99a7ac5ec7953e411cf70adc5a89c31f1d4d0213
   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 = xcalloc(1, sizeof(struct rename_conflict_info));
 232        ci->rename_type = rename_type;
 233        ci->pair1 = pair1;
 234        ci->branch1 = branch1;
 235        ci->branch2 = branch2;
 236
 237        ci->dst_entry1 = dst_entry1;
 238        dst_entry1->rename_conflict_info = ci;
 239        dst_entry1->processed = 0;
 240
 241        assert(!pair2 == !dst_entry2);
 242        if (dst_entry2) {
 243                ci->dst_entry2 = dst_entry2;
 244                ci->pair2 = pair2;
 245                dst_entry2->rename_conflict_info = ci;
 246        }
 247
 248        if (rename_type == RENAME_TWO_FILES_TO_ONE) {
 249                /*
 250                 * For each rename, there could have been
 251                 * modifications on the side of history where that
 252                 * file was not renamed.
 253                 */
 254                int ostage1 = o->branch1 == branch1 ? 3 : 2;
 255                int ostage2 = ostage1 ^ 1;
 256
 257                ci->ren1_other.path = pair1->one->path;
 258                oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid);
 259                ci->ren1_other.mode = src_entry1->stages[ostage1].mode;
 260
 261                ci->ren2_other.path = pair2->one->path;
 262                oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid);
 263                ci->ren2_other.mode = src_entry2->stages[ostage2].mode;
 264        }
 265}
 266
 267static int show(struct merge_options *o, int v)
 268{
 269        return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5;
 270}
 271
 272__attribute__((format (printf, 3, 4)))
 273static void output(struct merge_options *o, int v, const char *fmt, ...)
 274{
 275        va_list ap;
 276
 277        if (!show(o, v))
 278                return;
 279
 280        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 281
 282        va_start(ap, fmt);
 283        strbuf_vaddf(&o->obuf, fmt, ap);
 284        va_end(ap);
 285
 286        strbuf_addch(&o->obuf, '\n');
 287        if (!o->buffer_output)
 288                flush_output(o);
 289}
 290
 291static void output_commit_title(struct merge_options *o, struct commit *commit)
 292{
 293        struct merge_remote_desc *desc;
 294
 295        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 296        desc = merge_remote_util(commit);
 297        if (desc)
 298                strbuf_addf(&o->obuf, "virtual %s\n", desc->name);
 299        else {
 300                strbuf_add_unique_abbrev(&o->obuf, &commit->object.oid,
 301                                         DEFAULT_ABBREV);
 302                strbuf_addch(&o->obuf, ' ');
 303                if (parse_commit(commit) != 0)
 304                        strbuf_addstr(&o->obuf, _("(bad commit)\n"));
 305                else {
 306                        const char *title;
 307                        const char *msg = get_commit_buffer(commit, NULL);
 308                        int len = find_commit_subject(msg, &title);
 309                        if (len)
 310                                strbuf_addf(&o->obuf, "%.*s\n", len, title);
 311                        unuse_commit_buffer(commit, msg);
 312                }
 313        }
 314        flush_output(o);
 315}
 316
 317static int add_cacheinfo(struct merge_options *o,
 318                         unsigned int mode, const struct object_id *oid,
 319                         const char *path, int stage, int refresh, int options)
 320{
 321        struct cache_entry *ce;
 322        int ret;
 323
 324        ce = make_cache_entry(&the_index, mode, oid ? oid : &null_oid, path, stage, 0);
 325        if (!ce)
 326                return err(o, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
 327
 328        ret = add_cache_entry(ce, options);
 329        if (refresh) {
 330                struct cache_entry *nce;
 331
 332                nce = refresh_cache_entry(&the_index, ce, CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
 333                if (!nce)
 334                        return err(o, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
 335                if (nce != ce)
 336                        ret = add_cache_entry(nce, options);
 337        }
 338        return ret;
 339}
 340
 341static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
 342{
 343        parse_tree(tree);
 344        init_tree_desc(desc, tree->buffer, tree->size);
 345}
 346
 347static int unpack_trees_start(struct merge_options *o,
 348                              struct tree *common,
 349                              struct tree *head,
 350                              struct tree *merge)
 351{
 352        int rc;
 353        struct tree_desc t[3];
 354        struct index_state tmp_index = { NULL };
 355
 356        memset(&o->unpack_opts, 0, sizeof(o->unpack_opts));
 357        if (o->call_depth)
 358                o->unpack_opts.index_only = 1;
 359        else
 360                o->unpack_opts.update = 1;
 361        o->unpack_opts.merge = 1;
 362        o->unpack_opts.head_idx = 2;
 363        o->unpack_opts.fn = threeway_merge;
 364        o->unpack_opts.src_index = &the_index;
 365        o->unpack_opts.dst_index = &tmp_index;
 366        o->unpack_opts.aggressive = !merge_detect_rename(o);
 367        setup_unpack_trees_porcelain(&o->unpack_opts, "merge");
 368
 369        init_tree_desc_from_tree(t+0, common);
 370        init_tree_desc_from_tree(t+1, head);
 371        init_tree_desc_from_tree(t+2, merge);
 372
 373        rc = unpack_trees(3, t, &o->unpack_opts);
 374        cache_tree_free(&active_cache_tree);
 375
 376        /*
 377         * Update the_index to match the new results, AFTER saving a copy
 378         * in o->orig_index.  Update src_index to point to the saved copy.
 379         * (verify_uptodate() checks src_index, and the original index is
 380         * the one that had the necessary modification timestamps.)
 381         */
 382        o->orig_index = the_index;
 383        the_index = tmp_index;
 384        o->unpack_opts.src_index = &o->orig_index;
 385
 386        return rc;
 387}
 388
 389static void unpack_trees_finish(struct merge_options *o)
 390{
 391        discard_index(&o->orig_index);
 392        clear_unpack_trees_porcelain(&o->unpack_opts);
 393}
 394
 395struct tree *write_tree_from_memory(struct merge_options *o)
 396{
 397        struct tree *result = NULL;
 398
 399        if (unmerged_cache()) {
 400                int i;
 401                fprintf(stderr, "BUG: There are unmerged index entries:\n");
 402                for (i = 0; i < active_nr; i++) {
 403                        const struct cache_entry *ce = active_cache[i];
 404                        if (ce_stage(ce))
 405                                fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
 406                                        (int)ce_namelen(ce), ce->name);
 407                }
 408                BUG("unmerged index entries in merge-recursive.c");
 409        }
 410
 411        if (!active_cache_tree)
 412                active_cache_tree = cache_tree();
 413
 414        if (!cache_tree_fully_valid(active_cache_tree) &&
 415            cache_tree_update(&the_index, 0) < 0) {
 416                err(o, _("error building trees"));
 417                return NULL;
 418        }
 419
 420        result = lookup_tree(the_repository, &active_cache_tree->oid);
 421
 422        return result;
 423}
 424
 425static int save_files_dirs(const struct object_id *oid,
 426                           struct strbuf *base, const char *path,
 427                           unsigned int mode, int stage, void *context)
 428{
 429        struct path_hashmap_entry *entry;
 430        int baselen = base->len;
 431        struct merge_options *o = context;
 432
 433        strbuf_addstr(base, path);
 434
 435        FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
 436        hashmap_entry_init(entry, path_hash(entry->path));
 437        hashmap_add(&o->current_file_dir_set, entry);
 438
 439        strbuf_setlen(base, baselen);
 440        return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
 441}
 442
 443static void get_files_dirs(struct merge_options *o, struct tree *tree)
 444{
 445        struct pathspec match_all;
 446        memset(&match_all, 0, sizeof(match_all));
 447        read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
 448}
 449
 450static int get_tree_entry_if_blob(const struct object_id *tree,
 451                                  const char *path,
 452                                  struct object_id *hashy,
 453                                  unsigned int *mode_o)
 454{
 455        int ret;
 456
 457        ret = get_tree_entry(tree, path, hashy, mode_o);
 458        if (S_ISDIR(*mode_o)) {
 459                oidcpy(hashy, &null_oid);
 460                *mode_o = 0;
 461        }
 462        return ret;
 463}
 464
 465/*
 466 * Returns an index_entry instance which doesn't have to correspond to
 467 * a real cache entry in Git's index.
 468 */
 469static struct stage_data *insert_stage_data(const char *path,
 470                struct tree *o, struct tree *a, struct tree *b,
 471                struct string_list *entries)
 472{
 473        struct string_list_item *item;
 474        struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
 475        get_tree_entry_if_blob(&o->object.oid, path,
 476                               &e->stages[1].oid, &e->stages[1].mode);
 477        get_tree_entry_if_blob(&a->object.oid, path,
 478                               &e->stages[2].oid, &e->stages[2].mode);
 479        get_tree_entry_if_blob(&b->object.oid, path,
 480                               &e->stages[3].oid, &e->stages[3].mode);
 481        item = string_list_insert(entries, path);
 482        item->util = e;
 483        return e;
 484}
 485
 486/*
 487 * Create a dictionary mapping file names to stage_data objects. The
 488 * dictionary contains one entry for every path with a non-zero stage entry.
 489 */
 490static struct string_list *get_unmerged(void)
 491{
 492        struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
 493        int i;
 494
 495        unmerged->strdup_strings = 1;
 496
 497        for (i = 0; i < active_nr; i++) {
 498                struct string_list_item *item;
 499                struct stage_data *e;
 500                const struct cache_entry *ce = active_cache[i];
 501                if (!ce_stage(ce))
 502                        continue;
 503
 504                item = string_list_lookup(unmerged, ce->name);
 505                if (!item) {
 506                        item = string_list_insert(unmerged, ce->name);
 507                        item->util = xcalloc(1, sizeof(struct stage_data));
 508                }
 509                e = item->util;
 510                e->stages[ce_stage(ce)].mode = ce->ce_mode;
 511                oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
 512        }
 513
 514        return unmerged;
 515}
 516
 517static int string_list_df_name_compare(const char *one, const char *two)
 518{
 519        int onelen = strlen(one);
 520        int twolen = strlen(two);
 521        /*
 522         * Here we only care that entries for D/F conflicts are
 523         * adjacent, in particular with the file of the D/F conflict
 524         * appearing before files below the corresponding directory.
 525         * The order of the rest of the list is irrelevant for us.
 526         *
 527         * To achieve this, we sort with df_name_compare and provide
 528         * the mode S_IFDIR so that D/F conflicts will sort correctly.
 529         * We use the mode S_IFDIR for everything else for simplicity,
 530         * since in other cases any changes in their order due to
 531         * sorting cause no problems for us.
 532         */
 533        int cmp = df_name_compare(one, onelen, S_IFDIR,
 534                                  two, twolen, S_IFDIR);
 535        /*
 536         * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
 537         * that 'foo' comes before 'foo/bar'.
 538         */
 539        if (cmp)
 540                return cmp;
 541        return onelen - twolen;
 542}
 543
 544static void record_df_conflict_files(struct merge_options *o,
 545                                     struct string_list *entries)
 546{
 547        /* If there is a D/F conflict and the file for such a conflict
 548         * currently exists in the working tree, we want to allow it to be
 549         * removed to make room for the corresponding directory if needed.
 550         * The files underneath the directories of such D/F conflicts will
 551         * be processed before the corresponding file involved in the D/F
 552         * conflict.  If the D/F directory ends up being removed by the
 553         * merge, then we won't have to touch the D/F file.  If the D/F
 554         * directory needs to be written to the working copy, then the D/F
 555         * file will simply be removed (in make_room_for_path()) to make
 556         * room for the necessary paths.  Note that if both the directory
 557         * and the file need to be present, then the D/F file will be
 558         * reinstated with a new unique name at the time it is processed.
 559         */
 560        struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
 561        const char *last_file = NULL;
 562        int last_len = 0;
 563        int i;
 564
 565        /*
 566         * If we're merging merge-bases, we don't want to bother with
 567         * any working directory changes.
 568         */
 569        if (o->call_depth)
 570                return;
 571
 572        /* Ensure D/F conflicts are adjacent in the entries list. */
 573        for (i = 0; i < entries->nr; i++) {
 574                struct string_list_item *next = &entries->items[i];
 575                string_list_append(&df_sorted_entries, next->string)->util =
 576                                   next->util;
 577        }
 578        df_sorted_entries.cmp = string_list_df_name_compare;
 579        string_list_sort(&df_sorted_entries);
 580
 581        string_list_clear(&o->df_conflict_file_set, 1);
 582        for (i = 0; i < df_sorted_entries.nr; i++) {
 583                const char *path = df_sorted_entries.items[i].string;
 584                int len = strlen(path);
 585                struct stage_data *e = df_sorted_entries.items[i].util;
 586
 587                /*
 588                 * Check if last_file & path correspond to a D/F conflict;
 589                 * i.e. whether path is last_file+'/'+<something>.
 590                 * If so, record that it's okay to remove last_file to make
 591                 * room for path and friends if needed.
 592                 */
 593                if (last_file &&
 594                    len > last_len &&
 595                    memcmp(path, last_file, last_len) == 0 &&
 596                    path[last_len] == '/') {
 597                        string_list_insert(&o->df_conflict_file_set, last_file);
 598                }
 599
 600                /*
 601                 * Determine whether path could exist as a file in the
 602                 * working directory as a possible D/F conflict.  This
 603                 * will only occur when it exists in stage 2 as a
 604                 * file.
 605                 */
 606                if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
 607                        last_file = path;
 608                        last_len = len;
 609                } else {
 610                        last_file = NULL;
 611                }
 612        }
 613        string_list_clear(&df_sorted_entries, 0);
 614}
 615
 616struct rename {
 617        struct diff_filepair *pair;
 618        /*
 619         * Purpose of src_entry and dst_entry:
 620         *
 621         * If 'before' is renamed to 'after' then src_entry will contain
 622         * the versions of 'before' from the merge_base, HEAD, and MERGE in
 623         * stages 1, 2, and 3; dst_entry will contain the respective
 624         * versions of 'after' in corresponding locations.  Thus, we have a
 625         * total of six modes and oids, though some will be null.  (Stage 0
 626         * is ignored; we're interested in handling conflicts.)
 627         *
 628         * Since we don't turn on break-rewrites by default, neither
 629         * src_entry nor dst_entry can have all three of their stages have
 630         * non-null oids, meaning at most four of the six will be non-null.
 631         * Also, since this is a rename, both src_entry and dst_entry will
 632         * have at least one non-null oid, meaning at least two will be
 633         * non-null.  Of the six oids, a typical rename will have three be
 634         * non-null.  Only two implies a rename/delete, and four implies a
 635         * rename/add.
 636         */
 637        struct stage_data *src_entry;
 638        struct stage_data *dst_entry;
 639        unsigned add_turned_into_rename:1;
 640        unsigned processed:1;
 641};
 642
 643static int update_stages(struct merge_options *opt, const char *path,
 644                         const struct diff_filespec *o,
 645                         const struct diff_filespec *a,
 646                         const struct diff_filespec *b)
 647{
 648
 649        /*
 650         * NOTE: It is usually a bad idea to call update_stages on a path
 651         * before calling update_file on that same path, since it can
 652         * sometimes lead to spurious "refusing to lose untracked file..."
