merge-recursive.con commit doc/CodingGuidelines: URLs and paths as monospace (0dbd305)
   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 repository *repo,
 150                                      struct tree *one, struct tree *two,
 151                                      const char *subtree_shift)
 152{
 153        struct object_id shifted;
 154
 155        if (!*subtree_shift) {
 156                shift_tree(&one->object.oid, &two->object.oid, &shifted, 0);
 157        } else {
 158                shift_tree_by(&one->object.oid, &two->object.oid, &shifted,
 159                              subtree_shift);
 160        }
 161        if (oideq(&two->object.oid, &shifted))
 162                return two;
 163        return lookup_tree(repo, &shifted);
 164}
 165
 166static struct commit *make_virtual_commit(struct repository *repo,
 167                                          struct tree *tree,
 168                                          const char *comment)
 169{
 170        struct commit *commit = alloc_commit_node(repo);
 171
 172        set_merge_remote_desc(commit, comment, (struct object *)commit);
 173        commit->maybe_tree = tree;
 174        commit->object.parsed = 1;
 175        return commit;
 176}
 177
 178/*
 179 * Since we use get_tree_entry(), which does not put the read object into
 180 * the object pool, we cannot rely on a == b.
 181 */
 182static int oid_eq(const struct object_id *a, const struct object_id *b)
 183{
 184        if (!a && !b)
 185                return 2;
 186        return a && b && oideq(a, b);
 187}
 188
 189enum rename_type {
 190        RENAME_NORMAL = 0,
 191        RENAME_VIA_DIR,
 192        RENAME_ADD,
 193        RENAME_DELETE,
 194        RENAME_ONE_FILE_TO_ONE,
 195        RENAME_ONE_FILE_TO_TWO,
 196        RENAME_TWO_FILES_TO_ONE
 197};
 198
 199struct rename_conflict_info {
 200        enum rename_type rename_type;
 201        struct diff_filepair *pair1;
 202        struct diff_filepair *pair2;
 203        const char *branch1;
 204        const char *branch2;
 205        struct stage_data *dst_entry1;
 206        struct stage_data *dst_entry2;
 207        struct diff_filespec ren1_other;
 208        struct diff_filespec ren2_other;
 209};
 210
 211/*
 212 * Since we want to write the index eventually, we cannot reuse the index
 213 * for these (temporary) data.
 214 */
 215struct stage_data {
 216        struct {
 217                unsigned mode;
 218                struct object_id oid;
 219        } stages[4];
 220        struct rename_conflict_info *rename_conflict_info;
 221        unsigned processed:1;
 222};
 223
 224static inline void setup_rename_conflict_info(enum rename_type rename_type,
 225                                              struct diff_filepair *pair1,
 226                                              struct diff_filepair *pair2,
 227                                              const char *branch1,
 228                                              const char *branch2,
 229                                              struct stage_data *dst_entry1,
 230                                              struct stage_data *dst_entry2,
 231                                              struct merge_options *o,
 232                                              struct stage_data *src_entry1,
 233                                              struct stage_data *src_entry2)
 234{
 235        int ostage1 = 0, ostage2;
 236        struct rename_conflict_info *ci;
 237
 238        /*
 239         * When we have two renames involved, it's easiest to get the
 240         * correct things into stage 2 and 3, and to make sure that the
 241         * content merge puts HEAD before the other branch if we just
 242         * ensure that branch1 == o->branch1.  So, simply flip arguments
 243         * around if we don't have that.
 244         */
 245        if (dst_entry2 && branch1 != o->branch1) {
 246                setup_rename_conflict_info(rename_type,
 247                                           pair2,      pair1,
 248                                           branch2,    branch1,
 249                                           dst_entry2, dst_entry1,
 250                                           o,
 251                                           src_entry2, src_entry1);
 252                return;
 253        }
 254
 255        ci = xcalloc(1, sizeof(struct rename_conflict_info));
 256        ci->rename_type = rename_type;
 257        ci->pair1 = pair1;
 258        ci->branch1 = branch1;
 259        ci->branch2 = branch2;
 260
 261        ci->dst_entry1 = dst_entry1;
 262        dst_entry1->rename_conflict_info = ci;
 263        dst_entry1->processed = 0;
 264
 265        assert(!pair2 == !dst_entry2);
 266        if (dst_entry2) {
 267                ci->dst_entry2 = dst_entry2;
 268                ci->pair2 = pair2;
 269                dst_entry2->rename_conflict_info = ci;
 270        }
 271
 272        /*
 273         * For each rename, there could have been
 274         * modifications on the side of history where that
 275         * file was not renamed.
 276         */
 277        if (rename_type == RENAME_ADD ||
 278            rename_type == RENAME_TWO_FILES_TO_ONE) {
 279                ostage1 = o->branch1 == branch1 ? 3 : 2;
 280
 281                ci->ren1_other.path = pair1->one->path;
 282                oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid);
 283                ci->ren1_other.mode = src_entry1->stages[ostage1].mode;
 284        }
 285
 286        if (rename_type == RENAME_TWO_FILES_TO_ONE) {
 287                ostage2 = ostage1 ^ 1;
 288
 289                ci->ren2_other.path = pair2->one->path;
 290                oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid);
 291                ci->ren2_other.mode = src_entry2->stages[ostage2].mode;
 292        }
 293}
 294
 295static int show(struct merge_options *o, int v)
 296{
 297        return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5;
 298}
 299
 300__attribute__((format (printf, 3, 4)))
 301static void output(struct merge_options *o, int v, const char *fmt, ...)
 302{
 303        va_list ap;
 304
 305        if (!show(o, v))
 306                return;
 307
 308        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 309
 310        va_start(ap, fmt);
 311        strbuf_vaddf(&o->obuf, fmt, ap);
 312        va_end(ap);
 313
 314        strbuf_addch(&o->obuf, '\n');
 315        if (!o->buffer_output)
 316                flush_output(o);
 317}
 318
 319static void output_commit_title(struct merge_options *o, struct commit *commit)
 320{
 321        struct merge_remote_desc *desc;
 322
 323        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 324        desc = merge_remote_util(commit);
 325        if (desc)
 326                strbuf_addf(&o->obuf, "virtual %s\n", desc->name);
 327        else {
 328                strbuf_add_unique_abbrev(&o->obuf, &commit->object.oid,
 329                                         DEFAULT_ABBREV);
 330                strbuf_addch(&o->obuf, ' ');
 331                if (parse_commit(commit) != 0)
 332                        strbuf_addstr(&o->obuf, _("(bad commit)\n"));
 333                else {
 334                        const char *title;
 335                        const char *msg = get_commit_buffer(commit, NULL);
 336                        int len = find_commit_subject(msg, &title);
 337                        if (len)
 338                                strbuf_addf(&o->obuf, "%.*s\n", len, title);
 339                        unuse_commit_buffer(commit, msg);
 340                }
 341        }
 342        flush_output(o);
 343}
 344
 345static int add_cacheinfo(struct merge_options *o,
 346                         unsigned int mode, const struct object_id *oid,
 347                         const char *path, int stage, int refresh, int options)
 348{
 349        struct index_state *istate = o->repo->index;
 350        struct cache_entry *ce;
 351        int ret;
 352
 353        ce = make_cache_entry(istate, mode, oid ? oid : &null_oid, path, stage, 0);
 354        if (!ce)
 355                return err(o, _("add_cacheinfo failed for path '%s'; merge aborting."), path);
 356
 357        ret = add_index_entry(istate, ce, options);
 358        if (refresh) {
 359                struct cache_entry *nce;
 360
 361                nce = refresh_cache_entry(istate, ce,
 362                                          CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
 363                if (!nce)
 364                        return err(o, _("add_cacheinfo failed to refresh for path '%s'; merge aborting."), path);
 365                if (nce != ce)
 366                        ret = add_index_entry(istate, nce, options);
 367        }
 368        return ret;
 369}
 370
 371static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
 372{
 373        parse_tree(tree);
 374        init_tree_desc(desc, tree->buffer, tree->size);
 375}
 376
 377static int unpack_trees_start(struct merge_options *o,
 378                              struct tree *common,
 379                              struct tree *head,
 380                              struct tree *merge)
 381{
 382        int rc;
 383        struct tree_desc t[3];
 384        struct index_state tmp_index = { NULL };
 385
 386        memset(&o->unpack_opts, 0, sizeof(o->unpack_opts));
 387        if (o->call_depth)
 388                o->unpack_opts.index_only = 1;
 389        else
 390                o->unpack_opts.update = 1;
 391        o->unpack_opts.merge = 1;
 392        o->unpack_opts.head_idx = 2;
 393        o->unpack_opts.fn = threeway_merge;
 394        o->unpack_opts.src_index = o->repo->index;
 395        o->unpack_opts.dst_index = &tmp_index;
 396        o->unpack_opts.aggressive = !merge_detect_rename(o);
 397        setup_unpack_trees_porcelain(&o->unpack_opts, "merge");
 398
 399        init_tree_desc_from_tree(t+0, common);
 400        init_tree_desc_from_tree(t+1, head);
 401        init_tree_desc_from_tree(t+2, merge);
 402
 403        rc = unpack_trees(3, t, &o->unpack_opts);
 404        cache_tree_free(&o->repo->index->cache_tree);
 405
 406        /*
 407         * Update o->repo->index to match the new results, AFTER saving a copy
 408         * in o->orig_index.  Update src_index to point to the saved copy.
 409         * (verify_uptodate() checks src_index, and the original index is
 410         * the one that had the necessary modification timestamps.)
 411         */
 412        o->orig_index = *o->repo->index;
 413        *o->repo->index = tmp_index;
 414        o->unpack_opts.src_index = &o->orig_index;
 415
 416        return rc;
 417}
 418
 419static void unpack_trees_finish(struct merge_options *o)
 420{
 421        discard_index(&o->orig_index);
 422        clear_unpack_trees_porcelain(&o->unpack_opts);
 423}
 424
 425struct tree *write_tree_from_memory(struct merge_options *o)
 426{
 427        struct tree *result = NULL;
 428        struct index_state *istate = o->repo->index;
 429
 430        if (unmerged_index(istate)) {
 431                int i;
 432                fprintf(stderr, "BUG: There are unmerged index entries:\n");
 433                for (i = 0; i < istate->cache_nr; i++) {
 434                        const struct cache_entry *ce = istate->cache[i];
 435                        if (ce_stage(ce))
 436                                fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
 437                                        (int)ce_namelen(ce), ce->name);
 438                }
 439                BUG("unmerged index entries in merge-recursive.c");
 440        }
 441
 442        if (!istate->cache_tree)
 443                istate->cache_tree = cache_tree();
 444
 445        if (!cache_tree_fully_valid(istate->cache_tree) &&
 446            cache_tree_update(istate, 0) < 0) {
 447                err(o, _("error building trees"));
 448                return NULL;
 449        }
 450
 451        result = lookup_tree(o->repo, &istate->cache_tree->oid);
 452
 453        return result;
 454}
 455
 456static int save_files_dirs(const struct object_id *oid,
 457                           struct strbuf *base, const char *path,
 458                           unsigned int mode, int stage, void *context)
 459{
 460        struct path_hashmap_entry *entry;
 461        int baselen = base->len;
 462        struct merge_options *o = context;
 463
 464        strbuf_addstr(base, path);
 465
 466        FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
 467        hashmap_entry_init(entry, path_hash(entry->path));
 468        hashmap_add(&o->current_file_dir_set, entry);
 469
 470        strbuf_setlen(base, baselen);
 471        return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
 472}
 473
 474static void get_files_dirs(struct merge_options *o, struct tree *tree)
 475{
 476        struct pathspec match_all;
 477        memset(&match_all, 0, sizeof(match_all));
 478        read_tree_recursive(the_repository, tree, "", 0, 0,
 479                            &match_all, save_files_dirs, o);
 480}
 481
 482static int get_tree_entry_if_blob(const struct object_id *tree,
 483                                  const char *path,
 484                                  struct object_id *hashy,
 485                                  unsigned int *mode_o)
 486{
 487        int ret;
 488
 489        ret = get_tree_entry(tree, path, hashy, mode_o);
 490        if (S_ISDIR(*mode_o)) {
 491                oidcpy(hashy, &null_oid);
 492                *mode_o = 0;
 493        }
 494        return ret;
 495}
 496
 497/*
 498 * Returns an index_entry instance which doesn't have to correspond to
 499 * a real cache entry in Git's index.
 500 */
 501static struct stage_data *insert_stage_data(const char *path,
 502                struct tree *o, struct tree *a, struct tree *b,
 503                struct string_list *entries)
 504{
 505        struct string_list_item *item;
 506        struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
 507        get_tree_entry_if_blob(&o->object.oid, path,
 508                               &e->stages[1].oid, &e->stages[1].mode);
 509        get_tree_entry_if_blob(&a->object.oid, path,
 510                               &e->stages[2].oid, &e->stages[2].mode);
 511        get_tree_entry_if_blob(&b->object.oid, path,
 512                               &e->stages[3].oid, &e->stages[3].mode);
 513        item = string_list_insert(entries, path);
 514        item->util = e;
 515        return e;
 516}
 517
 518/*
 519 * Create a dictionary mapping file names to stage_data objects. The
 520 * dictionary contains one entry for every path with a non-zero stage entry.
 521 */
 522static struct string_list *get_unmerged(struct index_state *istate)
 523{
 524        struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
 525        int i;
 526
 527        unmerged->strdup_strings = 1;
 528
 529        for (i = 0; i < istate->cache_nr; i++) {
 530                struct string_list_item *item;
 531                struct stage_data *e;
 532                const struct cache_entry *ce = istate->cache[i];
 533                if (!ce_stage(ce))
 534                        continue;
 535
 536                item = string_list_lookup(unmerged, ce->name);
 537                if (!item) {
 538                        item = string_list_insert(unmerged, ce->name);
 539                        item->util = xcalloc(1, sizeof(struct stage_data));
 540                }
 541                e = item->util;
 542                e->stages[ce_stage(ce)].mode = ce->ce_mode;
 543                oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
 544        }
 545
 546        return unmerged;
 547}
 548
 549static int string_list_df_name_compare(const char *one, const char *two)
 550{
 551        int onelen = strlen(one);
 552        int twolen = strlen(two);
 553        /*
 554         * Here we only care that entries for D/F conflicts are
 555         * adjacent, in particular with the file of the D/F conflict
 556         * appearing before files below the corresponding directory.
 557         * The order of the rest of the list is irrelevant for us.
 558         *
 559         * To achieve this, we sort with df_name_compare and provide
 560         * the mode S_IFDIR so that D/F conflicts will sort correctly.
 561         * We use the mode S_IFDIR for everything else for simplicity,
 562         * since in other cases any changes in their order due to
 563         * sorting cause no problems for us.
 564         */
 565        int cmp = df_name_compare(one, onelen, S_IFDIR,
 566                                  two, twolen, S_IFDIR);
 567        /*
 568         * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
 569         * that 'foo' comes before 'foo/bar'.
 570         */
 571        if (cmp)
 572                return cmp;
 573        return onelen - twolen;
 574}
 575
 576static void record_df_conflict_files(struct merge_options *o,
 577                                     struct string_list *entries)
 578{
 579        /* If there is a D/F conflict and the file for such a conflict
 580         * currently exists in the working tree, we want to allow it to be
 581         * removed to make room for the corresponding directory if needed.
 582         * The files underneath the directories of such D/F conflicts will
 583         * be processed before the corresponding file involved in the D/F
 584         * conflict.  If the D/F directory ends up being removed by the
 585         * merge, then we won't have to touch the D/F file.  If the D/F
 586         * directory needs to be written to the working copy, then the D/F
 587         * file will simply be removed (in make_room_for_path()) to make
 588         * room for the necessary paths.  Note that if both the directory
 589         * and the file need to be present, then the D/F file will be
 590         * reinstated with a new unique name at the time it is processed.
 591         */
 592        struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
 593        const char *last_file = NULL;
 594        int last_len = 0;
 595        int i;
 596
 597        /*
 598         * If we're merging merge-bases, we don't want to bother with
 599         * any working directory changes.
 600         */
 601        if (o->call_depth)
 602                return;
 603
 604        /* Ensure D/F conflicts are adjacent in the entries list. */
 605        for (i = 0; i < entries->nr; i++) {
 606                struct string_list_item *next = &entries->items[i];
 607                string_list_append(&df_sorted_entries, next->string)->util =
 608                                   next->util;
 609        }
 610        df_sorted_entries.cmp = string_list_df_name_compare;
 611        string_list_sort(&df_sorted_entries);
 612
 613        string_list_clear(&o->df_conflict_file_set, 1);
 614        for (i = 0; i < df_sorted_entries.nr; i++) {
 615                const char *path = df_sorted_entries.items[i].string;
 616                int len = strlen(path);
 617                struct stage_data *e = df_sorted_entries.items[i].util;
 618
 619                /*
 620                 * Check if last_file & path correspond to a D/F conflict;
 621                 * i.e. whether path is last_file+'/'+<something>.
