merge-recursive.con commit merge-recursive: when comparing files, don't include trees (7d38849)
   1/*
   2 * Recursive Merge algorithm stolen from git-merge-recursive.py by
   3 * Fredrik Kuivinen.
   4 * The thieves were Alex Riesen and Johannes Schindelin, in June/July 2006
   5 */
   6#include "cache.h"
   7#include "config.h"
   8#include "advice.h"
   9#include "lockfile.h"
  10#include "cache-tree.h"
  11#include "commit.h"
  12#include "blob.h"
  13#include "builtin.h"
  14#include "tree-walk.h"
  15#include "diff.h"
  16#include "diffcore.h"
  17#include "tag.h"
  18#include "unpack-trees.h"
  19#include "string-list.h"
  20#include "xdiff-interface.h"
  21#include "ll-merge.h"
  22#include "attr.h"
  23#include "merge-recursive.h"
  24#include "dir.h"
  25#include "submodule.h"
  26
  27struct path_hashmap_entry {
  28        struct hashmap_entry e;
  29        char path[FLEX_ARRAY];
  30};
  31
  32static int path_hashmap_cmp(const void *cmp_data,
  33                            const void *entry,
  34                            const void *entry_or_key,
  35                            const void *keydata)
  36{
  37        const struct path_hashmap_entry *a = entry;
  38        const struct path_hashmap_entry *b = entry_or_key;
  39        const char *key = keydata;
  40
  41        if (ignore_case)
  42                return strcasecmp(a->path, key ? key : b->path);
  43        else
  44                return strcmp(a->path, key ? key : b->path);
  45}
  46
  47static unsigned int path_hash(const char *path)
  48{
  49        return ignore_case ? strihash(path) : strhash(path);
  50}
  51
  52static struct dir_rename_entry *dir_rename_find_entry(struct hashmap *hashmap,
  53                                                      char *dir)
  54{
  55        struct dir_rename_entry key;
  56
  57        if (dir == NULL)
  58                return NULL;
  59        hashmap_entry_init(&key, strhash(dir));
  60        key.dir = dir;
  61        return hashmap_get(hashmap, &key, NULL);
  62}
  63
  64static int dir_rename_cmp(const void *unused_cmp_data,
  65                          const void *entry,
  66                          const void *entry_or_key,
  67                          const void *unused_keydata)
  68{
  69        const struct dir_rename_entry *e1 = entry;
  70        const struct dir_rename_entry *e2 = entry_or_key;
  71
  72        return strcmp(e1->dir, e2->dir);
  73}
  74
  75static void dir_rename_init(struct hashmap *map)
  76{
  77        hashmap_init(map, dir_rename_cmp, NULL, 0);
  78}
  79
  80static void dir_rename_entry_init(struct dir_rename_entry *entry,
  81                                  char *directory)
  82{
  83        hashmap_entry_init(entry, strhash(directory));
  84        entry->dir = directory;
  85        entry->non_unique_new_dir = 0;
  86        strbuf_init(&entry->new_dir, 0);
  87        string_list_init(&entry->possible_new_dirs, 0);
  88}
  89
  90static struct collision_entry *collision_find_entry(struct hashmap *hashmap,
  91                                                    char *target_file)
  92{
  93        struct collision_entry key;
  94
  95        hashmap_entry_init(&key, strhash(target_file));
  96        key.target_file = target_file;
  97        return hashmap_get(hashmap, &key, NULL);
  98}
  99
 100static int collision_cmp(void *unused_cmp_data,
 101                         const struct collision_entry *e1,
 102                         const struct collision_entry *e2,
 103                         const void *unused_keydata)
 104{
 105        return strcmp(e1->target_file, e2->target_file);
 106}
 107
 108static void collision_init(struct hashmap *map)
 109{
 110        hashmap_init(map, (hashmap_cmp_fn) collision_cmp, NULL, 0);
 111}
 112
 113static void flush_output(struct merge_options *o)
 114{
 115        if (o->buffer_output < 2 && o->obuf.len) {
 116                fputs(o->obuf.buf, stdout);
 117                strbuf_reset(&o->obuf);
 118        }
 119}
 120
 121static int err(struct merge_options *o, const char *err, ...)
 122{
 123        va_list params;
 124
 125        if (o->buffer_output < 2)
 126                flush_output(o);
 127        else {
 128                strbuf_complete(&o->obuf, '\n');
 129                strbuf_addstr(&o->obuf, "error: ");
 130        }
 131        va_start(params, err);
 132        strbuf_vaddf(&o->obuf, err, params);
 133        va_end(params);
 134        if (o->buffer_output > 1)
 135                strbuf_addch(&o->obuf, '\n');
 136        else {
 137                error("%s", o->obuf.buf);
 138                strbuf_reset(&o->obuf);
 139        }
 140
 141        return -1;
 142}
 143
 144static struct tree *shift_tree_object(struct tree *one, struct tree *two,
 145                                      const char *subtree_shift)
 146{
 147        struct object_id shifted;
 148
 149        if (!*subtree_shift) {
 150                shift_tree(&one->object.oid, &two->object.oid, &shifted, 0);
 151        } else {
 152                shift_tree_by(&one->object.oid, &two->object.oid, &shifted,
 153                              subtree_shift);
 154        }
 155        if (!oidcmp(&two->object.oid, &shifted))
 156                return two;
 157        return lookup_tree(&shifted);
 158}
 159
 160static struct commit *make_virtual_commit(struct tree *tree, const char *comment)
 161{
 162        struct commit *commit = alloc_commit_node();
 163
 164        set_merge_remote_desc(commit, comment, (struct object *)commit);
 165        commit->tree = tree;
 166        commit->object.parsed = 1;
 167        return commit;
 168}
 169
 170/*
 171 * Since we use get_tree_entry(), which does not put the read object into
 172 * the object pool, we cannot rely on a == b.
 173 */
 174static int oid_eq(const struct object_id *a, const struct object_id *b)
 175{
 176        if (!a && !b)
 177                return 2;
 178        return a && b && oidcmp(a, b) == 0;
 179}
 180
 181enum rename_type {
 182        RENAME_NORMAL = 0,
 183        RENAME_DELETE,
 184        RENAME_ONE_FILE_TO_ONE,
 185        RENAME_ONE_FILE_TO_TWO,
 186        RENAME_TWO_FILES_TO_ONE
 187};
 188
 189struct rename_conflict_info {
 190        enum rename_type rename_type;
 191        struct diff_filepair *pair1;
 192        struct diff_filepair *pair2;
 193        const char *branch1;
 194        const char *branch2;
 195        struct stage_data *dst_entry1;
 196        struct stage_data *dst_entry2;
 197        struct diff_filespec ren1_other;
 198        struct diff_filespec ren2_other;
 199};
 200
 201/*
 202 * Since we want to write the index eventually, we cannot reuse the index
 203 * for these (temporary) data.
 204 */
 205struct stage_data {
 206        struct {
 207                unsigned mode;
 208                struct object_id oid;
 209        } stages[4];
 210        struct rename_conflict_info *rename_conflict_info;
 211        unsigned processed:1;
 212};
 213
 214static inline void setup_rename_conflict_info(enum rename_type rename_type,
 215                                              struct diff_filepair *pair1,
 216                                              struct diff_filepair *pair2,
 217                                              const char *branch1,
 218                                              const char *branch2,
 219                                              struct stage_data *dst_entry1,
 220                                              struct stage_data *dst_entry2,
 221                                              struct merge_options *o,
 222                                              struct stage_data *src_entry1,
 223                                              struct stage_data *src_entry2)
 224{
 225        struct rename_conflict_info *ci = xcalloc(1, sizeof(struct rename_conflict_info));
 226        ci->rename_type = rename_type;
 227        ci->pair1 = pair1;
 228        ci->branch1 = branch1;
 229        ci->branch2 = branch2;
 230
 231        ci->dst_entry1 = dst_entry1;
 232        dst_entry1->rename_conflict_info = ci;
 233        dst_entry1->processed = 0;
 234
 235        assert(!pair2 == !dst_entry2);
 236        if (dst_entry2) {
 237                ci->dst_entry2 = dst_entry2;
 238                ci->pair2 = pair2;
 239                dst_entry2->rename_conflict_info = ci;
 240        }
 241
 242        if (rename_type == RENAME_TWO_FILES_TO_ONE) {
 243                /*
 244                 * For each rename, there could have been
 245                 * modifications on the side of history where that
 246                 * file was not renamed.
