_path);
2774                                }
2775                        } else
2776                                try_merge = 1;
2777
2778                        if (clean_merge < 0)
2779                                goto cleanup_and_return;
2780                        if (try_merge) {
2781                                struct diff_filespec *one, *a, *b;
2782                                src_other.path = (char *)ren1_src;
2783
2784                                one = ren1->pair->one;
2785                                if (a_renames == renames1) {
2786                                        a = ren1->pair->two;
2787                                        b = &src_other;
2788                                } else {
2789                                        b = ren1->pair->two;
2790                                        a = &src_other;
2791                                }
2792                                update_entry(ren1->dst_entry, one, a, b);
2793                                setup_rename_conflict_info(RENAME_NORMAL,
2794                                                           ren1->pair,
2795                                                           NULL,
2796                                                           branch1,
2797                                                           NULL,
2798                                                           ren1->dst_entry,
2799                                                           NULL,
2800                                                           o,
2801                                                           NULL,
2802                                                           NULL);
2803                        }
2804                }
2805        }
2806cleanup_and_return:
2807        string_list_clear(&a_by_dst, 0);
2808        string_list_clear(&b_by_dst, 0);
2809
2810        return clean_merge;
2811}
2812
2813struct rename_info {
2814        struct string_list *head_renames;
2815        struct string_list *merge_renames;
2816};
2817
2818static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2819                                   struct hashmap *dir_renames)
2820{
2821        struct hashmap_iter iter;
2822        struct dir_rename_entry *e;
2823
2824        hashmap_iter_init(dir_renames, &iter);
2825        while ((e = hashmap_iter_next(&iter))) {
2826                free(e->dir);
2827                strbuf_release(&e->new_dir);
2828                /* possible_new_dirs already cleared in get_directory_renames */
2829        }
2830        hashmap_free(dir_renames, 1);
2831        free(dir_renames);
2832
2833        free(pairs->queue);
2834        free(pairs);
2835}
2836
2837static int detect_and_process_renames(struct merge_options *o,
2838                                      struct tree *common,
2839                                      struct tree *head,
2840                                      struct tree *merge,
2841                                      struct string_list *entries,
2842                                      struct rename_info *ri)
2843{
2844        struct diff_queue_struct *head_pairs, *merge_pairs;
2845        struct hashmap *dir_re_head, *dir_re_merge;
2846        int clean = 1;
2847
2848        ri->head_renames = NULL;
2849        ri->merge_renames = NULL;
2850
2851        if (!merge_detect_rename(o))
2852                return 1;
2853
2854        head_pairs = get_diffpairs(o, common, head);
2855        merge_pairs = get_diffpairs(o, common, merge);
2856
2857        if (o->detect_directory_renames) {
2858                dir_re_head = get_directory_renames(head_pairs, head);
2859                dir_re_merge = get_directory_renames(merge_pairs, merge);
2860
2861                handle_directory_level_conflicts(o,
2862                                                 dir_re_head, head,
2863                                                 dir_re_merge, merge);
2864        } else {
2865                dir_re_head  = xmalloc(sizeof(*dir_re_head));
2866                dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2867                dir_rename_init(dir_re_head);
2868                dir_rename_init(dir_re_merge);
2869        }
2870
2871        ri->head_renames  = get_renames(o, head_pairs,
2872                                        dir_re_merge, dir_re_head, head,
2873                                        common, head, merge, entries,
2874                                        &clean);
2875        if (clean < 0)
2876                goto cleanup;
2877        ri->merge_renames = get_renames(o, merge_pairs,
2878                                        dir_re_head, dir_re_merge, merge,
2879                                        common, head, merge, entries,
2880                                        &clean);
2881        if (clean < 0)
2882                goto cleanup;
2883        clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2884
2885cleanup:
2886        /*
2887         * Some cleanup is deferred until cleanup_renames() because the
2888         * data structures are still needed and referenced in
2889         * process_entry().  But there are a few things we can free now.