 622                 * If so, record that it's okay to remove last_file to make
 623                 * room for path and friends if needed.
 624                 */
 625                if (last_file &&
 626                    len > last_len &&
 627                    memcmp(path, last_file, last_len) == 0 &&
 628                    path[last_len] == '/') {
 629                        string_list_insert(&o->df_conflict_file_set, last_file);
 630                }
 631
 632                /*
 633                 * Determine whether path could exist as a file in the
 634                 * working directory as a possible D/F conflict.  This
 635                 * will only occur when it exists in stage 2 as a
 636                 * file.
 637                 */
 638                if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
 639                        last_file = path;
 640                        last_len = len;
 641                } else {
 642                        last_file = NULL;
 643                }
 644        }
 645        string_list_clear(&df_sorted_entries, 0);
 646}
 647
 648struct rename {
 649        struct diff_filepair *pair;
 650        /*
 651         * Purpose of src_entry and dst_entry:
 652         *
 653         * If 'before' is renamed to 'after' then src_entry will contain
 654         * the versions of 'before' from the merge_base, HEAD, and MERGE in
 655         * stages 1, 2, and 3; dst_entry will contain the respective
 656         * versions of 'after' in corresponding locations.  Thus, we have a
 657         * total of six modes and oids, though some will be null.  (Stage 0
 658         * is ignored; we're interested in handling conflicts.)
 659         *
 660         * Since we don't turn on break-rewrites by default, neither
 661         * src_entry nor dst_entry can have all three of their stages have
 662         * non-null oids, meaning at most four of the six will be non-null.
 663         * Also, since this is a rename, both src_entry and dst_entry will
 664         * have at least one non-null oid, meaning at least two will be
 665         * non-null.  Of the six oids, a typical rename will have three be
 666         * non-null.  Only two implies a rename/delete, and four implies a
 667         * rename/add.
 668         */
 669        struct stage_data *src_entry;
 670        struct stage_data *dst_entry;
 671        unsigned add_turned_into_rename:1;
 672        unsigned processed:1;
 673};
 674
 675static int update_stages(struct merge_options *opt, const char *path,
 676                         const struct diff_filespec *o,
 677                         const struct diff_filespec *a,
 678                         const struct diff_filespec *b)
 679{
 680
 681        /*
 682         * NOTE: It is usually a bad idea to call update_stages on a path
 683         * before calling update_file on that same path, since it can
 684         * sometimes lead to spurious "refusing to lose untracked file..."
 685         * messages from update_file (via make_room_for path via
 686         * would_lose_untracked).  Instead, reverse the order of the calls
 687         * (executing update_file first and then update_stages).
 688         */
 689        int clear = 1;
 690        int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
 691        if (clear)
 692                if (remove_file_from_index(opt->repo->index, path))
 693                        return -1;
 694        if (o)
 695                if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
 696                        return -1;
 697        if (a)
 698                if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
 699                        return -1;
 700        if (b)
 701                if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
 702                        return -1;
 703        return 0;
 704}
 705
 706static void update_entry(struct stage_data *entry,
 707                         struct diff_filespec *o,
 708                         struct diff_filespec *a,
 709                         struct diff_filespec *b)
 710{
 711        entry->processed = 0;
 712        entry->stages[1].mode = o->mode;
 713        entry->stages[2].mode = a->mode;
 714        entry->stages[3].mode = b->mode;
 715        oidcpy(&entry->stages[1].oid, &o->oid);
 716        oidcpy(&entry->stages[2].oid, &a->oid);
 717        oidcpy(&entry->stages[3].oid, &b->oid);
 718}
 719
 720static int remove_file(struct merge_options *o, int clean,
 721                       const char *path, int no_wd)
 722{
 723        int update_cache = o->call_depth || clean;
 724        int update_working_directory = !o->call_depth && !no_wd;
 725
 726        if (update_cache) {
 727                if (remove_file_from_index(o->repo->index, path))
 728                        return -1;
 729        }
 730        if (update_working_directory) {
 731                if (ignore_case) {
 732                        struct cache_entry *ce;
 733                        ce = index_file_exists(o->repo->index, path, strlen(path),
 734                                               ignore_case);
 735                        if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
 736                                return 0;
 737                }
 738                if (remove_path(path))
 739                        return -1;
 740        }
 741        return 0;
 742}
 743
 744/* add a string to a strbuf, but converting "/" to "_" */
 745static void add_flattened_path(struct strbuf *out, const char *s)
 746{
 747        size_t i = out->len;
 748        strbuf_addstr(out, s);
 749        for (; i < out->len; i++)
 750                if (out->buf[i] == '/')
 751                        out->buf[i] = '_';
 752}
 753
 754static char *unique_path(struct merge_options *o, const char *path, const char *branch)
 755{
 756        struct path_hashmap_entry *entry;
 757        struct strbuf newpath = STRBUF_INIT;
 758        int suffix = 0;
 759        size_t base_len;
 760
 761        strbuf_addf(&newpath, "%s~", path);
 762        add_flattened_path(&newpath, branch);
 763
 764        base_len = newpath.len;
 765        while (hashmap_get_from_hash(&o->current_file_dir_set,
 766                                     path_hash(newpath.buf), newpath.buf) ||
 767               (!o->call_depth && file_exists(newpath.buf))) {
 768                strbuf_setlen(&newpath, base_len);
 769                strbuf_addf(&newpath, "_%d", suffix++);
 770        }
 771
 772        FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
 773        hashmap_entry_init(entry, path_hash(entry->path));
 774        hashmap_add(&o->current_file_dir_set, entry);
 775        return strbuf_detach(&newpath, NULL);
 776}
 777
 778/**
 779 * Check whether a directory in the index is in the way of an incoming
 780 * file.  Return 1 if so.  If check_working_copy is non-zero, also
 781 * check the working directory.  If empty_ok is non-zero, also return
 782 * 0 in the case where the working-tree dir exists but is empty.
 783 */
 784static int dir_in_way(struct index_state *istate, const char *path,
 785                      int check_working_copy, int empty_ok)
 786{
 787        int pos;
 788        struct strbuf dirpath = STRBUF_INIT;
 789        struct stat st;
 790
 791        strbuf_addstr(&dirpath, path);
 792        strbuf_addch(&dirpath, '/');
 793
 794        pos = index_name_pos(istate, dirpath.buf, dirpath.len);
 795
 796        if (pos < 0)
 797                pos = -1 - pos;
 798        if (pos < istate->cache_nr &&
 799            !strncmp(dirpath.buf, istate->cache[pos]->name, dirpath.len)) {
 800                strbuf_release(&dirpath);
 801                return 1;
 802        }
 803
 804        strbuf_release(&dirpath);
 805        return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
 806                !(empty_ok && is_empty_dir(path));
 807}
 808
 809/*
 810 * Returns whether path was tracked in the index before the merge started,
 811 * and its oid and mode match the specified values
 812 */
 813static int was_tracked_and_matches(struct merge_options *o, const char *path,
 814                                   const struct object_id *oid, unsigned mode)
 815{
 816        int pos = index_name_pos(&o->orig_index, path, strlen(path));
 817        struct cache_entry *ce;
 818
 819        if (0 > pos)
 820                /* we were not tracking this path before the merge */
 821                return 0;
 822
 823        /* See if the file we were tracking before matches */
 824        ce = o->orig_index.cache[pos];
 825        return (oid_eq(&ce->oid, oid) && ce->ce_mode == mode);
 826}
 827
 828/*
 829 * Returns whether path was tracked in the index before the merge started
 830 */
 831static int was_tracked(struct merge_options *o, const char *path)
 832{
 833        int pos = index_name_pos(&o->orig_index, path, strlen(path));
 834
 835        if (0 <= pos)
 836                /* we were tracking this path before the merge */
 837                return 1;
 838
 839        return 0;
 840}
 841
 842static int would_lose_untracked(struct merge_options *o, const char *path)
 843{
 844        struct index_state *istate = o->repo->index;
 845
 846        /*
 847         * This may look like it can be simplified to:
 848         *   return !was_tracked(o, path) && file_exists(path)
 849         * but it can't.  This function needs to know whether path was in
 850         * the working tree due to EITHER having been tracked in the index
 851         * before the merge OR having been put into the working copy and
 852         * index by unpack_trees().  Due to that either-or requirement, we
 853         * check the current index instead of the original one.
 854         *
 855         * Note that we do not need to worry about merge-recursive itself
 856         * updating the index after unpack_trees() and before calling this
 857         * function, because we strictly require all code paths in
 858         * merge-recursive to update the working tree first and the index
 859         * second.  Doing otherwise would break
 860         * update_file()/would_lose_untracked(); see every comment in this
 861         * file which mentions "update_stages".
 862         */
 863        int pos = index_name_pos(istate, path, strlen(path));
 864
 865        if (pos < 0)
 866                pos = -1 - pos;
 867        while (pos < istate->cache_nr &&
 868               !strcmp(path, istate->cache[pos]->name)) {
 869                /*
 870                 * If stage #0, it is definitely tracked.
 871                 * If it has stage #2 then it was tracked
 872                 * before this merge started.  All other
 873                 * cases the path was not tracked.
 874                 */
 875                switch (ce_stage(istate->cache[pos])) {
 876                case 0:
 877                case 2:
 878                        return 0;
 879                }
 880                pos++;
 881        }
 882        return file_exists(path);
 883}
 884
 885static int was_dirty(struct merge_options *o, const char *path)
 886{
 887        struct cache_entry *ce;
 888        int dirty = 1;
 889
 890        if (o->call_depth || !was_tracked(o, path))
 891                return !dirty;
 892
 893        ce = index_file_exists(o->unpack_opts.src_index,
 894                               path, strlen(path), ignore_case);
 895        dirty = verify_uptodate(ce, &o->unpack_opts) != 0;
 896        return dirty;
 897}
 898
 899static int make_room_for_path(struct merge_options *o, const char *path)
 900{
 901        int status, i;
 902        const char *msg = _("failed to create path '%s'%s");
 903
 904        /* Unlink any D/F conflict files that are in the way */
 905        for (i = 0; i < o->df_conflict_file_set.nr; i++) {
 906                const char *df_path = o->df_conflict_file_set.items[i].string;
 907                size_t pathlen = strlen(path);
 908                size_t df_pathlen = strlen(df_path);
 909                if (df_pathlen < pathlen &&
 910                    path[df_pathlen] == '/' &&
 911                    strncmp(path, df_path, df_pathlen) == 0) {
 912                        output(o, 3,
 913                               _("Removing %s to make room for subdirectory\n"),
 914                               df_path);
 915                        unlink(df_path);
 916                        unsorted_string_list_delete_item(&o->df_conflict_file_set,
 917                                                         i, 0);
 918                        break;
 919                }
 920        }
 921
 922        /* Make sure leading directories are created */
 923        status = safe_create_leading_directories_const(path);
 924        if (status) {
 925                if (status == SCLD_EXISTS)
 926                        /* something else exists */
 927                        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 928                return err(o, msg, path, "");
 929        }
 930
 931        /*
 932         * Do not unlink a file in the work tree if we are not
 933         * tracking it.
 934         */
 935        if (would_lose_untracked(o, path))
 936                return err(o, _("refusing to lose untracked file at '%s'"),
 937                           path);
 938
 939        /* Successful unlink is good.. */
 940        if (!unlink(path))
 941                return 0;
 942        /* .. and so is no existing file */
 943        if (errno == ENOENT)
 944                return 0;
 945        /* .. but not some other error (who really cares what?) */
 946        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 947}
 948
 949static int update_file_flags(struct merge_options *o,
 950                             const struct object_id *oid,
 951                             unsigned mode,
 952                             const char *path,
 953                             int update_cache,
 954                             int update_wd)
 955{
 956        int ret = 0;
 957
 958        if (o->call_depth)
 959                update_wd = 0;
 960
 961        if (update_wd) {
 962                enum object_type type;
 963                void *buf;
 964                unsigned long size;
 965
 966                if (S_ISGITLINK(mode)) {
 967                        /*
 968                         * We may later decide to recursively descend into
 969                         * the submodule directory and update its index
 970                         * and/or work tree, but we do not do that now.
 971                         */
 972                        update_wd = 0;
 973                        goto update_index;
 974                }
 975
 976                buf = read_object_file(oid, &type, &size);
 977                if (!buf)
 978                        return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
 979                if (type != OBJ_BLOB) {
 980                        ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
 981                        goto free_buf;
 982                }
 983                if (S_ISREG(mode)) {
 984                        struct strbuf strbuf = STRBUF_INIT;
 985                        if (convert_to_working_tree(o->repo->index, path, buf, size, &strbuf)) {
 986                                free(buf);
 987                                size = strbuf.len;
 988                                buf = strbuf_detach(&strbuf, NULL);
 989                        }
 990                }
 991
 992                if (make_room_for_path(o, path) < 0) {
 993                        update_wd = 0;
 994                        goto free_buf;
 995                }
 996                if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
 997                        int fd;
 998                        if (mode & 0100)
 999                                mode = 0777;
1000                        else
1001                                mode = 0666;
1002                        fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
1003                        if (fd < 0) {
1004                                ret = err(o, _("failed to open '%s': %s"),
1005                                          path, strerror(errno));
1006                                goto free_buf;
1007                        }
1008                        write_in_full(fd, buf, size);
1009                        close(fd);
1010                } else if (S_ISLNK(mode)) {
1011                        char *lnk = xmemdupz(buf, size);
1012                        safe_create_leading_directories_const(path);
1013                        unlink(path);
1014                        if (symlink(lnk, path))
1015                                ret = err(o, _("failed to symlink '%s': %s"),
1016                                          path, strerror(errno));
1017                        free(lnk);
1018                } else
1019                        ret = err(o,
1020                                  _("do not know what to do with %06o %s '%s'"),
1021                                  mode, oid_to_hex(oid), path);
1022        free_buf:
1023                free(buf);
1024        }
1025update_index:
1026        if (!ret && update_cache)
1027                if (add_cacheinfo(o, mode, oid, path, 0, update_wd,
1028                                  ADD_CACHE_OK_TO_ADD))
1029                        return -1;
1030        return ret;
1031}
1032
1033static int update_file(struct merge_options *o,
1034                       int clean,
1035                       const struct object_id *oid,
1036                       unsigned mode,
1037                       const char *path)
1038{
1039        return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
1040}
1041
1042/* Low level file merging, update and removal */
1043
1044struct merge_file_info {
1045        struct object_id oid;
1046        unsigned mode;
1047        unsigned clean:1,
1048                 merge:1;
1049};
1050
1051static int merge_3way(struct merge_options *o,
1052                      mmbuffer_t *result_buf,
1053                      const struct diff_filespec *one,
1054                      const struct diff_filespec *a,
1055                      const struct diff_filespec *b,
1056                      const char *branch1,
1057                      const char *branch2,
1058                      const int extra_marker_size)
1059{
1060        mmfile_t orig, src1, src2;
1061        struct ll_merge_options ll_opts = {0};
1062        char *base_name, *name1, *name2;
1063        int merge_status;
1064
1065        ll_opts.renormalize = o->renormalize;
1066        ll_opts.extra_marker_size = extra_marker_size;
1067        ll_opts.xdl_opts = o->xdl_opts;
1068
1069        if (o->call_depth) {
1070                ll_opts.virtual_ancestor = 1;
1071                ll_opts.variant = 0;
1072        } else {
1073                switch (o->recursive_variant) {
1074                case MERGE_RECURSIVE_OURS:
1075                        ll_opts.variant = XDL_MERGE_FAVOR_OURS;
1076                        break;
1077                case MERGE_RECURSIVE_THEIRS:
1078                        ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
1079                        break;
1080                default:
1081                        ll_opts.variant = 0;
1082                        break;
1083                }
1084        }
1085
1086        if (strcmp(a->path, b->path) ||
1087            (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
1088                base_name = o->ancestor == NULL ? NULL :
1089                        mkpathdup("%s:%s", o->ancestor, one->path);
1090                name1 = mkpathdup("%s:%s", branch1, a->path);
1091                name2 = mkpathdup("%s:%s", branch2, b->path);
1092        } else {
1093                base_name = o->ancestor == NULL ? NULL :
1094                        mkpathdup("%s", o->ancestor);
1095                name1 = mkpathdup("%s", branch1);
1096                name2 = mkpathdup("%s", branch2);
1097        }
1098
1099        read_mmblob(&orig, &one->oid);
1100        read_mmblob(&src1, &a->oid);
1101        read_mmblob(&src2, &b->oid);
1102
1103        merge_status = ll_merge(result_buf, a->path, &orig, base_name,
1104                                &src1, name1, &src2, name2,
1105                                o->repo->index, &ll_opts);
1106
1107        free(base_name);
1108        free(name1);
1109        free(name2);
1110        free(orig.ptr);
1111        free(src1.ptr);
1112        free(src2.ptr);
1113        return merge_status;
1114}
1115
1116static int find_first_merges(struct repository *repo,
1117                             struct object_array *result, const char *path,
1118                             struct commit *a, struct commit *b)
1119{
1120        int i, j;
1121        struct object_array merges = OBJECT_ARRAY_INIT;
1122        struct commit *commit;
1123        int contains_another;
1124
1125        char merged_revision[42];
1126        const char *rev_args[] = { "rev-list", "--merges", "--ancestry-path",
1127                                   "--all", merged_revision, NULL };
1128        struct rev_info revs;
1129        struct setup_revision_opt rev_opts;
1130
1131        memset(result, 0, sizeof(struct object_array));
1132        memset(&rev_opts, 0, sizeof(rev_opts));
1133
1134        /* get all revisions that merge commit a */
1135        xsnprintf(merged_revision, sizeof(merged_revision), "^%s",
1136                  oid_to_hex(&a->object.oid));
1137        repo_init_revisions(repo, &revs, NULL);
1138        rev_opts.submodule = path;
1139        /* FIXME: can't handle linked worktrees in submodules yet */
1140        revs.single_worktree = path != NULL;
1141        setup_revisions(ARRAY_SIZE(rev_args)-1, rev_args, &revs, &rev_opts);
1142
1143        /* save all revisions from the above list that contain b */
1144        if (prepare_revision_walk(&revs))
1145                die("revision walk setup failed");
1146        while ((commit = get_revision(&revs)) != NULL) {
1147                struct object *o = &(commit->object);
1148                if (in_merge_bases(b, commit))
1149                        add_object_array(o, NULL, &merges);
1150        }
1151        reset_revision_walk();
1152
1153        /* Now we've got all merges that contain a and b. Prune all
1154         * merges that contain another found merge and save them in
1155         * result.