 247                 */
 248                int ostage1 = o->branch1 == branch1 ? 3 : 2;
 249                int ostage2 = ostage1 ^ 1;
 250
 251                ci->ren1_other.path = pair1->one->path;
 252                oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid);
 253                ci->ren1_other.mode = src_entry1->stages[ostage1].mode;
 254
 255                ci->ren2_other.path = pair2->one->path;
 256                oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid);
 257                ci->ren2_other.mode = src_entry2->stages[ostage2].mode;
 258        }
 259}
 260
 261static int show(struct merge_options *o, int v)
 262{
 263        return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5;
 264}
 265
 266__attribute__((format (printf, 3, 4)))
 267static void output(struct merge_options *o, int v, const char *fmt, ...)
 268{
 269        va_list ap;
 270
 271        if (!show(o, v))
 272                return;
 273
 274        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 275
 276        va_start(ap, fmt);
 277        strbuf_vaddf(&o->obuf, fmt, ap);
 278        va_end(ap);
 279
 280        strbuf_addch(&o->obuf, '\n');
 281        if (!o->buffer_output)
 282                flush_output(o);
 283}
 284
 285static void output_commit_title(struct merge_options *o, struct commit *commit)
 286{
 287        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 288        if (commit->util)
 289                strbuf_addf(&o->obuf, "virtual %s\n",
 290                        merge_remote_util(commit)->name);
 291        else {
 292                strbuf_add_unique_abbrev(&o->obuf, commit->object.oid.hash,
 293                                         DEFAULT_ABBREV);
 294                strbuf_addch(&o->obuf, ' ');
 295                if (parse_commit(commit) != 0)
 296                        strbuf_addstr(&o->obuf, _("(bad commit)\n"));
 297                else {
 298                        const char *title;
 299                        const char *msg = get_commit_buffer(commit, NULL);
 300                        int len = find_commit_subject(msg, &title);
 301                        if (len)
 302                                strbuf_addf(&o->obuf, "%.*s\n", len, title);
 303                        unuse_commit_buffer(commit, msg);
 304                }
 305        }
 306        flush_output(o);
 307}
 308
 309static int add_cacheinfo(struct merge_options *o,
 310                unsigned int mode, const struct object_id *oid,
 311                const char *path, int stage, int refresh, int options)
 312{
 313        struct cache_entry *ce;
 314        int ret;
 315
 316        ce = make_cache_entry(mode, oid ? oid->hash : null_sha1, path, stage, 0);
 317        if (!ce)
 318                return err(o, _("addinfo_cache failed for path '%s'"), path);
 319
 320        ret = add_cache_entry(ce, options);
 321        if (refresh) {
 322                struct cache_entry *nce;
 323
 324                nce = refresh_cache_entry(ce, CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
 325                if (!nce)
 326                        return err(o, _("addinfo_cache failed for path '%s'"), path);
 327                if (nce != ce)
 328                        ret = add_cache_entry(nce, options);
 329        }
 330        return ret;
 331}
 332
 333static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
 334{
 335        parse_tree(tree);
 336        init_tree_desc(desc, tree->buffer, tree->size);
 337}
 338
 339static int git_merge_trees(int index_only,
 340                           struct tree *common,
 341                           struct tree *head,
 342                           struct tree *merge)
 343{
 344        int rc;
 345        struct tree_desc t[3];
 346        struct unpack_trees_options opts;
 347
 348        memset(&opts, 0, sizeof(opts));
 349        if (index_only)
 350                opts.index_only = 1;
 351        else
 352                opts.update = 1;
 353        opts.merge = 1;
 354        opts.head_idx = 2;
 355        opts.fn = threeway_merge;
 356        opts.src_index = &the_index;
 357        opts.dst_index = &the_index;
 358        setup_unpack_trees_porcelain(&opts, "merge");
 359
 360        init_tree_desc_from_tree(t+0, common);
 361        init_tree_desc_from_tree(t+1, head);
 362        init_tree_desc_from_tree(t+2, merge);
 363
 364        rc = unpack_trees(3, t, &opts);
 365        cache_tree_free(&active_cache_tree);
 366        return rc;
 367}
 368
 369struct tree *write_tree_from_memory(struct merge_options *o)
 370{
 371        struct tree *result = NULL;
 372
 373        if (unmerged_cache()) {
 374                int i;
 375                fprintf(stderr, "BUG: There are unmerged index entries:\n");
 376                for (i = 0; i < active_nr; i++) {
 377                        const struct cache_entry *ce = active_cache[i];
 378                        if (ce_stage(ce))
 379                                fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
 380                                        (int)ce_namelen(ce), ce->name);
 381                }
 382                die("BUG: unmerged index entries in merge-recursive.c");
 383        }
 384
 385        if (!active_cache_tree)
 386                active_cache_tree = cache_tree();
 387
 388        if (!cache_tree_fully_valid(active_cache_tree) &&
 389            cache_tree_update(&the_index, 0) < 0) {
 390                err(o, _("error building trees"));
 391                return NULL;
 392        }
 393
 394        result = lookup_tree(&active_cache_tree->oid);
 395
 396        return result;
 397}
 398
 399static int save_files_dirs(const unsigned char *sha1,
 400                struct strbuf *base, const char *path,
 401                unsigned int mode, int stage, void *context)
 402{
 403        struct path_hashmap_entry *entry;
 404        int baselen = base->len;
 405        struct merge_options *o = context;
 406
 407        strbuf_addstr(base, path);
 408
 409        FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
 410        hashmap_entry_init(entry, path_hash(entry->path));
 411        hashmap_add(&o->current_file_dir_set, entry);
 412
 413        strbuf_setlen(base, baselen);
 414        return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
 415}
 416
 417static void get_files_dirs(struct merge_options *o, struct tree *tree)
 418{
 419        struct pathspec match_all;
 420        memset(&match_all, 0, sizeof(match_all));
 421        read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
 422}
 423
 424static int get_tree_entry_if_blob(const unsigned char *tree,
 425                                  const char *path,
 426                                  unsigned char *hashy,
 427                                  unsigned int *mode_o)
 428{
 429        int ret;
 430
 431        ret = get_tree_entry(tree, path, hashy, mode_o);
 432        if (S_ISDIR(*mode_o)) {
 433                hashcpy(hashy, null_sha1);
 434                *mode_o = 0;
 435        }
 436        return ret;
 437}
 438
 439/*
 440 * Returns an index_entry instance which doesn't have to correspond to
 441 * a real cache entry in Git's index.
 442 */
 443static struct stage_data *insert_stage_data(const char *path,
 444                struct tree *o, struct tree *a, struct tree *b,
 445                struct string_list *entries)
 446{
 447        struct string_list_item *item;
 448        struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
 449        get_tree_entry_if_blob(o->object.oid.hash, path,
 450                               e->stages[1].oid.hash, &e->stages[1].mode);
 451        get_tree_entry_if_blob(a->object.oid.hash, path,
 452                               e->stages[2].oid.hash, &e->stages[2].mode);
 453        get_tree_entry_if_blob(b->object.oid.hash, path,
 454                               e->stages[3].oid.hash, &e->stages[3].mode);
 455        item = string_list_insert(entries, path);
 456        item->util = e;
 457        return e;
 458}
 459
 460/*
 461 * Create a dictionary mapping file names to stage_data objects. The
 462 * dictionary contains one entry for every path with a non-zero stage entry.