2890         */
2891        initial_cleanup_rename(head_pairs, dir_re_head);
2892        initial_cleanup_rename(merge_pairs, dir_re_merge);
2893
2894        return clean;
2895}
2896
2897static void final_cleanup_rename(struct string_list *rename)
2898{
2899        const struct rename *re;
2900        int i;
2901
2902        if (rename == NULL)
2903                return;
2904
2905        for (i = 0; i < rename->nr; i++) {
2906                re = rename->items[i].util;
2907                diff_free_filepair(re->pair);
2908        }
2909        string_list_clear(rename, 1);
2910        free(rename);
2911}
2912
2913static void final_cleanup_renames(struct rename_info *re_info)
2914{
2915        final_cleanup_rename(re_info->head_renames);
2916        final_cleanup_rename(re_info->merge_renames);
2917}
2918
2919static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2920{
2921        return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2922}
2923
2924static int read_oid_strbuf(struct merge_options *o,
2925                           const struct object_id *oid,
2926                           struct strbuf *dst)
2927{
2928        void *buf;
2929        enum object_type type;
2930        unsigned long size;
2931        buf = read_object_file(oid, &type, &size);
2932        if (!buf)
2933                return err(o, _("cannot read object %s"), oid_to_hex(oid));
2934        if (type != OBJ_BLOB) {
2935                free(buf);
2936                return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2937        }
2938        strbuf_attach(dst, buf, size, size + 1);
2939        return 0;
2940}
2941
2942static int blob_unchanged(struct merge_options *opt,
2943                          const struct object_id *o_oid,
2944                          unsigned o_mode,
2945                          const struct object_id *a_oid,
2946                          unsigned a_mode,
2947                          int renormalize, const char *path)
2948{
2949        struct strbuf o = STRBUF_INIT;
2950        struct strbuf a = STRBUF_INIT;
2951        int ret = 0; /* assume changed for safety */
2952
2953        if (a_mode != o_mode)
2954                return 0;
2955        if (oid_eq(o_oid, a_oid))
2956                return 1;
2957        if (!renormalize)
2958                return 0;
2959
2960        assert(o_oid && a_oid);
2961        if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2962                goto error_return;
2963        /*
2964         * Note: binary | is used so that both renormalizations are
2965         * performed.  Comparison can be skipped if both files are
2966         * unchanged since their sha1s have already been compared.
2967         */
2968        if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2969            renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2970                ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2971
2972error_return:
2973        strbuf_release(&o);
2974        strbuf_release(&a);
2975        return ret;
2976}
2977
2978static int handle_modify_delete(struct merge_options *o,
2979                                const char *path,
2980                                struct object_id *o_oid, int o_mode,
2981                                struct object_id *a_oid, int a_mode,
2982                                struct object_id *b_oid, int b_mode)
2983{
2984        const char *modify_branch, *delete_branch;
2985        struct object_id *changed_oid;
2986        int changed_mode;
2987
2988        if (a_oid) {
2989                modify_branch = o->branch1;
2990                delete_branch = o->branch2;
2991                changed_oid = a_oid;
2992                changed_mode = a_mode;
2993        } else {
2994                modify_branch = o->branch2;
2995                delete_branch = o->branch1;
2996                changed_oid = b_oid;
2997                changed_mode = b_mode;
2998        }
2999
3000        return handle_change_delete(o,
3001                                    path, NULL,
3002                                    o_oid, o_mode,
3003                                    changed_oid, changed_mode,
3004                                    modify_branch, delete_branch,
3005                                    _("modify"), _("modified"));
3006}
3007
3008static int handle_content_merge(struct merge_options *o,
3009                                const char *path,
3010                                int is_dirty,
3011                                struct object_id *o_oid, int o_mode,
3012                                struct object_id *a_oid, int a_mode,
3013                                struct object_id *b_oid, int b_mode,
3014                                struct rename_conflict_info *rename_conflict_info)
3015{
3016        const char *reason = _("content");
3017        const char *path1 = NULL, *path2 = NULL;
3018        struct merge_file_info mfi;
3019        struct diff_filespec one, a, b;
3020        unsigned df_conflict_remains = 0;
3021
3022        if (!o_oid) {
3023                reason = _("add/add");
3024                o_oid = (struct object_id *)&null_oid;
3025        }
3026        one.path = a.path = b.path = (char *)path;
3027        oidcpy(&one.oid, o_oid);
3028        one.mode = o_mode;
3029        oidcpy(&a.oid, a_oid);
3030        a.mode = a_mode;
3031        oidcpy(&b.oid, b_oid);
3032        b.mode = b_mode;
3033
3034        if (rename_conflict_info) {
3035                struct diff_filepair *pair1 = rename_conflict_info->pair1;
3036
3037                path1 = (o->branch1 == rename_conflict_info->branch1) ?