 653         * messages from update_file (via make_room_for path via
 654         * would_lose_untracked).  Instead, reverse the order of the calls
 655         * (executing update_file first and then update_stages).
 656         */
 657        int clear = 1;
 658        int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
 659        if (clear)
 660                if (remove_file_from_cache(path))
 661                        return -1;
 662        if (o)
 663                if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
 664                        return -1;
 665        if (a)
 666                if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
 667                        return -1;
 668        if (b)
 669                if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
 670                        return -1;
 671        return 0;
 672}
 673
 674static int update_stages_for_stage_data(struct merge_options *opt,
 675                                        const char *path,
 676                                        const struct stage_data *stage_data)
 677{
 678        struct diff_filespec o, a, b;
 679
 680        o.mode = stage_data->stages[1].mode;
 681        oidcpy(&o.oid, &stage_data->stages[1].oid);
 682
 683        a.mode = stage_data->stages[2].mode;
 684        oidcpy(&a.oid, &stage_data->stages[2].oid);
 685
 686        b.mode = stage_data->stages[3].mode;
 687        oidcpy(&b.oid, &stage_data->stages[3].oid);
 688
 689        return update_stages(opt, path,
 690                             is_null_oid(&o.oid) ? NULL : &o,
 691                             is_null_oid(&a.oid) ? NULL : &a,
 692                             is_null_oid(&b.oid) ? NULL : &b);
 693}
 694
 695static void update_entry(struct stage_data *entry,
 696                         struct diff_filespec *o,
 697                         struct diff_filespec *a,
 698                         struct diff_filespec *b)
 699{
 700        entry->processed = 0;
 701        entry->stages[1].mode = o->mode;
 702        entry->stages[2].mode = a->mode;
 703        entry->stages[3].mode = b->mode;
 704        oidcpy(&entry->stages[1].oid, &o->oid);
 705        oidcpy(&entry->stages[2].oid, &a->oid);
 706        oidcpy(&entry->stages[3].oid, &b->oid);
 707}
 708
 709static int remove_file(struct merge_options *o, int clean,
 710                       const char *path, int no_wd)
 711{
 712        int update_cache = o->call_depth || clean;
 713        int update_working_directory = !o->call_depth && !no_wd;
 714
 715        if (update_cache) {
 716                if (remove_file_from_cache(path))
 717                        return -1;
 718        }
 719        if (update_working_directory) {
 720                if (ignore_case) {
 721                        struct cache_entry *ce;
 722                        ce = cache_file_exists(path, strlen(path), ignore_case);
 723                        if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
 724                                return 0;
 725                }
 726                if (remove_path(path))
 727                        return -1;
 728        }
 729        return 0;
 730}
 731
 732/* add a string to a strbuf, but converting "/" to "_" */
 733static void add_flattened_path(struct strbuf *out, const char *s)
 734{
 735        size_t i = out->len;
 736        strbuf_addstr(out, s);
 737        for (; i < out->len; i++)
 738                if (out->buf[i] == '/')
 739                        out->buf[i] = '_';
 740}
 741
 742static char *unique_path(struct merge_options *o, const char *path, const char *branch)
 743{
 744        struct path_hashmap_entry *entry;
 745        struct strbuf newpath = STRBUF_INIT;
 746        int suffix = 0;
 747        size_t base_len;
 748
 749        strbuf_addf(&newpath, "%s~", path);
 750        add_flattened_path(&newpath, branch);
 751
 752        base_len = newpath.len;
 753        while (hashmap_get_from_hash(&o->current_file_dir_set,
 754                                     path_hash(newpath.buf), newpath.buf) ||
 755               (!o->call_depth && file_exists(newpath.buf))) {
 756                strbuf_setlen(&newpath, base_len);
 757                strbuf_addf(&newpath, "_%d", suffix++);
 758        }
 759
 760        FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
 761        hashmap_entry_init(entry, path_hash(entry->path));
 762        hashmap_add(&o->current_file_dir_set, entry);
 763        return strbuf_detach(&newpath, NULL);
 764}
 765
 766/**
 767 * Check whether a directory in the index is in the way of an incoming
 768 * file.  Return 1 if so.  If check_working_copy is non-zero, also
 769 * check the working directory.  If empty_ok is non-zero, also return
 770 * 0 in the case where the working-tree dir exists but is empty.
 771 */
 772static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
 773{
 774        int pos;
 775        struct strbuf dirpath = STRBUF_INIT;
 776        struct stat st;
 777
 778        strbuf_addstr(&dirpath, path);
 779        strbuf_addch(&dirpath, '/');
 780
 781        pos = cache_name_pos(dirpath.buf, dirpath.len);
 782
 783        if (pos < 0)
 784                pos = -1 - pos;
 785        if (pos < active_nr &&
 786            !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
 787                strbuf_release(&dirpath);
 788                return 1;
 789        }
 790
 791        strbuf_release(&dirpath);
 792        return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
 793                !(empty_ok && is_empty_dir(path));
 794}
 795
 796/*
 797 * Returns whether path was tracked in the index before the merge started,
 798 * and its oid and mode match the specified values
 799 */
 800static int was_tracked_and_matches(struct merge_options *o, const char *path,
 801                                   const struct object_id *oid, unsigned mode)
 802{
 803        int pos = index_name_pos(&o->orig_index, path, strlen(path));
 804        struct cache_entry *ce;
 805
 806        if (0 > pos)
 807                /* we were not tracking this path before the merge */
 808                return 0;
 809
 810        /* See if the file we were tracking before matches */
 811        ce = o->orig_index.cache[pos];
 812        return (oid_eq(&ce->oid, oid) && ce->ce_mode == mode);
 813}
 814
 815/*
 816 * Returns whether path was tracked in the index before the merge started
 817 */
 818static int was_tracked(struct merge_options *o, const char *path)
 819{
 820        int pos = index_name_pos(&o->orig_index, path, strlen(path));
 821
 822        if (0 <= pos)
 823                /* we were tracking this path before the merge */
 824                return 1;
 825
 826        return 0;
 827}
 828
 829static int would_lose_untracked(const char *path)
 830{
 831        /*
 832         * This may look like it can be simplified to:
 833         *   return !was_tracked(o, path) && file_exists(path)
 834         * but it can't.  This function needs to know whether path was in
 835         * the working tree due to EITHER having been tracked in the index
 836         * before the merge OR having been put into the working copy and
 837         * index by unpack_trees().  Due to that either-or requirement, we
 838         * check the current index instead of the original one.
 839         *
 840         * Note that we do not need to worry about merge-recursive itself
 841         * updating the index after unpack_trees() and before calling this
 842         * function, because we strictly require all code paths in
 843         * merge-recursive to update the working tree first and the index
 844         * second.  Doing otherwise would break
 845         * update_file()/would_lose_untracked(); see every comment in this
 846         * file which mentions "update_stages".
 847         */
 848        int pos = cache_name_pos(path, strlen(path));
 849
 850        if (pos < 0)
 851                pos = -1 - pos;
 852        while (pos < active_nr &&
 853               !strcmp(path, active_cache[pos]->name)) {
 854                /*
 855                 * If stage #0, it is definitely tracked.
 856                 * If it has stage #2 then it was tracked
 857                 * before this merge started.  All other
 858                 * cases the path was not tracked.
 859                 */
 860                switch (ce_stage(active_cache[pos])) {
 861                case 0:
 862                case 2:
 863                        return 0;
 864                }
 865                pos++;
 866        }
 867        return file_exists(path);
 868}
 869
 870static int was_dirty(struct merge_options *o, const char *path)
 871{
 872        struct cache_entry *ce;
 873        int dirty = 1;
 874
 875        if (o->call_depth || !was_tracked(o, path))
 876                return !dirty;
 877
 878        ce = index_file_exists(o->unpack_opts.src_index,
 879                               path, strlen(path), ignore_case);
 880        dirty = verify_uptodate(ce, &o->unpack_opts) != 0;
 881        return dirty;
 882}
 883
 884static int make_room_for_path(struct merge_options *o, const char *path)
 885{
 886        int status, i;
 887        const char *msg = _("failed to create path '%s'%s");
 888
 889        /* Unlink any D/F conflict files that are in the way */
 890        for (i = 0; i < o->df_conflict_file_set.nr; i++) {
 891                const char *df_path = o->df_conflict_file_set.items[i].string;
 892                size_t pathlen = strlen(path);
 893                size_t df_pathlen = strlen(df_path);
 894                if (df_pathlen < pathlen &&
 895                    path[df_pathlen] == '/' &&
 896                    strncmp(path, df_path, df_pathlen) == 0) {
 897                        output(o, 3,
 898                               _("Removing %s to make room for subdirectory\n"),
 899                               df_path);
 900                        unlink(df_path);
 901                        unsorted_string_list_delete_item(&o->df_conflict_file_set,
 902                                                         i, 0);
 903                        break;
 904                }
 905        }
 906
 907        /* Make sure leading directories are created */
 908        status = safe_create_leading_directories_const(path);
 909        if (status) {
 910                if (status == SCLD_EXISTS)
 911                        /* something else exists */
 912                        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 913                return err(o, msg, path, "");
 914        }
 915
 916        /*
 917         * Do not unlink a file in the work tree if we are not
 918         * tracking it.
 919         */
 920        if (would_lose_untracked(path))
 921                return err(o, _("refusing to lose untracked file at '%s'"),
 922                           path);
 923
 924        /* Successful unlink is good.. */
 925        if (!unlink(path))
 926                return 0;
 927        /* .. and so is no existing file */
 928        if (errno == ENOENT)
 929                return 0;
 930        /* .. but not some other error (who really cares what?) */
 931        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 932}
 933
 934static int update_file_flags(struct merge_options *o,
 935                             const struct object_id *oid,
 936                             unsigned mode,
 937                             const char *path,
 938                             int update_cache,
 939                             int update_wd)
 940{
 941        int ret = 0;
 942
 943        if (o->call_depth)
 944                update_wd = 0;
 945
 946        if (update_wd) {
 947                enum object_type type;
 948                void *buf;
 949                unsigned long size;
 950
 951                if (S_ISGITLINK(mode)) {
 952                        /*
 953                         * We may later decide to recursively descend into
 954                         * the submodule directory and update its index
 955                         * and/or work tree, but we do not do that now.
 956                         */
 957                        update_wd = 0;
 958                        goto update_index;
 959                }
 960
 961                buf = read_object_file(oid, &type, &size);
 962                if (!buf)
 963                        return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
 964                if (type != OBJ_BLOB) {
 965                        ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
 966                        goto free_buf;
 967                }
 968                if (S_ISREG(mode)) {
 969                        struct strbuf strbuf = STRBUF_INIT;
 970                        if (convert_to_working_tree(&the_index, path, buf, size, &strbuf)) {
 971                                free(buf);
 972                                size = strbuf.len;
 973                                buf = strbuf_detach(&strbuf, NULL);
 974                        }
 975                }
 976
 977                if (make_room_for_path(o, path) < 0) {
 978                        update_wd = 0;
 979                        goto free_buf;
 980                }
 981                if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
 982                        int fd;
 983                        if (mode & 0100)
 984                                mode = 0777;
 985                        else
 986                                mode = 0666;
 987                        fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
 988                        if (fd < 0) {
 989                                ret = err(o, _("failed to open '%s': %s"),
 990                                          path, strerror(errno));
 991                                goto free_buf;
 992                        }
 993                        write_in_full(fd, buf, size);
 994                        close(fd);
 995                } else if (S_ISLNK(mode)) {
 996                        char *lnk = xmemdupz(buf, size);
 997                        safe_create_leading_directories_const(path);
 998                        unlink(path);
 999                        if (symlink(lnk, path))
1000                                ret = err(o, _("failed to symlink '%s': %s"),
1001                                          path, strerror(errno));
1002                        free(lnk);
1003                } else
1004                        ret = err(o,
1005                                  _("do not know what to do with %06o %s '%s'"),
1006                                  mode, oid_to_hex(oid), path);
1007        free_buf:
1008                free(buf);
1009        }
1010update_index:
1011        if (!ret && update_cache)
1012                if (add_cacheinfo(o, mode, oid, path, 0, update_wd,
1013                                  ADD_CACHE_OK_TO_ADD))
1014                        return -1;
1015        return ret;
1016}
1017
1018static int update_file(struct merge_options *o,
1019                       int clean,
1020                       const struct object_id *oid,
1021                       unsigned mode,
1022                       const char *path)
1023{
1024        return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
1025}
1026
1027/* Low level file merging, update and removal */
1028
1029struct merge_file_info {
1030        struct object_id oid;
1031        unsigned mode;
1032        unsigned clean:1,
1033                 merge:1;
1034};
1035
1036static int merge_3way(struct merge_options *o,
1037                      mmbuffer_t *result_buf,
1038                      const struct diff_filespec *one,
1039                      const struct diff_filespec *a,
1040                      const struct diff_filespec *b,
1041                      const char *branch1,
1042                      const char *branch2)
1043{
1044        mmfile_t orig, src1, src2;
1045        struct ll_merge_options ll_opts = {0};
1046        char *base_name, *name1, *name2;
1047        int merge_status;
1048
1049        ll_opts.renormalize = o->renormalize;
1050        ll_opts.xdl_opts = o->xdl_opts;
1051
1052        if (o->call_depth) {
1053                ll_opts.virtual_ancestor = 1;
1054                ll_opts.variant = 0;
1055        } else {
1056                switch (o->recursive_variant) {
1057                case MERGE_RECURSIVE_OURS:
1058                        ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1059                        break;
1060                case MERGE_RECURSIVE_THEIRS:
1061                        ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1062                        break;
1063                default:
1064                        ll_opts.variant = 0;
1065                        break;
1066                }
1067        }
1068
1069        if (strcmp(a->path, b->path) ||
1070            (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1071                base_name = o->ancestor == NULL ? NULL :
1072                        mkpathdup("%s:%s", o->ancestor, one->path);
1073                name1 = mkpathdup("%s:%s", branch1, a->path);
1074                name2 = mkpathdup("%s:%s", branch2, b->path);
1075        } else {
1076                base_name = o->ancestor == NULL ? NULL :
1077                        mkpathdup("%s", o->ancestor);
1078                name1 = mkpathdup("%s", branch1);
1079                name2 = mkpathdup("%s", branch2);
1080        }
1081
1082        read_mmblob(&orig, &one->oid);
1083        read_mmblob(&src1, &a->oid);
1084        read_mmblob(&src2, &b->oid);
1085
1086        merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1087                                &src1, name1, &src2, name2, &ll_opts);
1088
1089        free(base_name);
1090        free(name1);
1091        free(name2);
1092        free(orig.ptr);
1093        free(src1.ptr);
1094        free(src2.ptr);
1095        return merge_status;
1096}
1097
1098static int find_first_merges(struct object_array *result, const char *path,
1099                             struct commit *a, struct commit *b)
1100{
1101        int i, j;
1102        struct object_array merges = OBJECT_ARRAY_INIT;
1103        struct commit *commit;
1104        int contains_another;
1105
1106        char merged_revision[42];
1107        const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1108                                   "--all", merged_revision, NULL };
1109        struct rev_info revs;
1110        struct setup_revision_opt rev_opts;
1111
1112        memset(result, 0, sizeof(struct object_array));
1113        memset(&rev_opts, 0, sizeof(rev_opts));
1114
1115        /* get all revisions that merge commit a */
1116        xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1117                  oid_to_hex(&a->object.oid));
1118        init_revisions(&revs, NULL);
1119        rev_opts.submodule = path;
1120        /* FIXME: can't handle linked worktrees in submodules yet */
1121        revs.single_worktree = path != NULL;
1122        setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1123
1124        /* save all revisions from the above list that contain b */
1125        if (prepare_revision_walk(&revs))
1126                die("revision walk setup failed");
1127        while ((commit = get_revision(&revs)) != NULL) {
1128                struct object *o = &(commit->object);
1129                if (in_merge_bases(b, commit))
1130                        add_object_array(o, NULL, &merges);
1131        }
1132        reset_revision_walk();
1133
1134        /* Now we've got all merges that contain a and b. Prune all
1135         * merges that contain another found merge and save them in
1136         * result.