 463 */
 464static struct string_list *get_unmerged(void)
 465{
 466        struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
 467        int i;
 468
 469        unmerged->strdup_strings = 1;
 470
 471        for (i = 0; i < active_nr; i++) {
 472                struct string_list_item *item;
 473                struct stage_data *e;
 474                const struct cache_entry *ce = active_cache[i];
 475                if (!ce_stage(ce))
 476                        continue;
 477
 478                item = string_list_lookup(unmerged, ce->name);
 479                if (!item) {
 480                        item = string_list_insert(unmerged, ce->name);
 481                        item->util = xcalloc(1, sizeof(struct stage_data));
 482                }
 483                e = item->util;
 484                e->stages[ce_stage(ce)].mode = ce->ce_mode;
 485                oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
 486        }
 487
 488        return unmerged;
 489}
 490
 491static int string_list_df_name_compare(const char *one, const char *two)
 492{
 493        int onelen = strlen(one);
 494        int twolen = strlen(two);
 495        /*
 496         * Here we only care that entries for D/F conflicts are
 497         * adjacent, in particular with the file of the D/F conflict
 498         * appearing before files below the corresponding directory.
 499         * The order of the rest of the list is irrelevant for us.
 500         *
 501         * To achieve this, we sort with df_name_compare and provide
 502         * the mode S_IFDIR so that D/F conflicts will sort correctly.
 503         * We use the mode S_IFDIR for everything else for simplicity,
 504         * since in other cases any changes in their order due to
 505         * sorting cause no problems for us.
 506         */
 507        int cmp = df_name_compare(one, onelen, S_IFDIR,
 508                                  two, twolen, S_IFDIR);
 509        /*
 510         * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
 511         * that 'foo' comes before 'foo/bar'.
 512         */
 513        if (cmp)
 514                return cmp;
 515        return onelen - twolen;
 516}
 517
 518static void record_df_conflict_files(struct merge_options *o,
 519                                     struct string_list *entries)
 520{
 521        /* If there is a D/F conflict and the file for such a conflict
 522         * currently exist in the working tree, we want to allow it to be
 523         * removed to make room for the corresponding directory if needed.
 524         * The files underneath the directories of such D/F conflicts will
 525         * be processed before the corresponding file involved in the D/F
 526         * conflict.  If the D/F directory ends up being removed by the
 527         * merge, then we won't have to touch the D/F file.  If the D/F
 528         * directory needs to be written to the working copy, then the D/F
 529         * file will simply be removed (in make_room_for_path()) to make
 530         * room for the necessary paths.  Note that if both the directory
 531         * and the file need to be present, then the D/F file will be
 532         * reinstated with a new unique name at the time it is processed.
 533         */
 534        struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
 535        const char *last_file = NULL;
 536        int last_len = 0;
 537        int i;
 538
 539        /*
 540         * If we're merging merge-bases, we don't want to bother with
 541         * any working directory changes.
 542         */
 543        if (o->call_depth)
 544                return;
 545
 546        /* Ensure D/F conflicts are adjacent in the entries list. */
 547        for (i = 0; i < entries->nr; i++) {
 548                struct string_list_item *next = &entries->items[i];
 549                string_list_append(&df_sorted_entries, next->string)->util =
 550                                   next->util;
 551        }
 552        df_sorted_entries.cmp = string_list_df_name_compare;
 553        string_list_sort(&df_sorted_entries);
 554
 555        string_list_clear(&o->df_conflict_file_set, 1);
 556        for (i = 0; i < df_sorted_entries.nr; i++) {
 557                const char *path = df_sorted_entries.items[i].string;
 558                int len = strlen(path);
 559                struct stage_data *e = df_sorted_entries.items[i].util;
 560
 561                /*
 562                 * Check if last_file & path correspond to a D/F conflict;
 563                 * i.e. whether path is last_file+'/'+<something>.
 564                 * If so, record that it's okay to remove last_file to make
 565                 * room for path and friends if needed.
 566                 */
 567                if (last_file &&
 568                    len > last_len &&
 569                    memcmp(path, last_file, last_len) == 0 &&
 570                    path[last_len] == '/') {
 571                        string_list_insert(&o->df_conflict_file_set, last_file);
 572                }
 573
 574                /*
 575                 * Determine whether path could exist as a file in the
 576                 * working directory as a possible D/F conflict.  This
 577                 * will only occur when it exists in stage 2 as a
 578                 * file.
 579                 */
 580                if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
 581                        last_file = path;
 582                        last_len = len;
 583                } else {
 584                        last_file = NULL;
 585                }
 586        }
 587        string_list_clear(&df_sorted_entries, 0);
 588}
 589
 590struct rename {
 591        struct diff_filepair *pair;
 592        /*
 593         * Purpose of src_entry and dst_entry:
 594         *
 595         * If 'before' is renamed to 'after' then src_entry will contain
 596         * the versions of 'before' from the merge_base, HEAD, and MERGE in
 597         * stages 1, 2, and 3; dst_entry will contain the respective
 598         * versions of 'after' in corresponding locations.  Thus, we have a
 599         * total of six modes and oids, though some will be null.  (Stage 0
 600         * is ignored; we're interested in handling conflicts.)
 601         *
 602         * Since we don't turn on break-rewrites by default, neither
 603         * src_entry nor dst_entry can have all three of their stages have
 604         * non-null oids, meaning at most four of the six will be non-null.
 605         * Also, since this is a rename, both src_entry and dst_entry will
 606         * have at least one non-null oid, meaning at least two will be
 607         * non-null.  Of the six oids, a typical rename will have three be
 608         * non-null.  Only two implies a rename/delete, and four implies a
 609         * rename/add.
 610         */
 611        struct stage_data *src_entry;
 612        struct stage_data *dst_entry;
 613        unsigned processed:1;
 614};
 615
 616static int update_stages(struct merge_options *opt, const char *path,
 617                         const struct diff_filespec *o,
 618                         const struct diff_filespec *a,
 619                         const struct diff_filespec *b)
 620{
 621
 622        /*
 623         * NOTE: It is usually a bad idea to call update_stages on a path
 624         * before calling update_file on that same path, since it can
 625         * sometimes lead to spurious "refusing to lose untracked file..."
 626         * messages from update_file (via make_room_for path via
 627         * would_lose_untracked).  Instead, reverse the order of the calls
 628         * (executing update_file first and then update_stages).
 629         */
 630        int clear = 1;
 631        int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
 632        if (clear)
 633                if (remove_file_from_cache(path))
 634                        return -1;
 635        if (o)
 636                if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
 637                        return -1;
 638        if (a)
 639                if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
 640                        return -1;
 641        if (b)
 642                if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
 643                        return -1;
 644        return 0;
 645}
 646
 647static void update_entry(struct stage_data *entry,
 648                         struct diff_filespec *o,
 649                         struct diff_filespec *a,
 650                         struct diff_filespec *b)
 651{
 652        entry->processed = 0;
 653        entry->stages[1].mode = o->mode;
 654        entry->stages[2].mode = a->mode;
 655        entry->stages[3].mode = b->mode;
 656        oidcpy(&entry->stages[1].oid, &o->oid);
 657        oidcpy(&entry->stages[2].oid, &a->oid);
 658        oidcpy(&entry->stages[3].oid, &b->oid);
 659}
 660
 661static int remove_file(struct merge_options *o, int clean,
 662                       const char *path, int no_wd)
 663{
 664        int update_cache = o->call_depth || clean;
 665        int update_working_directory = !o->call_depth && !no_wd;
 666
 667        if (update_cache) {
 668                if (remove_file_from_cache(path))
 669                        return -1;
 670        }
 671        if (update_working_directory) {
 672                if (ignore_case) {
 673                        struct cache_entry *ce;
 674                        ce = cache_file_exists(path, strlen(path), ignore_case);
 675                        if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
 676                                return 0;
 677                }
 678                if (remove_path(path))
 679                        return -1;
 680        }
 681        return 0;
 682}
 683
 684/* add a string to a strbuf, but converting "/" to "_" */
 685static void add_flattened_path(struct strbuf *out, const char *s)
 686{
 687        size_t i = out->len;
 688        strbuf_addstr(out, s);
 689        for (; i < out->len; i++)
 690                if (out->buf[i] == '/')
 691                        out->buf[i] = '_';
 692}
 693
 694static char *unique_path(struct merge_options *o, const char *path, const char *branch)
 695{
 696        struct path_hashmap_entry *entry;
 697        struct strbuf newpath = STRBUF_INIT;
 698        int suffix = 0;
 699        size_t base_len;
 700
 701        strbuf_addf(&newpath, "%s~", path);
 702        add_flattened_path(&newpath, branch);
 703
 704        base_len = newpath.len;
 705        while (hashmap_get_from_hash(&o->current_file_dir_set,
 706                                     path_hash(newpath.buf), newpath.buf) ||
 707               (!o->call_depth && file_exists(newpath.buf))) {
 708                strbuf_setlen(&newpath, base_len);
 709                strbuf_addf(&newpath, "_%d", suffix++);
 710        }
 711
 712        FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
 713        hashmap_entry_init(entry, path_hash(entry->path));
 714        hashmap_add(&o->current_file_dir_set, entry);
 715        return strbuf_detach(&newpath, NULL);
 716}
 717
 718/**
 719 * Check whether a directory in the index is in the way of an incoming
 720 * file.  Return 1 if so.  If check_working_copy is non-zero, also
 721 * check the working directory.  If empty_ok is non-zero, also return
 722 * 0 in the case where the working-tree dir exists but is empty.