3038                        pair1->two->path : pair1->one->path;
3039                /* If rename_conflict_info->pair2 != NULL, we are in
3040                 * RENAME_ONE_FILE_TO_ONE case.  Otherwise, we have a
3041                 * normal rename.
3042                 */
3043                path2 = (rename_conflict_info->pair2 ||
3044                         o->branch2 == rename_conflict_info->branch1) ?
3045                        pair1->two->path : pair1->one->path;
3046                one.path = pair1->one->path;
3047                a.path = (char *)path1;
3048                b.path = (char *)path2;
3049
3050                if (dir_in_way(path, !o->call_depth,
3051                               S_ISGITLINK(pair1->two->mode)))
3052                        df_conflict_remains = 1;
3053        }
3054        if (merge_mode_and_contents(o, &one, &a, &b, path,
3055                                    o->branch1, o->branch2, &mfi))
3056                return -1;
3057
3058        /*
3059         * We can skip updating the working tree file iff:
3060         *   a) The merge is clean
3061         *   b) The merge matches what was in HEAD (content, mode, pathname)
3062         *   c) The target path is usable (i.e. not involved in D/F conflict)
3063         */
3064        if (mfi.clean &&
3065            was_tracked_and_matches(o, path, &mfi.oid, mfi.mode) &&
3066            !df_conflict_remains) {
3067                int pos;
3068                struct cache_entry *ce;
3069
3070                output(o, 3, _("Skipped %s (merged same as existing)"), path);
3071                if (add_cacheinfo(o, mfi.mode, &mfi.oid, path,
3072                                  0, (!o->call_depth && !is_dirty), 0))
3073                        return -1;
3074                /*
3075                 * However, add_cacheinfo() will delete the old cache entry
3076                 * and add a new one.  We need to copy over any skip_worktree
3077                 * flag to avoid making the file appear as if it were
3078                 * deleted by the user.