1137         */
1138        for (i = 0; i < merges.nr; i++) {
1139                struct commit *m1 = (struct commit *) merges.objects[i].item;
1140
1141                contains_another = 0;
1142                for (j = 0; j < merges.nr; j++) {
1143                        struct commit *m2 = (struct commit *) merges.objects[j].item;
1144                        if (i != j && in_merge_bases(m2, m1)) {
1145                                contains_another = 1;
1146                                break;
1147                        }
1148                }
1149
1150                if (!contains_another)
1151                        add_object_array(merges.objects[i].item, NULL, result);
1152        }
1153
1154        object_array_clear(&merges);
1155        return result->nr;
1156}
1157
1158static void print_commit(struct commit *commit)
1159{
1160        struct strbuf sb = STRBUF_INIT;
1161        struct pretty_print_context ctx = {0};
1162        ctx.date_mode.type = DATE_NORMAL;
1163        format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1164        fprintf(stderr, "%s\n", sb.buf);
1165        strbuf_release(&sb);
1166}
1167
1168static int merge_submodule(struct merge_options *o,
1169                           struct object_id *result, const char *path,
1170                           const struct object_id *base, const struct object_id *a,
1171                           const struct object_id *b)
1172{
1173        struct commit *commit_base, *commit_a, *commit_b;
1174        int parent_count;
1175        struct object_array merges;
1176
1177        int i;
1178        int search = !o->call_depth;
1179
1180        /* store a in result in case we fail */
1181        oidcpy(result, a);
1182
1183        /* we can not handle deletion conflicts */
1184        if (is_null_oid(base))
1185                return 0;
1186        if (is_null_oid(a))
1187                return 0;
1188        if (is_null_oid(b))
1189                return 0;
1190
1191        if (add_submodule_odb(path)) {
1192                output(o, 1, _("Failed to merge submodule %s (not checked out)"), path);
1193                return 0;
1194        }
1195
1196        if (!(commit_base = lookup_commit_reference(the_repository, base)) ||
1197            !(commit_a = lookup_commit_reference(the_repository, a)) ||
1198            !(commit_b = lookup_commit_reference(the_repository, b))) {
1199                output(o, 1, _("Failed to merge submodule %s (commits not present)"), path);
1200                return 0;
1201        }
1202
1203        /* check whether both changes are forward */
1204        if (!in_merge_bases(commit_base, commit_a) ||
1205            !in_merge_bases(commit_base, commit_b)) {
1206                output(o, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1207                return 0;
1208        }
1209
1210        /* Case #1: a is contained in b or vice versa */
1211        if (in_merge_bases(commit_a, commit_b)) {
1212                oidcpy(result, b);
1213                if (show(o, 3)) {
1214                        output(o, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1215                        output_commit_title(o, commit_b);
1216                } else if (show(o, 2))
1217                        output(o, 2, _("Fast-forwarding submodule %s"), path);
1218                else
1219                        ; /* no output */
1220
1221                return 1;
1222        }
1223        if (in_merge_bases(commit_b, commit_a)) {
1224                oidcpy(result, a);
1225                if (show(o, 3)) {
1226                        output(o, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1227                        output_commit_title(o, commit_a);
1228                } else if (show(o, 2))
1229                        output(o, 2, _("Fast-forwarding submodule %s"), path);
1230                else
1231                        ; /* no output */
1232
1233                return 1;
1234        }
1235
1236        /*
1237         * Case #2: There are one or more merges that contain a and b in
1238         * the submodule. If there is only one, then present it as a
1239         * suggestion to the user, but leave it marked unmerged so the
1240         * user needs to confirm the resolution.
1241         */
1242
1243        /* Skip the search if makes no sense to the calling context.  */
1244        if (!search)
1245                return 0;
1246
1247        /* find commit which merges them */
1248        parent_count = find_first_merges(&merges, path, commit_a, commit_b);
1249        switch (parent_count) {
1250        case 0:
1251                output(o, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1252                break;
1253
1254        case 1:
1255                output(o, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1256                output(o, 2, _("Found a possible merge resolution for the submodule:\n"));
1257                print_commit((struct commit *) merges.objects[0].item);
1258                output(o, 2, _(
1259                       "If this is correct simply add it to the index "
1260                       "for example\n"
1261                       "by using:\n\n"
1262                       "  git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1263                       "which will accept this suggestion.\n"),
1264                       oid_to_hex(&merges.objects[0].item->oid), path);
1265                break;
1266
1267        default:
1268                output(o, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1269                for (i = 0; i < merges.nr; i++)
1270                        print_commit((struct commit *) merges.objects[i].item);
1271        }
1272
1273        object_array_clear(&merges);
1274        return 0;
1275}
1276
1277static int merge_file_1(struct merge_options *o,
1278                        const struct diff_filespec *one,
1279                        const struct diff_filespec *a,
1280                        const struct diff_filespec *b,
1281                        const char *filename,
1282                        const char *branch1,
1283                        const char *branch2,
1284                        struct merge_file_info *result)
1285{
1286        result->merge = 0;
1287        result->clean = 1;
1288
1289        if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1290                result->clean = 0;
1291                if (S_ISREG(a->mode)) {
1292                        result->mode = a->mode;
1293                        oidcpy(&result->oid, &a->oid);
1294                } else {
1295                        result->mode = b->mode;
1296                        oidcpy(&result->oid, &b->oid);
1297                }
1298        } else {
1299                if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1300                        result->merge = 1;
1301
1302                /*
1303                 * Merge modes
1304                 */
1305                if (a->mode == b->mode || a->mode == one->mode)
1306                        result->mode = b->mode;
1307                else {
1308                        result->mode = a->mode;
1309                        if (b->mode != one->mode) {
1310                                result->clean = 0;
1311                                result->merge = 1;
1312                        }
1313                }
1314
1315                if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1316                        oidcpy(&result->oid, &b->oid);
1317                else if (oid_eq(&b->oid, &one->oid))
1318                        oidcpy(&result->oid, &a->oid);
1319                else if (S_ISREG(a->mode)) {
1320                        mmbuffer_t result_buf;
1321                        int ret = 0, merge_status;
1322
1323                        merge_status = merge_3way(o, &result_buf, one, a, b,
1324                                                  branch1, branch2);
1325
1326                        if ((merge_status < 0) || !result_buf.ptr)
1327                                ret = err(o, _("Failed to execute internal merge"));
1328
1329                        if (!ret &&
1330                            write_object_file(result_buf.ptr, result_buf.size,
1331                                              blob_type, &result->oid))
1332                                ret = err(o, _("Unable to add %s to database"),
1333                                          a->path);
1334
1335                        free(result_buf.ptr);
1336                        if (ret)
1337                                return ret;
1338                        result->clean = (merge_status == 0);
1339                } else if (S_ISGITLINK(a->mode)) {
1340                        result->clean = merge_submodule(o, &result->oid,
1341                                                        one->path,
1342                                                        &one->oid,
1343                                                        &a->oid,
1344                                                        &b->oid);
1345                } else if (S_ISLNK(a->mode)) {
1346                        switch (o->recursive_variant) {
1347                        case MERGE_RECURSIVE_NORMAL:
1348                                oidcpy(&result->oid, &a->oid);
1349                                if (!oid_eq(&a->oid, &b->oid))
1350                                        result->clean = 0;
1351                                break;
1352                        case MERGE_RECURSIVE_OURS:
1353                                oidcpy(&result->oid, &a->oid);
1354                                break;
1355                        case MERGE_RECURSIVE_THEIRS:
1356                                oidcpy(&result->oid, &b->oid);
1357                                break;
1358                        }
1359                } else
1360                        BUG("unsupported object type in the tree");
1361        }
1362
1363        if (result->merge)
1364                output(o, 2, _("Auto-merging %s"), filename);
1365
1366        return 0;
1367}
1368
1369static int merge_file_one(struct merge_options *o,
1370                          const char *path,
1371                          const struct object_id *o_oid, int o_mode,
1372                          const struct object_id *a_oid, int a_mode,
1373                          const struct object_id *b_oid, int b_mode,
1374                          const char *branch1,
1375                          const char *branch2,
1376                          struct merge_file_info *mfi)
1377{
1378        struct diff_filespec one, a, b;
1379
1380        one.path = a.path = b.path = (char *)path;
1381        oidcpy(&one.oid, o_oid);
1382        one.mode = o_mode;
1383        oidcpy(&a.oid, a_oid);
1384        a.mode = a_mode;
1385        oidcpy(&b.oid, b_oid);
1386        b.mode = b_mode;
1387        return merge_file_1(o, &one, &a, &b, path, branch1, branch2, mfi);
1388}
1389
1390static int handle_rename_via_dir(struct merge_options *o,
1391                                 struct diff_filepair *pair,
1392                                 const char *rename_branch,
1393                                 const char *other_branch)
1394{
1395        /*
1396         * Handle file adds that need to be renamed due to directory rename
1397         * detection.  This differs from handle_rename_normal, because
1398         * there is no content merge to do; just move the file into the
1399         * desired final location.
1400         */
1401        const struct diff_filespec *dest = pair->two;
1402
1403        if (!o->call_depth && would_lose_untracked(dest->path)) {
1404                char *alt_path = unique_path(o, dest->path, rename_branch);
1405
1406                output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1407                               "writing to %s instead."),
1408                       dest->path, alt_path);
1409                /*
1410                 * Write the file in worktree at alt_path, but not in the
1411                 * index.  Instead, write to dest->path for the index but
1412                 * only at the higher appropriate stage.
1413                 */
1414                if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1415                        return -1;
1416                free(alt_path);
1417                return update_stages(o, dest->path, NULL,
1418                                     rename_branch == o->branch1 ? dest : NULL,
1419                                     rename_branch == o->branch1 ? NULL : dest);
1420        }
1421
1422        /* Update dest->path both in index and in worktree */
1423        if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1424                return -1;
1425        return 0;
1426}
1427
1428static int handle_change_delete(struct merge_options *o,
1429                                const char *path, const char *old_path,
1430                                const struct object_id *o_oid, int o_mode,
1431                                const struct object_id *changed_oid,
1432                                int changed_mode,
1433                                const char *change_branch,
1434                                const char *delete_branch,
1435                                const char *change, const char *change_past)
1436{
1437        char *alt_path = NULL;
1438        const char *update_path = path;
1439        int ret = 0;
1440
1441        if (dir_in_way(path, !o->call_depth, 0) ||
1442            (!o->call_depth && would_lose_untracked(path))) {
1443                update_path = alt_path = unique_path(o, path, change_branch);
1444        }
1445
1446        if (o->call_depth) {
1447                /*
1448                 * We cannot arbitrarily accept either a_sha or b_sha as
1449                 * correct; since there is no true "middle point" between
1450                 * them, simply reuse the base version for virtual merge base.
1451                 */
1452                ret = remove_file_from_cache(path);
1453                if (!ret)
1454                        ret = update_file(o, 0, o_oid, o_mode, update_path);
1455        } else {
1456                /*
1457                 * Despite the four nearly duplicate messages and argument
1458                 * lists below and the ugliness of the nested if-statements,
1459                 * having complete messages makes the job easier for
1460                 * translators.
1461                 *
1462                 * The slight variance among the cases is due to the fact
1463                 * that:
1464                 *   1) directory/file conflicts (in effect if
1465                 *      !alt_path) could cause us to need to write the
1466                 *      file to a different path.
1467                 *   2) renames (in effect if !old_path) could mean that
1468                 *      there are two names for the path that the user
1469                 *      may know the file by.
1470                 */
1471                if (!alt_path) {
1472                        if (!old_path) {
1473                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1474                                       "and %s in %s. Version %s of %s left in tree."),
1475                                       change, path, delete_branch, change_past,
1476                                       change_branch, change_branch, path);
1477                        } else {
1478                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1479                                       "and %s to %s in %s. Version %s of %s left in tree."),
1480                                       change, old_path, delete_branch, change_past, path,
1481                                       change_branch, change_branch, path);
1482                        }
1483                } else {
1484                        if (!old_path) {
1485                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1486                                       "and %s in %s. Version %s of %s left in tree at %s."),
1487                                       change, path, delete_branch, change_past,
1488                                       change_branch, change_branch, path, alt_path);
1489                        } else {
1490                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1491                                       "and %s to %s in %s. Version %s of %s left in tree at %s."),
1492                                       change, old_path, delete_branch, change_past, path,
1493                                       change_branch, change_branch, path, alt_path);
1494                        }
1495                }
1496                /*
1497                 * No need to call update_file() on path when change_branch ==
1498                 * o->branch1 && !alt_path, since that would needlessly touch
1499                 * path.  We could call update_file_flags() with update_cache=0
1500                 * and update_wd=0, but that's a no-op.
1501                 */
1502                if (change_branch != o->branch1 || alt_path)
1503                        ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1504        }
1505        free(alt_path);
1506
1507        return ret;
1508}
1509
1510static int handle_rename_delete(struct merge_options *o,
1511                                struct diff_filepair *pair,
1512                                const char *rename_branch,
1513                                const char *delete_branch)
1514{
1515        const struct diff_filespec *orig = pair->one;
1516        const struct diff_filespec *dest = pair->two;
1517
1518        if (handle_change_delete(o,
1519                                 o->call_depth ? orig->path : dest->path,
1520                                 o->call_depth ? NULL : orig->path,
1521                                 &orig->oid, orig->mode,
1522                                 &dest->oid, dest->mode,
1523                                 rename_branch, delete_branch,
1524                                 _("rename"), _("renamed")))
1525                return -1;
1526
1527        if (o->call_depth)
1528                return remove_file_from_cache(dest->path);
1529        else
1530                return update_stages(o, dest->path, NULL,
1531                                     rename_branch == o->branch1 ? dest : NULL,
1532                                     rename_branch == o->branch1 ? NULL : dest);
1533}
1534
1535static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1536                                                 struct stage_data *entry,
1537                                                 int stage)
1538{
1539        struct object_id *oid = &entry->stages[stage].oid;
1540        unsigned mode = entry->stages[stage].mode;
1541        if (mode == 0 || is_null_oid(oid))
1542                return NULL;
1543        oidcpy(&target->oid, oid);
1544        target->mode = mode;
1545        return target;
1546}
1547
1548static int handle_file(struct merge_options *o,
1549                        struct diff_filespec *rename,
1550                        int stage,
1551                        struct rename_conflict_info *ci)
1552{
1553        char *dst_name = rename->path;
1554        struct stage_data *dst_entry;
1555        const char *cur_branch, *other_branch;
1556        struct diff_filespec other;
1557        struct diff_filespec *add;
1558        int ret;
1559
1560        if (stage == 2) {
1561                dst_entry = ci->dst_entry1;
1562                cur_branch = ci->branch1;
1563                other_branch = ci->branch2;
1564        } else {
1565                dst_entry = ci->dst_entry2;
1566                cur_branch = ci->branch2;
1567                other_branch = ci->branch1;
1568        }
1569
1570        add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1571        if (add) {
1572                int ren_src_was_dirty = was_dirty(o, rename->path);
1573                char *add_name = unique_path(o, rename->path, other_branch);
1574                if (update_file(o, 0, &add->oid, add->mode, add_name))
1575                        return -1;
1576
1577                if (ren_src_was_dirty) {
1578                        output(o, 1, _("Refusing to lose dirty file at %s"),
1579                               rename->path);
1580                }
1581                /*
1582                 * Because the double negatives somehow keep confusing me...