 723 */
 724static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
 725{
 726        int pos;
 727        struct strbuf dirpath = STRBUF_INIT;
 728        struct stat st;
 729
 730        strbuf_addstr(&dirpath, path);
 731        strbuf_addch(&dirpath, '/');
 732
 733        pos = cache_name_pos(dirpath.buf, dirpath.len);
 734
 735        if (pos < 0)
 736                pos = -1 - pos;
 737        if (pos < active_nr &&
 738            !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
 739                strbuf_release(&dirpath);
 740                return 1;
 741        }
 742
 743        strbuf_release(&dirpath);
 744        return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
 745                !(empty_ok && is_empty_dir(path));
 746}
 747
 748static int was_tracked(const char *path)
 749{
 750        int pos = cache_name_pos(path, strlen(path));
 751
 752        if (0 <= pos)
 753                /* we have been tracking this path */
 754                return 1;
 755
 756        /*
 757         * Look for an unmerged entry for the path,
 758         * specifically stage #2, which would indicate
 759         * that "our" side before the merge started
 760         * had the path tracked (and resulted in a conflict).
 761         */
 762        for (pos = -1 - pos;
 763             pos < active_nr && !strcmp(path, active_cache[pos]->name);
 764             pos++)
 765                if (ce_stage(active_cache[pos]) == 2)
 766                        return 1;
 767        return 0;
 768}
 769
 770static int would_lose_untracked(const char *path)
 771{
 772        return !was_tracked(path) && file_exists(path);
 773}
 774
 775static int make_room_for_path(struct merge_options *o, const char *path)
 776{
 777        int status, i;
 778        const char *msg = _("failed to create path '%s'%s");
 779
 780        /* Unlink any D/F conflict files that are in the way */
 781        for (i = 0; i < o->df_conflict_file_set.nr; i++) {
 782                const char *df_path = o->df_conflict_file_set.items[i].string;
 783                size_t pathlen = strlen(path);
 784                size_t df_pathlen = strlen(df_path);
 785                if (df_pathlen < pathlen &&
 786                    path[df_pathlen] == '/' &&
 787                    strncmp(path, df_path, df_pathlen) == 0) {
 788                        output(o, 3,
 789                               _("Removing %s to make room for subdirectory\n"),
 790                               df_path);
 791                        unlink(df_path);
 792                        unsorted_string_list_delete_item(&o->df_conflict_file_set,
 793                                                         i, 0);
 794                        break;
 795                }
 796        }
 797
 798        /* Make sure leading directories are created */
 799        status = safe_create_leading_directories_const(path);
 800        if (status) {
 801                if (status == SCLD_EXISTS)
 802                        /* something else exists */
 803                        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 804                return err(o, msg, path, "");
 805        }
 806
 807        /*
 808         * Do not unlink a file in the work tree if we are not
 809         * tracking it.
 810         */
 811        if (would_lose_untracked(path))
 812                return err(o, _("refusing to lose untracked file at '%s'"),
 813                             path);
 814
 815        /* Successful unlink is good.. */
 816        if (!unlink(path))
 817                return 0;
 818        /* .. and so is no existing file */
 819        if (errno == ENOENT)
 820                return 0;
 821        /* .. but not some other error (who really cares what?) */
 822        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 823}
 824
 825static int update_file_flags(struct merge_options *o,
 826                             const struct object_id *oid,
 827                             unsigned mode,
 828                             const char *path,
 829                             int update_cache,
 830                             int update_wd)
 831{
 832        int ret = 0;
 833
 834        if (o->call_depth)
 835                update_wd = 0;
 836
 837        if (update_wd) {
 838                enum object_type type;
 839                void *buf;
 840                unsigned long size;
 841
 842                if (S_ISGITLINK(mode)) {
 843                        /*
 844                         * We may later decide to recursively descend into
 845                         * the submodule directory and update its index
 846                         * and/or work tree, but we do not do that now.
 847                         */
 848                        update_wd = 0;
 849                        goto update_index;
 850                }
 851
 852                buf = read_sha1_file(oid->hash, &type, &size);
 853                if (!buf)
 854                        return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
 855                if (type != OBJ_BLOB) {
 856                        ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
 857                        goto free_buf;
 858                }
 859                if (S_ISREG(mode)) {
 860                        struct strbuf strbuf = STRBUF_INIT;
 861                        if (convert_to_working_tree(path, buf, size, &strbuf)) {
 862                                free(buf);
 863                                size = strbuf.len;
 864                                buf = strbuf_detach(&strbuf, NULL);
 865                        }
 866                }
 867
 868                if (make_room_for_path(o, path) < 0) {
 869                        update_wd = 0;
 870                        goto free_buf;
 871                }
 872                if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
 873                        int fd;
 874                        if (mode & 0100)
 875                                mode = 0777;
 876                        else
 877                                mode = 0666;
 878                        fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
 879                        if (fd < 0) {
 880                                ret = err(o, _("failed to open '%s': %s"),
 881                                          path, strerror(errno));
 882                                goto free_buf;
 883                        }
 884                        write_in_full(fd, buf, size);
 885                        close(fd);
 886                } else if (S_ISLNK(mode)) {
 887                        char *lnk = xmemdupz(buf, size);
 888                        safe_create_leading_directories_const(path);
 889                        unlink(path);
 890                        if (symlink(lnk, path))
 891                                ret = err(o, _("failed to symlink '%s': %s"),
 892                                        path, strerror(errno));
 893                        free(lnk);
 894                } else
 895                        ret = err(o,
 896                                  _("do not know what to do with %06o %s '%s'"),
 897                                  mode, oid_to_hex(oid), path);
 898 free_buf:
 899                free(buf);
 900        }
 901 update_index:
 902        if (!ret && update_cache)
 903                add_cacheinfo(o, mode, oid, path, 0, update_wd, ADD_CACHE_OK_TO_ADD);
 904        return ret;
 905}
 906
 907static int update_file(struct merge_options *o,
 908                       int clean,
 909                       const struct object_id *oid,
 910                       unsigned mode,
 911                       const char *path)
 912{
 913        return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
 914}
 915
 916/* Low level file merging, update and removal */
 917
 918struct merge_file_info {
 919        struct object_id oid;
 920        unsigned mode;
 921        unsigned clean:1,
 922                 merge:1;
 923};
 924
 925static int merge_3way(struct merge_options *o,
 926                      mmbuffer_t *result_buf,
 927                      const struct diff_filespec *one,
 928                      const struct diff_filespec *a,
 929                      const struct diff_filespec *b,
 930                      const char *branch1,
 931                      const char *branch2)
 932{
 933        mmfile_t orig, src1, src2;
 934        struct ll_merge_options ll_opts = {0};
 935        char *base_name, *name1, *name2;
 936        int merge_status;
 937
 938        ll_opts.renormalize = o->renormalize;
 939        ll_opts.xdl_opts = o->xdl_opts;
 940
 941        if (o->call_depth) {
 942                ll_opts.virtual_ancestor = 1;
 943                ll_opts.variant = 0;
 944        } else {
 945                switch (o->recursive_variant) {
 946                case MERGE_RECURSIVE_OURS:
 947                        ll_opts.variant = XDL_MERGE_FAVOR_OURS;
 948                        break;
 949                case MERGE_RECURSIVE_THEIRS:
 950                        ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
 951                        break;
 952                default:
 953                        ll_opts.variant = 0;
 954                        break;
 955                }
 956        }
 957
 958        if (strcmp(a->path, b->path) ||
 959            (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
 960                base_name = o->ancestor == NULL ? NULL :
 961                        mkpathdup("%s:%s", o->ancestor, one->path);
 962                name1 = mkpathdup("%s:%s", branch1, a->path);
 963                name2 = mkpathdup("%s:%s", branch2, b->path);
 964        } else {
 965                base_name = o->ancestor == NULL ? NULL :