3079                 */
3080                pos = index_name_pos(&o->orig_index, path, strlen(path));
3081                ce = o->orig_index.cache[pos];
3082                if (ce_skip_worktree(ce)) {
3083                        pos = index_name_pos(&the_index, path, strlen(path));
3084                        ce = the_index.cache[pos];
3085                        ce->ce_flags |= CE_SKIP_WORKTREE;
3086                }
3087                return mfi.clean;
3088        }
3089
3090        if (!mfi.clean) {
3091                if (S_ISGITLINK(mfi.mode))
3092                        reason = _("submodule");
3093                output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
3094                                reason, path);
3095                if (rename_conflict_info && !df_conflict_remains)
3096                        if (update_stages(o, path, &one, &a, &b))
3097                                return -1;
3098        }
3099
3100        if (df_conflict_remains || is_dirty) {
3101                char *new_path;
3102                if (o->call_depth) {
3103                        remove_file_from_cache(path);
3104                } else {
3105                        if (!mfi.clean) {
3106                                if (update_stages(o, path, &one, &a, &b))
3107                                        return -1;
3108                        } else {
3109                                int file_from_stage2 = was_tracked(o, path);
3110                                struct diff_filespec merged;
3111                                oidcpy(&merged.oid, &mfi.oid);
3112                                merged.mode = mfi.mode;
3113
3114                                if (update_stages(o, path, NULL,
3115                                                  file_from_stage2 ? &merged : NULL,
3116                                                  file_from_stage2 ? NULL : &merged))
3117                                        return -1;
3118                        }
3119
3120                }
3121                new_path = unique_path(o, path, rename_conflict_info->branch1);
3122                if (is_dirty) {
3123                        output(o, 1, _("Refusing to lose dirty file at %s"),
3124                               path);
3125                }
3126                output(o, 1, _("Adding as %s instead"), new_path);
3127                if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
3128                        free(new_path);
3129                        return -1;
3130                }
3131                free(new_path);
3132                mfi.clean = 0;
3133        } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
3134                return -1;
3135        return !is_dirty && mfi.clean;
3136}
3137
3138static int handle_rename_normal(struct merge_options *o,
3139                                const char *path,
3140                                struct object_id *o_oid, unsigned int o_mode,
3141                                struct object_id *a_oid, unsigned int a_mode,
3142                                struct object_id *b_oid, unsigned int b_mode,
3143                                struct rename_conflict_info *ci)
3144{
3145        /* Merge the content and write it out */
3146        return handle_content_merge(o, path, was_dirty(o, path),
3147                                    o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
3148                                    ci);
3149}
3150
3151/* Per entry merge function */
3152static int process_entry(struct merge_options *o,
3153                         const char *path, struct stage_data *entry)
3154{
3155        int clean_merge = 1;
3156        int normalize = o->renormalize;
3157        unsigned o_mode = entry->stages[1].mode;
3158        unsigned a_mode = entry->stages[2].mode;
3159        unsigned b_mode = entry->stages[3].mode;
3160        struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
3161        struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
3162        struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
3163
3164        entry->processed = 1;
3165        if (entry->rename_conflict_info) {
3166                struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
3167                switch (conflict_info->rename_type) {
3168                case RENAME_NORMAL:
3169                case RENAME_ONE_FILE_TO_ONE:
3170                        clean_merge = handle_rename_normal(o,
3171                                                           path,
3172                                                           o_oid, o_mode,
3173                                                           a_oid, a_mode,
3174                                                           b_oid, b_mode,
3175                                                           conflict_info);
3176                        break;
3177                case RENAME_VIA_DIR:
3178                        clean_merge = 1;
3179                        if (handle_rename_via_dir(o,
3180                                                  conflict_info->pair1,
3181                                                  conflict_info->branch1,
3182                                                  conflict_info->branch2))
3183                                clean_merge = -1;
3184                        break;
3185                case RENAME_DELETE:
3186                        clean_merge = 0;
3187                        if (handle_rename_delete(o,
3188                                                 conflict_info->pair1,
3189                                                 conflict_info->branch1,
3190                                                 conflict_info->branch2))
3191                                clean_merge = -1;
3192                        break;
3193                case RENAME_ONE_FILE_TO_TWO:
3194                        clean_merge = 0;
3195                        if (handle_rename_rename_1to2(o, conflict_info))
3196                                clean_merge = -1;
3197                        break;
3198                case RENAME_TWO_FILES_TO_ONE:
3199                        clean_merge = 0;
3200                        if (handle_rename_rename_2to1(o, conflict_info))
3201                                clean_merge = -1;
3202                        break;
3203                default:
3204                        entry->processed = 0;