1583                 *    1) update_wd iff !ren_src_was_dirty.
1584                 *    2) no_wd iff !update_wd
1585                 *    3) so, no_wd == !!ren_src_was_dirty == ren_src_was_dirty
1586                 */
1587                remove_file(o, 0, rename->path, ren_src_was_dirty);
1588                dst_name = unique_path(o, rename->path, cur_branch);
1589        } else {
1590                if (dir_in_way(rename->path, !o->call_depth, 0)) {
1591                        dst_name = unique_path(o, rename->path, cur_branch);
1592                        output(o, 1, _("%s is a directory in %s adding as %s instead"),
1593                               rename->path, other_branch, dst_name);
1594                } else if (!o->call_depth &&
1595                           would_lose_untracked(rename->path)) {
1596                        dst_name = unique_path(o, rename->path, cur_branch);
1597                        output(o, 1, _("Refusing to lose untracked file at %s; "
1598                                       "adding as %s instead"),
1599                               rename->path, dst_name);
1600                }
1601        }
1602        if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1603                ; /* fall through, do allow dst_name to be released */
1604        else if (stage == 2)
1605                ret = update_stages(o, rename->path, NULL, rename, add);
1606        else
1607                ret = update_stages(o, rename->path, NULL, add, rename);
1608
1609        if (dst_name != rename->path)
1610                free(dst_name);
1611
1612        return ret;
1613}
1614
1615static int handle_rename_rename_1to2(struct merge_options *o,
1616                                     struct rename_conflict_info *ci)
1617{
1618        /* One file was renamed in both branches, but to different names. */
1619        struct diff_filespec *one = ci->pair1->one;
1620        struct diff_filespec *a = ci->pair1->two;
1621        struct diff_filespec *b = ci->pair2->two;
1622
1623        output(o, 1, _("CONFLICT (rename/rename): "
1624               "Rename \"%s\"->\"%s\" in branch \"%s\" "
1625               "rename \"%s\"->\"%s\" in \"%s\"%s"),
1626               one->path, a->path, ci->branch1,
1627               one->path, b->path, ci->branch2,
1628               o->call_depth ? _(" (left unresolved)") : "");
1629        if (o->call_depth) {
1630                struct merge_file_info mfi;
1631                struct diff_filespec other;
1632                struct diff_filespec *add;
1633                if (merge_file_one(o, one->path,
1634                                 &one->oid, one->mode,
1635                                 &a->oid, a->mode,
1636                                 &b->oid, b->mode,
1637                                 ci->branch1, ci->branch2, &mfi))
1638                        return -1;
1639
1640                /*
1641                 * FIXME: For rename/add-source conflicts (if we could detect
1642                 * such), this is wrong.  We should instead find a unique
1643                 * pathname and then either rename the add-source file to that
1644                 * unique path, or use that unique path instead of src here.
1645                 */
1646                if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1647                        return -1;
1648
1649                /*
1650                 * Above, we put the merged content at the merge-base's
1651                 * path.  Now we usually need to delete both a->path and
1652                 * b->path.  However, the rename on each side of the merge
1653                 * could also be involved in a rename/add conflict.  In
1654                 * such cases, we should keep the added file around,
1655                 * resolving the conflict at that path in its favor.
1656                 */
1657                add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1658                if (add) {
1659                        if (update_file(o, 0, &add->oid, add->mode, a->path))
1660                                return -1;
1661                }
1662                else
1663                        remove_file_from_cache(a->path);
1664                add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1665                if (add) {
1666                        if (update_file(o, 0, &add->oid, add->mode, b->path))
1667                                return -1;
1668                }
1669                else
1670                        remove_file_from_cache(b->path);
1671        } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1672                return -1;
1673
1674        return 0;
1675}
1676
1677static int handle_rename_rename_2to1(struct merge_options *o,
1678                                     struct rename_conflict_info *ci)
1679{
1680        /* Two files, a & b, were renamed to the same thing, c. */
1681        struct diff_filespec *a = ci->pair1->one;
1682        struct diff_filespec *b = ci->pair2->one;
1683        struct diff_filespec *c1 = ci->pair1->two;
1684        struct diff_filespec *c2 = ci->pair2->two;
1685        char *path = c1->path; /* == c2->path */
1686        char *path_side_1_desc;
1687        char *path_side_2_desc;
1688        struct merge_file_info mfi_c1;
1689        struct merge_file_info mfi_c2;
1690        int ret;
1691
1692        output(o, 1, _("CONFLICT (rename/rename): "
1693               "Rename %s->%s in %s. "
1694               "Rename %s->%s in %s"),
1695               a->path, c1->path, ci->branch1,
1696               b->path, c2->path, ci->branch2);
1697
1698        remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1699        remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1700
1701        path_side_1_desc = xstrfmt("%s (was %s)", path, a->path);
1702        path_side_2_desc = xstrfmt("%s (was %s)", path, b->path);
1703        if (merge_file_1(o, a, c1, &ci->ren1_other, path_side_1_desc,
1704                         o->branch1, o->branch2, &mfi_c1) ||
1705            merge_file_1(o, b, &ci->ren2_other, c2, path_side_2_desc,
1706                         o->branch1, o->branch2, &mfi_c2))
1707                return -1;
1708        free(path_side_1_desc);
1709        free(path_side_2_desc);
1710
1711        if (o->call_depth) {
1712                /*
1713                 * If mfi_c1.clean && mfi_c2.clean, then it might make
1714                 * sense to do a two-way merge of those results.  But, I
1715                 * think in all cases, it makes sense to have the virtual
1716                 * merge base just undo the renames; they can be detected
1717                 * again later for the non-recursive merge.
1718                 */
1719                remove_file(o, 0, path, 0);
1720                ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1721                if (!ret)
1722                        ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1723                                          b->path);
1724        } else {
1725                char *new_path1 = unique_path(o, path, ci->branch1);
1726                char *new_path2 = unique_path(o, path, ci->branch2);
1727                output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1728                       a->path, new_path1, b->path, new_path2);
1729                if (was_dirty(o, path))
1730                        output(o, 1, _("Refusing to lose dirty file at %s"),
1731                               path);
1732                else if (would_lose_untracked(path))
1733                        /*
1734                         * Only way we get here is if both renames were from
1735                         * a directory rename AND user had an untracked file
1736                         * at the location where both files end up after the
1737                         * two directory renames.  See testcase 10d of t6043.
1738                         */
1739                        output(o, 1, _("Refusing to lose untracked file at "
1740                                       "%s, even though it's in the way."),
1741                               path);
1742                else
1743                        remove_file(o, 0, path, 0);
1744                ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1745                if (!ret)
1746                        ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1747                                          new_path2);
1748                /*
1749                 * unpack_trees() actually populates the index for us for
1750                 * "normal" rename/rename(2to1) situtations so that the
1751                 * correct entries are at the higher stages, which would
1752                 * make the call below to update_stages_for_stage_data
1753                 * unnecessary.  However, if either of the renames came
1754                 * from a directory rename, then unpack_trees() will not
1755                 * have gotten the right data loaded into the index, so we
1756                 * need to do so now.  (While it'd be tempting to move this
1757                 * call to update_stages_for_stage_data() to
1758                 * apply_directory_rename_modifications(), that would break
1759                 * our intermediate calls to would_lose_untracked() since
1760                 * those rely on the current in-memory index.  See also the
1761                 * big "NOTE" in update_stages()).
1762                 */
1763                if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1764                        ret = -1;
1765
1766                free(new_path2);
1767                free(new_path1);
1768        }
1769
1770        return ret;
1771}
1772
1773/*
1774 * Get the diff_filepairs changed between o_tree and tree.
1775 */
1776static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1777                                               struct tree *o_tree,
1778                                               struct tree *tree)
1779{
1780        struct diff_queue_struct *ret;
1781        struct diff_options opts;
1782
1783        diff_setup(&opts);
1784        opts.flags.recursive = 1;
1785        opts.flags.rename_empty = 0;
1786        opts.detect_rename = merge_detect_rename(o);
1787        /*
1788         * We do not have logic to handle the detection of copies.  In
1789         * fact, it may not even make sense to add such logic: would we
1790         * really want a change to a base file to be propagated through
1791         * multiple other files by a merge?
1792         */
1793        if (opts.detect_rename > DIFF_DETECT_RENAME)
1794                opts.detect_rename = DIFF_DETECT_RENAME;
1795        opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1796                            o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1797                            1000;
1798        opts.rename_score = o->rename_score;
1799        opts.show_rename_progress = o->show_rename_progress;
1800        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1801        diff_setup_done(&opts);
1802        diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1803        diffcore_std(&opts);
1804        if (opts.needed_rename_limit > o->needed_rename_limit)
1805                o->needed_rename_limit = opts.needed_rename_limit;
1806
1807        ret = xmalloc(sizeof(*ret));
1808        *ret = diff_queued_diff;
1809
1810        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1811        diff_queued_diff.nr = 0;
1812        diff_queued_diff.queue = NULL;
1813        diff_flush(&opts);
1814        return ret;
1815}
1816
1817static int tree_has_path(struct tree *tree, const char *path)
1818{
1819        struct object_id hashy;
1820        unsigned int mode_o;
1821
1822        return !get_tree_entry(&tree->object.oid, path,
1823                               &hashy, &mode_o);
1824}
1825
1826/*
1827 * Return a new string that replaces the beginning portion (which matches
1828 * entry->dir), with entry->new_dir.  In perl-speak:
1829 *   new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1830 * NOTE:
1831 *   Caller must ensure that old_path starts with entry->dir + '/'.
1832 */
1833static char *apply_dir_rename(struct dir_rename_entry *entry,
1834                              const char *old_path)
1835{
1836        struct strbuf new_path = STRBUF_INIT;
1837        int oldlen, newlen;
1838
1839        if (entry->non_unique_new_dir)
1840                return NULL;
1841
1842        oldlen = strlen(entry->dir);
1843        newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1844        strbuf_grow(&new_path, newlen);
1845        strbuf_addbuf(&new_path, &entry->new_dir);
1846        strbuf_addstr(&new_path, &old_path[oldlen]);
1847
1848        return strbuf_detach(&new_path, NULL);
1849}
1850
1851static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1852                                    char **old_dir, char **new_dir)
1853{
1854        char *end_of_old, *end_of_new;
1855        int old_len, new_len;
1856
1857        *old_dir = NULL;
1858        *new_dir = NULL;
1859
1860        /*
1861         * For
1862         *    "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1863         * the "e/foo.c" part is the same, we just want to know that
1864         *    "a/b/c/d" was renamed to "a/b/some/thing/else"
1865         * so, for this example, this function returns "a/b/c/d" in
1866         * *old_dir and "a/b/some/thing/else" in *new_dir.
1867         *
1868         * Also, if the basename of the file changed, we don't care.  We
1869         * want to know which portion of the directory, if any, changed.
1870         */
1871        end_of_old = strrchr(old_path, '/');
1872        end_of_new = strrchr(new_path, '/');
1873
1874        if (end_of_old == NULL || end_of_new == NULL)
1875                return;
1876        while (*--end_of_new == *--end_of_old &&
1877               end_of_old != old_path &&
1878               end_of_new != new_path)
1879                ; /* Do nothing; all in the while loop */
1880        /*
1881         * We've found the first non-matching character in the directory
1882         * paths.  That means the current directory we were comparing
1883         * represents the rename.  Move end_of_old and end_of_new back
1884         * to the full directory name.
1885         */
1886        if (*end_of_old == '/')
1887                end_of_old++;
1888        if (*end_of_old != '/')
1889                end_of_new++;
1890        end_of_old = strchr(end_of_old, '/');
1891        end_of_new = strchr(end_of_new, '/');
1892
1893        /*
1894         * It may have been the case that old_path and new_path were the same
1895         * directory all along.  Don't claim a rename if they're the same.
1896         */
1897        old_len = end_of_old - old_path;
1898        new_len = end_of_new - new_path;
1899
1900        if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1901                *old_dir = xstrndup(old_path, old_len);
1902                *new_dir = xstrndup(new_path, new_len);
1903        }
1904}
1905
1906static void remove_hashmap_entries(struct hashmap *dir_renames,
1907                                   struct string_list *items_to_remove)
1908{
1909        int i;
1910        struct dir_rename_entry *entry;
1911
1912        for (i = 0; i < items_to_remove->nr; i++) {
1913                entry = items_to_remove->items[i].util;
1914                hashmap_remove(dir_renames, entry, NULL);
1915        }
1916        string_list_clear(items_to_remove, 0);
1917}
1918
1919/*
1920 * See if there is a directory rename for path, and if there are any file
1921 * level conflicts for the renamed location.  If there is a rename and
1922 * there are no conflicts, return the new name.  Otherwise, return NULL.
1923 */
1924static char *handle_path_level_conflicts(struct merge_options *o,
1925                                         const char *path,
1926                                         struct dir_rename_entry *entry,
1927                                         struct hashmap *collisions,
1928                                         struct tree *tree)
1929{
1930        char *new_path = NULL;
1931        struct collision_entry *collision_ent;
1932        int clean = 1;
1933        struct strbuf collision_paths = STRBUF_INIT;
1934
1935        /*
1936         * entry has the mapping of old directory name to new directory name
1937         * that we want to apply to path.
1938         */
1939        new_path = apply_dir_rename(entry, path);
1940
1941        if (!new_path) {
1942                /* This should only happen when entry->non_unique_new_dir set */
1943                if (!entry->non_unique_new_dir)
1944                        BUG("entry->non_unqiue_dir not set and !new_path");
1945                output(o, 1, _("CONFLICT (directory rename split): "
1946                               "Unclear where to place %s because directory "
1947                               "%s was renamed to multiple other directories, "
1948                               "with no destination getting a majority of the "
1949                               "files."),
1950                       path, entry->dir);
1951                clean = 0;
1952                return NULL;
1953        }
1954
1955        /*
1956         * The caller needs to have ensured that it has pre-populated
1957         * collisions with all paths that map to new_path.  Do a quick check
1958         * to ensure that's the case.