1156         */
1157        for (i = 0; i < merges.nr; i++) {
1158                struct commit *m1 = (struct commit *) merges.objects[i].item;
1159
1160                contains_another = 0;
1161                for (j = 0; j < merges.nr; j++) {
1162                        struct commit *m2 = (struct commit *) merges.objects[j].item;
1163                        if (i != j && in_merge_bases(m2, m1)) {
1164                                contains_another = 1;
1165                                break;
1166                        }
1167                }
1168
1169                if (!contains_another)
1170                        add_object_array(merges.objects[i].item, NULL, result);
1171        }
1172
1173        object_array_clear(&merges);
1174        return result->nr;
1175}
1176
1177static void print_commit(struct commit *commit)
1178{
1179        struct strbuf sb = STRBUF_INIT;
1180        struct pretty_print_context ctx = {0};
1181        ctx.date_mode.type = DATE_NORMAL;
1182        format_commit_message(commit, " %h: %m %s", &sb, &ctx);
1183        fprintf(stderr, "%s\n", sb.buf);
1184        strbuf_release(&sb);
1185}
1186
1187static int merge_submodule(struct merge_options *o,
1188                           struct object_id *result, const char *path,
1189                           const struct object_id *base, const struct object_id *a,
1190                           const struct object_id *b)
1191{
1192        struct commit *commit_base, *commit_a, *commit_b;
1193        int parent_count;
1194        struct object_array merges;
1195
1196        int i;
1197        int search = !o->call_depth;
1198
1199        /* store a in result in case we fail */
1200        oidcpy(result, a);
1201
1202        /* we can not handle deletion conflicts */
1203        if (is_null_oid(base))
1204                return 0;
1205        if (is_null_oid(a))
1206                return 0;
1207        if (is_null_oid(b))
1208                return 0;
1209
1210        if (add_submodule_odb(path)) {
1211                output(o, 1, _("Failed to merge submodule %s (not checked out)"), path);
1212                return 0;
1213        }
1214
1215        if (!(commit_base = lookup_commit_reference(o->repo, base)) ||
1216            !(commit_a = lookup_commit_reference(o->repo, a)) ||
1217            !(commit_b = lookup_commit_reference(o->repo, b))) {
1218                output(o, 1, _("Failed to merge submodule %s (commits not present)"), path);
1219                return 0;
1220        }
1221
1222        /* check whether both changes are forward */
1223        if (!in_merge_bases(commit_base, commit_a) ||
1224            !in_merge_bases(commit_base, commit_b)) {
1225                output(o, 1, _("Failed to merge submodule %s (commits don't follow merge-base)"), path);
1226                return 0;
1227        }
1228
1229        /* Case #1: a is contained in b or vice versa */
1230        if (in_merge_bases(commit_a, commit_b)) {
1231                oidcpy(result, b);
1232                if (show(o, 3)) {
1233                        output(o, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1234                        output_commit_title(o, commit_b);
1235                } else if (show(o, 2))
1236                        output(o, 2, _("Fast-forwarding submodule %s"), path);
1237                else
1238                        ; /* no output */
1239
1240                return 1;
1241        }
1242        if (in_merge_bases(commit_b, commit_a)) {
1243                oidcpy(result, a);
1244                if (show(o, 3)) {
1245                        output(o, 3, _("Fast-forwarding submodule %s to the following commit:"), path);
1246                        output_commit_title(o, commit_a);
1247                } else if (show(o, 2))
1248                        output(o, 2, _("Fast-forwarding submodule %s"), path);
1249                else
1250                        ; /* no output */
1251
1252                return 1;
1253        }
1254
1255        /*
1256         * Case #2: There are one or more merges that contain a and b in
1257         * the submodule. If there is only one, then present it as a
1258         * suggestion to the user, but leave it marked unmerged so the
1259         * user needs to confirm the resolution.
1260         */
1261
1262        /* Skip the search if makes no sense to the calling context.  */
1263        if (!search)
1264                return 0;
1265
1266        /* find commit which merges them */
1267        parent_count = find_first_merges(o->repo, &merges, path,
1268                                         commit_a, commit_b);
1269        switch (parent_count) {
1270        case 0:
1271                output(o, 1, _("Failed to merge submodule %s (merge following commits not found)"), path);
1272                break;
1273
1274        case 1:
1275                output(o, 1, _("Failed to merge submodule %s (not fast-forward)"), path);
1276                output(o, 2, _("Found a possible merge resolution for the submodule:\n"));
1277                print_commit((struct commit *) merges.objects[0].item);
1278                output(o, 2, _(
1279                       "If this is correct simply add it to the index "
1280                       "for example\n"
1281                       "by using:\n\n"
1282                       "  git update-index --cacheinfo 160000 %s \"%s\"\n\n"
1283                       "which will accept this suggestion.\n"),
1284                       oid_to_hex(&merges.objects[0].item->oid), path);
1285                break;
1286
1287        default:
1288                output(o, 1, _("Failed to merge submodule %s (multiple merges found)"), path);
1289                for (i = 0; i < merges.nr; i++)
1290                        print_commit((struct commit *) merges.objects[i].item);
1291        }
1292
1293        object_array_clear(&merges);
1294        return 0;
1295}
1296
1297static int merge_mode_and_contents(struct merge_options *o,
1298                                   const struct diff_filespec *one,
1299                                   const struct diff_filespec *a,
1300                                   const struct diff_filespec *b,
1301                                   const char *filename,
1302                                   const char *branch1,
1303                                   const char *branch2,
1304                                   const int extra_marker_size,
1305                                   struct merge_file_info *result)
1306{
1307        if (o->branch1 != branch1) {
1308                /*
1309                 * It's weird getting a reverse merge with HEAD on the bottom
1310                 * side of the conflict markers and the other branch on the
1311                 * top.  Fix that.
1312                 */
1313                return merge_mode_and_contents(o, one, b, a,
1314                                               filename,
1315                                               branch2, branch1,
1316                                               extra_marker_size, result);
1317        }
1318
1319        result->merge = 0;
1320        result->clean = 1;
1321
1322        if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1323                result->clean = 0;
1324                if (S_ISREG(a->mode)) {
1325                        result->mode = a->mode;
1326                        oidcpy(&result->oid, &a->oid);
1327                } else {
1328                        result->mode = b->mode;
1329                        oidcpy(&result->oid, &b->oid);
1330                }
1331        } else {
1332                if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1333                        result->merge = 1;
1334
1335                /*
1336                 * Merge modes
1337                 */
1338                if (a->mode == b->mode || a->mode == one->mode)
1339                        result->mode = b->mode;
1340                else {
1341                        result->mode = a->mode;
1342                        if (b->mode != one->mode) {
1343                                result->clean = 0;
1344                                result->merge = 1;
1345                        }
1346                }
1347
1348                if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1349                        oidcpy(&result->oid, &b->oid);
1350                else if (oid_eq(&b->oid, &one->oid))
1351                        oidcpy(&result->oid, &a->oid);
1352                else if (S_ISREG(a->mode)) {
1353                        mmbuffer_t result_buf;
1354                        int ret = 0, merge_status;
1355
1356                        merge_status = merge_3way(o, &result_buf, one, a, b,
1357                                                  branch1, branch2,
1358                                                  extra_marker_size);
1359
1360                        if ((merge_status < 0) || !result_buf.ptr)
1361                                ret = err(o, _("Failed to execute internal merge"));
1362
1363                        if (!ret &&
1364                            write_object_file(result_buf.ptr, result_buf.size,
1365                                              blob_type, &result->oid))
1366                                ret = err(o, _("Unable to add %s to database"),
1367                                          a->path);
1368
1369                        free(result_buf.ptr);
1370                        if (ret)
1371                                return ret;
1372                        result->clean = (merge_status == 0);
1373                } else if (S_ISGITLINK(a->mode)) {
1374                        result->clean = merge_submodule(o, &result->oid,
1375                                                        one->path,
1376                                                        &one->oid,
1377                                                        &a->oid,
1378                                                        &b->oid);
1379                } else if (S_ISLNK(a->mode)) {
1380                        switch (o->recursive_variant) {
1381                        case MERGE_RECURSIVE_NORMAL:
1382                                oidcpy(&result->oid, &a->oid);
1383                                if (!oid_eq(&a->oid, &b->oid))
1384                                        result->clean = 0;
1385                                break;
1386                        case MERGE_RECURSIVE_OURS:
1387                                oidcpy(&result->oid, &a->oid);
1388                                break;
1389                        case MERGE_RECURSIVE_THEIRS:
1390                                oidcpy(&result->oid, &b->oid);
1391                                break;
1392                        }
1393                } else
1394                        BUG("unsupported object type in the tree");
1395        }
1396
1397        if (result->merge)
1398                output(o, 2, _("Auto-merging %s"), filename);
1399
1400        return 0;
1401}
1402
1403static int handle_rename_via_dir(struct merge_options *o,
1404                                 struct diff_filepair *pair,
1405                                 const char *rename_branch,
1406                                 const char *other_branch)
1407{
1408        /*
1409         * Handle file adds that need to be renamed due to directory rename
1410         * detection.  This differs from handle_rename_normal, because
1411         * there is no content merge to do; just move the file into the
1412         * desired final location.
1413         */
1414        const struct diff_filespec *dest = pair->two;
1415
1416        if (!o->call_depth && would_lose_untracked(o, dest->path)) {
1417                char *alt_path = unique_path(o, dest->path, rename_branch);
1418
1419                output(o, 1, _("Error: Refusing to lose untracked file at %s; "
1420                               "writing to %s instead."),
1421                       dest->path, alt_path);
1422                /*
1423                 * Write the file in worktree at alt_path, but not in the
1424                 * index.  Instead, write to dest->path for the index but
1425                 * only at the higher appropriate stage.
1426                 */
1427                if (update_file(o, 0, &dest->oid, dest->mode, alt_path))
1428                        return -1;
1429                free(alt_path);
1430                return update_stages(o, dest->path, NULL,
1431                                     rename_branch == o->branch1 ? dest : NULL,
1432                                     rename_branch == o->branch1 ? NULL : dest);
1433        }
1434
1435        /* Update dest->path both in index and in worktree */
1436        if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1437                return -1;
1438        return 0;
1439}
1440
1441static int handle_change_delete(struct merge_options *o,
1442                                const char *path, const char *old_path,
1443                                const struct object_id *o_oid, int o_mode,
1444                                const struct object_id *changed_oid,
1445                                int changed_mode,
1446                                const char *change_branch,
1447                                const char *delete_branch,
1448                                const char *change, const char *change_past)
1449{
1450        char *alt_path = NULL;
1451        const char *update_path = path;
1452        int ret = 0;
1453
1454        if (dir_in_way(o->repo->index, path, !o->call_depth, 0) ||
1455            (!o->call_depth && would_lose_untracked(o, path))) {
1456                update_path = alt_path = unique_path(o, path, change_branch);
1457        }
1458
1459        if (o->call_depth) {
1460                /*
1461                 * We cannot arbitrarily accept either a_sha or b_sha as
1462                 * correct; since there is no true "middle point" between
1463                 * them, simply reuse the base version for virtual merge base.
1464                 */
1465                ret = remove_file_from_index(o->repo->index, path);
1466                if (!ret)
1467                        ret = update_file(o, 0, o_oid, o_mode, update_path);
1468        } else {
1469                /*
1470                 * Despite the four nearly duplicate messages and argument
1471                 * lists below and the ugliness of the nested if-statements,
1472                 * having complete messages makes the job easier for
1473                 * translators.
1474                 *
1475                 * The slight variance among the cases is due to the fact
1476                 * that:
1477                 *   1) directory/file conflicts (in effect if
1478                 *      !alt_path) could cause us to need to write the
1479                 *      file to a different path.
1480                 *   2) renames (in effect if !old_path) could mean that
1481                 *      there are two names for the path that the user
1482                 *      may know the file by.
1483                 */
1484                if (!alt_path) {
1485                        if (!old_path) {
1486                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1487                                       "and %s in %s. Version %s of %s left in tree."),
1488                                       change, path, delete_branch, change_past,
1489                                       change_branch, change_branch, path);
1490                        } else {
1491                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1492                                       "and %s to %s in %s. Version %s of %s left in tree."),
1493                                       change, old_path, delete_branch, change_past, path,
1494                                       change_branch, change_branch, path);
1495                        }
1496                } else {
1497                        if (!old_path) {
1498                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1499                                       "and %s in %s. Version %s of %s left in tree at %s."),
1500                                       change, path, delete_branch, change_past,
1501                                       change_branch, change_branch, path, alt_path);
1502                        } else {
1503                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1504                                       "and %s to %s in %s. Version %s of %s left in tree at %s."),
1505                                       change, old_path, delete_branch, change_past, path,
1506                                       change_branch, change_branch, path, alt_path);
1507                        }
1508                }
1509                /*
1510                 * No need to call update_file() on path when change_branch ==
1511                 * o->branch1 && !alt_path, since that would needlessly touch
1512                 * path.  We could call update_file_flags() with update_cache=0
1513                 * and update_wd=0, but that's a no-op.