3205                        break;
3206                }
3207        } else if (o_oid && (!a_oid || !b_oid)) {
3208                /* Case A: Deleted in one */
3209                if ((!a_oid && !b_oid) ||
3210                    (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
3211                    (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
3212                        /* Deleted in both or deleted in one and
3213                         * unchanged in the other */
3214                        if (a_oid)
3215                                output(o, 2, _("Removing %s"), path);
3216                        /* do not touch working file if it did not exist */
3217                        remove_file(o, 1, path, !a_oid);
3218                } else {
3219                        /* Modify/delete; deleted side may have put a directory in the way */
3220                        clean_merge = 0;
3221                        if (handle_modify_delete(o, path, o_oid, o_mode,
3222                                                 a_oid, a_mode, b_oid, b_mode))
3223                                clean_merge = -1;
3224                }
3225        } else if ((!o_oid && a_oid && !b_oid) ||
3226                   (!o_oid && !a_oid && b_oid)) {
3227                /* Case B: Added in one. */
3228                /* [nothing|directory] -> ([nothing|directory], file) */
3229
3230                const char *add_branch;
3231                const char *other_branch;
3232                unsigned mode;
3233                const struct object_id *oid;
3234                const char *conf;
3235
3236                if (a_oid) {
3237                        add_branch = o->branch1;
3238                        other_branch = o->branch2;
3239                        mode = a_mode;
3240                        oid = a_oid;
3241                        conf = _("file/directory");
3242                } else {
3243                        add_branch = o->branch2;
3244                        other_branch = o->branch1;
3245                        mode = b_mode;
3246                        oid = b_oid;
3247                        conf = _("directory/file");
3248                }
3249                if (dir_in_way(path,
3250                               !o->call_depth && !S_ISGITLINK(a_mode),
3251                               0)) {
3252                        char *new_path = unique_path(o, path, add_branch);
3253                        clean_merge = 0;
3254                        output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3255                               "Adding %s as %s"),
3256                               conf, path, other_branch, path, new_path);
3257                        if (update_file(o, 0, oid, mode, new_path))
3258                                clean_merge = -1;
3259                        else if (o->call_depth)
3260                                remove_file_from_cache(path);
3261                        free(new_path);
3262                } else {
3263                        output(o, 2, _("Adding %s"), path);
3264                        /* do not overwrite file if already present */
3265                        if (update_file_flags(o, oid, mode, path, 1, !a_oid))
3266                                clean_merge = -1;
3267                }
3268        } else if (a_oid && b_oid) {
3269                /* Case C: Added in both (check for same permissions) and */
3270                /* case D: Modified in both, but differently. */
3271                int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3272                clean_merge = handle_content_merge(o, path, is_dirty,
3273                                                   o_oid, o_mode,
3274                                                   a_oid, a_mode,
3275                                                   b_oid, b_mode,
3276                                                   NULL);
3277        } else if (!o_oid && !a_oid && !b_oid) {
3278                /*
3279                 * this entry was deleted altogether. a_mode == 0 means
3280                 * we had that path and want to actively remove it.
3281                 */
3282                remove_file(o, 1, path, !a_mode);
3283        } else
3284                BUG("fatal merge failure, shouldn't happen.");
3285
3286        return clean_merge;
3287}
3288
3289int merge_trees(struct merge_options *o,
3290                struct tree *head,
3291                struct tree *merge,
3292                struct tree *common,
3293                struct tree **result)
3294{
3295        int code, clean;
3296        struct strbuf sb = STRBUF_INIT;
3297
3298        if (!o->call_depth && index_has_changes(&the_index, head, &sb)) {
3299                err(o, _("Your local changes to the following files would be overwritten by merge:\n  %s"),
3300                    sb.buf);
3301                return -1;
3302        }
3303
3304        if (o->subtree_shift) {
3305                merge = shift_tree_object(head, merge, o->subtree_shift);
3306                common = shift_tree_object(head, common, o->subtree_shift);
3307        }
3308
3309        if (oid_eq(&common->object.oid, &merge->object.oid)) {
3310                output(o, 0, _("Already up to date!"));
3311                *result = head;
3312                return 1;
3313        }
3314
3315        code = unpack_trees_start(o, common, head, merge);
3316
3317        if (code != 0) {
3318                if (show(o, 4) || o->call_depth)
3319                        err(o, _("merging of trees %s and %s failed"),
3320                            oid_to_hex(&head->object.oid),
3321                            oid_to_hex(&merge->object.oid));
3322                unpack_trees_finish(o);
3323                return -1;
3324        }
3325
3326        if (unmerged_cache()) {
3327                struct string_list *entries;
3328                struct rename_info re_info;
3329                int i;
3330                /*
3331                 * Only need the hashmap while processing entries, so
3332                 * initialize it here and free it when we are done running
3333                 * through the entries. Keeping it in the merge_options as
3334                 * opposed to decaring a local hashmap is for convenience
3335                 * so that we don't have to pass it to around.