1959         */
1960        collision_ent = collision_find_entry(collisions, new_path);
1961        if (collision_ent == NULL)
1962                BUG("collision_ent is NULL");
1963
1964        /*
1965         * Check for one-sided add/add/.../add conflicts, i.e.
1966         * where implicit renames from the other side doing
1967         * directory rename(s) can affect this side of history
1968         * to put multiple paths into the same location.  Warn
1969         * and bail on directory renames for such paths.
1970         */
1971        if (collision_ent->reported_already) {
1972                clean = 0;
1973        } else if (tree_has_path(tree, new_path)) {
1974                collision_ent->reported_already = 1;
1975                strbuf_add_separated_string_list(&collision_paths, ", ",
1976                                                 &collision_ent->source_files);
1977                output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1978                               "file/dir at %s in the way of implicit "
1979                               "directory rename(s) putting the following "
1980                               "path(s) there: %s."),
1981                       new_path, collision_paths.buf);
1982                clean = 0;
1983        } else if (collision_ent->source_files.nr > 1) {
1984                collision_ent->reported_already = 1;
1985                strbuf_add_separated_string_list(&collision_paths, ", ",
1986                                                 &collision_ent->source_files);
1987                output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1988                               "more than one path to %s; implicit directory "
1989                               "renames tried to put these paths there: %s"),
1990                       new_path, collision_paths.buf);
1991                clean = 0;
1992        }
1993
1994        /* Free memory we no longer need */
1995        strbuf_release(&collision_paths);
1996        if (!clean && new_path) {
1997                free(new_path);
1998                return NULL;
1999        }
2000
2001        return new_path;
2002}
2003
2004/*
2005 * There are a couple things we want to do at the directory level:
2006 *   1. Check for both sides renaming to the same thing, in order to avoid
2007 *      implicit renaming of files that should be left in place.  (See
2008 *      testcase 6b in t6043 for details.)
2009 *   2. Prune directory renames if there are still files left in the
2010 *      the original directory.  These represent a partial directory rename,
2011 *      i.e. a rename where only some of the files within the directory
2012 *      were renamed elsewhere.  (Technically, this could be done earlier
2013 *      in get_directory_renames(), except that would prevent us from
2014 *      doing the previous check and thus failing testcase 6b.)
2015 *   3. Check for rename/rename(1to2) conflicts (at the directory level).
2016 *      In the future, we could potentially record this info as well and
2017 *      omit reporting rename/rename(1to2) conflicts for each path within
2018 *      the affected directories, thus cleaning up the merge output.
2019 *   NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2020 *         directory level, because merging directories is fine.  If it
2021 *         causes conflicts for files within those merged directories, then
2022 *         that should be detected at the individual path level.
2023 */
2024static void handle_directory_level_conflicts(struct merge_options *o,
2025                                             struct hashmap *dir_re_head,
2026                                             struct tree *head,
2027                                             struct hashmap *dir_re_merge,
2028                                             struct tree *merge)
2029{
2030        struct hashmap_iter iter;
2031        struct dir_rename_entry *head_ent;
2032        struct dir_rename_entry *merge_ent;
2033
2034        struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2035        struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2036
2037        hashmap_iter_init(dir_re_head, &iter);
2038        while ((head_ent = hashmap_iter_next(&iter))) {
2039                merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2040                if (merge_ent &&
2041                    !head_ent->non_unique_new_dir &&
2042                    !merge_ent->non_unique_new_dir &&
2043                    !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2044                        /* 1. Renamed identically; remove it from both sides */
2045                        string_list_append(&remove_from_head,
2046                                           head_ent->dir)->util = head_ent;
2047                        strbuf_release(&head_ent->new_dir);
2048                        string_list_append(&remove_from_merge,
2049                                           merge_ent->dir)->util = merge_ent;
2050                        strbuf_release(&merge_ent->new_dir);
2051                } else if (tree_has_path(head, head_ent->dir)) {
2052                        /* 2. This wasn't a directory rename after all */
2053                        string_list_append(&remove_from_head,
2054                                           head_ent->dir)->util = head_ent;
2055                        strbuf_release(&head_ent->new_dir);
2056                }
2057        }
2058
2059        remove_hashmap_entries(dir_re_head, &remove_from_head);
2060        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2061
2062        hashmap_iter_init(dir_re_merge, &iter);
2063        while ((merge_ent = hashmap_iter_next(&iter))) {
2064                head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2065                if (tree_has_path(merge, merge_ent->dir)) {
2066                        /* 2. This wasn't a directory rename after all */
2067                        string_list_append(&remove_from_merge,
2068                                           merge_ent->dir)->util = merge_ent;
2069                } else if (head_ent &&
2070                           !head_ent->non_unique_new_dir &&
2071                           !merge_ent->non_unique_new_dir) {
2072                        /* 3. rename/rename(1to2) */
2073                        /*
2074                         * We can assume it's not rename/rename(1to1) because
2075                         * that was case (1), already checked above.  So we
2076                         * know that head_ent->new_dir and merge_ent->new_dir
2077                         * are different strings.
2078                         */
2079                        output(o, 1, _("CONFLICT (rename/rename): "
2080                                       "Rename directory %s->%s in %s. "
2081                                       "Rename directory %s->%s in %s"),
2082                               head_ent->dir, head_ent->new_dir.buf, o->branch1,
2083                               head_ent->dir, merge_ent->new_dir.buf, o->branch2);
2084                        string_list_append(&remove_from_head,
2085                                           head_ent->dir)->util = head_ent;
2086                        strbuf_release(&head_ent->new_dir);
2087                        string_list_append(&remove_from_merge,
2088                                           merge_ent->dir)->util = merge_ent;
2089                        strbuf_release(&merge_ent->new_dir);
2090                }
2091        }
2092
2093        remove_hashmap_entries(dir_re_head, &remove_from_head);
2094        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2095}
2096
2097static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
2098                                             struct tree *tree)
2099{
2100        struct hashmap *dir_renames;
2101        struct hashmap_iter iter;
2102        struct dir_rename_entry *entry;
2103        int i;
2104
2105        /*
2106         * Typically, we think of a directory rename as all files from a
2107         * certain directory being moved to a target directory.  However,
2108         * what if someone first moved two files from the original
2109         * directory in one commit, and then renamed the directory
2110         * somewhere else in a later commit?  At merge time, we just know
2111         * that files from the original directory went to two different
2112         * places, and that the bulk of them ended up in the same place.
2113         * We want each directory rename to represent where the bulk of the
2114         * files from that directory end up; this function exists to find
2115         * where the bulk of the files went.
2116         *
2117         * The first loop below simply iterates through the list of file
2118         * renames, finding out how often each directory rename pair
2119         * possibility occurs.
2120         */
2121        dir_renames = xmalloc(sizeof(*dir_renames));
2122        dir_rename_init(dir_renames);
2123        for (i = 0; i < pairs->nr; ++i) {
2124                struct string_list_item *item;
2125                int *count;
2126                struct diff_filepair *pair = pairs->queue[i];
2127                char *old_dir, *new_dir;
2128
2129                /* File not part of directory rename if it wasn't renamed */
2130                if (pair->status != 'R')
2131                        continue;
2132
2133                get_renamed_dir_portion(pair->one->path, pair->two->path,
2134                                        &old_dir,        &new_dir);
2135                if (!old_dir)
2136                        /* Directory didn't change at all; ignore this one. */
2137                        continue;
2138
2139                entry = dir_rename_find_entry(dir_renames, old_dir);
2140                if (!entry) {
2141                        entry = xmalloc(sizeof(*entry));
2142                        dir_rename_entry_init(entry, old_dir);
2143                        hashmap_put(dir_renames, entry);
2144                } else {
2145                        free(old_dir);
2146                }
2147                item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2148                if (!item) {
2149                        item = string_list_insert(&entry->possible_new_dirs,
2150                                                  new_dir);
2151                        item->util = xcalloc(1, sizeof(int));
2152                } else {
2153                        free(new_dir);
2154                }
2155                count = item->util;
2156                *count += 1;
2157        }
2158
2159        /*
2160         * For each directory with files moved out of it, we find out which
2161         * target directory received the most files so we can declare it to
2162         * be the "winning" target location for the directory rename.  This
2163         * winner gets recorded in new_dir.  If there is no winner
2164         * (multiple target directories received the same number of files),
2165         * we set non_unique_new_dir.  Once we've determined the winner (or
2166         * that there is no winner), we no longer need possible_new_dirs.
2167         */
2168        hashmap_iter_init(dir_renames, &iter);
2169        while ((entry = hashmap_iter_next(&iter))) {
2170                int max = 0;
2171                int bad_max = 0;
2172                char *best = NULL;
2173
2174                for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2175                        int *count = entry->possible_new_dirs.items[i].util;
2176
2177                        if (*count == max)
2178                                bad_max = max;
2179                        else if (*count > max) {
2180                                max = *count;
2181                                best = entry->possible_new_dirs.items[i].string;
2182                        }
2183                }
2184                if (bad_max == max)
2185                        entry->non_unique_new_dir = 1;
2186                else {
2187                        assert(entry->new_dir.len == 0);
2188                        strbuf_addstr(&entry->new_dir, best);
2189                }
2190                /*
2191                 * The relevant directory sub-portion of the original full
2192                 * filepaths were xstrndup'ed before inserting into
2193                 * possible_new_dirs, and instead of manually iterating the
2194                 * list and free'ing each, just lie and tell
2195                 * possible_new_dirs that it did the strdup'ing so that it
2196                 * will free them for us.
2197                 */
2198                entry->possible_new_dirs.strdup_strings = 1;
2199                string_list_clear(&entry->possible_new_dirs, 1);
2200        }
2201
2202        return dir_renames;
2203}
2204
2205static struct dir_rename_entry *check_dir_renamed(const char *path,
2206                                                  struct hashmap *dir_renames)
2207{
2208        char *temp = xstrdup(path);
2209        char *end;
2210        struct dir_rename_entry *entry = NULL;;
2211
2212        while ((end = strrchr(temp, '/'))) {
2213                *end = '\0';
2214                entry = dir_rename_find_entry(dir_renames, temp);
2215                if (entry)
2216                        break;
2217        }
2218        free(temp);
2219        return entry;
2220}
2221
2222static void compute_collisions(struct hashmap *collisions,
2223                               struct hashmap *dir_renames,
2224                               struct diff_queue_struct *pairs)
2225{
2226        int i;
2227
2228        /*
2229         * Multiple files can be mapped to the same path due to directory
2230         * renames done by the other side of history.  Since that other
2231         * side of history could have merged multiple directories into one,
2232         * if our side of history added the same file basename to each of
2233         * those directories, then all N of them would get implicitly
2234         * renamed by the directory rename detection into the same path,
2235         * and we'd get an add/add/.../add conflict, and all those adds
2236         * from *this* side of history.  This is not representable in the
2237         * index, and users aren't going to easily be able to make sense of
2238         * it.  So we need to provide a good warning about what's
2239         * happening, and fall back to no-directory-rename detection
2240         * behavior for those paths.
2241         *
2242         * See testcases 9e and all of section 5 from t6043 for examples.
2243         */
2244        collision_init(collisions);
2245
2246        for (i = 0; i < pairs->nr; ++i) {
2247                struct dir_rename_entry *dir_rename_ent;
2248                struct collision_entry *collision_ent;
2249                char *new_path;
2250                struct diff_filepair *pair = pairs->queue[i];
2251
2252                if (pair->status != 'A' && pair->status != 'R')
2253                        continue;
2254                dir_rename_ent = check_dir_renamed(pair->two->path,
2255                                                   dir_renames);
2256                if (!dir_rename_ent)
2257                        continue;
2258
2259                new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2260                if (!new_path)
2261                        /*
2262                         * dir_rename_ent->non_unique_new_path is true, which
2263                         * means there is no directory rename for us to use,
2264                         * which means it won't cause us any additional
2265                         * collisions.
2266                         */
2267                        continue;
2268                collision_ent = collision_find_entry(collisions, new_path);
2269                if (!collision_ent) {
2270                        collision_ent = xcalloc(1,
2271                                                sizeof(struct collision_entry));
2272                        hashmap_entry_init(collision_ent, strhash(new_path));
2273                        hashmap_put(collisions, collision_ent);
2274                        collision_ent->target_file = new_path;
2275                } else {
2276                        free(new_path);
2277                }
2278                string_list_insert(&collision_ent->source_files,
2279                                   pair->two->path);
2280        }
2281}
2282
2283static char *check_for_directory_rename(struct merge_options *o,
2284                                        const char *path,
2285                                        struct tree *tree,
2286                                        struct hashmap *dir_renames,
2287                                        struct hashmap *dir_rename_exclusions,
2288                                        struct hashmap *collisions,
2289                                        int *clean_merge)
2290{
2291        char *new_path = NULL;
2292        struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2293        struct dir_rename_entry *oentry = NULL;
2294
2295        if (!entry)
2296                return new_path;
2297
2298        /*
2299         * This next part is a little weird.  We do not want to do an
2300         * implicit rename into a directory we renamed on our side, because
2301         * that will result in a spurious rename/rename(1to2) conflict.  An
2302         * example:
2303         *   Base commit: dumbdir/afile, otherdir/bfile
2304         *   Side 1:      smrtdir/afile, otherdir/bfile
2305         *   Side 2:      dumbdir/afile, dumbdir/bfile
2306         * Here, while working on Side 1, we could notice that otherdir was
2307         * renamed/merged to dumbdir, and change the diff_filepair for
2308         * otherdir/bfile into a rename into dumbdir/bfile.  However, Side
2309         * 2 will notice the rename from dumbdir to smrtdir, and do the
2310         * transitive rename to move it from dumbdir/bfile to
2311         * smrtdir/bfile.  That gives us bfile in dumbdir vs being in
2312         * smrtdir, a rename/rename(1to2) conflict.  We really just want
2313         * the file to end up in smrtdir.  And the way to achieve that is
2314         * to not let Side1 do the rename to dumbdir, since we know that is
2315         * the source of one of our directory renames.
2316         *
2317         * That's why oentry and dir_rename_exclusions is here.
2318         *
2319         * As it turns out, this also prevents N-way transient rename
2320         * confusion; See testcases 9c and 9d of t6043.
2321         */
2322        oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2323        if (oentry) {
2324                output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2325                               "to %s, because %s itself was renamed."),
2326                       entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2327        } else {
2328                new_path = handle_path_level_conflicts(o, path, entry,
2329                                                       collisions, tree);
2330                *clean_merge &= (new_path != NULL);
2331        }
2332
2333        return new_path;
2334}
2335
2336static void apply_directory_rename_modifications(struct merge_options *o,
2337                                                 struct diff_filepair *pair,
2338                                                 char *new_path,
2339                                                 struct rename *re,
2340                                                 struct tree *tree,
2341                                                 struct tree *o_tree,
2342                                                 struct tree *a_tree,
2343                                                 struct tree *b_tree,
2344                                                 struct string_list *entries,
2345                                                 int *clean)
2346{
2347        struct string_list_item *item;
2348        int stage = (tree == a_tree ? 2 : 3);
2349        int update_wd;
2350
2351        /*
2352         * In all cases where we can do directory rename detection,
2353         * unpack_trees() will have read pair->two->path into the
2354         * index and the working copy.  We need to remove it so that
2355         * we can instead place it at new_path.  It is guaranteed to
2356         * not be untracked (unpack_trees() would have errored out
2357         * saying the file would have been overwritten), but it might
2358         * be dirty, though.