 966                        mkpathdup("%s", o->ancestor);
 967                name1 = mkpathdup("%s", branch1);
 968                name2 = mkpathdup("%s", branch2);
 969        }
 970
 971        read_mmblob(&orig, &one->oid);
 972        read_mmblob(&src1, &a->oid);
 973        read_mmblob(&src2, &b->oid);
 974
 975        merge_status = ll_merge(result_buf, a->path, &orig, base_name,
 976                                &src1, name1, &src2, name2, &ll_opts);
 977
 978        free(base_name);
 979        free(name1);
 980        free(name2);
 981        free(orig.ptr);
 982        free(src1.ptr);
 983        free(src2.ptr);
 984        return merge_status;
 985}
 986
 987static int merge_file_1(struct merge_options *o,
 988                                           const struct diff_filespec *one,
 989                                           const struct diff_filespec *a,
 990                                           const struct diff_filespec *b,
 991                                           const char *branch1,
 992                                           const char *branch2,
 993                                           struct merge_file_info *result)
 994{
 995        result->merge = 0;
 996        result->clean = 1;
 997
 998        if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
 999                result->clean = 0;
1000                if (S_ISREG(a->mode)) {
1001                        result->mode = a->mode;
1002                        oidcpy(&result->oid, &a->oid);
1003                } else {
1004                        result->mode = b->mode;
1005                        oidcpy(&result->oid, &b->oid);
1006                }
1007        } else {
1008                if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1009                        result->merge = 1;
1010
1011                /*
1012                 * Merge modes
1013                 */
1014                if (a->mode == b->mode || a->mode == one->mode)
1015                        result->mode = b->mode;
1016                else {
1017                        result->mode = a->mode;
1018                        if (b->mode != one->mode) {
1019                                result->clean = 0;
1020                                result->merge = 1;
1021                        }
1022                }
1023
1024                if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1025                        oidcpy(&result->oid, &b->oid);
1026                else if (oid_eq(&b->oid, &one->oid))
1027                        oidcpy(&result->oid, &a->oid);
1028                else if (S_ISREG(a->mode)) {
1029                        mmbuffer_t result_buf;
1030                        int ret = 0, merge_status;
1031
1032                        merge_status = merge_3way(o, &result_buf, one, a, b,
1033                                                  branch1, branch2);
1034
1035                        if ((merge_status < 0) || !result_buf.ptr)
1036                                ret = err(o, _("Failed to execute internal merge"));
1037
1038                        if (!ret && write_sha1_file(result_buf.ptr, result_buf.size,
1039                                                    blob_type, result->oid.hash))
1040                                ret = err(o, _("Unable to add %s to database"),
1041                                          a->path);
1042
1043                        free(result_buf.ptr);
1044                        if (ret)
1045                                return ret;
1046                        result->clean = (merge_status == 0);
1047                } else if (S_ISGITLINK(a->mode)) {
1048                        result->clean = merge_submodule(&result->oid,
1049                                                       one->path,
1050                                                       &one->oid,
1051                                                       &a->oid,
1052                                                       &b->oid,
1053                                                       !o->call_depth);
1054                } else if (S_ISLNK(a->mode)) {
1055                        oidcpy(&result->oid, &a->oid);
1056
1057                        if (!oid_eq(&a->oid, &b->oid))
1058                                result->clean = 0;
1059                } else
1060                        die("BUG: unsupported object type in the tree");
1061        }
1062
1063        return 0;
1064}
1065
1066static int merge_file_special_markers(struct merge_options *o,
1067                           const struct diff_filespec *one,
1068                           const struct diff_filespec *a,
1069                           const struct diff_filespec *b,
1070                           const char *branch1,
1071                           const char *filename1,
1072                           const char *branch2,
1073                           const char *filename2,
1074                           struct merge_file_info *mfi)
1075{
1076        char *side1 = NULL;
1077        char *side2 = NULL;
1078        int ret;
1079
1080        if (filename1)
1081                side1 = xstrfmt("%s:%s", branch1, filename1);
1082        if (filename2)
1083                side2 = xstrfmt("%s:%s", branch2, filename2);
1084
1085        ret = merge_file_1(o, one, a, b,
1086                           side1 ? side1 : branch1,
1087                           side2 ? side2 : branch2, mfi);
1088        free(side1);
1089        free(side2);
1090        return ret;
1091}
1092
1093static int merge_file_one(struct merge_options *o,
1094                                         const char *path,
1095                                         const struct object_id *o_oid, int o_mode,
1096                                         const struct object_id *a_oid, int a_mode,
1097                                         const struct object_id *b_oid, int b_mode,
1098                                         const char *branch1,
1099                                         const char *branch2,
1100                                         struct merge_file_info *mfi)
1101{
1102        struct diff_filespec one, a, b;
1103
1104        one.path = a.path = b.path = (char *)path;
1105        oidcpy(&one.oid, o_oid);
1106        one.mode = o_mode;
1107        oidcpy(&a.oid, a_oid);
1108        a.mode = a_mode;
1109        oidcpy(&b.oid, b_oid);
1110        b.mode = b_mode;
1111        return merge_file_1(o, &one, &a, &b, branch1, branch2, mfi);
1112}
1113
1114static int handle_change_delete(struct merge_options *o,
1115                                 const char *path, const char *old_path,
1116                                 const struct object_id *o_oid, int o_mode,
1117                                 const struct object_id *changed_oid,
1118                                 int changed_mode,
1119                                 const char *change_branch,
1120                                 const char *delete_branch,
1121                                 const char *change, const char *change_past)
1122{
1123        char *alt_path = NULL;
1124        const char *update_path = path;
1125        int ret = 0;
1126
1127        if (dir_in_way(path, !o->call_depth, 0)) {
1128                update_path = alt_path = unique_path(o, path, change_branch);
1129        }
1130
1131        if (o->call_depth) {
1132                /*
1133                 * We cannot arbitrarily accept either a_sha or b_sha as
1134                 * correct; since there is no true "middle point" between
1135                 * them, simply reuse the base version for virtual merge base.
1136                 */
1137                ret = remove_file_from_cache(path);
1138                if (!ret)
1139                        ret = update_file(o, 0, o_oid, o_mode, update_path);
1140        } else {
1141                if (!alt_path) {
1142                        if (!old_path) {
1143                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1144                                       "and %s in %s. Version %s of %s left in tree."),
1145                                       change, path, delete_branch, change_past,
1146                                       change_branch, change_branch, path);
1147                        } else {
1148                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1149                                       "and %s to %s in %s. Version %s of %s left in tree."),
1150                                       change, old_path, delete_branch, change_past, path,
1151                                       change_branch, change_branch, path);
1152                        }
1153                } else {
1154                        if (!old_path) {
1155                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1156                                       "and %s in %s. Version %s of %s left in tree at %s."),
1157                                       change, path, delete_branch, change_past,
1158                                       change_branch, change_branch, path, alt_path);
1159                        } else {
1160                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1161                                       "and %s to %s in %s. Version %s of %s left in tree at %s."),
1162                                       change, old_path, delete_branch, change_past, path,
1163                                       change_branch, change_branch, path, alt_path);
1164                        }
1165                }
1166                /*
1167                 * No need to call update_file() on path when change_branch ==
1168                 * o->branch1 && !alt_path, since that would needlessly touch
1169                 * path.  We could call update_file_flags() with update_cache=0
1170                 * and update_wd=0, but that's a no-op.