1514                 */
1515                if (change_branch != o->branch1 || alt_path)
1516                        ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1517        }
1518        free(alt_path);
1519
1520        return ret;
1521}
1522
1523static int handle_rename_delete(struct merge_options *o,
1524                                struct diff_filepair *pair,
1525                                const char *rename_branch,
1526                                const char *delete_branch)
1527{
1528        const struct diff_filespec *orig = pair->one;
1529        const struct diff_filespec *dest = pair->two;
1530
1531        if (handle_change_delete(o,
1532                                 o->call_depth ? orig->path : dest->path,
1533                                 o->call_depth ? NULL : orig->path,
1534                                 &orig->oid, orig->mode,
1535                                 &dest->oid, dest->mode,
1536                                 rename_branch, delete_branch,
1537                                 _("rename"), _("renamed")))
1538                return -1;
1539
1540        if (o->call_depth)
1541                return remove_file_from_index(o->repo->index, dest->path);
1542        else
1543                return update_stages(o, dest->path, NULL,
1544                                     rename_branch == o->branch1 ? dest : NULL,
1545                                     rename_branch == o->branch1 ? NULL : dest);
1546}
1547
1548static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1549                                                 struct stage_data *entry,
1550                                                 int stage)
1551{
1552        struct object_id *oid = &entry->stages[stage].oid;
1553        unsigned mode = entry->stages[stage].mode;
1554        if (mode == 0 || is_null_oid(oid))
1555                return NULL;
1556        oidcpy(&target->oid, oid);
1557        target->mode = mode;
1558        return target;
1559}
1560
1561static int handle_file_collision(struct merge_options *o,
1562                                 const char *collide_path,
1563                                 const char *prev_path1,
1564                                 const char *prev_path2,
1565                                 const char *branch1, const char *branch2,
1566                                 const struct object_id *a_oid,
1567                                 unsigned int a_mode,
1568                                 const struct object_id *b_oid,
1569                                 unsigned int b_mode)
1570{
1571        struct merge_file_info mfi;
1572        struct diff_filespec null, a, b;
1573        char *alt_path = NULL;
1574        const char *update_path = collide_path;
1575
1576        /*
1577         * It's easiest to get the correct things into stage 2 and 3, and
1578         * to make sure that the content merge puts HEAD before the other
1579         * branch if we just ensure that branch1 == o->branch1.  So, simply
1580         * flip arguments around if we don't have that.
1581         */
1582        if (branch1 != o->branch1) {
1583                return handle_file_collision(o, collide_path,
1584                                             prev_path2, prev_path1,
1585                                             branch2, branch1,
1586                                             b_oid, b_mode,
1587                                             a_oid, a_mode);
1588        }
1589
1590        /*
1591         * In the recursive case, we just opt to undo renames
1592         */
1593        if (o->call_depth && (prev_path1 || prev_path2)) {
1594                /* Put first file (a_oid, a_mode) in its original spot */
1595                if (prev_path1) {
1596                        if (update_file(o, 1, a_oid, a_mode, prev_path1))
1597                                return -1;
1598                } else {
1599                        if (update_file(o, 1, a_oid, a_mode, collide_path))
1600                                return -1;
1601                }
1602
1603                /* Put second file (b_oid, b_mode) in its original spot */
1604                if (prev_path2) {
1605                        if (update_file(o, 1, b_oid, b_mode, prev_path2))
1606                                return -1;
1607                } else {
1608                        if (update_file(o, 1, b_oid, b_mode, collide_path))
1609                                return -1;
1610                }
1611
1612                /* Don't leave something at collision path if unrenaming both */
1613                if (prev_path1 && prev_path2)
1614                        remove_file(o, 1, collide_path, 0);
1615
1616                return 0;
1617        }
1618
1619        /* Remove rename sources if rename/add or rename/rename(2to1) */
1620        if (prev_path1)
1621                remove_file(o, 1, prev_path1,
1622                            o->call_depth || would_lose_untracked(o, prev_path1));
1623        if (prev_path2)
1624                remove_file(o, 1, prev_path2,
1625                            o->call_depth || would_lose_untracked(o, prev_path2));
1626
1627        /*
1628         * Remove the collision path, if it wouldn't cause dirty contents
1629         * or an untracked file to get lost.  We'll either overwrite with
1630         * merged contents, or just write out to differently named files.
1631         */
1632        if (was_dirty(o, collide_path)) {
1633                output(o, 1, _("Refusing to lose dirty file at %s"),
1634                       collide_path);
1635                update_path = alt_path = unique_path(o, collide_path, "merged");
1636        } else if (would_lose_untracked(o, collide_path)) {
1637                /*
1638                 * Only way we get here is if both renames were from
1639                 * a directory rename AND user had an untracked file
1640                 * at the location where both files end up after the
1641                 * two directory renames.  See testcase 10d of t6043.
1642                 */
1643                output(o, 1, _("Refusing to lose untracked file at "
1644                               "%s, even though it's in the way."),
1645                       collide_path);
1646                update_path = alt_path = unique_path(o, collide_path, "merged");
1647        } else {
1648                /*
1649                 * FIXME: It's possible that the two files are identical
1650                 * and that the current working copy happens to match, in
1651                 * which case we are unnecessarily touching the working
1652                 * tree file.  It's not a likely enough scenario that I
1653                 * want to code up the checks for it and a better fix is
1654                 * available if we restructure how unpack_trees() and
1655                 * merge-recursive interoperate anyway, so punting for
1656                 * now...
1657                 */
1658                remove_file(o, 0, collide_path, 0);
1659        }
1660
1661        /* Store things in diff_filespecs for functions that need it */
1662        memset(&a, 0, sizeof(struct diff_filespec));
1663        memset(&b, 0, sizeof(struct diff_filespec));
1664        null.path = a.path = b.path = (char *)collide_path;
1665        oidcpy(&null.oid, &null_oid);
1666        null.mode = 0;
1667        oidcpy(&a.oid, a_oid);
1668        a.mode = a_mode;
1669        a.oid_valid = 1;
1670        oidcpy(&b.oid, b_oid);
1671        b.mode = b_mode;
1672        b.oid_valid = 1;
1673
1674        if (merge_mode_and_contents(o, &null, &a, &b, collide_path,
1675                                    branch1, branch2, o->call_depth * 2, &mfi))
1676                return -1;
1677        mfi.clean &= !alt_path;
1678        if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, update_path))
1679                return -1;
1680        if (!mfi.clean && !o->call_depth &&
1681            update_stages(o, collide_path, NULL, &a, &b))
1682                return -1;
1683        free(alt_path);
1684        /*
1685         * FIXME: If both a & b both started with conflicts (only possible
1686         * if they came from a rename/rename(2to1)), but had IDENTICAL
1687         * contents including those conflicts, then in the next line we claim
1688         * it was clean.  If someone cares about this case, we should have the
1689         * caller notify us if we started with conflicts.
1690         */
1691        return mfi.clean;
1692}
1693
1694static int handle_rename_add(struct merge_options *o,
1695                             struct rename_conflict_info *ci)
1696{
1697        /* a was renamed to c, and a separate c was added. */
1698        struct diff_filespec *a = ci->pair1->one;
1699        struct diff_filespec *c = ci->pair1->two;
1700        char *path = c->path;
1701        char *prev_path_desc;
1702        struct merge_file_info mfi;
1703
1704        int other_stage = (ci->branch1 == o->branch1 ? 3 : 2);
1705
1706        output(o, 1, _("CONFLICT (rename/add): "
1707               "Rename %s->%s in %s.  Added %s in %s"),
1708               a->path, c->path, ci->branch1,
1709               c->path, ci->branch2);
1710
1711        prev_path_desc = xstrfmt("version of %s from %s", path, a->path);
1712        if (merge_mode_and_contents(o, a, c, &ci->ren1_other, prev_path_desc,
1713                                    o->branch1, o->branch2,
1714                                    1 + o->call_depth * 2, &mfi))
1715                return -1;
1716        free(prev_path_desc);
1717
1718        return handle_file_collision(o,
1719                                     c->path, a->path, NULL,
1720                                     ci->branch1, ci->branch2,
1721                                     &mfi.oid, mfi.mode,
1722                                     &ci->dst_entry1->stages[other_stage].oid,
1723                                     ci->dst_entry1->stages[other_stage].mode);
1724}
1725
1726static char *find_path_for_conflict(struct merge_options *o,
1727                                    const char *path,
1728                                    const char *branch1,
1729                                    const char *branch2)
1730{
1731        char *new_path = NULL;
1732        if (dir_in_way(o->repo->index, path, !o->call_depth, 0)) {
1733                new_path = unique_path(o, path, branch1);
1734                output(o, 1, _("%s is a directory in %s adding "
1735                               "as %s instead"),
1736                       path, branch2, new_path);
1737        } else if (would_lose_untracked(o, path)) {
1738                new_path = unique_path(o, path, branch1);
1739                output(o, 1, _("Refusing to lose untracked file"
1740                               " at %s; adding as %s instead"),
1741                       path, new_path);
1742        }
1743
1744        return new_path;
1745}
1746
1747static int handle_rename_rename_1to2(struct merge_options *o,
1748                                     struct rename_conflict_info *ci)
1749{
1750        /* One file was renamed in both branches, but to different names. */
1751        struct merge_file_info mfi;
1752        struct diff_filespec other;
1753        struct diff_filespec *add;
1754        struct diff_filespec *one = ci->pair1->one;
1755        struct diff_filespec *a = ci->pair1->two;
1756        struct diff_filespec *b = ci->pair2->two;
1757        char *path_desc;
1758
1759        output(o, 1, _("CONFLICT (rename/rename): "
1760               "Rename \"%s\"->\"%s\" in branch \"%s\" "
1761               "rename \"%s\"->\"%s\" in \"%s\"%s"),
1762               one->path, a->path, ci->branch1,
1763               one->path, b->path, ci->branch2,
1764               o->call_depth ? _(" (left unresolved)") : "");
1765
1766        path_desc = xstrfmt("%s and %s, both renamed from %s",
1767                            a->path, b->path, one->path);
1768        if (merge_mode_and_contents(o, one, a, b, path_desc,
1769                                    ci->branch1, ci->branch2,
1770                                    o->call_depth * 2, &mfi))
1771                return -1;
1772        free(path_desc);
1773
1774        if (o->call_depth) {
1775                /*
1776                 * FIXME: For rename/add-source conflicts (if we could detect
1777                 * such), this is wrong.  We should instead find a unique
1778                 * pathname and then either rename the add-source file to that
1779                 * unique path, or use that unique path instead of src here.
1780                 */
1781                if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1782                        return -1;
1783
1784                /*
1785                 * Above, we put the merged content at the merge-base's
1786                 * path.  Now we usually need to delete both a->path and
1787                 * b->path.  However, the rename on each side of the merge
1788                 * could also be involved in a rename/add conflict.  In
1789                 * such cases, we should keep the added file around,
1790                 * resolving the conflict at that path in its favor.
1791                 */
1792                add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1793                if (add) {
1794                        if (update_file(o, 0, &add->oid, add->mode, a->path))
1795                                return -1;
1796                }
1797                else
1798                        remove_file_from_index(o->repo->index, a->path);
1799                add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1800                if (add) {
1801                        if (update_file(o, 0, &add->oid, add->mode, b->path))
1802                                return -1;
1803                }
1804                else
1805                        remove_file_from_index(o->repo->index, b->path);
1806        } else {
1807                /*
1808                 * For each destination path, we need to see if there is a
1809                 * rename/add collision.  If not, we can write the file out
1810                 * to the specified location.
1811                 */
1812                add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1813                if (add) {
1814                        if (handle_file_collision(o, a->path,
1815                                                  NULL, NULL,
1816                                                  ci->branch1, ci->branch2,
1817                                                  &mfi.oid, mfi.mode,
1818                                                  &add->oid, add->mode) < 0)
1819                                return -1;
1820                } else {
1821                        char *new_path = find_path_for_conflict(o, a->path,
1822                                                                ci->branch1,
1823                                                                ci->branch2);
1824                        if (update_file(o, 0, &mfi.oid, mfi.mode, new_path ? new_path : a->path))
1825                                return -1;
1826                        free(new_path);
1827                        if (update_stages(o, a->path, NULL, a, NULL))
1828                                return -1;
1829                }
1830
1831                add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1832                if (add) {
1833                        if (handle_file_collision(o, b->path,
1834                                                  NULL, NULL,
1835                                                  ci->branch1, ci->branch2,
1836                                                  &add->oid, add->mode,
1837                                                  &mfi.oid, mfi.mode) < 0)
1838                                return -1;
1839                } else {
1840                        char *new_path = find_path_for_conflict(o, b->path,
1841                                                                ci->branch2,
1842                                                                ci->branch1);
1843                        if (update_file(o, 0, &mfi.oid, mfi.mode, new_path ? new_path : b->path))
1844                                return -1;
1845                        free(new_path);
1846                        if (update_stages(o, b->path, NULL, NULL, b))
1847                                return -1;
1848                }
1849        }
1850
1851        return 0;
1852}
1853
1854static int handle_rename_rename_2to1(struct merge_options *o,
1855                                     struct rename_conflict_info *ci)
1856{
1857        /* Two files, a & b, were renamed to the same thing, c. */
1858        struct diff_filespec *a = ci->pair1->one;
1859        struct diff_filespec *b = ci->pair2->one;
1860        struct diff_filespec *c1 = ci->pair1->two;
1861        struct diff_filespec *c2 = ci->pair2->two;
1862        char *path = c1->path; /* == c2->path */
1863        char *path_side_1_desc;
1864        char *path_side_2_desc;
1865        struct merge_file_info mfi_c1;
1866        struct merge_file_info mfi_c2;
1867
1868        output(o, 1, _("CONFLICT (rename/rename): "
1869               "Rename %s->%s in %s. "
1870               "Rename %s->%s in %s"),
1871               a->path, c1->path, ci->branch1,
1872               b->path, c2->path, ci->branch2);
1873
1874        path_side_1_desc = xstrfmt("version of %s from %s", path, a->path);
1875        path_side_2_desc = xstrfmt("version of %s from %s", path, b->path);
1876        if (merge_mode_and_contents(o, a, c1, &ci->ren1_other, path_side_1_desc,
1877                                    o->branch1, o->branch2,
1878                                    1 + o->call_depth * 2, &mfi_c1) ||
1879            merge_mode_and_contents(o, b, &ci->ren2_other, c2, path_side_2_desc,
1880                                    o->branch1, o->branch2,
1881                                    1 + o->call_depth * 2, &mfi_c2))
1882                return -1;
1883        free(path_side_1_desc);
1884        free(path_side_2_desc);
1885
1886        return handle_file_collision(o, path, a->path, b->path,
1887                                     ci->branch1, ci->branch2,
1888                                     &mfi_c1.oid, mfi_c1.mode,
1889                                     &mfi_c2.oid, mfi_c2.mode);
1890}
1891
1892/*
1893 * Get the diff_filepairs changed between o_tree and tree.
1894 */
1895static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1896                                               struct tree *o_tree,
1897                                               struct tree *tree)
1898{
1899        struct diff_queue_struct *ret;
1900        struct diff_options opts;
1901
1902        repo_diff_setup(o->repo, &opts);
1903        opts.flags.recursive = 1;
1904        opts.flags.rename_empty = 0;
1905        opts.detect_rename = merge_detect_rename(o);
1906        /*
1907         * We do not have logic to handle the detection of copies.  In
1908         * fact, it may not even make sense to add such logic: would we
1909         * really want a change to a base file to be propagated through
1910         * multiple other files by a merge?
1911         */
1912        if (opts.detect_rename > DIFF_DETECT_RENAME)
1913                opts.detect_rename = DIFF_DETECT_RENAME;
1914        opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1915                            o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1916                            1000;
1917        opts.rename_score = o->rename_score;
1918        opts.show_rename_progress = o->show_rename_progress;
1919        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1920        diff_setup_done(&opts);
1921        diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1922        diffcore_std(&opts);
1923        if (opts.needed_rename_limit > o->needed_rename_limit)
1924                o->needed_rename_limit = opts.needed_rename_limit;
1925
1926        ret = xmalloc(sizeof(*ret));
1927        *ret = diff_queued_diff;
1928
1929        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1930        diff_queued_diff.nr = 0;
1931        diff_queued_diff.queue = NULL;
1932        diff_flush(&opts);
1933        return ret;
1934}
1935
1936static int tree_has_path(struct tree *tree, const char *path)
1937{
1938        struct object_id hashy;
1939        unsigned int mode_o;
1940
1941        return !get_tree_entry(&tree->object.oid, path,
1942                               &hashy, &mode_o);
1943}
1944
1945/*
1946 * Return a new string that replaces the beginning portion (which matches
1947 * entry->dir), with entry->new_dir.  In perl-speak:
1948 *   new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1949 * NOTE:
1950 *   Caller must ensure that old_path starts with entry->dir + '/'.
1951 */
1952static char *apply_dir_rename(struct dir_rename_entry *entry,
1953                              const char *old_path)
1954{
1955        struct strbuf new_path = STRBUF_INIT;
1956        int oldlen, newlen;
1957
1958        if (entry->non_unique_new_dir)
1959                return NULL;
1960
1961        oldlen = strlen(entry->dir);
1962        newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1963        strbuf_grow(&new_path, newlen);
1964        strbuf_addbuf(&new_path, &entry->new_dir);
1965        strbuf_addstr(&new_path, &old_path[oldlen]);
1966
1967        return strbuf_detach(&new_path, NULL);
1968}
1969
1970static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1971                                    char **old_dir, char **new_dir)
1972{
1973        char *end_of_old, *end_of_new;
1974        int old_len, new_len;
1975
1976        *old_dir = NULL;
1977        *new_dir = NULL;
1978
1979        /*
1980         * For
1981         *    "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1982         * the "e/foo.c" part is the same, we just want to know that
1983         *    "a/b/c/d" was renamed to "a/b/some/thing/else"
1984         * so, for this example, this function returns "a/b/c/d" in
1985         * *old_dir and "a/b/some/thing/else" in *new_dir.