2359         */
2360        update_wd = !was_dirty(o, pair->two->path);
2361        if (!update_wd)
2362                output(o, 1, _("Refusing to lose dirty file at %s"),
2363                       pair->two->path);
2364        remove_file(o, 1, pair->two->path, !update_wd);
2365
2366        /* Find or create a new re->dst_entry */
2367        item = string_list_lookup(entries, new_path);
2368        if (item) {
2369                /*
2370                 * Since we're renaming on this side of history, and it's
2371                 * due to a directory rename on the other side of history
2372                 * (which we only allow when the directory in question no
2373                 * longer exists on the other side of history), the
2374                 * original entry for re->dst_entry is no longer
2375                 * necessary...
2376                 */
2377                re->dst_entry->processed = 1;
2378
2379                /*
2380                 * ...because we'll be using this new one.
2381                 */
2382                re->dst_entry = item->util;
2383        } else {
2384                /*
2385                 * re->dst_entry is for the before-dir-rename path, and we
2386                 * need it to hold information for the after-dir-rename
2387                 * path.  Before creating a new entry, we need to mark the
2388                 * old one as unnecessary (...unless it is shared by
2389                 * src_entry, i.e. this didn't use to be a rename, in which
2390                 * case we can just allow the normal processing to happen
2391                 * for it).
2392                 */
2393                if (pair->status == 'R')
2394                        re->dst_entry->processed = 1;
2395
2396                re->dst_entry = insert_stage_data(new_path,
2397                                                  o_tree, a_tree, b_tree,
2398                                                  entries);
2399                item = string_list_insert(entries, new_path);
2400                item->util = re->dst_entry;
2401        }
2402
2403        /*
2404         * Update the stage_data with the information about the path we are
2405         * moving into place.  That slot will be empty and available for us
2406         * to write to because of the collision checks in
2407         * handle_path_level_conflicts().  In other words,
2408         * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2409         * open for us to write to.
2410         *
2411         * It may be tempting to actually update the index at this point as
2412         * well, using update_stages_for_stage_data(), but as per the big
2413         * "NOTE" in update_stages(), doing so will modify the current
2414         * in-memory index which will break calls to would_lose_untracked()
2415         * that we need to make.  Instead, we need to just make sure that
2416         * the various handle_rename_*() functions update the index
2417         * explicitly rather than relying on unpack_trees() to have done it.
2418         */
2419        get_tree_entry(&tree->object.oid,
2420                       pair->two->path,
2421                       &re->dst_entry->stages[stage].oid,
2422                       &re->dst_entry->stages[stage].mode);
2423
2424        /* Update pair status */
2425        if (pair->status == 'A') {
2426                /*
2427                 * Recording rename information for this add makes it look
2428                 * like a rename/delete conflict.  Make sure we can
2429                 * correctly handle this as an add that was moved to a new
2430                 * directory instead of reporting a rename/delete conflict.
2431                 */
2432                re->add_turned_into_rename = 1;
2433        }
2434        /*
2435         * We don't actually look at pair->status again, but it seems
2436         * pedagogically correct to adjust it.
2437         */
2438        pair->status = 'R';
2439
2440        /*
2441         * Finally, record the new location.
2442         */
2443        pair->two->path = new_path;
2444}
2445
2446/*
2447 * Get information of all renames which occurred in 'pairs', making use of
2448 * any implicit directory renames inferred from the other side of history.
2449 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2450 * to be able to associate the correct cache entries with the rename
2451 * information; tree is always equal to either a_tree or b_tree.
2452 */
2453static struct string_list *get_renames(struct merge_options *o,
2454                                       struct diff_queue_struct *pairs,
2455                                       struct hashmap *dir_renames,
2456                                       struct hashmap *dir_rename_exclusions,
2457                                       struct tree *tree,
2458                                       struct tree *o_tree,
2459                                       struct tree *a_tree,
2460                                       struct tree *b_tree,
2461                                       struct string_list *entries,
2462                                       int *clean_merge)
2463{
2464        int i;
2465        struct hashmap collisions;
2466        struct hashmap_iter iter;
2467        struct collision_entry *e;
2468        struct string_list *renames;
2469
2470        compute_collisions(&collisions, dir_renames, pairs);
2471        renames = xcalloc(1, sizeof(struct string_list));
2472
2473        for (i = 0; i < pairs->nr; ++i) {
2474                struct string_list_item *item;
2475                struct rename *re;
2476                struct diff_filepair *pair = pairs->queue[i];
2477                char *new_path; /* non-NULL only with directory renames */
2478
2479                if (pair->status != 'A' && pair->status != 'R') {
2480                        diff_free_filepair(pair);
2481                        continue;
2482                }
2483                new_path = check_for_directory_rename(o, pair->two->path, tree,
2484                                                      dir_renames,
2485                                                      dir_rename_exclusions,
2486                                                      &collisions,
2487                                                      clean_merge);
2488                if (pair->status != 'R' && !new_path) {
2489                        diff_free_filepair(pair);
2490                        continue;
2491                }
2492
2493                re = xmalloc(sizeof(*re));
2494                re->processed = 0;
2495                re->add_turned_into_rename = 0;
2496                re->pair = pair;
2497                item = string_list_lookup(entries, re->pair->one->path);
2498                if (!item)
2499                        re->src_entry = insert_stage_data(re->pair->one->path,
2500                                        o_tree, a_tree, b_tree, entries);
2501                else
2502                        re->src_entry = item->util;
2503
2504                item = string_list_lookup(entries, re->pair->two->path);
2505                if (!item)
2506                        re->dst_entry = insert_stage_data(re->pair->two->path,
2507                                        o_tree, a_tree, b_tree, entries);
2508                else
2509                        re->dst_entry = item->util;
2510                item = string_list_insert(renames, pair->one->path);
2511                item->util = re;
2512                if (new_path)
2513                        apply_directory_rename_modifications(o, pair, new_path,
2514                                                             re, tree, o_tree,
2515                                                             a_tree, b_tree,
2516                                                             entries,
2517                                                             clean_merge);
2518        }
2519
2520        hashmap_iter_init(&collisions, &iter);
2521        while ((e = hashmap_iter_next(&iter))) {
2522                free(e->target_file);
2523                string_list_clear(&e->source_files, 0);
2524        }
2525        hashmap_free(&collisions, 1);
2526        return renames;
2527}
2528
2529static int process_renames(struct merge_options *o,
2530                           struct string_list *a_renames,
2531                           struct string_list *b_renames)
2532{
2533        int clean_merge = 1, i, j;
2534        struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2535        struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2536        const struct rename *sre;
2537
2538        for (i = 0; i < a_renames->nr; i++) {
2539                sre = a_renames->items[i].util;
2540                string_list_insert(&a_by_dst, sre->pair->two->path)->util
2541                        = (void *)sre;
2542        }
2543        for (i = 0; i < b_renames->nr; i++) {
2544                sre = b_renames->items[i].util;
2545                string_list_insert(&b_by_dst, sre->pair->two->path)->util
2546                        = (void *)sre;
2547        }
2548
2549        for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2550                struct string_list *renames1, *renames2Dst;
2551                struct rename *ren1 = NULL, *ren2 = NULL;
2552                const char *branch1, *branch2;
2553                const char *ren1_src, *ren1_dst;
2554                struct string_list_item *lookup;
2555
2556                if (i >= a_renames->nr) {
2557                        ren2 = b_renames->items[j++].util;
2558                } else if (j >= b_renames->nr) {
2559                        ren1 = a_renames->items[i++].util;
2560                } else {
2561                        int compare = strcmp(a_renames->items[i].string,
2562                                             b_renames->items[j].string);
2563                        if (compare <= 0)
2564                                ren1 = a_renames->items[i++].util;
2565                        if (compare >= 0)
2566                                ren2 = b_renames->items[j++].util;
2567                }
2568
2569                /* TODO: refactor, so that 1/2 are not needed */
2570                if (ren1) {
2571                        renames1 = a_renames;
2572                        renames2Dst = &b_by_dst;
2573                        branch1 = o->branch1;
2574                        branch2 = o->branch2;
2575                } else {
2576                        renames1 = b_renames;
2577                        renames2Dst = &a_by_dst;
2578                        branch1 = o->branch2;
2579                        branch2 = o->branch1;
2580                        SWAP(ren2, ren1);
2581                }
2582
2583                if (ren1->processed)
2584                        continue;
2585                ren1->processed = 1;
2586                ren1->dst_entry->processed = 1;
2587                /* BUG: We should only mark src_entry as processed if we
2588                 * are not dealing with a rename + add-source case.
2589                 */
2590                ren1->src_entry->processed = 1;
2591
2592                ren1_src = ren1->pair->one->path;
2593                ren1_dst = ren1->pair->two->path;
2594
2595                if (ren2) {
2596                        /* One file renamed on both sides */
2597                        const char *ren2_src = ren2->pair->one->path;
2598                        const char *ren2_dst = ren2->pair->two->path;
2599                        enum rename_type rename_type;
2600                        if (strcmp(ren1_src, ren2_src) != 0)
2601                                BUG("ren1_src != ren2_src");
2602                        ren2->dst_entry->processed = 1;
2603                        ren2->processed = 1;
2604                        if (strcmp(ren1_dst, ren2_dst) != 0) {
2605                                rename_type = RENAME_ONE_FILE_TO_TWO;
2606                                clean_merge = 0;
2607                        } else {
2608                                rename_type = RENAME_ONE_FILE_TO_ONE;
2609                                /* BUG: We should only remove ren1_src in
2610                                 * the base stage (think of rename +
2611                                 * add-source cases).
2612                                 */
2613                                remove_file(o, 1, ren1_src, 1);
2614                                update_entry(ren1->dst_entry,
2615                                             ren1->pair->one,
2616                                             ren1->pair->two,
2617                                             ren2->pair->two);
2618                        }
2619                        setup_rename_conflict_info(rename_type,
2620                                                   ren1->pair,
2621                                                   ren2->pair,
2622                                                   branch1,
2623                                                   branch2,
2624                                                   ren1->dst_entry,
2625                                                   ren2->dst_entry,
2626                                                   o,
2627                                                   NULL,
2628                                                   NULL);
2629                } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2630                        /* Two different files renamed to the same thing */
2631                        char *ren2_dst;
2632                        ren2 = lookup->util;
2633                        ren2_dst = ren2->pair->two->path;
2634                        if (strcmp(ren1_dst, ren2_dst) != 0)
2635                                BUG("ren1_dst != ren2_dst");
2636
2637                        clean_merge = 0;
2638                        ren2->processed = 1;
2639                        /*
2640                         * BUG: We should only mark src_entry as processed
2641                         * if we are not dealing with a rename + add-source
2642                         * case.
2643                         */
2644                        ren2->src_entry->processed = 1;
2645
2646                        setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2647                                                   ren1->pair,
2648                                                   ren2->pair,
2649                                                   branch1,
2650                                                   branch2,
2651                                                   ren1->dst_entry,
2652                                                   ren2->dst_entry,
2653                                                   o,
2654                                                   ren1->src_entry,
2655                                                   ren2->src_entry);
2656
2657                } else {
2658                        /* Renamed in 1, maybe changed in 2 */
2659                        /* we only use sha1 and mode of these */
2660                        struct diff_filespec src_other, dst_other;
2661                        int try_merge;
2662
2663                        /*
2664                         * unpack_trees loads entries from common-commit
2665                         * into stage 1, from head-commit into stage 2, and
2666                         * from merge-commit into stage 3.  We keep track
2667                         * of which side corresponds to the rename.
2668                         */
2669                        int renamed_stage = a_renames == renames1 ? 2 : 3;
2670                        int other_stage =   a_renames == renames1 ? 3 : 2;
2671
2672                        /* BUG: We should only remove ren1_src in the base
2673                         * stage and in other_stage (think of rename +
2674                         * add-source case).
2675                         */
2676                        remove_file(o, 1, ren1_src,
2677                                    renamed_stage == 2 || !was_tracked(o, ren1_src));
2678
2679                        oidcpy(&src_other.oid,
2680                               &ren1->src_entry->stages[other_stage].oid);
2681                        src_other.mode = ren1->src_entry->stages[other_stage].mode;
2682                        oidcpy(&dst_other.oid,
2683                               &ren1->dst_entry->stages[other_stage].oid);
2684                        dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2685                        try_merge = 0;
2686
2687                        if (oid_eq(&src_other.oid, &null_oid) &&
2688                            ren1->add_turned_into_rename) {
2689                                setup_rename_conflict_info(RENAME_VIA_DIR,
2690                                                           ren1->pair,
2691                                                           NULL,
2692                                                           branch1,
2693                                                           branch2,
2694                                                           ren1->dst_entry,
2695                                                           NULL,
2696                                                           o,
2697                                                           NULL,
2698                                                           NULL);
2699                        } else if (oid_eq(&src_other.oid, &null_oid)) {
2700                                setup_rename_conflict_info(RENAME_DELETE,
2701                                                           ren1->pair,
2702                                                           NULL,
2703                                                           branch1,
2704                                                           branch2,
2705                                                           ren1->dst_entry,
2706                                                           NULL,
2707                                                           o,
2708                                                           NULL,
2709                                                           NULL);
2710                        } else if ((dst_other.mode == ren1->pair->two->mode) &&
2711                                   oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2712                                /*
2713                                 * Added file on the other side identical to
2714                                 * the file being renamed: clean merge.
2715                                 * Also, there is no need to overwrite the
2716                                 * file already in the working copy, so call
2717                                 * update_file_flags() instead of
2718                                 * update_file().