1171                 */
1172                if (change_branch != o->branch1 || alt_path)
1173                        ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1174        }
1175        free(alt_path);
1176
1177        return ret;
1178}
1179
1180static int conflict_rename_delete(struct merge_options *o,
1181                                   struct diff_filepair *pair,
1182                                   const char *rename_branch,
1183                                   const char *delete_branch)
1184{
1185        const struct diff_filespec *orig = pair->one;
1186        const struct diff_filespec *dest = pair->two;
1187
1188        if (handle_change_delete(o,
1189                                 o->call_depth ? orig->path : dest->path,
1190                                 o->call_depth ? NULL : orig->path,
1191                                 &orig->oid, orig->mode,
1192                                 &dest->oid, dest->mode,
1193                                 rename_branch, delete_branch,
1194                                 _("rename"), _("renamed")))
1195                return -1;
1196
1197        if (o->call_depth)
1198                return remove_file_from_cache(dest->path);
1199        else
1200                return update_stages(o, dest->path, NULL,
1201                                     rename_branch == o->branch1 ? dest : NULL,
1202                                     rename_branch == o->branch1 ? NULL : dest);
1203}
1204
1205static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1206                                                 struct stage_data *entry,
1207                                                 int stage)
1208{
1209        struct object_id *oid = &entry->stages[stage].oid;
1210        unsigned mode = entry->stages[stage].mode;
1211        if (mode == 0 || is_null_oid(oid))
1212                return NULL;
1213        oidcpy(&target->oid, oid);
1214        target->mode = mode;
1215        return target;
1216}
1217
1218static int handle_file(struct merge_options *o,
1219                        struct diff_filespec *rename,
1220                        int stage,
1221                        struct rename_conflict_info *ci)
1222{
1223        char *dst_name = rename->path;
1224        struct stage_data *dst_entry;
1225        const char *cur_branch, *other_branch;
1226        struct diff_filespec other;
1227        struct diff_filespec *add;
1228        int ret;
1229
1230        if (stage == 2) {
1231                dst_entry = ci->dst_entry1;
1232                cur_branch = ci->branch1;
1233                other_branch = ci->branch2;
1234        } else {
1235                dst_entry = ci->dst_entry2;
1236                cur_branch = ci->branch2;
1237                other_branch = ci->branch1;
1238        }
1239
1240        add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1241        if (add) {
1242                char *add_name = unique_path(o, rename->path, other_branch);
1243                if (update_file(o, 0, &add->oid, add->mode, add_name))
1244                        return -1;
1245
1246                remove_file(o, 0, rename->path, 0);
1247                dst_name = unique_path(o, rename->path, cur_branch);
1248        } else {
1249                if (dir_in_way(rename->path, !o->call_depth, 0)) {
1250                        dst_name = unique_path(o, rename->path, cur_branch);
1251                        output(o, 1, _("%s is a directory in %s adding as %s instead"),
1252                               rename->path, other_branch, dst_name);
1253                }
1254        }
1255        if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1256                ; /* fall through, do allow dst_name to be released */
1257        else if (stage == 2)
1258                ret = update_stages(o, rename->path, NULL, rename, add);
1259        else
1260                ret = update_stages(o, rename->path, NULL, add, rename);
1261
1262        if (dst_name != rename->path)
1263                free(dst_name);
1264
1265        return ret;
1266}
1267
1268static int conflict_rename_rename_1to2(struct merge_options *o,
1269                                        struct rename_conflict_info *ci)
1270{
1271        /* One file was renamed in both branches, but to different names. */
1272        struct diff_filespec *one = ci->pair1->one;
1273        struct diff_filespec *a = ci->pair1->two;
1274        struct diff_filespec *b = ci->pair2->two;
1275
1276        output(o, 1, _("CONFLICT (rename/rename): "
1277               "Rename \"%s\"->\"%s\" in branch \"%s\" "
1278               "rename \"%s\"->\"%s\" in \"%s\"%s"),
1279               one->path, a->path, ci->branch1,
1280               one->path, b->path, ci->branch2,
1281               o->call_depth ? _(" (left unresolved)") : "");
1282        if (o->call_depth) {
1283                struct merge_file_info mfi;
1284                struct diff_filespec other;
1285                struct diff_filespec *add;
1286                if (merge_file_one(o, one->path,
1287                                 &one->oid, one->mode,
1288                                 &a->oid, a->mode,
1289                                 &b->oid, b->mode,
1290                                 ci->branch1, ci->branch2, &mfi))
1291                        return -1;
1292
1293                /*
1294                 * FIXME: For rename/add-source conflicts (if we could detect
1295                 * such), this is wrong.  We should instead find a unique
1296                 * pathname and then either rename the add-source file to that
1297                 * unique path, or use that unique path instead of src here.
1298                 */
1299                if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1300                        return -1;
1301
1302                /*
1303                 * Above, we put the merged content at the merge-base's
1304                 * path.  Now we usually need to delete both a->path and
1305                 * b->path.  However, the rename on each side of the merge
1306                 * could also be involved in a rename/add conflict.  In
1307                 * such cases, we should keep the added file around,
1308                 * resolving the conflict at that path in its favor.
1309                 */
1310                add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1311                if (add) {
1312                        if (update_file(o, 0, &add->oid, add->mode, a->path))
1313                                return -1;
1314                }
1315                else
1316                        remove_file_from_cache(a->path);
1317                add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1318                if (add) {
1319                        if (update_file(o, 0, &add->oid, add->mode, b->path))
1320                                return -1;
1321                }
1322                else
1323                        remove_file_from_cache(b->path);
1324        } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1325                return -1;
1326
1327        return 0;
1328}
1329
1330static int conflict_rename_rename_2to1(struct merge_options *o,
1331                                        struct rename_conflict_info *ci)
1332{
1333        /* Two files, a & b, were renamed to the same thing, c. */
1334        struct diff_filespec *a = ci->pair1->one;
1335        struct diff_filespec *b = ci->pair2->one;
1336        struct diff_filespec *c1 = ci->pair1->two;
1337        struct diff_filespec *c2 = ci->pair2->two;
1338        char *path = c1->path; /* == c2->path */
1339        struct merge_file_info mfi_c1;
1340        struct merge_file_info mfi_c2;
1341        int ret;
1342
1343        output(o, 1, _("CONFLICT (rename/rename): "
1344               "Rename %s->%s in %s. "
1345               "Rename %s->%s in %s"),
1346               a->path, c1->path, ci->branch1,
1347               b->path, c2->path, ci->branch2);
1348
1349        remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1350        remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1351
1352        if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1353                                       o->branch1, c1->path,
1354                                       o->branch2, ci->ren1_other.path, &mfi_c1) ||
1355            merge_file_special_markers(o, b, &ci->ren2_other, c2,
1356                                       o->branch1, ci->ren2_other.path,
1357                                       o->branch2, c2->path, &mfi_c2))
1358                return -1;
1359
1360        if (o->call_depth) {
1361                /*
1362                 * If mfi_c1.clean && mfi_c2.clean, then it might make
1363                 * sense to do a two-way merge of those results.  But, I
1364                 * think in all cases, it makes sense to have the virtual
1365                 * merge base just undo the renames; they can be detected
1366                 * again later for the non-recursive merge.
1367                 */
1368                remove_file(o, 0, path, 0);
1369                ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1370                if (!ret)
1371                        ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1372                                          b->path);
1373        } else {
1374                char *new_path1 = unique_path(o, path, ci->branch1);
1375                char *new_path2 = unique_path(o, path, ci->branch2);
1376                output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1377                       a->path, new_path1, b->path, new_path2);
1378                remove_file(o, 0, path, 0);
1379                ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1380                if (!ret)
1381                        ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1382                                          new_path2);
1383                free(new_path2);
1384                free(new_path1);
1385        }
1386
1387        return ret;
1388}
1389
1390/*
1391 * Get the diff_filepairs changed between o_tree and tree.