1986         *
1987         * Also, if the basename of the file changed, we don't care.  We
1988         * want to know which portion of the directory, if any, changed.
1989         */
1990        end_of_old = strrchr(old_path, '/');
1991        end_of_new = strrchr(new_path, '/');
1992
1993        if (end_of_old == NULL || end_of_new == NULL)
1994                return;
1995        while (*--end_of_new == *--end_of_old &&
1996               end_of_old != old_path &&
1997               end_of_new != new_path)
1998                ; /* Do nothing; all in the while loop */
1999        /*
2000         * We've found the first non-matching character in the directory
2001         * paths.  That means the current directory we were comparing
2002         * represents the rename.  Move end_of_old and end_of_new back
2003         * to the full directory name.
2004         */
2005        if (*end_of_old == '/')
2006                end_of_old++;
2007        if (*end_of_old != '/')
2008                end_of_new++;
2009        end_of_old = strchr(end_of_old, '/');
2010        end_of_new = strchr(end_of_new, '/');
2011
2012        /*
2013         * It may have been the case that old_path and new_path were the same
2014         * directory all along.  Don't claim a rename if they're the same.
2015         */
2016        old_len = end_of_old - old_path;
2017        new_len = end_of_new - new_path;
2018
2019        if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
2020                *old_dir = xstrndup(old_path, old_len);
2021                *new_dir = xstrndup(new_path, new_len);
2022        }
2023}
2024
2025static void remove_hashmap_entries(struct hashmap *dir_renames,
2026                                   struct string_list *items_to_remove)
2027{
2028        int i;
2029        struct dir_rename_entry *entry;
2030
2031        for (i = 0; i < items_to_remove->nr; i++) {
2032                entry = items_to_remove->items[i].util;
2033                hashmap_remove(dir_renames, entry, NULL);
2034        }
2035        string_list_clear(items_to_remove, 0);
2036}
2037
2038/*
2039 * See if there is a directory rename for path, and if there are any file
2040 * level conflicts for the renamed location.  If there is a rename and
2041 * there are no conflicts, return the new name.  Otherwise, return NULL.
2042 */
2043static char *handle_path_level_conflicts(struct merge_options *o,
2044                                         const char *path,
2045                                         struct dir_rename_entry *entry,
2046                                         struct hashmap *collisions,
2047                                         struct tree *tree)
2048{
2049        char *new_path = NULL;
2050        struct collision_entry *collision_ent;
2051        int clean = 1;
2052        struct strbuf collision_paths = STRBUF_INIT;
2053
2054        /*
2055         * entry has the mapping of old directory name to new directory name
2056         * that we want to apply to path.
2057         */
2058        new_path = apply_dir_rename(entry, path);
2059
2060        if (!new_path) {
2061                /* This should only happen when entry->non_unique_new_dir set */
2062                if (!entry->non_unique_new_dir)
2063                        BUG("entry->non_unqiue_dir not set and !new_path");
2064                output(o, 1, _("CONFLICT (directory rename split): "
2065                               "Unclear where to place %s because directory "
2066                               "%s was renamed to multiple other directories, "
2067                               "with no destination getting a majority of the "
2068                               "files."),
2069                       path, entry->dir);
2070                clean = 0;
2071                return NULL;
2072        }
2073
2074        /*
2075         * The caller needs to have ensured that it has pre-populated
2076         * collisions with all paths that map to new_path.  Do a quick check
2077         * to ensure that's the case.
2078         */
2079        collision_ent = collision_find_entry(collisions, new_path);
2080        if (collision_ent == NULL)
2081                BUG("collision_ent is NULL");
2082
2083        /*
2084         * Check for one-sided add/add/.../add conflicts, i.e.
2085         * where implicit renames from the other side doing
2086         * directory rename(s) can affect this side of history
2087         * to put multiple paths into the same location.  Warn
2088         * and bail on directory renames for such paths.
2089         */
2090        if (collision_ent->reported_already) {
2091                clean = 0;
2092        } else if (tree_has_path(tree, new_path)) {
2093                collision_ent->reported_already = 1;
2094                strbuf_add_separated_string_list(&collision_paths, ", ",
2095                                                 &collision_ent->source_files);
2096                output(o, 1, _("CONFLICT (implicit dir rename): Existing "
2097                               "file/dir at %s in the way of implicit "
2098                               "directory rename(s) putting the following "
2099                               "path(s) there: %s."),
2100                       new_path, collision_paths.buf);
2101                clean = 0;
2102        } else if (collision_ent->source_files.nr > 1) {
2103                collision_ent->reported_already = 1;
2104                strbuf_add_separated_string_list(&collision_paths, ", ",
2105                                                 &collision_ent->source_files);
2106                output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
2107                               "more than one path to %s; implicit directory "
2108                               "renames tried to put these paths there: %s"),
2109                       new_path, collision_paths.buf);
2110                clean = 0;
2111        }
2112
2113        /* Free memory we no longer need */
2114        strbuf_release(&collision_paths);
2115        if (!clean && new_path) {
2116                free(new_path);
2117                return NULL;
2118        }
2119
2120        return new_path;
2121}
2122
2123/*
2124 * There are a couple things we want to do at the directory level:
2125 *   1. Check for both sides renaming to the same thing, in order to avoid
2126 *      implicit renaming of files that should be left in place.  (See
2127 *      testcase 6b in t6043 for details.)
2128 *   2. Prune directory renames if there are still files left in the
2129 *      the original directory.  These represent a partial directory rename,
2130 *      i.e. a rename where only some of the files within the directory
2131 *      were renamed elsewhere.  (Technically, this could be done earlier
2132 *      in get_directory_renames(), except that would prevent us from
2133 *      doing the previous check and thus failing testcase 6b.)
2134 *   3. Check for rename/rename(1to2) conflicts (at the directory level).
2135 *      In the future, we could potentially record this info as well and
2136 *      omit reporting rename/rename(1to2) conflicts for each path within
2137 *      the affected directories, thus cleaning up the merge output.
2138 *   NOTE: We do NOT check for rename/rename(2to1) conflicts at the
2139 *         directory level, because merging directories is fine.  If it
2140 *         causes conflicts for files within those merged directories, then
2141 *         that should be detected at the individual path level.
2142 */
2143static void handle_directory_level_conflicts(struct merge_options *o,
2144                                             struct hashmap *dir_re_head,
2145                                             struct tree *head,
2146                                             struct hashmap *dir_re_merge,
2147                                             struct tree *merge)
2148{
2149        struct hashmap_iter iter;
2150        struct dir_rename_entry *head_ent;
2151        struct dir_rename_entry *merge_ent;
2152
2153        struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
2154        struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
2155
2156        hashmap_iter_init(dir_re_head, &iter);
2157        while ((head_ent = hashmap_iter_next(&iter))) {
2158                merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
2159                if (merge_ent &&
2160                    !head_ent->non_unique_new_dir &&
2161                    !merge_ent->non_unique_new_dir &&
2162                    !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
2163                        /* 1. Renamed identically; remove it from both sides */
2164                        string_list_append(&remove_from_head,
2165                                           head_ent->dir)->util = head_ent;
2166                        strbuf_release(&head_ent->new_dir);
2167                        string_list_append(&remove_from_merge,
2168                                           merge_ent->dir)->util = merge_ent;
2169                        strbuf_release(&merge_ent->new_dir);
2170                } else if (tree_has_path(head, head_ent->dir)) {
2171                        /* 2. This wasn't a directory rename after all */
2172                        string_list_append(&remove_from_head,
2173                                           head_ent->dir)->util = head_ent;
2174                        strbuf_release(&head_ent->new_dir);
2175                }
2176        }
2177
2178        remove_hashmap_entries(dir_re_head, &remove_from_head);
2179        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2180
2181        hashmap_iter_init(dir_re_merge, &iter);
2182        while ((merge_ent = hashmap_iter_next(&iter))) {
2183                head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
2184                if (tree_has_path(merge, merge_ent->dir)) {
2185                        /* 2. This wasn't a directory rename after all */
2186                        string_list_append(&remove_from_merge,
2187                                           merge_ent->dir)->util = merge_ent;
2188                } else if (head_ent &&
2189                           !head_ent->non_unique_new_dir &&
2190                           !merge_ent->non_unique_new_dir) {
2191                        /* 3. rename/rename(1to2) */
2192                        /*
2193                         * We can assume it's not rename/rename(1to1) because
2194                         * that was case (1), already checked above.  So we
2195                         * know that head_ent->new_dir and merge_ent->new_dir
2196                         * are different strings.
2197                         */
2198                        output(o, 1, _("CONFLICT (rename/rename): "
2199                                       "Rename directory %s->%s in %s. "
2200                                       "Rename directory %s->%s in %s"),
2201                               head_ent->dir, head_ent->new_dir.buf, o->branch1,
2202                               head_ent->dir, merge_ent->new_dir.buf, o->branch2);
2203                        string_list_append(&remove_from_head,
2204                                           head_ent->dir)->util = head_ent;
2205                        strbuf_release(&head_ent->new_dir);
2206                        string_list_append(&remove_from_merge,
2207                                           merge_ent->dir)->util = merge_ent;
2208                        strbuf_release(&merge_ent->new_dir);
2209                }
2210        }
2211
2212        remove_hashmap_entries(dir_re_head, &remove_from_head);
2213        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
2214}
2215
2216static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
2217                                             struct tree *tree)
2218{
2219        struct hashmap *dir_renames;
2220        struct hashmap_iter iter;
2221        struct dir_rename_entry *entry;
2222        int i;
2223
2224        /*
2225         * Typically, we think of a directory rename as all files from a
2226         * certain directory being moved to a target directory.  However,
2227         * what if someone first moved two files from the original
2228         * directory in one commit, and then renamed the directory
2229         * somewhere else in a later commit?  At merge time, we just know
2230         * that files from the original directory went to two different
2231         * places, and that the bulk of them ended up in the same place.
2232         * We want each directory rename to represent where the bulk of the
2233         * files from that directory end up; this function exists to find
2234         * where the bulk of the files went.
2235         *
2236         * The first loop below simply iterates through the list of file
2237         * renames, finding out how often each directory rename pair
2238         * possibility occurs.
2239         */
2240        dir_renames = xmalloc(sizeof(*dir_renames));
2241        dir_rename_init(dir_renames);
2242        for (i = 0; i < pairs->nr; ++i) {
2243                struct string_list_item *item;
2244                int *count;
2245                struct diff_filepair *pair = pairs->queue[i];
2246                char *old_dir, *new_dir;
2247
2248                /* File not part of directory rename if it wasn't renamed */
2249                if (pair->status != 'R')
2250                        continue;
2251
2252                get_renamed_dir_portion(pair->one->path, pair->two->path,
2253                                        &old_dir,        &new_dir);
2254                if (!old_dir)
2255                        /* Directory didn't change at all; ignore this one. */
2256                        continue;
2257
2258                entry = dir_rename_find_entry(dir_renames, old_dir);
2259                if (!entry) {
2260                        entry = xmalloc(sizeof(*entry));
2261                        dir_rename_entry_init(entry, old_dir);
2262                        hashmap_put(dir_renames, entry);
2263                } else {
2264                        free(old_dir);
2265                }
2266                item = string_list_lookup(&entry->possible_new_dirs, new_dir);
2267                if (!item) {
2268                        item = string_list_insert(&entry->possible_new_dirs,
2269                                                  new_dir);
2270                        item->util = xcalloc(1, sizeof(int));
2271                } else {
2272                        free(new_dir);
2273                }
2274                count = item->util;
2275                *count += 1;
2276        }
2277
2278        /*
2279         * For each directory with files moved out of it, we find out which
2280         * target directory received the most files so we can declare it to
2281         * be the "winning" target location for the directory rename.  This
2282         * winner gets recorded in new_dir.  If there is no winner
2283         * (multiple target directories received the same number of files),
2284         * we set non_unique_new_dir.  Once we've determined the winner (or
2285         * that there is no winner), we no longer need possible_new_dirs.
2286         */
2287        hashmap_iter_init(dir_renames, &iter);
2288        while ((entry = hashmap_iter_next(&iter))) {
2289                int max = 0;
2290                int bad_max = 0;
2291                char *best = NULL;
2292
2293                for (i = 0; i < entry->possible_new_dirs.nr; i++) {
2294                        int *count = entry->possible_new_dirs.items[i].util;
2295
2296                        if (*count == max)
2297                                bad_max = max;
2298                        else if (*count > max) {
2299                                max = *count;
2300                                best = entry->possible_new_dirs.items[i].string;
2301                        }
2302                }
2303                if (bad_max == max)
2304                        entry->non_unique_new_dir = 1;
2305                else {
2306                        assert(entry->new_dir.len == 0);
2307                        strbuf_addstr(&entry->new_dir, best);
2308                }
2309                /*
2310                 * The relevant directory sub-portion of the original full
2311                 * filepaths were xstrndup'ed before inserting into
2312                 * possible_new_dirs, and instead of manually iterating the
2313                 * list and free'ing each, just lie and tell
2314                 * possible_new_dirs that it did the strdup'ing so that it
2315                 * will free them for us.
2316                 */
2317                entry->possible_new_dirs.strdup_strings = 1;
2318                string_list_clear(&entry->possible_new_dirs, 1);
2319        }
2320
2321        return dir_renames;
2322}
2323
2324static struct dir_rename_entry *check_dir_renamed(const char *path,
2325                                                  struct hashmap *dir_renames)
2326{
2327        char *temp = xstrdup(path);
2328        char *end;
2329        struct dir_rename_entry *entry = NULL;
2330
2331        while ((end = strrchr(temp, '/'))) {
2332                *end = '\0';
2333                entry = dir_rename_find_entry(dir_renames, temp);
2334                if (entry)
2335                        break;
2336        }
2337        free(temp);
2338        return entry;
2339}
2340
2341static void compute_collisions(struct hashmap *collisions,
2342                               struct hashmap *dir_renames,
2343                               struct diff_queue_struct *pairs)
2344{
2345        int i;
2346
2347        /*
2348         * Multiple files can be mapped to the same path due to directory
2349         * renames done by the other side of history.  Since that other
2350         * side of history could have merged multiple directories into one,
2351         * if our side of history added the same file basename to each of
2352         * those directories, then all N of them would get implicitly
2353         * renamed by the directory rename detection into the same path,
2354         * and we'd get an add/add/.../add conflict, and all those adds
2355         * from *this* side of history.  This is not representable in the
2356         * index, and users aren't going to easily be able to make sense of
2357         * it.  So we need to provide a good warning about what's
2358         * happening, and fall back to no-directory-rename detection
2359         * behavior for those paths.
2360         *
2361         * See testcases 9e and all of section 5 from t6043 for examples.
2362         */
2363        collision_init(collisions);
2364
2365        for (i = 0; i < pairs->nr; ++i) {
2366                struct dir_rename_entry *dir_rename_ent;
2367                struct collision_entry *collision_ent;
2368                char *new_path;
2369                struct diff_filepair *pair = pairs->queue[i];
2370
2371                if (pair->status != 'A' && pair->status != 'R')
2372                        continue;
2373                dir_rename_ent = check_dir_renamed(pair->two->path,
2374                                                   dir_renames);
2375                if (!dir_rename_ent)
2376                        continue;
2377
2378                new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
2379                if (!new_path)
2380                        /*
2381                         * dir_rename_ent->non_unique_new_path is true, which
2382                         * means there is no directory rename for us to use,
2383                         * which means it won't cause us any additional
2384                         * collisions.