3336                 */
3337                hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3338                get_files_dirs(o, head);
3339                get_files_dirs(o, merge);
3340
3341                entries = get_unmerged();
3342                clean = detect_and_process_renames(o, common, head, merge,
3343                                                   entries, &re_info);
3344                record_df_conflict_files(o, entries);
3345                if (clean < 0)
3346                        goto cleanup;
3347                for (i = entries->nr-1; 0 <= i; i--) {
3348                        const char *path = entries->items[i].string;
3349                        struct stage_data *e = entries->items[i].util;
3350                        if (!e->processed) {
3351                                int ret = process_entry(o, path, e);
3352                                if (!ret)
3353                                        clean = 0;
3354                                else if (ret < 0) {
3355                                        clean = ret;
3356                                        goto cleanup;
3357                                }
3358                        }
3359                }
3360                for (i = 0; i < entries->nr; i++) {
3361                        struct stage_data *e = entries->items[i].util;
3362                        if (!e->processed)
3363                                BUG("unprocessed path??? %s",
3364                                    entries->items[i].string);
3365                }
3366
3367        cleanup:
3368                final_cleanup_renames(&re_info);
3369
3370                string_list_clear(entries, 1);
3371                free(entries);
3372
3373                hashmap_free(&o->current_file_dir_set, 1);
3374
3375                if (clean < 0) {
3376                        unpack_trees_finish(o);
3377                        return clean;
3378                }
3379        }
3380        else
3381                clean = 1;
3382
3383        unpack_trees_finish(o);
3384
3385        if (o->call_depth && !(*result = write_tree_from_memory(o)))
3386                return -1;
3387
3388        return clean;
3389}
3390
3391static struct commit_list *reverse_commit_list(struct commit_list *list)
3392{
3393        struct commit_list *next = NULL, *current, *backup;
3394        for (current = list; current; current = backup) {
3395                backup = current->next;
3396                current->next = next;
3397                next = current;
3398        }
3399        return next;
3400}
3401
3402/*
3403 * Merge the commits h1 and h2, return the resulting virtual
3404 * commit object and a flag indicating the cleanness of the merge.
3405 */
3406int merge_recursive(struct merge_options *o,
3407                    struct commit *h1,
3408                    struct commit *h2,
3409                    struct commit_list *ca,
3410                    struct commit **result)
3411{
3412        struct commit_list *iter;
3413        struct commit *merged_common_ancestors;
3414        struct tree *mrtree;
3415        int clean;
3416
3417        if (show(o, 4)) {
3418                output(o, 4, _("Merging:"));
3419                output_commit_title(o, h1);
3420                output_commit_title(o, h2);
3421        }
3422
3423        if (!ca) {
3424                ca = get_merge_bases(h1, h2);
3425                ca = reverse_commit_list(ca);
3426        }
3427
3428        if (show(o, 5)) {
3429                unsigned cnt = commit_list_count(ca);
3430
3431                output(o, 5, Q_("found %u common ancestor:",
3432                                "found %u common ancestors:", cnt), cnt);
3433                for (iter = ca; iter; iter = iter->next)
3434                        output_commit_title(o, iter->item);
3435        }
3436
3437        merged_common_ancestors = pop_commit(&ca);
3438        if (merged_common_ancestors == NULL) {
3439                /* if there is no common ancestor, use an empty tree */
3440                struct tree *tree;
3441
3442                tree = lookup_tree(the_repository, the_repository->hash_algo->empty_tree);
3443                merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3444        }
3445
3446        for (iter = ca; iter; iter = iter->next) {
3447                const char *saved_b1, *saved_b2;
3448                o->call_depth++;
3449                /*
3450                 * When the merge fails, the result contains files
3451                 * with conflict markers. The cleanness flag is
3452                 * ignored (unless indicating an error), it was never
3453                 * actually used, as result of merge_trees has always
3454                 * overwritten it: the committed "conflicts" were
3455                 * already resolved.