1392 */
1393static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1394                                               struct tree *o_tree,
1395                                               struct tree *tree)
1396{
1397        struct diff_queue_struct *ret;
1398        struct diff_options opts;
1399
1400        diff_setup(&opts);
1401        opts.flags.recursive = 1;
1402        opts.flags.rename_empty = 0;
1403        opts.detect_rename = DIFF_DETECT_RENAME;
1404        opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1405                            o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1406                            1000;
1407        opts.rename_score = o->rename_score;
1408        opts.show_rename_progress = o->show_rename_progress;
1409        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1410        diff_setup_done(&opts);
1411        diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1412        diffcore_std(&opts);
1413        if (opts.needed_rename_limit > o->needed_rename_limit)
1414                o->needed_rename_limit = opts.needed_rename_limit;
1415
1416        ret = xmalloc(sizeof(*ret));
1417        *ret = diff_queued_diff;
1418
1419        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1420        diff_queued_diff.nr = 0;
1421        diff_queued_diff.queue = NULL;
1422        diff_flush(&opts);
1423        return ret;
1424}
1425
1426static int tree_has_path(struct tree *tree, const char *path)
1427{
1428        unsigned char hashy[GIT_MAX_RAWSZ];
1429        unsigned int mode_o;
1430
1431        return !get_tree_entry(tree->object.oid.hash, path,
1432                               hashy, &mode_o);
1433}
1434
1435/*
1436 * Return a new string that replaces the beginning portion (which matches
1437 * entry->dir), with entry->new_dir.  In perl-speak:
1438 *   new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1439 * NOTE:
1440 *   Caller must ensure that old_path starts with entry->dir + '/'.
1441 */
1442static char *apply_dir_rename(struct dir_rename_entry *entry,
1443                              const char *old_path)
1444{
1445        struct strbuf new_path = STRBUF_INIT;
1446        int oldlen, newlen;
1447
1448        if (entry->non_unique_new_dir)
1449                return NULL;
1450
1451        oldlen = strlen(entry->dir);
1452        newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1453        strbuf_grow(&new_path, newlen);
1454        strbuf_addbuf(&new_path, &entry->new_dir);
1455        strbuf_addstr(&new_path, &old_path[oldlen]);
1456
1457        return strbuf_detach(&new_path, NULL);
1458}
1459
1460static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1461                                    char **old_dir, char **new_dir)
1462{
1463        char *end_of_old, *end_of_new;
1464        int old_len, new_len;
1465
1466        *old_dir = NULL;
1467        *new_dir = NULL;
1468
1469        /*
1470         * For
1471         *    "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1472         * the "e/foo.c" part is the same, we just want to know that
1473         *    "a/b/c/d" was renamed to "a/b/some/thing/else"
1474         * so, for this example, this function returns "a/b/c/d" in
1475         * *old_dir and "a/b/some/thing/else" in *new_dir.
1476         *
1477         * Also, if the basename of the file changed, we don't care.  We
1478         * want to know which portion of the directory, if any, changed.
1479         */
1480        end_of_old = strrchr(old_path, '/');
1481        end_of_new = strrchr(new_path, '/');
1482
1483        if (end_of_old == NULL || end_of_new == NULL)
1484                return;
1485        while (*--end_of_new == *--end_of_old &&
1486               end_of_old != old_path &&
1487               end_of_new != new_path)
1488                ; /* Do nothing; all in the while loop */
1489        /*
1490         * We've found the first non-matching character in the directory
1491         * paths.  That means the current directory we were comparing
1492         * represents the rename.  Move end_of_old and end_of_new back
1493         * to the full directory name.
1494         */
1495        if (*end_of_old == '/')
1496                end_of_old++;
1497        if (*end_of_old != '/')
1498                end_of_new++;
1499        end_of_old = strchr(end_of_old, '/');
1500        end_of_new = strchr(end_of_new, '/');
1501
1502        /*
1503         * It may have been the case that old_path and new_path were the same
1504         * directory all along.  Don't claim a rename if they're the same.
1505         */
1506        old_len = end_of_old - old_path;
1507        new_len = end_of_new - new_path;
1508
1509        if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1510                *old_dir = xstrndup(old_path, old_len);
1511                *new_dir = xstrndup(new_path, new_len);
1512        }
1513}
1514
1515static void remove_hashmap_entries(struct hashmap *dir_renames,
1516                                   struct string_list *items_to_remove)
1517{
1518        int i;
1519        struct dir_rename_entry *entry;
1520
1521        for (i = 0; i < items_to_remove->nr; i++) {
1522                entry = items_to_remove->items[i].util;
1523                hashmap_remove(dir_renames, entry, NULL);
1524        }
1525        string_list_clear(items_to_remove, 0);
1526}
1527
1528/*
1529 * See if there is a directory rename for path, and if there are any file
1530 * level conflicts for the renamed location.  If there is a rename and
1531 * there are no conflicts, return the new name.  Otherwise, return NULL.
1532 */
1533static char *handle_path_level_conflicts(struct merge_options *o,
1534                                         const char *path,
1535                                         struct dir_rename_entry *entry,
1536                                         struct hashmap *collisions,
1537                                         struct tree *tree)
1538{
1539        char *new_path = NULL;
1540        struct collision_entry *collision_ent;
1541        int clean = 1;
1542        struct strbuf collision_paths = STRBUF_INIT;
1543
1544        /*
1545         * entry has the mapping of old directory name to new directory name
1546         * that we want to apply to path.
1547         */
1548        new_path = apply_dir_rename(entry, path);
1549
1550        if (!new_path) {
1551                /* This should only happen when entry->non_unique_new_dir set */
1552                if (!entry->non_unique_new_dir)
1553                        BUG("entry->non_unqiue_dir not set and !new_path");
1554                output(o, 1, _("CONFLICT (directory rename split): "
1555                               "Unclear where to place %s because directory "
1556                               "%s was renamed to multiple other directories, "
1557                               "with no destination getting a majority of the "
1558                               "files."),
1559                       path, entry->dir);
1560                clean = 0;
1561                return NULL;
1562        }
1563
1564        /*
1565         * The caller needs to have ensured that it has pre-populated
1566         * collisions with all paths that map to new_path.  Do a quick check
1567         * to ensure that's the case.
1568         */
1569        collision_ent = collision_find_entry(collisions, new_path);
1570        if (collision_ent == NULL)
1571                BUG("collision_ent is NULL");
1572
1573        /*
1574         * Check for one-sided add/add/.../add conflicts, i.e.
1575         * where implicit renames from the other side doing
1576         * directory rename(s) can affect this side of history
1577         * to put multiple paths into the same location.  Warn
1578         * and bail on directory renames for such paths.
1579         */
1580        if (collision_ent->reported_already) {
1581                clean = 0;
1582        } else if (tree_has_path(tree, new_path)) {
1583                collision_ent->reported_already = 1;
1584                strbuf_add_separated_string_list(&collision_paths, ", ",
1585                                                 &collision_ent->source_files);
1586                output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1587                               "file/dir at %s in the way of implicit "
1588                               "directory rename(s) putting the following "
1589                               "path(s) there: %s."),
1590                       new_path, collision_paths.buf);
1591                clean = 0;
1592        } else if (collision_ent->source_files.nr > 1) {
1593                collision_ent->reported_already = 1;
1594                strbuf_add_separated_string_list(&collision_paths, ", ",
1595                                                 &collision_ent->source_files);
1596                output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1597                               "more than one path to %s; implicit directory "
1598                               "renames tried to put these paths there: %s"),
1599                       new_path, collision_paths.buf);
1600                clean = 0;
1601        }
1602
1603        /* Free memory we no longer need */
1604        strbuf_release(&collision_paths);
1605        if (!clean && new_path) {
1606                free(new_path);
1607                return NULL;
1608        }
1609
1610        return new_path;
1611}
1612
1613/*
1614 * There are a couple things we want to do at the directory level:
1615 *   1. Check for both sides renaming to the same thing, in order to avoid
1616 *      implicit renaming of files that should be left in place.  (See
1617 *      testcase 6b in t6043 for details.)
1618 *   2. Prune directory renames if there are still files left in the
1619 *      the original directory.  These represent a partial directory rename,
1620 *      i.e. a rename where only some of the files within the directory
1621 *      were renamed elsewhere.  (Technically, this could be done earlier
1622 *      in get_directory_renames(), except that would prevent us from
1623 *      doing the previous check and thus failing testcase 6b.)
1624 *   3. Check for rename/rename(1to2) conflicts (at the directory level).
1625 *      In the future, we could potentially record this info as well and
1626 *      omit reporting rename/rename(1to2) conflicts for each path within
1627 *      the affected directories, thus cleaning up the merge output.