2719                                 */
2720                                if (update_file_flags(o,
2721                                                      &ren1->pair->two->oid,
2722                                                      ren1->pair->two->mode,
2723                                                      ren1_dst,
2724                                                      1, /* update_cache */
2725                                                      0  /* update_wd    */))
2726                                        clean_merge = -1;
2727                        } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2728                                clean_merge = 0;
2729                                try_merge = 1;
2730                                output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2731                                       "%s added in %s"),
2732                                       ren1_src, ren1_dst, branch1,
2733                                       ren1_dst, branch2);
2734                                if (o->call_depth) {
2735                                        struct merge_file_info mfi;
2736                                        if (merge_file_one(o, ren1_dst, &null_oid, 0,
2737                                                           &ren1->pair->two->oid,
2738                                                           ren1->pair->two->mode,
2739                                                           &dst_other.oid,
2740                                                           dst_other.mode,
2741                                                           branch1, branch2, &mfi)) {
2742                                                clean_merge = -1;
2743                                                goto cleanup_and_return;
2744                                        }
2745                                        output(o, 1, _("Adding merged %s"), ren1_dst);
2746                                        if (update_file(o, 0, &mfi.oid,
2747                                                        mfi.mode, ren1_dst))
2748                                                clean_merge = -1;
2749                                        try_merge = 0;
2750                                } else {
2751                                        char *new_path = unique_path(o, ren1_dst, branch2);
2752                                        output(o, 1, _("Adding as %s instead"), new_path);
2753                                        if (update_file(o, 0, &dst_other.oid,
2754                                                        dst_other.mode, new_path))
2755                                                clean_merge = -1;
2756                                        free(new_path);
2757                                }
2758                        } else
2759                                try_merge = 1;
2760
2761                        if (clean_merge < 0)
2762                                goto cleanup_and_return;
2763                        if (try_merge) {
2764                                struct diff_filespec *one, *a, *b;
2765                                src_other.path = (char *)ren1_src;
2766
2767                                one = ren1->pair->one;
2768                                if (a_renames == renames1) {
2769                                        a = ren1->pair->two;
2770                                        b = &src_other;
2771                                } else {
2772                                        b = ren1->pair->two;
2773                                        a = &src_other;
2774                                }
2775                                update_entry(ren1->dst_entry, one, a, b);
2776                                setup_rename_conflict_info(RENAME_NORMAL,
2777                                                           ren1->pair,
2778                                                           NULL,
2779                                                           branch1,
2780                                                           NULL,
2781                                                           ren1->dst_entry,
2782                                                           NULL,
2783                                                           o,
2784                                                           NULL,
2785                                                           NULL);
2786                        }
2787                }
2788        }
2789cleanup_and_return:
2790        string_list_clear(&a_by_dst, 0);
2791        string_list_clear(&b_by_dst, 0);
2792
2793        return clean_merge;
2794}
2795
2796struct rename_info {
2797        struct string_list *head_renames;
2798        struct string_list *merge_renames;
2799};
2800
2801static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2802                                   struct hashmap *dir_renames)
2803{
2804        struct hashmap_iter iter;
2805        struct dir_rename_entry *e;
2806
2807        hashmap_iter_init(dir_renames, &iter);
2808        while ((e = hashmap_iter_next(&iter))) {
2809                free(e->dir);
2810                strbuf_release(&e->new_dir);
2811                /* possible_new_dirs already cleared in get_directory_renames */
2812        }
2813        hashmap_free(dir_renames, 1);
2814        free(dir_renames);
2815
2816        free(pairs->queue);
2817        free(pairs);
2818}
2819
2820static int detect_and_process_renames(struct merge_options *o,
2821                                      struct tree *common,
2822                                      struct tree *head,
2823                                      struct tree *merge,
2824                                      struct string_list *entries,
2825                                      struct rename_info *ri)
2826{
2827        struct diff_queue_struct *head_pairs, *merge_pairs;
2828        struct hashmap *dir_re_head, *dir_re_merge;
2829        int clean = 1;
2830
2831        ri->head_renames = NULL;
2832        ri->merge_renames = NULL;
2833
2834        if (!merge_detect_rename(o))
2835                return 1;
2836
2837        head_pairs = get_diffpairs(o, common, head);
2838        merge_pairs = get_diffpairs(o, common, merge);
2839
2840        if (o->detect_directory_renames) {
2841                dir_re_head = get_directory_renames(head_pairs, head);
2842                dir_re_merge = get_directory_renames(merge_pairs, merge);
2843
2844                handle_directory_level_conflicts(o,
2845                                                 dir_re_head, head,
2846                                                 dir_re_merge, merge);
2847        } else {
2848                dir_re_head  = xmalloc(sizeof(*dir_re_head));
2849                dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2850                dir_rename_init(dir_re_head);
2851                dir_rename_init(dir_re_merge);
2852        }
2853
2854        ri->head_renames  = get_renames(o, head_pairs,
2855                                        dir_re_merge, dir_re_head, head,
2856                                        common, head, merge, entries,
2857                                        &clean);
2858        if (clean < 0)
2859                goto cleanup;
2860        ri->merge_renames = get_renames(o, merge_pairs,
2861                                        dir_re_head, dir_re_merge, merge,
2862                                        common, head, merge, entries,
2863                                        &clean);
2864        if (clean < 0)
2865                goto cleanup;
2866        clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2867
2868cleanup:
2869        /*
2870         * Some cleanup is deferred until cleanup_renames() because the
2871         * data structures are still needed and referenced in
2872         * process_entry().  But there are a few things we can free now.
2873         */
2874        initial_cleanup_rename(head_pairs, dir_re_head);
2875        initial_cleanup_rename(merge_pairs, dir_re_merge);
2876
2877        return clean;
2878}
2879
2880static void final_cleanup_rename(struct string_list *rename)
2881{
2882        const struct rename *re;
2883        int i;
2884
2885        if (rename == NULL)
2886                return;
2887
2888        for (i = 0; i < rename->nr; i++) {
2889                re = rename->items[i].util;
2890                diff_free_filepair(re->pair);
2891        }
2892        string_list_clear(rename, 1);
2893        free(rename);
2894}
2895
2896static void final_cleanup_renames(struct rename_info *re_info)
2897{
2898        final_cleanup_rename(re_info->head_renames);
2899        final_cleanup_rename(re_info->merge_renames);
2900}
2901
2902static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2903{
2904        return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2905}
2906
2907static int read_oid_strbuf(struct merge_options *o,
2908                           const struct object_id *oid,
2909                           struct strbuf *dst)
2910{
2911        void *buf;
2912        enum object_type type;
2913        unsigned long size;
2914        buf = read_object_file(oid, &type, &size);
2915        if (!buf)
2916                return err(o, _("cannot read object %s"), oid_to_hex(oid));
2917        if (type != OBJ_BLOB) {
2918                free(buf);
2919                return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2920        }
2921        strbuf_attach(dst, buf, size, size + 1);
2922        return 0;
2923}
2924
2925static int blob_unchanged(struct merge_options *opt,
2926                          const struct object_id *o_oid,
2927                          unsigned o_mode,
2928                          const struct object_id *a_oid,
2929                          unsigned a_mode,
2930                          int renormalize, const char *path)
2931{
2932        struct strbuf o = STRBUF_INIT;
2933        struct strbuf a = STRBUF_INIT;
2934        int ret = 0; /* assume changed for safety */
2935
2936        if (a_mode != o_mode)
2937                return 0;
2938        if (oid_eq(o_oid, a_oid))
2939                return 1;
2940        if (!renormalize)
2941                return 0;
2942
2943        assert(o_oid && a_oid);
2944        if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2945                goto error_return;
2946        /*
2947         * Note: binary | is used so that both renormalizations are
2948         * performed.  Comparison can be skipped if both files are
2949         * unchanged since their sha1s have already been compared.
2950         */
2951        if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2952            renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2953                ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2954
2955error_return:
2956        strbuf_release(&o);
2957        strbuf_release(&a);
2958        return ret;
2959}
2960
2961static int handle_modify_delete(struct merge_options *o,
2962                                const char *path,
2963                                struct object_id *o_oid, int o_mode,
2964                                struct object_id *a_oid, int a_mode,
2965                                struct object_id *b_oid, int b_mode)
2966{
2967        const char *modify_branch, *delete_branch;
2968        struct object_id *changed_oid;
2969        int changed_mode;
2970
2971        if (a_oid) {
2972                modify_branch = o->branch1;
2973                delete_branch = o->branch2;
2974                changed_oid = a_oid;
2975                changed_mode = a_mode;
2976        } else {
2977                modify_branch = o->branch2;
2978                delete_branch = o->branch1;
2979                changed_oid = b_oid;
2980                changed_mode = b_mode;
2981        }
2982
2983        return handle_change_delete(o,
2984                                    path, NULL,
2985                                    o_oid, o_mode,
2986                                    changed_oid, changed_mode,
2987                                    modify_branch, delete_branch,
2988                                    _("modify"), _("modified"));
2989}
2990
2991static int merge_content(struct merge_options *o,
2992                         const char *path,
2993                         int is_dirty,
2994                         struct object_id *o_oid, int o_mode,
2995                         struct object_id *a_oid, int a_mode,
2996                         struct object_id *b_oid, int b_mode,
2997                         struct rename_conflict_info *rename_conflict_info)
2998{
2999        const char *reason = _("content");
3000        const char *path1 = NULL, *path2 = NULL;
3001        struct merge_file_info mfi;
3002        struct diff_filespec one, a, b;
3003        unsigned df_conflict_remains = 0;
3004
3005        if (!o_oid) {
3006                reason = _("add/add");
3007                o_oid = (struct object_id *)&null_oid;
3008        }
3009        one.path = a.path = b.path = (char *)path;
3010        oidcpy(&one.oid, o_oid);
3011        one.mode = o_mode;
3012        oidcpy(&a.oid, a_oid);
3013        a.mode = a_mode;
3014        oidcpy(&b.oid, b_oid);
3015        b.mode = b_mode;
3016
3017        if (rename_conflict_info) {
3018                struct diff_filepair *pair1 = rename_conflict_info->pair1;
3019
3020                path1 = (o->branch1 == rename_conflict_info->branch1) ?
3021                        pair1->two->path : pair1->one->path;
3022                /* If rename_conflict_info->pair2 != NULL, we are in
3023                 * RENAME_ONE_FILE_TO_ONE case.  Otherwise, we have a
3024                 * normal rename.
3025                 */
3026                path2 = (rename_conflict_info->pair2 ||
3027                         o->branch2 == rename_conflict_info->branch1) ?
3028                        pair1->two->path : pair1->one->path;
3029                one.path = pair1->one->path;
3030                a.path = (char *)path1;
3031                b.path = (char *)path2;
3032
3033                if (dir_in_way(path, !o->call_depth,
3034                               S_ISGITLINK(pair1->two->mode)))
3035                        df_conflict_remains = 1;
3036        }
3037        if (merge_file_1(o, &one, &a, &b, path,
3038                         o->branch1, o->branch2, &mfi))
3039                return -1;
3040
3041        /*
3042         * We can skip updating the working tree file iff:
3043         *   a) The merge is clean
3044         *   b) The merge matches what was in HEAD (content, mode, pathname)
3045         *   c) The target path is usable (i.e. not involved in D/F conflict)
3046         */
3047        if (mfi.clean &&
3048            was_tracked_and_matches(o, path, &mfi.oid, mfi.mode) &&
3049            !df_conflict_remains) {
3050                int pos;
3051                struct cache_entry *ce;
3052
3053                output(o, 3, _("Skipped %s (merged same as existing)"), path);
3054                if (add_cacheinfo(o, mfi.mode, &mfi.oid, path,
3055                                  0, (!o->call_depth && !is_dirty), 0))
3056                        return -1;
3057                /*
3058                 * However, add_cacheinfo() will delete the old cache entry
3059                 * and add a new one.  We need to copy over any skip_worktree
3060                 * flag to avoid making the file appear as if it were
3061                 * deleted by the user.
3062                 */
3063                pos = index_name_pos(&o->orig_index, path, strlen(path));
3064                ce = o->orig_index.cache[pos];
3065                if (ce_skip_worktree(ce)) {
3066                        pos = index_name_pos(&the_index, path, strlen(path));
3067                        ce = the_index.cache[pos];
3068                        ce->ce_flags |= CE_SKIP_WORKTREE;
3069                }
3070                return mfi.clean;
3071        }
3072
3073        if (!mfi.clean) {
3074                if (S_ISGITLINK(mfi.mode))
3075                        reason = _("submodule");
3076                output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
3077                                reason, path);
3078                if (rename_conflict_info && !df_conflict_remains)
3079                        if (update_stages(o, path, &one, &a, &b))
3080                                return -1;
3081        }
3082
3083        if (df_conflict_remains || is_dirty) {
3084                char *new_path;
3085                if (o->call_depth) {
3086                        remove_file_from_cache(path);
3087                } else {
3088                        if (!mfi.clean) {
3089                                if (update_stages(o, path, &one, &a, &b))
3090                                        return -1;
3091                        } else {
3092                                int file_from_stage2 = was_tracked(o, path);
3093                                struct diff_filespec merged;
3094                                oidcpy(&merged.oid, &mfi.oid);
3095                                merged.mode = mfi.mode;
3096
3097                                if (update_stages(o, path, NULL,
3098                                                  file_from_stage2 ? &merged : NULL,
3099                                                  file_from_stage2 ? NULL : &merged))
3100                                        return -1;
3101                        }
3102
3103                }
3104                new_path = unique_path(o, path, rename_conflict_info->branch1);
3105                if (is_dirty) {
3106                        output(o, 1, _("Refusing to lose dirty file at %s"),
3107                               path);
3108                }
3109                output(o, 1, _("Adding as %s instead"), new_path);
3110                if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
3111                        free(new_path);
3112                        return -1;
3113                }
3114                free(new_path);
3115                mfi.clean = 0;
3116        } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
3117                return -1;
3118        return !is_dirty && mfi.clean;
3119}
3120
3121static int handle_rename_normal(struct merge_options *o,
3122                                const char *path,
3123                                struct object_id *o_oid, unsigned int o_mode,
3124                                struct object_id *a_oid, unsigned int a_mode,
3125                                struct object_id *b_oid, unsigned int b_mode,
3126                                struct rename_conflict_info *ci)
3127{
3128        /* Merge the content and write it out */
3129        return merge_content(o, path, was_dirty(o, path),
3130                             o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
3131                             ci);
3132}
3133
3134/* Per entry merge function */
3135static int process_entry(struct merge_options *o,
3136                         const char *path, struct stage_data *entry)
3137{
3138        int clean_merge = 1;
3139        int normalize = o->renormalize;
3140        unsigned o_mode = entry->stages[1].mode;
3141        unsigned a_mode = entry->stages[2].mode;
3142        unsigned b_mode = entry->stages[3].mode;
3143        struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
3144        struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
3145        struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
3146
3147        entry->processed = 1;
3148        if (entry->rename_conflict_info) {
3149                struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
3150                switch (conflict_info->rename_type) {
3151                case RENAME_NORMAL:
3152                case RENAME_ONE_FILE_TO_ONE:
3153                        clean_merge = handle_rename_normal(o,
3154                                                           path,
3155                                                           o_oid, o_mode,
3156                                                           a_oid, a_mode,
3157                                                           b_oid, b_mode,
3158                                                           conflict_info);
3159                        break;
3160                case RENAME_VIA_DIR:
3161                        clean_merge = 1;
3162                        if (handle_rename_via_dir(o,
3163                                                  conflict_info->pair1,
3164                                                  conflict_info->branch1,
3165                                                  conflict_info->branch2))
3166                                clean_merge = -1;
3167                        break;
3168                case RENAME_DELETE:
3169                        clean_merge = 0;
3170                        if (handle_rename_delete(o,
3171                                                 conflict_info->pair1,
3172                                                 conflict_info->branch1,
3173                                                 conflict_info->branch2))
3174                                clean_merge = -1;
3175                        break;
3176                case RENAME_ONE_FILE_TO_TWO:
3177                        clean_merge = 0;
3178                        if (handle_rename_rename_1to2(o, conflict_info))
3179                                clean_merge = -1;
3180                        break;
3181                case RENAME_TWO_FILES_TO_ONE:
3182                        clean_merge = 0;
3183                        if (handle_rename_rename_2to1(o, conflict_info))
3184                                clean_merge = -1;
3185                        break;
3186                default:
3187                        entry->processed = 0;
3188                        break;
3189                }
3190        } else if (o_oid && (!a_oid || !b_oid)) {
3191                /* Case A: Deleted in one */
3192                if ((!a_oid && !b_oid) ||
3193                    (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
3194                    (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
3195                        /* Deleted in both or deleted in one and
3196                         * unchanged in the other */
3197                        if (a_oid)
3198                                output(o, 2, _("Removing %s"), path);
3199                        /* do not touch working file if it did not exist */
3200                        remove_file(o, 1, path, !a_oid);
3201                } else {
3202                        /* Modify/delete; deleted side may have put a directory in the way */
3203                        clean_merge = 0;
3204                        if (handle_modify_delete(o, path, o_oid, o_mode,
3205                                                 a_oid, a_mode, b_oid, b_mode))
3206                                clean_merge = -1;
3207                }
3208        } else if ((!o_oid && a_oid && !b_oid) ||
3209                   (!o_oid && !a_oid && b_oid)) {
3210                /* Case B: Added in one. */
3211                /* [nothing|directory] -> ([nothing|directory], file) */
3212
3213                const char *add_branch;
3214                const char *other_branch;
3215                unsigned mode;
3216                const struct object_id *oid;
3217                const char *conf;
3218
3219                if (a_oid) {
3220                        add_branch = o->branch1;
3221                        other_branch = o->branch2;
3222                        mode = a_mode;
3223                        oid = a_oid;
3224                        conf = _("file/directory");
3225                } else {
3226                        add_branch = o->branch2;
3227                        other_branch = o->branch1;
3228                        mode = b_mode;
3229                        oid = b_oid;
3230                        conf = _("directory/file");
3231                }
3232                if (dir_in_way(path,
3233                               !o->call_depth && !S_ISGITLINK(a_mode),
3234                               0)) {
3235                        char *new_path = unique_path(o, path, add_branch);
3236                        clean_merge = 0;
3237                        output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3238                               "Adding %s as %s"),
3239                               conf, path, other_branch, path, new_path);
3240                        if (update_file(o, 0, oid, mode, new_path))
3241                                clean_merge = -1;
3242                        else if (o->call_depth)
3243                                remove_file_from_cache(path);
3244                        free(new_path);
3245                } else {
3246                        output(o, 2, _("Adding %s"), path);
3247                        /* do not overwrite file if already present */
3248                        if (update_file_flags(o, oid, mode, path, 1, !a_oid))
3249                                clean_merge = -1;
3250                }
3251        } else if (a_oid && b_oid) {
3252                /* Case C: Added in both (check for same permissions) and */
3253                /* case D: Modified in both, but differently. */
3254                int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3255                clean_merge = merge_content(o, path, is_dirty,
3256                                            o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
3257                                            NULL);
3258        } else if (!o_oid && !a_oid && !b_oid) {
3259                /*
3260                 * this entry was deleted altogether. a_mode == 0 means
3261                 * we had that path and want to actively remove it.