2385                         */
2386                        continue;
2387                collision_ent = collision_find_entry(collisions, new_path);
2388                if (!collision_ent) {
2389                        collision_ent = xcalloc(1,
2390                                                sizeof(struct collision_entry));
2391                        hashmap_entry_init(collision_ent, strhash(new_path));
2392                        hashmap_put(collisions, collision_ent);
2393                        collision_ent->target_file = new_path;
2394                } else {
2395                        free(new_path);
2396                }
2397                string_list_insert(&collision_ent->source_files,
2398                                   pair->two->path);
2399        }
2400}
2401
2402static char *check_for_directory_rename(struct merge_options *o,
2403                                        const char *path,
2404                                        struct tree *tree,
2405                                        struct hashmap *dir_renames,
2406                                        struct hashmap *dir_rename_exclusions,
2407                                        struct hashmap *collisions,
2408                                        int *clean_merge)
2409{
2410        char *new_path = NULL;
2411        struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
2412        struct dir_rename_entry *oentry = NULL;
2413
2414        if (!entry)
2415                return new_path;
2416
2417        /*
2418         * This next part is a little weird.  We do not want to do an
2419         * implicit rename into a directory we renamed on our side, because
2420         * that will result in a spurious rename/rename(1to2) conflict.  An
2421         * example:
2422         *   Base commit: dumbdir/afile, otherdir/bfile
2423         *   Side 1:      smrtdir/afile, otherdir/bfile
2424         *   Side 2:      dumbdir/afile, dumbdir/bfile
2425         * Here, while working on Side 1, we could notice that otherdir was
2426         * renamed/merged to dumbdir, and change the diff_filepair for
2427         * otherdir/bfile into a rename into dumbdir/bfile.  However, Side
2428         * 2 will notice the rename from dumbdir to smrtdir, and do the
2429         * transitive rename to move it from dumbdir/bfile to
2430         * smrtdir/bfile.  That gives us bfile in dumbdir vs being in
2431         * smrtdir, a rename/rename(1to2) conflict.  We really just want
2432         * the file to end up in smrtdir.  And the way to achieve that is
2433         * to not let Side1 do the rename to dumbdir, since we know that is
2434         * the source of one of our directory renames.
2435         *
2436         * That's why oentry and dir_rename_exclusions is here.
2437         *
2438         * As it turns out, this also prevents N-way transient rename
2439         * confusion; See testcases 9c and 9d of t6043.
2440         */
2441        oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
2442        if (oentry) {
2443                output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
2444                               "to %s, because %s itself was renamed."),
2445                       entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
2446        } else {
2447                new_path = handle_path_level_conflicts(o, path, entry,
2448                                                       collisions, tree);
2449                *clean_merge &= (new_path != NULL);
2450        }
2451
2452        return new_path;
2453}
2454
2455static void apply_directory_rename_modifications(struct merge_options *o,
2456                                                 struct diff_filepair *pair,
2457                                                 char *new_path,
2458                                                 struct rename *re,
2459                                                 struct tree *tree,
2460                                                 struct tree *o_tree,
2461                                                 struct tree *a_tree,
2462                                                 struct tree *b_tree,
2463                                                 struct string_list *entries,
2464                                                 int *clean)
2465{
2466        struct string_list_item *item;
2467        int stage = (tree == a_tree ? 2 : 3);
2468        int update_wd;
2469
2470        /*
2471         * In all cases where we can do directory rename detection,
2472         * unpack_trees() will have read pair->two->path into the
2473         * index and the working copy.  We need to remove it so that
2474         * we can instead place it at new_path.  It is guaranteed to
2475         * not be untracked (unpack_trees() would have errored out
2476         * saying the file would have been overwritten), but it might
2477         * be dirty, though.
2478         */
2479        update_wd = !was_dirty(o, pair->two->path);
2480        if (!update_wd)
2481                output(o, 1, _("Refusing to lose dirty file at %s"),
2482                       pair->two->path);
2483        remove_file(o, 1, pair->two->path, !update_wd);
2484
2485        /* Find or create a new re->dst_entry */
2486        item = string_list_lookup(entries, new_path);
2487        if (item) {
2488                /*
2489                 * Since we're renaming on this side of history, and it's
2490                 * due to a directory rename on the other side of history
2491                 * (which we only allow when the directory in question no
2492                 * longer exists on the other side of history), the
2493                 * original entry for re->dst_entry is no longer
2494                 * necessary...
2495                 */
2496                re->dst_entry->processed = 1;
2497
2498                /*
2499                 * ...because we'll be using this new one.
2500                 */
2501                re->dst_entry = item->util;
2502        } else {
2503                /*
2504                 * re->dst_entry is for the before-dir-rename path, and we
2505                 * need it to hold information for the after-dir-rename
2506                 * path.  Before creating a new entry, we need to mark the
2507                 * old one as unnecessary (...unless it is shared by
2508                 * src_entry, i.e. this didn't use to be a rename, in which
2509                 * case we can just allow the normal processing to happen
2510                 * for it).
2511                 */
2512                if (pair->status == 'R')
2513                        re->dst_entry->processed = 1;
2514
2515                re->dst_entry = insert_stage_data(new_path,
2516                                                  o_tree, a_tree, b_tree,
2517                                                  entries);
2518                item = string_list_insert(entries, new_path);
2519                item->util = re->dst_entry;
2520        }
2521
2522        /*
2523         * Update the stage_data with the information about the path we are
2524         * moving into place.  That slot will be empty and available for us
2525         * to write to because of the collision checks in
2526         * handle_path_level_conflicts().  In other words,
2527         * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2528         * open for us to write to.
2529         *
2530         * It may be tempting to actually update the index at this point as
2531         * well, using update_stages_for_stage_data(), but as per the big
2532         * "NOTE" in update_stages(), doing so will modify the current
2533         * in-memory index which will break calls to would_lose_untracked()
2534         * that we need to make.  Instead, we need to just make sure that
2535         * the various handle_rename_*() functions update the index
2536         * explicitly rather than relying on unpack_trees() to have done it.
2537         */
2538        get_tree_entry(&tree->object.oid,
2539                       pair->two->path,
2540                       &re->dst_entry->stages[stage].oid,
2541                       &re->dst_entry->stages[stage].mode);
2542
2543        /* Update pair status */
2544        if (pair->status == 'A') {
2545                /*
2546                 * Recording rename information for this add makes it look
2547                 * like a rename/delete conflict.  Make sure we can
2548                 * correctly handle this as an add that was moved to a new
2549                 * directory instead of reporting a rename/delete conflict.
2550                 */
2551                re->add_turned_into_rename = 1;
2552        }
2553        /*
2554         * We don't actually look at pair->status again, but it seems
2555         * pedagogically correct to adjust it.
2556         */
2557        pair->status = 'R';
2558
2559        /*
2560         * Finally, record the new location.
2561         */
2562        pair->two->path = new_path;
2563}
2564
2565/*
2566 * Get information of all renames which occurred in 'pairs', making use of
2567 * any implicit directory renames inferred from the other side of history.
2568 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2569 * to be able to associate the correct cache entries with the rename
2570 * information; tree is always equal to either a_tree or b_tree.
2571 */
2572static struct string_list *get_renames(struct merge_options *o,
2573                                       struct diff_queue_struct *pairs,
2574                                       struct hashmap *dir_renames,
2575                                       struct hashmap *dir_rename_exclusions,
2576                                       struct tree *tree,
2577                                       struct tree *o_tree,
2578                                       struct tree *a_tree,
2579                                       struct tree *b_tree,
2580                                       struct string_list *entries,
2581                                       int *clean_merge)
2582{
2583        int i;
2584        struct hashmap collisions;
2585        struct hashmap_iter iter;
2586        struct collision_entry *e;
2587        struct string_list *renames;
2588
2589        compute_collisions(&collisions, dir_renames, pairs);
2590        renames = xcalloc(1, sizeof(struct string_list));
2591
2592        for (i = 0; i < pairs->nr; ++i) {
2593                struct string_list_item *item;
2594                struct rename *re;
2595                struct diff_filepair *pair = pairs->queue[i];
2596                char *new_path; /* non-NULL only with directory renames */
2597
2598                if (pair->status != 'A' && pair->status != 'R') {
2599                        diff_free_filepair(pair);
2600                        continue;
2601                }
2602                new_path = check_for_directory_rename(o, pair->two->path, tree,
2603                                                      dir_renames,
2604                                                      dir_rename_exclusions,
2605                                                      &collisions,
2606                                                      clean_merge);
2607                if (pair->status != 'R' && !new_path) {
2608                        diff_free_filepair(pair);
2609                        continue;
2610                }
2611
2612                re = xmalloc(sizeof(*re));
2613                re->processed = 0;
2614                re->add_turned_into_rename = 0;
2615                re->pair = pair;
2616                item = string_list_lookup(entries, re->pair->one->path);
2617                if (!item)
2618                        re->src_entry = insert_stage_data(re->pair->one->path,
2619                                        o_tree, a_tree, b_tree, entries);
2620                else
2621                        re->src_entry = item->util;
2622
2623                item = string_list_lookup(entries, re->pair->two->path);
2624                if (!item)
2625                        re->dst_entry = insert_stage_data(re->pair->two->path,
2626                                        o_tree, a_tree, b_tree, entries);
2627                else
2628                        re->dst_entry = item->util;
2629                item = string_list_insert(renames, pair->one->path);
2630                item->util = re;
2631                if (new_path)
2632                        apply_directory_rename_modifications(o, pair, new_path,
2633                                                             re, tree, o_tree,
2634                                                             a_tree, b_tree,
2635                                                             entries,
2636                                                             clean_merge);
2637        }
2638
2639        hashmap_iter_init(&collisions, &iter);
2640        while ((e = hashmap_iter_next(&iter))) {
2641                free(e->target_file);
2642                string_list_clear(&e->source_files, 0);
2643        }
2644        hashmap_free(&collisions, 1);
2645        return renames;
2646}
2647
2648static int process_renames(struct merge_options *o,
2649                           struct string_list *a_renames,
2650                           struct string_list *b_renames)
2651{
2652        int clean_merge = 1, i, j;
2653        struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2654        struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2655        const struct rename *sre;
2656
2657        for (i = 0; i < a_renames->nr; i++) {
2658                sre = a_renames->items[i].util;
2659                string_list_insert(&a_by_dst, sre->pair->two->path)->util
2660                        = (void *)sre;
2661        }
2662        for (i = 0; i < b_renames->nr; i++) {
2663                sre = b_renames->items[i].util;
2664                string_list_insert(&b_by_dst, sre->pair->two->path)->util
2665                        = (void *)sre;
2666        }
2667
2668        for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2669                struct string_list *renames1, *renames2Dst;
2670                struct rename *ren1 = NULL, *ren2 = NULL;
2671                const char *branch1, *branch2;
2672                const char *ren1_src, *ren1_dst;
2673                struct string_list_item *lookup;
2674
2675                if (i >= a_renames->nr) {
2676                        ren2 = b_renames->items[j++].util;
2677                } else if (j >= b_renames->nr) {
2678                        ren1 = a_renames->items[i++].util;
2679                } else {
2680                        int compare = strcmp(a_renames->items[i].string,
2681                                             b_renames->items[j].string);
2682                        if (compare <= 0)
2683                                ren1 = a_renames->items[i++].util;
2684                        if (compare >= 0)
2685                                ren2 = b_renames->items[j++].util;
2686                }
2687
2688                /* TODO: refactor, so that 1/2 are not needed */
2689                if (ren1) {
2690                        renames1 = a_renames;
2691                        renames2Dst = &b_by_dst;
2692                        branch1 = o->branch1;
2693                        branch2 = o->branch2;
2694                } else {
2695                        renames1 = b_renames;
2696                        renames2Dst = &a_by_dst;
2697                        branch1 = o->branch2;
2698                        branch2 = o->branch1;
2699                        SWAP(ren2, ren1);
2700                }
2701
2702                if (ren1->processed)
2703                        continue;
2704                ren1->processed = 1;
2705                ren1->dst_entry->processed = 1;
2706                /* BUG: We should only mark src_entry as processed if we
2707                 * are not dealing with a rename + add-source case.
2708                 */
2709                ren1->src_entry->processed = 1;
2710
2711                ren1_src = ren1->pair->one->path;
2712                ren1_dst = ren1->pair->two->path;
2713
2714                if (ren2) {
2715                        /* One file renamed on both sides */
2716                        const char *ren2_src = ren2->pair->one->path;
2717                        const char *ren2_dst = ren2->pair->two->path;
2718                        enum rename_type rename_type;
2719                        if (strcmp(ren1_src, ren2_src) != 0)
2720                                BUG("ren1_src != ren2_src");
2721                        ren2->dst_entry->processed = 1;
2722                        ren2->processed = 1;
2723                        if (strcmp(ren1_dst, ren2_dst) != 0) {
2724                                rename_type = RENAME_ONE_FILE_TO_TWO;
2725                                clean_merge = 0;
2726                        } else {
2727                                rename_type = RENAME_ONE_FILE_TO_ONE;
2728                                /* BUG: We should only remove ren1_src in
2729                                 * the base stage (think of rename +
2730                                 * add-source cases).
2731                                 */
2732                                remove_file(o, 1, ren1_src, 1);
2733                                update_entry(ren1->dst_entry,
2734                                             ren1->pair->one,
2735                                             ren1->pair->two,
2736                                             ren2->pair->two);
2737                        }
2738                        setup_rename_conflict_info(rename_type,
2739                                                   ren1->pair,
2740                                                   ren2->pair,
2741                                                   branch1,
2742                                                   branch2,
2743                                                   ren1->dst_entry,
2744                                                   ren2->dst_entry,
2745                                                   o,
2746                                                   NULL,
2747                                                   NULL);
2748                } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2749                        /* Two different files renamed to the same thing */
2750                        char *ren2_dst;
2751                        ren2 = lookup->util;
2752                        ren2_dst = ren2->pair->two->path;
2753                        if (strcmp(ren1_dst, ren2_dst) != 0)
2754                                BUG("ren1_dst != ren2_dst");
2755
2756                        clean_merge = 0;
2757                        ren2->processed = 1;
2758                        /*
2759                         * BUG: We should only mark src_entry as processed
2760                         * if we are not dealing with a rename + add-source
2761                         * case.
2762                         */
2763                        ren2->src_entry->processed = 1;
2764
2765                        setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2766                                                   ren1->pair,
2767                                                   ren2->pair,
2768                                                   branch1,
2769                                                   branch2,
2770                                                   ren1->dst_entry,
2771                                                   ren2->dst_entry,
2772                                                   o,
2773                                                   ren1->src_entry,
2774                                                   ren2->src_entry);
2775
2776                } else {
2777                        /* Renamed in 1, maybe changed in 2 */
2778                        /* we only use sha1 and mode of these */
2779                        struct diff_filespec src_other, dst_other;
2780                        int try_merge;
2781
2782                        /*
2783                         * unpack_trees loads entries from common-commit
2784                         * into stage 1, from head-commit into stage 2, and
2785                         * from merge-commit into stage 3.  We keep track
2786                         * of which side corresponds to the rename.
2787                         */
2788                        int renamed_stage = a_renames == renames1 ? 2 : 3;
2789                        int other_stage =   a_renames == renames1 ? 3 : 2;
2790
2791                        /* BUG: We should only remove ren1_src in the base
2792                         * stage and in other_stage (think of rename +
2793                         * add-source case).
2794                         */
2795                        remove_file(o, 1, ren1_src,
2796                                    renamed_stage == 2 || !was_tracked(o, ren1_src));
2797
2798                        oidcpy(&src_other.oid,
2799                               &ren1->src_entry->stages[other_stage].oid);
2800                        src_other.mode = ren1->src_entry->stages[other_stage].mode;
2801                        oidcpy(&dst_other.oid,
2802                               &ren1->dst_entry->stages[other_stage].oid);
2803                        dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2804                        try_merge = 0;
2805
2806                        if (oid_eq(&src_other.oid, &null_oid) &&
2807                            ren1->add_turned_into_rename) {
2808                                setup_rename_conflict_info(RENAME_VIA_DIR,
2809                                                           ren1->pair,
2810                                                           NULL,
2811                                                           branch1,
2812                                                           branch2,
2813                                                           ren1->dst_entry,
2814                                                           NULL,
2815                                                           o,
2816                                                           NULL,
2817                                                           NULL);
2818                        } else if (oid_eq(&src_other.oid, &null_oid)) {
2819                                setup_rename_conflict_info(RENAME_DELETE,
2820                                                           ren1->pair,
2821                                                           NULL,
2822                                                           branch1,
2823                                                           branch2,
2824                                                           ren1->dst_entry,
2825                                                           NULL,
2826                                                           o,
2827                                                           NULL,
2828                                                           NULL);
2829                        } else if ((dst_other.mode == ren1->pair->two->mode) &&
2830                                   oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2831                                /*
2832                                 * Added file on the other side identical to
2833                                 * the file being renamed: clean merge.
2834                                 * Also, there is no need to overwrite the
2835                                 * file already in the working copy, so call
2836                                 * update_file_flags() instead of
2837                                 * update_file().