3456                 */
3457                discard_cache();
3458                saved_b1 = o->branch1;
3459                saved_b2 = o->branch2;
3460                o->branch1 = "Temporary merge branch 1";
3461                o->branch2 = "Temporary merge branch 2";
3462                if (merge_recursive(o, merged_common_ancestors, iter->item,
3463                                    NULL, &merged_common_ancestors) < 0)
3464                        return -1;
3465                o->branch1 = saved_b1;
3466                o->branch2 = saved_b2;
3467                o->call_depth--;
3468
3469                if (!merged_common_ancestors)
3470                        return err(o, _("merge returned no commit"));
3471        }
3472
3473        discard_cache();
3474        if (!o->call_depth)
3475                read_cache();
3476
3477        o->ancestor = "merged common ancestors";
3478        clean = merge_trees(o, get_commit_tree(h1), get_commit_tree(h2),
3479                            get_commit_tree(merged_common_ancestors),
3480                            &mrtree);
3481        if (clean < 0) {
3482                flush_output(o);
3483                return clean;
3484        }
3485
3486        if (o->call_depth) {
3487                *result = make_virtual_commit(mrtree, "merged tree");
3488                commit_list_insert(h1, &(*result)->parents);
3489                commit_list_insert(h2, &(*result)->parents->next);
3490        }
3491        flush_output(o);
3492        if (!o->call_depth && o->buffer_output < 2)
3493                strbuf_release(&o->obuf);
3494        if (show(o, 2))
3495                diff_warn_rename_limit("merge.renamelimit",
3496                                       o->needed_rename_limit, 0);
3497        return clean;
3498}
3499
3500static struct commit *get_ref(const struct object_id *oid, const char *name)
3501{
3502        struct object *object;
3503
3504        object = deref_tag(the_repository, parse_object(the_repository, oid),
3505                           name,
3506                           strlen(name));
3507        if (!object)
3508                return NULL;
3509        if (object->type == OBJ_TREE)
3510                return make_virtual_commit((struct tree*)object, name);
3511        if (object->type != OBJ_COMMIT)
3512                return NULL;
3513        if (parse_commit((struct commit *)object))
3514                return NULL;
3515        return (struct commit *)object;
3516}
3517
3518int merge_recursive_generic(struct merge_options *o,
3519                            const struct object_id *head,
3520                            const struct object_id *merge,
3521                            int num_base_list,
3522                            const struct object_id **base_list,
3523                            struct commit **result)
3524{
3525        int clean;
3526        struct lock_file lock = LOCK_INIT;
3527        struct commit *head_commit = get_ref(head, o->branch1);
3528        struct commit *next_commit = get_ref(merge, o->branch2);
3529        struct commit_list *ca = NULL;
3530
3531        if (base_list) {
3532                int i;
3533                for (i = 0; i < num_base_list; ++i) {
3534                        struct commit *base;
3535                        if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3536                                return err(o, _("Could not parse object '%s'"),
3537                                           oid_to_hex(base_list[i]));
3538                        commit_list_insert(base, &ca);
3539                }
3540        }
3541
3542        hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3543        clean = merge_recursive(o, head_commit, next_commit, ca,
3544                                result);
3545        if (clean < 0) {
3546                rollback_lock_file(&lock);
3547                return clean;
3548        }
3549
3550        if (write_locked_index(&the_index, &lock,
3551                               COMMIT_LOCK | SKIP_IF_UNCHANGED))
3552                return err(o, _("Unable to write index."));
3553
3554        return clean ? 0 : 1;
3555}
3556