3262                 */
3263                remove_file(o, 1, path, !a_mode);
3264        } else
3265                BUG("fatal merge failure, shouldn't happen.");
3266
3267        return clean_merge;
3268}
3269
3270int merge_trees(struct merge_options *o,
3271                struct tree *head,
3272                struct tree *merge,
3273                struct tree *common,
3274                struct tree **result)
3275{
3276        int code, clean;
3277        struct strbuf sb = STRBUF_INIT;
3278
3279        if (!o->call_depth && index_has_changes(&the_index, head, &sb)) {
3280                err(o, _("Your local changes to the following files would be overwritten by merge:\n  %s"),
3281                    sb.buf);
3282                return -1;
3283        }
3284
3285        if (o->subtree_shift) {
3286                merge = shift_tree_object(head, merge, o->subtree_shift);
3287                common = shift_tree_object(head, common, o->subtree_shift);
3288        }
3289
3290        if (oid_eq(&common->object.oid, &merge->object.oid)) {
3291                output(o, 0, _("Already up to date!"));
3292                *result = head;
3293                return 1;
3294        }
3295
3296        code = unpack_trees_start(o, common, head, merge);
3297
3298        if (code != 0) {
3299                if (show(o, 4) || o->call_depth)
3300                        err(o, _("merging of trees %s and %s failed"),
3301                            oid_to_hex(&head->object.oid),
3302                            oid_to_hex(&merge->object.oid));
3303                unpack_trees_finish(o);
3304                return -1;
3305        }
3306
3307        if (unmerged_cache()) {
3308                struct string_list *entries;
3309                struct rename_info re_info;
3310                int i;
3311                /*
3312                 * Only need the hashmap while processing entries, so
3313                 * initialize it here and free it when we are done running
3314                 * through the entries. Keeping it in the merge_options as
3315                 * opposed to decaring a local hashmap is for convenience
3316                 * so that we don't have to pass it to around.
3317                 */
3318                hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3319                get_files_dirs(o, head);
3320                get_files_dirs(o, merge);
3321
3322                entries = get_unmerged();
3323                clean = detect_and_process_renames(o, common, head, merge,
3324                                                   entries, &re_info);
3325                record_df_conflict_files(o, entries);
3326                if (clean < 0)
3327                        goto cleanup;
3328                for (i = entries->nr-1; 0 <= i; i--) {
3329                        const char *path = entries->items[i].string;
3330                        struct stage_data *e = entries->items[i].util;
3331                        if (!e->processed) {
3332                                int ret = process_entry(o, path, e);
3333                                if (!ret)
3334                                        clean = 0;
3335                                else if (ret < 0) {
3336                                        clean = ret;
3337                                        goto cleanup;
3338                                }
3339                        }
3340                }
3341                for (i = 0; i < entries->nr; i++) {
3342                        struct stage_data *e = entries->items[i].util;
3343                        if (!e->processed)
3344                                BUG("unprocessed path??? %s",
3345                                    entries->items[i].string);
3346                }
3347
3348        cleanup:
3349                final_cleanup_renames(&re_info);
3350
3351                string_list_clear(entries, 1);
3352                free(entries);
3353
3354                hashmap_free(&o->current_file_dir_set, 1);
3355
3356                if (clean < 0) {
3357                        unpack_trees_finish(o);
3358                        return clean;
3359                }
3360        }
3361        else
3362                clean = 1;
3363
3364        unpack_trees_finish(o);
3365
3366        if (o->call_depth && !(*result = write_tree_from_memory(o)))
3367                return -1;
3368
3369        return clean;
3370}
3371
3372static struct commit_list *reverse_commit_list(struct commit_list *list)
3373{
3374        struct commit_list *next = NULL, *current, *backup;
3375        for (current = list; current; current = backup) {
3376                backup = current->next;
3377                current->next = next;
3378                next = current;
3379        }
3380        return next;
3381}
3382
3383/*
3384 * Merge the commits h1 and h2, return the resulting virtual
3385 * commit object and a flag indicating the cleanness of the merge.
3386 */
3387int merge_recursive(struct merge_options *o,
3388                    struct commit *h1,
3389                    struct commit *h2,
3390                    struct commit_list *ca,
3391                    struct commit **result)
3392{
3393        struct commit_list *iter;
3394        struct commit *merged_common_ancestors;
3395        struct tree *mrtree;
3396        int clean;
3397
3398        if (show(o, 4)) {
3399                output(o, 4, _("Merging:"));
3400                output_commit_title(o, h1);
3401                output_commit_title(o, h2);
3402        }
3403
3404        if (!ca) {
3405                ca = get_merge_bases(h1, h2);
3406                ca = reverse_commit_list(ca);
3407        }
3408
3409        if (show(o, 5)) {
3410                unsigned cnt = commit_list_count(ca);
3411
3412                output(o, 5, Q_("found %u common ancestor:",
3413                                "found %u common ancestors:", cnt), cnt);
3414                for (iter = ca; iter; iter = iter->next)
3415                        output_commit_title(o, iter->item);
3416        }
3417
3418        merged_common_ancestors = pop_commit(&ca);
3419        if (merged_common_ancestors == NULL) {
3420                /* if there is no common ancestor, use an empty tree */
3421                struct tree *tree;
3422
3423                tree = lookup_tree(the_repository, the_repository->hash_algo->empty_tree);
3424                merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3425        }
3426
3427        for (iter = ca; iter; iter = iter->next) {
3428                const char *saved_b1, *saved_b2;
3429                o->call_depth++;
3430                /*
3431                 * When the merge fails, the result contains files
3432                 * with conflict markers. The cleanness flag is
3433                 * ignored (unless indicating an error), it was never
3434                 * actually used, as result of merge_trees has always
3435                 * overwritten it: the committed "conflicts" were
3436                 * already resolved.
3437                 */
3438                discard_cache();
3439                saved_b1 = o->branch1;
3440                saved_b2 = o->branch2;
3441                o->branch1 = "Temporary merge branch 1";
3442                o->branch2 = "Temporary merge branch 2";
3443                if (merge_recursive(o, merged_common_ancestors, iter->item,
3444                                    NULL, &merged_common_ancestors) < 0)
3445                        return -1;
3446                o->branch1 = saved_b1;
3447                o->branch2 = saved_b2;
3448                o->call_depth--;
3449
3450                if (!merged_common_ancestors)
3451                        return err(o, _("merge returned no commit"));
3452        }
3453
3454        discard_cache();
3455        if (!o->call_depth)
3456                read_cache();
3457
3458        o->ancestor = "merged common ancestors";
3459        clean = merge_trees(o, get_commit_tree(h1), get_commit_tree(h2),
3460                            get_commit_tree(merged_common_ancestors),
3461                            &mrtree);
3462        if (clean < 0) {
3463                flush_output(o);
3464                return clean;
3465        }
3466
3467        if (o->call_depth) {
3468                *result = make_virtual_commit(mrtree, "merged tree");
3469                commit_list_insert(h1, &(*result)->parents);
3470                commit_list_insert(h2, &(*result)->parents->next);
3471        }
3472        flush_output(o);
3473        if (!o->call_depth && o->buffer_output < 2)
3474                strbuf_release(&o->obuf);
3475        if (show(o, 2))
3476                diff_warn_rename_limit("merge.renamelimit",
3477                                       o->needed_rename_limit, 0);
3478        return clean;
3479}
3480
3481static struct commit *get_ref(const struct object_id *oid, const char *name)
3482{
3483        struct object *object;
3484
3485        object = deref_tag(the_repository, parse_object(the_repository, oid),
3486                           name,
3487                           strlen(name));
3488        if (!object)
3489                return NULL;
3490        if (object->type == OBJ_TREE)
3491                return make_virtual_commit((struct tree*)object, name);
3492        if (object->type != OBJ_COMMIT)
3493                return NULL;
3494        if (parse_commit((struct commit *)object))
3495                return NULL;
3496        return (struct commit *)object;
3497}
3498
3499int merge_recursive_generic(struct merge_options *o,
3500                            const struct object_id *head,
3501                            const struct object_id *merge,
3502                            int num_base_list,
3503                            const struct object_id **base_list,
3504                            struct commit **result)
3505{
3506        int clean;
3507        struct lock_file lock = LOCK_INIT;
3508        struct commit *head_commit = get_ref(head, o->branch1);
3509        struct commit *next_commit = get_ref(merge, o->branch2);
3510        struct commit_list *ca = NULL;
3511
3512        if (base_list) {
3513                int i;
3514                for (i = 0; i < num_base_list; ++i) {
3515                        struct commit *base;
3516                        if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3517                                return err(o, _("Could not parse object '%s'"),
3518                                           oid_to_hex(base_list[i]));
3519                        commit_list_insert(base, &ca);
3520                }
3521        }
3522
3523        hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3524        clean = merge_recursive(o, head_commit, next_commit, ca,
3525                                result);
3526        if (clean < 0) {
3527                rollback_lock_file(&lock);
3528                return clean;
3529        }
3530
3531        if (write_locked_index(&the_index, &lock,
3532                               COMMIT_LOCK | SKIP_IF_UNCHANGED))
3533                return err(o, _("Unable to write index."));
3534
3535        return clean ? 0 : 1;
3536}
3537
3538static void merge_recursive_config(struct merge_options *o)
3539{
3540        char *value = NULL;
3541        git_config_get_int("merge.verbosity", &o->verbosity);
3542        git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3543        git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3544        if (!git_config_get_string("diff.renames", &value)) {
3545                o->diff_detect_rename = git_config_rename("diff.renames", value);
3546                free(value);
3547        }
3548        if (!git_config_get_string("merge.renames", &value)) {
3549                o->merge_detect_rename = git_config_rename("merge.renames", value);
3550                free(value);
3551        }
3552        git_config(git_xmerge_config, NULL);
3553}
3554
3555void init_merge_options(struct merge_options *o)
3556{
3557        const char *merge_verbosity;
3558        memset(o, 0, sizeof(struct merge_options));
3559        o->verbosity = 2;
3560        o->buffer_output = 1;
3561        o->diff_rename_limit = -1;
3562        o->merge_rename_limit = -1;
3563        o->renormalize = 0;
3564        o->diff_detect_rename = -1;
3565        o->merge_detect_rename = -1;
3566        o->detect_directory_renames = 1;
3567        merge_recursive_config(o);
3568        merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3569        if (merge_verbosity)
3570                o->verbosity = strtol(merge_verbosity, NULL, 10);
3571        if (o->verbosity >= 5)
3572                o->buffer_output = 0;
3573        strbuf_init(&o->obuf, 0);
3574        string_list_init(&o->df_conflict_file_set, 1);
3575}
3576
3577int parse_merge_opt(struct merge_options *o, const char *s)
3578{
3579        const char *arg;
3580
3581        if (!s || !*s)
3582                return -1;
3583        if (!strcmp(s, "ours"))
3584                o->recursive_variant = MERGE_RECURSIVE_OURS;
3585        else if (!strcmp(s, "theirs"))
3586                o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3587        else if (!strcmp(s, "subtree"))
3588                o->subtree_shift = "";
3589        else if (skip_prefix(s, "subtree=", &arg))
3590                o->subtree_shift = arg;
3591        else if (!strcmp(s, "patience"))
3592                o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3593        else if (!strcmp(s, "histogram"))
3594                o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3595        else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3596                long value = parse_algorithm_value(arg);
3597                if (value < 0)
3598                        return -1;
3599                /* clear out previous settings */
3600                DIFF_XDL_CLR(o, NEED_MINIMAL);
3601                o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3602                o->xdl_opts |= value;
3603        }
3604        else if (!strcmp(s, "ignore-space-change"))
3605                DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3606        else if (!strcmp(s, "ignore-all-space"))
3607                DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3608        else if (!strcmp(s, "ignore-space-at-eol"))
3609                DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3610        else if (!strcmp(s, "ignore-cr-at-eol"))
3611                DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3612        else if (!strcmp(s, "renormalize"))
3613                o->renormalize = 1;
3614        else if (!strcmp(s, "no-renormalize"))
3615                o->renormalize = 0;
3616        else if (!strcmp(s, "no-renames"))
3617                o->merge_detect_rename = 0;
3618        else if (!strcmp(s, "find-renames")) {
3619                o->merge_detect_rename = 1;
3620                o->rename_score = 0;
3621        }
3622        else if (skip_prefix(s, "find-renames=", &arg) ||
3623                 skip_prefix(s, "rename-threshold=", &arg)) {
3624                if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3625                        return -1;
3626                o->merge_detect_rename = 1;
3627        }
3628        else
3629                return -1;
3630        return 0;
3631}