2838                                 */
2839                                if (update_file_flags(o,
2840                                                      &ren1->pair->two->oid,
2841                                                      ren1->pair->two->mode,
2842                                                      ren1_dst,
2843                                                      1, /* update_cache */
2844                                                      0  /* update_wd    */))
2845                                        clean_merge = -1;
2846                        } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2847                                /*
2848                                 * Probably not a clean merge, but it's
2849                                 * premature to set clean_merge to 0 here,
2850                                 * because if the rename merges cleanly and
2851                                 * the merge exactly matches the newly added
2852                                 * file, then the merge will be clean.
2853                                 */
2854                                setup_rename_conflict_info(RENAME_ADD,
2855                                                           ren1->pair,
2856                                                           NULL,
2857                                                           branch1,
2858                                                           branch2,
2859                                                           ren1->dst_entry,
2860                                                           NULL,
2861                                                           o,
2862                                                           ren1->src_entry,
2863                                                           NULL);
2864                        } else
2865                                try_merge = 1;
2866
2867                        if (clean_merge < 0)
2868                                goto cleanup_and_return;
2869                        if (try_merge) {
2870                                struct diff_filespec *one, *a, *b;
2871                                src_other.path = (char *)ren1_src;
2872
2873                                one = ren1->pair->one;
2874                                if (a_renames == renames1) {
2875                                        a = ren1->pair->two;
2876                                        b = &src_other;
2877                                } else {
2878                                        b = ren1->pair->two;
2879                                        a = &src_other;
2880                                }
2881                                update_entry(ren1->dst_entry, one, a, b);
2882                                setup_rename_conflict_info(RENAME_NORMAL,
2883                                                           ren1->pair,
2884                                                           NULL,
2885                                                           branch1,
2886                                                           NULL,
2887                                                           ren1->dst_entry,
2888                                                           NULL,
2889                                                           o,
2890                                                           NULL,
2891                                                           NULL);
2892                        }
2893                }
2894        }
2895cleanup_and_return:
2896        string_list_clear(&a_by_dst, 0);
2897        string_list_clear(&b_by_dst, 0);
2898
2899        return clean_merge;
2900}
2901
2902struct rename_info {
2903        struct string_list *head_renames;
2904        struct string_list *merge_renames;
2905};
2906
2907static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2908                                   struct hashmap *dir_renames)
2909{
2910        struct hashmap_iter iter;
2911        struct dir_rename_entry *e;
2912
2913        hashmap_iter_init(dir_renames, &iter);
2914        while ((e = hashmap_iter_next(&iter))) {
2915                free(e->dir);
2916                strbuf_release(&e->new_dir);
2917                /* possible_new_dirs already cleared in get_directory_renames */
2918        }
2919        hashmap_free(dir_renames, 1);
2920        free(dir_renames);
2921
2922        free(pairs->queue);
2923        free(pairs);
2924}
2925
2926static int detect_and_process_renames(struct merge_options *o,
2927                                      struct tree *common,
2928                                      struct tree *head,
2929                                      struct tree *merge,
2930                                      struct string_list *entries,
2931                                      struct rename_info *ri)
2932{
2933        struct diff_queue_struct *head_pairs, *merge_pairs;
2934        struct hashmap *dir_re_head, *dir_re_merge;
2935        int clean = 1;
2936
2937        ri->head_renames = NULL;
2938        ri->merge_renames = NULL;
2939
2940        if (!merge_detect_rename(o))
2941                return 1;
2942
2943        head_pairs = get_diffpairs(o, common, head);
2944        merge_pairs = get_diffpairs(o, common, merge);
2945
2946        if (o->detect_directory_renames) {
2947                dir_re_head = get_directory_renames(head_pairs, head);
2948                dir_re_merge = get_directory_renames(merge_pairs, merge);
2949
2950                handle_directory_level_conflicts(o,
2951                                                 dir_re_head, head,
2952                                                 dir_re_merge, merge);
2953        } else {
2954                dir_re_head  = xmalloc(sizeof(*dir_re_head));
2955                dir_re_merge = xmalloc(sizeof(*dir_re_merge));
2956                dir_rename_init(dir_re_head);
2957                dir_rename_init(dir_re_merge);
2958        }
2959
2960        ri->head_renames  = get_renames(o, head_pairs,
2961                                        dir_re_merge, dir_re_head, head,
2962                                        common, head, merge, entries,
2963                                        &clean);
2964        if (clean < 0)
2965                goto cleanup;
2966        ri->merge_renames = get_renames(o, merge_pairs,
2967                                        dir_re_head, dir_re_merge, merge,
2968                                        common, head, merge, entries,
2969                                        &clean);
2970        if (clean < 0)
2971                goto cleanup;
2972        clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2973
2974cleanup:
2975        /*
2976         * Some cleanup is deferred until cleanup_renames() because the
2977         * data structures are still needed and referenced in
2978         * process_entry().  But there are a few things we can free now.
2979         */
2980        initial_cleanup_rename(head_pairs, dir_re_head);
2981        initial_cleanup_rename(merge_pairs, dir_re_merge);
2982
2983        return clean;
2984}
2985
2986static void final_cleanup_rename(struct string_list *rename)
2987{
2988        const struct rename *re;
2989        int i;
2990
2991        if (rename == NULL)
2992                return;
2993
2994        for (i = 0; i < rename->nr; i++) {
2995                re = rename->items[i].util;
2996                diff_free_filepair(re->pair);
2997        }
2998        string_list_clear(rename, 1);
2999        free(rename);
3000}
3001
3002static void final_cleanup_renames(struct rename_info *re_info)
3003{
3004        final_cleanup_rename(re_info->head_renames);
3005        final_cleanup_rename(re_info->merge_renames);
3006}
3007
3008static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
3009{
3010        return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
3011}
3012
3013static int read_oid_strbuf(struct merge_options *o,
3014                           const struct object_id *oid,
3015                           struct strbuf *dst)
3016{
3017        void *buf;
3018        enum object_type type;
3019        unsigned long size;
3020        buf = read_object_file(oid, &type, &size);
3021        if (!buf)
3022                return err(o, _("cannot read object %s"), oid_to_hex(oid));
3023        if (type != OBJ_BLOB) {
3024                free(buf);
3025                return err(o, _("object %s is not a blob"), oid_to_hex(oid));
3026        }
3027        strbuf_attach(dst, buf, size, size + 1);
3028        return 0;
3029}
3030
3031static int blob_unchanged(struct merge_options *opt,
3032                          const struct object_id *o_oid,
3033                          unsigned o_mode,
3034                          const struct object_id *a_oid,
3035                          unsigned a_mode,
3036                          int renormalize, const char *path)
3037{
3038        struct strbuf o = STRBUF_INIT;
3039        struct strbuf a = STRBUF_INIT;
3040        int ret = 0; /* assume changed for safety */
3041
3042        if (a_mode != o_mode)
3043                return 0;
3044        if (oid_eq(o_oid, a_oid))
3045                return 1;
3046        if (!renormalize)
3047                return 0;
3048
3049        assert(o_oid && a_oid);
3050        if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
3051                goto error_return;
3052        /*
3053         * Note: binary | is used so that both renormalizations are
3054         * performed.  Comparison can be skipped if both files are
3055         * unchanged since their sha1s have already been compared.
3056         */
3057        if (renormalize_buffer(opt->repo->index, path, o.buf, o.len, &o) |
3058            renormalize_buffer(opt->repo->index, path, a.buf, a.len, &a))
3059                ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
3060
3061error_return:
3062        strbuf_release(&o);
3063        strbuf_release(&a);
3064        return ret;
3065}
3066
3067static int handle_modify_delete(struct merge_options *o,
3068                                const char *path,
3069                                struct object_id *o_oid, int o_mode,
3070                                struct object_id *a_oid, int a_mode,
3071                                struct object_id *b_oid, int b_mode)
3072{
3073        const char *modify_branch, *delete_branch;
3074        struct object_id *changed_oid;
3075        int changed_mode;
3076
3077        if (a_oid) {
3078                modify_branch = o->branch1;
3079                delete_branch = o->branch2;
3080                changed_oid = a_oid;
3081                changed_mode = a_mode;
3082        } else {
3083                modify_branch = o->branch2;
3084                delete_branch = o->branch1;
3085                changed_oid = b_oid;
3086                changed_mode = b_mode;
3087        }
3088
3089        return handle_change_delete(o,
3090                                    path, NULL,
3091                                    o_oid, o_mode,
3092                                    changed_oid, changed_mode,
3093                                    modify_branch, delete_branch,
3094                                    _("modify"), _("modified"));
3095}
3096
3097static int handle_content_merge(struct merge_options *o,
3098                                const char *path,
3099                                int is_dirty,
3100                                struct object_id *o_oid, int o_mode,
3101                                struct object_id *a_oid, int a_mode,
3102                                struct object_id *b_oid, int b_mode,
3103                                struct rename_conflict_info *rename_conflict_info)
3104{
3105        const char *reason = _("content");
3106        const char *path1 = NULL, *path2 = NULL;
3107        struct merge_file_info mfi;
3108        struct diff_filespec one, a, b;
3109        unsigned df_conflict_remains = 0;
3110
3111        if (!o_oid) {
3112                reason = _("add/add");
3113                o_oid = (struct object_id *)&null_oid;
3114        }
3115        one.path = a.path = b.path = (char *)path;
3116        oidcpy(&one.oid, o_oid);
3117        one.mode = o_mode;
3118        oidcpy(&a.oid, a_oid);
3119        a.mode = a_mode;
3120        oidcpy(&b.oid, b_oid);
3121        b.mode = b_mode;
3122
3123        if (rename_conflict_info) {
3124                struct diff_filepair *pair1 = rename_conflict_info->pair1;
3125
3126                path1 = (o->branch1 == rename_conflict_info->branch1) ?
3127                        pair1->two->path : pair1->one->path;
3128                /* If rename_conflict_info->pair2 != NULL, we are in
3129                 * RENAME_ONE_FILE_TO_ONE case.  Otherwise, we have a
3130                 * normal rename.
3131                 */
3132                path2 = (rename_conflict_info->pair2 ||
3133                         o->branch2 == rename_conflict_info->branch1) ?
3134                        pair1->two->path : pair1->one->path;
3135                one.path = pair1->one->path;
3136                a.path = (char *)path1;
3137                b.path = (char *)path2;
3138
3139                if (dir_in_way(o->repo->index, path, !o->call_depth,
3140                               S_ISGITLINK(pair1->two->mode)))
3141                        df_conflict_remains = 1;
3142        }
3143        if (merge_mode_and_contents(o, &one, &a, &b, path,
3144                                    o->branch1, o->branch2,
3145                                    o->call_depth * 2, &mfi))
3146                return -1;
3147
3148        /*
3149         * We can skip updating the working tree file iff:
3150         *   a) The merge is clean
3151         *   b) The merge matches what was in HEAD (content, mode, pathname)
3152         *   c) The target path is usable (i.e. not involved in D/F conflict)
3153         */
3154        if (mfi.clean &&
3155            was_tracked_and_matches(o, path, &mfi.oid, mfi.mode) &&
3156            !df_conflict_remains) {
3157                int pos;
3158                struct cache_entry *ce;
3159
3160                output(o, 3, _("Skipped %s (merged same as existing)"), path);
3161                if (add_cacheinfo(o, mfi.mode, &mfi.oid, path,
3162                                  0, (!o->call_depth && !is_dirty), 0))
3163                        return -1;
3164                /*
3165                 * However, add_cacheinfo() will delete the old cache entry
3166                 * and add a new one.  We need to copy over any skip_worktree
3167                 * flag to avoid making the file appear as if it were
3168                 * deleted by the user.
3169                 */
3170                pos = index_name_pos(&o->orig_index, path, strlen(path));
3171                ce = o->orig_index.cache[pos];
3172                if (ce_skip_worktree(ce)) {
3173                        pos = index_name_pos(o->repo->index, path, strlen(path));
3174                        ce = o->repo->index->cache[pos];
3175                        ce->ce_flags |= CE_SKIP_WORKTREE;
3176                }
3177                return mfi.clean;
3178        }
3179
3180        if (!mfi.clean) {
3181                if (S_ISGITLINK(mfi.mode))
3182                        reason = _("submodule");
3183                output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
3184                                reason, path);
3185                if (rename_conflict_info && !df_conflict_remains)
3186                        if (update_stages(o, path, &one, &a, &b))
3187                                return -1;
3188        }
3189
3190        if (df_conflict_remains || is_dirty) {
3191                char *new_path;
3192                if (o->call_depth) {
3193                        remove_file_from_index(o->repo->index, path);
3194                } else {
3195                        if (!mfi.clean) {
3196                                if (update_stages(o, path, &one, &a, &b))
3197                                        return -1;
3198                        } else {
3199                                int file_from_stage2 = was_tracked(o, path);
3200                                struct diff_filespec merged;
3201                                oidcpy(&merged.oid, &mfi.oid);
3202                                merged.mode = mfi.mode;
3203
3204                                if (update_stages(o, path, NULL,
3205                                                  file_from_stage2 ? &merged : NULL,
3206                                                  file_from_stage2 ? NULL : &merged))
3207                                        return -1;
3208                        }
3209
3210                }
3211                new_path = unique_path(o, path, rename_conflict_info->branch1);
3212                if (is_dirty) {
3213                        output(o, 1, _("Refusing to lose dirty file at %s"),
3214                               path);
3215                }
3216                output(o, 1, _("Adding as %s instead"), new_path);
3217                if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
3218                        free(new_path);
3219                        return -1;
3220                }
3221                free(new_path);
3222                mfi.clean = 0;
3223        } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
3224                return -1;
3225        return !is_dirty && mfi.clean;
3226}
3227
3228static int handle_rename_normal(struct merge_options *o,
3229                                const char *path,
3230                                struct object_id *o_oid, unsigned int o_mode,
3231                                struct object_id *a_oid, unsigned int a_mode,
3232                                struct object_id *b_oid, unsigned int b_mode,
3233                                struct rename_conflict_info *ci)
3234{
3235        /* Merge the content and write it out */
3236        return handle_content_merge(o, path, was_dirty(o, path),
3237                                    o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
3238                                    ci);
3239}
3240
3241/* Per entry merge function */
3242static int process_entry(struct merge_options *o,
3243                         const char *path, struct stage_data *entry)
3244{
3245        int clean_merge = 1;
3246        int normalize = o->renormalize;
3247        unsigned o_mode = entry->stages[1].mode;
3248        unsigned a_mode = entry->stages[2].mode;
3249        unsigned b_mode = entry->stages[3].mode;
3250        struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
3251        struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
3252        struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
3253
3254        entry->processed = 1;
3255        if (entry->rename_conflict_info) {
3256                struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
3257                switch (conflict_info->rename_type) {
3258                case RENAME_NORMAL:
3259                case RENAME_ONE_FILE_TO_ONE:
3260                        clean_merge = handle_rename_normal(o,
3261                                                           path,
3262                                                           o_oid, o_mode,
3263                                                           a_oid, a_mode,
3264                                                           b_oid, b_mode,
3265                                                           conflict_info);
3266                        break;
3267                case RENAME_VIA_DIR:
3268                        clean_merge = 1;
3269                        if (handle_rename_via_dir(o,
3270                                                  conflict_info->pair1,
3271                                                  conflict_info->branch1,
3272                                                  conflict_info->branch2))
3273                                clean_merge = -1;
3274                        break;
3275                case RENAME_ADD:
3276                        /*
3277                         * Probably unclean merge, but if the renamed file
3278                         * merges cleanly and the result can then be
3279                         * two-way merged cleanly with the added file, I
3280                         * guess it's a clean merge?
3281                         */
3282                        clean_merge = handle_rename_add(o, conflict_info);
3283                        break;
3284                case RENAME_DELETE:
3285                        clean_merge = 0;
3286                        if (handle_rename_delete(o,
3287                                                 conflict_info->pair1,
3288                                                 conflict_info->branch1,
3289                                                 conflict_info->branch2))
3290                                clean_merge = -1;
3291                        break;
3292                case RENAME_ONE_FILE_TO_TWO:
3293                        clean_merge = 0;
3294                        if (handle_rename_rename_1to2(o, conflict_info))
3295                                clean_merge = -1;
3296                        break;
3297                case RENAME_TWO_FILES_TO_ONE:
3298                        /*
3299                         * Probably unclean merge, but if the two renamed
3300                         * files merge cleanly and the two resulting files
3301                         * can then be two-way merged cleanly, I guess it's
3302                         * a clean merge?