3557static void merge_recursive_config(struct merge_options *o)
3558{
3559        char *value = NULL;
3560        git_config_get_int("merge.verbosity", &o->verbosity);
3561        git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3562        git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3563        if (!git_config_get_string("diff.renames", &value)) {
3564                o->diff_detect_rename = git_config_rename("diff.renames", value);
3565                free(value);
3566        }
3567        if (!git_config_get_string("merge.renames", &value)) {
3568                o->merge_detect_rename = git_config_rename("merge.renames", value);
3569                free(value);
3570        }
3571        git_config(git_xmerge_config, NULL);
3572}
3573
3574void init_merge_options(struct merge_options *o)
3575{
3576        const char *merge_verbosity;
3577        memset(o, 0, sizeof(struct merge_options));
3578        o->verbosity = 2;
3579        o->buffer_output = 1;
3580        o->diff_rename_limit = -1;
3581        o->merge_rename_limit = -1;
3582        o->renormalize = 0;
3583        o->diff_detect_rename = -1;
3584        o->merge_detect_rename = -1;
3585        o->detect_directory_renames = 1;
3586        merge_recursive_config(o);
3587        merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3588        if (merge_verbosity)
3589                o->verbosity = strtol(merge_verbosity, NULL, 10);
3590        if (o->verbosity >= 5)
3591                o->buffer_output = 0;
3592        strbuf_init(&o->obuf, 0);
3593        string_list_init(&o->df_conflict_file_set, 1);
3594}
3595
3596int parse_merge_opt(struct merge_options *o, const char *s)
3597{
3598        const char *arg;
3599
3600        if (!s || !*s)
3601                return -1;
3602        if (!strcmp(s, "ours"))
3603                o->recursive_variant = MERGE_RECURSIVE_OURS;
3604        else if (!strcmp(s, "theirs"))
3605                o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3606        else if (!strcmp(s, "subtree"))
3607                o->subtree_shift = "";
3608        else if (skip_prefix(s, "subtree=", &arg))
3609                o->subtree_shift = arg;
3610        else if (!strcmp(s, "patience"))
3611                o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3612        else if (!strcmp(s, "histogram"))
3613                o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3614        else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3615                long value = parse_algorithm_value(arg);
3616                if (value < 0)
3617                        return -1;
3618                /* clear out previous settings */
3619                DIFF_XDL_CLR(o, NEED_MINIMAL);
3620                o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3621                o->xdl_opts |= value;
3622        }
3623        else if (!strcmp(s, "ignore-space-change"))
3624                DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3625        else if (!strcmp(s, "ignore-all-space"))
3626                DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3627        else if (!strcmp(s, "ignore-space-at-eol"))
3628                DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3629        else if (!strcmp(s, "ignore-cr-at-eol"))
3630                DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3631        else if (!strcmp(s, "renormalize"))
3632                o->renormalize = 1;
3633        else if (!strcmp(s, "no-renormalize"))
3634                o->renormalize = 0;
3635        else if (!strcmp(s, "no-renames"))
3636                o->merge_detect_rename = 0;
3637        else if (!strcmp(s, "find-renames")) {
3638                o->merge_detect_rename = 1;
3639                o->rename_score = 0;
3640        }
3641        else if (skip_prefix(s, "find-renames=", &arg) ||
3642                 skip_prefix(s, "rename-threshold=", &arg)) {
3643                if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3644                        return -1;
3645                o->merge_detect_rename = 1;
3646        }
3647        else
3648                return -1;
3649        return 0;
3650}