3303                         */
3304                        clean_merge = handle_rename_rename_2to1(o,
3305                                                                conflict_info);
3306                        break;
3307                default:
3308                        entry->processed = 0;
3309                        break;
3310                }
3311        } else if (o_oid && (!a_oid || !b_oid)) {
3312                /* Case A: Deleted in one */
3313                if ((!a_oid && !b_oid) ||
3314                    (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
3315                    (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
3316                        /* Deleted in both or deleted in one and
3317                         * unchanged in the other */
3318                        if (a_oid)
3319                                output(o, 2, _("Removing %s"), path);
3320                        /* do not touch working file if it did not exist */
3321                        remove_file(o, 1, path, !a_oid);
3322                } else {
3323                        /* Modify/delete; deleted side may have put a directory in the way */
3324                        clean_merge = 0;
3325                        if (handle_modify_delete(o, path, o_oid, o_mode,
3326                                                 a_oid, a_mode, b_oid, b_mode))
3327                                clean_merge = -1;
3328                }
3329        } else if ((!o_oid && a_oid && !b_oid) ||
3330                   (!o_oid && !a_oid && b_oid)) {
3331                /* Case B: Added in one. */
3332                /* [nothing|directory] -> ([nothing|directory], file) */
3333
3334                const char *add_branch;
3335                const char *other_branch;
3336                unsigned mode;
3337                const struct object_id *oid;
3338                const char *conf;
3339
3340                if (a_oid) {
3341                        add_branch = o->branch1;
3342                        other_branch = o->branch2;
3343                        mode = a_mode;
3344                        oid = a_oid;
3345                        conf = _("file/directory");
3346                } else {
3347                        add_branch = o->branch2;
3348                        other_branch = o->branch1;
3349                        mode = b_mode;
3350                        oid = b_oid;
3351                        conf = _("directory/file");
3352                }
3353                if (dir_in_way(o->repo->index, path,
3354                               !o->call_depth && !S_ISGITLINK(a_mode),
3355                               0)) {
3356                        char *new_path = unique_path(o, path, add_branch);
3357                        clean_merge = 0;
3358                        output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
3359                               "Adding %s as %s"),
3360                               conf, path, other_branch, path, new_path);
3361                        if (update_file(o, 0, oid, mode, new_path))
3362                                clean_merge = -1;
3363                        else if (o->call_depth)
3364                                remove_file_from_index(o->repo->index, path);
3365                        free(new_path);
3366                } else {
3367                        output(o, 2, _("Adding %s"), path);
3368                        /* do not overwrite file if already present */
3369                        if (update_file_flags(o, oid, mode, path, 1, !a_oid))
3370                                clean_merge = -1;
3371                }
3372        } else if (a_oid && b_oid) {
3373                if (!o_oid) {
3374                        /* Case C: Added in both (check for same permissions) */
3375                        output(o, 1,
3376                               _("CONFLICT (add/add): Merge conflict in %s"),
3377                               path);
3378                        clean_merge = handle_file_collision(o,
3379                                                            path, NULL, NULL,
3380                                                            o->branch1,
3381                                                            o->branch2,
3382                                                            a_oid, a_mode,
3383                                                            b_oid, b_mode);
3384                } else {
3385                        /* case D: Modified in both, but differently. */
3386                        int is_dirty = 0; /* unpack_trees would have bailed if dirty */
3387                        clean_merge = handle_content_merge(o, path,
3388                                                           is_dirty,
3389                                                           o_oid, o_mode,
3390                                                           a_oid, a_mode,
3391                                                           b_oid, b_mode,
3392                                                           NULL);
3393                }
3394        } else if (!o_oid && !a_oid && !b_oid) {
3395                /*
3396                 * this entry was deleted altogether. a_mode == 0 means
3397                 * we had that path and want to actively remove it.
3398                 */
3399                remove_file(o, 1, path, !a_mode);
3400        } else
3401                BUG("fatal merge failure, shouldn't happen.");
3402
3403        return clean_merge;
3404}
3405
3406int merge_trees(struct merge_options *o,
3407                struct tree *head,
3408                struct tree *merge,
3409                struct tree *common,
3410                struct tree **result)
3411{
3412        struct index_state *istate = o->repo->index;
3413        int code, clean;
3414        struct strbuf sb = STRBUF_INIT;
3415
3416        if (!o->call_depth && repo_index_has_changes(o->repo, head, &sb)) {
3417                err(o, _("Your local changes to the following files would be overwritten by merge:\n  %s"),
3418                    sb.buf);
3419                return -1;
3420        }
3421
3422        if (o->subtree_shift) {
3423                merge = shift_tree_object(o->repo, head, merge, o->subtree_shift);
3424                common = shift_tree_object(o->repo, head, common, o->subtree_shift);
3425        }
3426
3427        if (oid_eq(&common->object.oid, &merge->object.oid)) {
3428                output(o, 0, _("Already up to date!"));
3429                *result = head;
3430                return 1;
3431        }
3432
3433        code = unpack_trees_start(o, common, head, merge);
3434
3435        if (code != 0) {
3436                if (show(o, 4) || o->call_depth)
3437                        err(o, _("merging of trees %s and %s failed"),
3438                            oid_to_hex(&head->object.oid),
3439                            oid_to_hex(&merge->object.oid));
3440                unpack_trees_finish(o);
3441                return -1;
3442        }
3443
3444        if (unmerged_index(istate)) {
3445                struct string_list *entries;
3446                struct rename_info re_info;
3447                int i;
3448                /*
3449                 * Only need the hashmap while processing entries, so
3450                 * initialize it here and free it when we are done running
3451                 * through the entries. Keeping it in the merge_options as
3452                 * opposed to decaring a local hashmap is for convenience
3453                 * so that we don't have to pass it to around.
3454                 */
3455                hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
3456                get_files_dirs(o, head);
3457                get_files_dirs(o, merge);
3458
3459                entries = get_unmerged(o->repo->index);
3460                clean = detect_and_process_renames(o, common, head, merge,
3461                                                   entries, &re_info);
3462                record_df_conflict_files(o, entries);
3463                if (clean < 0)
3464                        goto cleanup;
3465                for (i = entries->nr-1; 0 <= i; i--) {
3466                        const char *path = entries->items[i].string;
3467                        struct stage_data *e = entries->items[i].util;
3468                        if (!e->processed) {
3469                                int ret = process_entry(o, path, e);
3470                                if (!ret)
3471                                        clean = 0;
3472                                else if (ret < 0) {
3473                                        clean = ret;
3474                                        goto cleanup;
3475                                }
3476                        }
3477                }
3478                for (i = 0; i < entries->nr; i++) {
3479                        struct stage_data *e = entries->items[i].util;
3480                        if (!e->processed)
3481                                BUG("unprocessed path??? %s",
3482                                    entries->items[i].string);
3483                }
3484
3485        cleanup:
3486                final_cleanup_renames(&re_info);
3487
3488                string_list_clear(entries, 1);
3489                free(entries);
3490
3491                hashmap_free(&o->current_file_dir_set, 1);
3492
3493                if (clean < 0) {
3494                        unpack_trees_finish(o);
3495                        return clean;
3496                }
3497        }
3498        else
3499                clean = 1;
3500
3501        unpack_trees_finish(o);
3502
3503        if (o->call_depth && !(*result = write_tree_from_memory(o)))
3504                return -1;
3505
3506        return clean;
3507}
3508
3509static struct commit_list *reverse_commit_list(struct commit_list *list)
3510{
3511        struct commit_list *next = NULL, *current, *backup;
3512        for (current = list; current; current = backup) {
3513                backup = current->next;
3514                current->next = next;
3515                next = current;
3516        }
3517        return next;
3518}
3519
3520/*
3521 * Merge the commits h1 and h2, return the resulting virtual
3522 * commit object and a flag indicating the cleanness of the merge.
3523 */
3524int merge_recursive(struct merge_options *o,
3525                    struct commit *h1,
3526                    struct commit *h2,
3527                    struct commit_list *ca,
3528                    struct commit **result)
3529{
3530        struct commit_list *iter;
3531        struct commit *merged_common_ancestors;
3532        struct tree *mrtree;
3533        int clean;
3534
3535        if (show(o, 4)) {
3536                output(o, 4, _("Merging:"));
3537                output_commit_title(o, h1);
3538                output_commit_title(o, h2);
3539        }
3540
3541        if (!ca) {
3542                ca = get_merge_bases(h1, h2);
3543                ca = reverse_commit_list(ca);
3544        }
3545
3546        if (show(o, 5)) {
3547                unsigned cnt = commit_list_count(ca);
3548
3549                output(o, 5, Q_("found %u common ancestor:",
3550                                "found %u common ancestors:", cnt), cnt);
3551                for (iter = ca; iter; iter = iter->next)
3552                        output_commit_title(o, iter->item);
3553        }
3554
3555        merged_common_ancestors = pop_commit(&ca);
3556        if (merged_common_ancestors == NULL) {
3557                /* if there is no common ancestor, use an empty tree */
3558                struct tree *tree;
3559
3560                tree = lookup_tree(o->repo, o->repo->hash_algo->empty_tree);
3561                merged_common_ancestors = make_virtual_commit(o->repo, tree, "ancestor");
3562        }
3563
3564        for (iter = ca; iter; iter = iter->next) {
3565                const char *saved_b1, *saved_b2;
3566                o->call_depth++;
3567                /*
3568                 * When the merge fails, the result contains files
3569                 * with conflict markers. The cleanness flag is
3570                 * ignored (unless indicating an error), it was never
3571                 * actually used, as result of merge_trees has always
3572                 * overwritten it: the committed "conflicts" were
3573                 * already resolved.
3574                 */
3575                discard_index(o->repo->index);
3576                saved_b1 = o->branch1;
3577                saved_b2 = o->branch2;
3578                o->branch1 = "Temporary merge branch 1";
3579                o->branch2 = "Temporary merge branch 2";
3580                if (merge_recursive(o, merged_common_ancestors, iter->item,
3581                                    NULL, &merged_common_ancestors) < 0)
3582                        return -1;
3583                o->branch1 = saved_b1;
3584                o->branch2 = saved_b2;
3585                o->call_depth--;
3586
3587                if (!merged_common_ancestors)
3588                        return err(o, _("merge returned no commit"));
3589        }
3590
3591        discard_index(o->repo->index);
3592        if (!o->call_depth)
3593                repo_read_index(o->repo);
3594
3595        o->ancestor = "merged common ancestors";
3596        clean = merge_trees(o, get_commit_tree(h1), get_commit_tree(h2),
3597                            get_commit_tree(merged_common_ancestors),
3598                            &mrtree);
3599        if (clean < 0) {
3600                flush_output(o);
3601                return clean;
3602        }
3603
3604        if (o->call_depth) {
3605                *result = make_virtual_commit(o->repo, mrtree, "merged tree");
3606                commit_list_insert(h1, &(*result)->parents);
3607                commit_list_insert(h2, &(*result)->parents->next);
3608        }
3609        flush_output(o);
3610        if (!o->call_depth && o->buffer_output < 2)
3611                strbuf_release(&o->obuf);
3612        if (show(o, 2))
3613                diff_warn_rename_limit("merge.renamelimit",
3614                                       o->needed_rename_limit, 0);
3615        return clean;
3616}
3617
3618static struct commit *get_ref(struct repository *repo, const struct object_id *oid,
3619                              const char *name)
3620{
3621        struct object *object;
3622
3623        object = deref_tag(repo, parse_object(repo, oid),
3624                           name, strlen(name));
3625        if (!object)
3626                return NULL;
3627        if (object->type == OBJ_TREE)
3628                return make_virtual_commit(repo, (struct tree*)object, name);
3629        if (object->type != OBJ_COMMIT)
3630                return NULL;
3631        if (parse_commit((struct commit *)object))
3632                return NULL;
3633        return (struct commit *)object;
3634}
3635
3636int merge_recursive_generic(struct merge_options *o,
3637                            const struct object_id *head,
3638                            const struct object_id *merge,
3639                            int num_base_list,
3640                            const struct object_id **base_list,
3641                            struct commit **result)
3642{
3643        int clean;
3644        struct lock_file lock = LOCK_INIT;
3645        struct commit *head_commit = get_ref(o->repo, head, o->branch1);
3646        struct commit *next_commit = get_ref(o->repo, merge, o->branch2);
3647        struct commit_list *ca = NULL;
3648
3649        if (base_list) {
3650                int i;
3651                for (i = 0; i < num_base_list; ++i) {
3652                        struct commit *base;
3653                        if (!(base = get_ref(o->repo, base_list[i], oid_to_hex(base_list[i]))))
3654                                return err(o, _("Could not parse object '%s'"),
3655                                           oid_to_hex(base_list[i]));
3656                        commit_list_insert(base, &ca);
3657                }
3658        }
3659
3660        repo_hold_locked_index(o->repo, &lock, LOCK_DIE_ON_ERROR);
3661        clean = merge_recursive(o, head_commit, next_commit, ca,
3662                                result);
3663        if (clean < 0) {
3664                rollback_lock_file(&lock);
3665                return clean;
3666        }
3667
3668        if (write_locked_index(o->repo->index, &lock,
3669                               COMMIT_LOCK | SKIP_IF_UNCHANGED))
3670                return err(o, _("Unable to write index."));
3671
3672        return clean ? 0 : 1;
3673}
3674
3675static void merge_recursive_config(struct merge_options *o)
3676{
3677        char *value = NULL;
3678        git_config_get_int("merge.verbosity", &o->verbosity);
3679        git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3680        git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3681        if (!git_config_get_string("diff.renames", &value)) {
3682                o->diff_detect_rename = git_config_rename("diff.renames", value);
3683                free(value);
3684        }
3685        if (!git_config_get_string("merge.renames", &value)) {
3686                o->merge_detect_rename = git_config_rename("merge.renames", value);
3687                free(value);
3688        }
3689        git_config(git_xmerge_config, NULL);
3690}
3691
3692void init_merge_options(struct merge_options *o,
3693                        struct repository *repo)
3694{
3695        const char *merge_verbosity;
3696        memset(o, 0, sizeof(struct merge_options));
3697        o->repo = repo;
3698        o->verbosity = 2;
3699        o->buffer_output = 1;
3700        o->diff_rename_limit = -1;
3701        o->merge_rename_limit = -1;
3702        o->renormalize = 0;
3703        o->diff_detect_rename = -1;
3704        o->merge_detect_rename = -1;
3705        o->detect_directory_renames = 1;
3706        merge_recursive_config(o);
3707        merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3708        if (merge_verbosity)
3709                o->verbosity = strtol(merge_verbosity, NULL, 10);
3710        if (o->verbosity >= 5)
3711                o->buffer_output = 0;
3712        strbuf_init(&o->obuf, 0);
3713        string_list_init(&o->df_conflict_file_set, 1);
3714}
3715
3716int parse_merge_opt(struct merge_options *o, const char *s)
3717{
3718        const char *arg;
3719
3720        if (!s || !*s)
3721                return -1;
3722        if (!strcmp(s, "ours"))
3723                o->recursive_variant = MERGE_RECURSIVE_OURS;
3724        else if (!strcmp(s, "theirs"))
3725                o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3726        else if (!strcmp(s, "subtree"))
3727                o->subtree_shift = "";
3728        else if (skip_prefix(s, "subtree=", &arg))
3729                o->subtree_shift = arg;
3730        else if (!strcmp(s, "patience"))
3731                o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3732        else if (!strcmp(s, "histogram"))
3733                o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3734        else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3735                long value = parse_algorithm_value(arg);
3736                if (value < 0)
3737                        return -1;
3738                /* clear out previous settings */
3739                DIFF_XDL_CLR(o, NEED_MINIMAL);
3740                o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3741                o->xdl_opts |= value;
3742        }
3743        else if (!strcmp(s, "ignore-space-change"))
3744                DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3745        else if (!strcmp(s, "ignore-all-space"))
3746                DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3747        else if (!strcmp(s, "ignore-space-at-eol"))
3748                DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3749        else if (!strcmp(s, "ignore-cr-at-eol"))
3750                DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3751        else if (!strcmp(s, "renormalize"))
3752                o->renormalize = 1;
3753        else if (!strcmp(s, "no-renormalize"))
3754                o->renormalize = 0;
3755        else if (!strcmp(s, "no-renames"))
3756                o->merge_detect_rename = 0;
3757        else if (!strcmp(s, "find-renames")) {
3758                o->merge_detect_rename = 1;
3759                o->rename_score = 0;
3760        }
3761        else if (skip_prefix(s, "find-renames=", &arg) ||
3762                 skip_prefix(s, "rename-threshold=", &arg)) {
3763                if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3764                        return -1;
3765                o->merge_detect_rename = 1;
3766        }
3767        else
3768                return -1;
3769        return 0;
3770}