1628 *   NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1629 *         directory level, because merging directories is fine.  If it
1630 *         causes conflicts for files within those merged directories, then
1631 *         that should be detected at the individual path level.
1632 */
1633static void handle_directory_level_conflicts(struct merge_options *o,
1634                                             struct hashmap *dir_re_head,
1635                                             struct tree *head,
1636                                             struct hashmap *dir_re_merge,
1637                                             struct tree *merge)
1638{
1639        struct hashmap_iter iter;
1640        struct dir_rename_entry *head_ent;
1641        struct dir_rename_entry *merge_ent;
1642
1643        struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1644        struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1645
1646        hashmap_iter_init(dir_re_head, &iter);
1647        while ((head_ent = hashmap_iter_next(&iter))) {
1648                merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1649                if (merge_ent &&
1650                    !head_ent->non_unique_new_dir &&
1651                    !merge_ent->non_unique_new_dir &&
1652                    !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1653                        /* 1. Renamed identically; remove it from both sides */
1654                        string_list_append(&remove_from_head,
1655                                           head_ent->dir)->util = head_ent;
1656                        strbuf_release(&head_ent->new_dir);
1657                        string_list_append(&remove_from_merge,
1658                                           merge_ent->dir)->util = merge_ent;
1659                        strbuf_release(&merge_ent->new_dir);
1660                } else if (tree_has_path(head, head_ent->dir)) {
1661                        /* 2. This wasn't a directory rename after all */
1662                        string_list_append(&remove_from_head,
1663                                           head_ent->dir)->util = head_ent;
1664                        strbuf_release(&head_ent->new_dir);
1665                }
1666        }
1667
1668        remove_hashmap_entries(dir_re_head, &remove_from_head);
1669        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1670
1671        hashmap_iter_init(dir_re_merge, &iter);
1672        while ((merge_ent = hashmap_iter_next(&iter))) {
1673                head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1674                if (tree_has_path(merge, merge_ent->dir)) {
1675                        /* 2. This wasn't a directory rename after all */
1676                        string_list_append(&remove_from_merge,
1677                                           merge_ent->dir)->util = merge_ent;
1678                } else if (head_ent &&
1679                           !head_ent->non_unique_new_dir &&
1680                           !merge_ent->non_unique_new_dir) {
1681                        /* 3. rename/rename(1to2) */
1682                        /*
1683                         * We can assume it's not rename/rename(1to1) because
1684                         * that was case (1), already checked above.  So we
1685                         * know that head_ent->new_dir and merge_ent->new_dir
1686                         * are different strings.
1687                         */
1688                        output(o, 1, _("CONFLICT (rename/rename): "
1689                                       "Rename directory %s->%s in %s. "
1690                                       "Rename directory %s->%s in %s"),
1691                               head_ent->dir, head_ent->new_dir.buf, o->branch1,
1692                               head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1693                        string_list_append(&remove_from_head,
1694                                           head_ent->dir)->util = head_ent;
1695                        strbuf_release(&head_ent->new_dir);
1696                        string_list_append(&remove_from_merge,
1697                                           merge_ent->dir)->util = merge_ent;
1698                        strbuf_release(&merge_ent->new_dir);
1699                }
1700        }
1701
1702        remove_hashmap_entries(dir_re_head, &remove_from_head);
1703        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1704}
1705
1706static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1707                                             struct tree *tree)
1708{
1709        struct hashmap *dir_renames;
1710        struct hashmap_iter iter;
1711        struct dir_rename_entry *entry;
1712        int i;
1713
1714        /*
1715         * Typically, we think of a directory rename as all files from a
1716         * certain directory being moved to a target directory.  However,
1717         * what if someone first moved two files from the original
1718         * directory in one commit, and then renamed the directory
1719         * somewhere else in a later commit?  At merge time, we just know
1720         * that files from the original directory went to two different
1721         * places, and that the bulk of them ended up in the same place.
1722         * We want each directory rename to represent where the bulk of the
1723         * files from that directory end up; this function exists to find
1724         * where the bulk of the files went.
1725         *
1726         * The first loop below simply iterates through the list of file
1727         * renames, finding out how often each directory rename pair
1728         * possibility occurs.
1729         */
1730        dir_renames = xmalloc(sizeof(struct hashmap));
1731        dir_rename_init(dir_renames);
1732        for (i = 0; i < pairs->nr; ++i) {
1733                struct string_list_item *item;
1734                int *count;
1735                struct diff_filepair *pair = pairs->queue[i];
1736                char *old_dir, *new_dir;
1737
1738                /* File not part of directory rename if it wasn't renamed */
1739                if (pair->status != 'R')
1740                        continue;
1741
1742                get_renamed_dir_portion(pair->one->path, pair->two->path,
1743                                        &old_dir,        &new_dir);
1744                if (!old_dir)
1745                        /* Directory didn't change at all; ignore this one. */
1746                        continue;
1747
1748                entry = dir_rename_find_entry(dir_renames, old_dir);
1749                if (!entry) {
1750                        entry = xmalloc(sizeof(struct dir_rename_entry));
1751                        dir_rename_entry_init(entry, old_dir);
1752                        hashmap_put(dir_renames, entry);
1753                } else {
1754                        free(old_dir);
1755                }
1756                item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1757                if (!item) {
1758                        item = string_list_insert(&entry->possible_new_dirs,
1759                                                  new_dir);
1760                        item->util = xcalloc(1, sizeof(int));
1761                } else {
1762                        free(new_dir);
1763                }
1764                count = item->util;
1765                *count += 1;
1766        }
1767
1768        /*
1769         * For each directory with files moved out of it, we find out which
1770         * target directory received the most files so we can declare it to
1771         * be the "winning" target location for the directory rename.  This
1772         * winner gets recorded in new_dir.  If there is no winner
1773         * (multiple target directories received the same number of files),
1774         * we set non_unique_new_dir.  Once we've determined the winner (or
1775         * that there is no winner), we no longer need possible_new_dirs.
1776         */
1777        hashmap_iter_init(dir_renames, &iter);
1778        while ((entry = hashmap_iter_next(&iter))) {
1779                int max = 0;
1780                int bad_max = 0;
1781                char *best = NULL;
1782
1783                for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1784                        int *count = entry->possible_new_dirs.items[i].util;
1785
1786                        if (*count == max)
1787                                bad_max = max;
1788                        else if (*count > max) {
1789                                max = *count;
1790                                best = entry->possible_new_dirs.items[i].string;
1791                        }
1792                }
1793                if (bad_max == max)
1794                        entry->non_unique_new_dir = 1;
1795                else {
1796                        assert(entry->new_dir.len == 0);
1797                        strbuf_addstr(&entry->new_dir, best);
1798                }
1799                /*
1800                 * The relevant directory sub-portion of the original full
1801                 * filepaths were xstrndup'ed before inserting into
1802                 * possible_new_dirs, and instead of manually iterating the
1803                 * list and free'ing each, just lie and tell
1804                 * possible_new_dirs that it did the strdup'ing so that it
1805                 * will free them for us.
1806                 */
1807                entry->possible_new_dirs.strdup_strings = 1;
1808                string_list_clear(&entry->possible_new_dirs, 1);
1809        }
1810
1811        return dir_renames;
1812}
1813
1814static struct dir_rename_entry *check_dir_renamed(const char *path,
1815                                                  struct hashmap *dir_renames)
1816{
1817        char temp[PATH_MAX];
1818        char *end;
1819        struct dir_rename_entry *entry;
1820
1821        strcpy(temp, path);
1822        while ((end = strrchr(temp, '/'))) {
1823                *end = '\0';
1824                entry = dir_rename_find_entry(dir_renames, temp);
1825                if (entry)
1826                        return entry;
1827        }
1828        return NULL;
1829}
1830
1831static void compute_collisions(struct hashmap *collisions,
1832                               struct hashmap *dir_renames,
1833                               struct diff_queue_struct *pairs)
1834{
1835        int i;
1836
1837        /*
1838         * Multiple files can be mapped to the same path due to directory
1839         * renames done by the other side of history.  Since that other
1840         * side of history could have merged multiple directories into one,
1841         * if our side of history added the same file basename to each of
1842         * those directories, then all N of them would get implicitly
1843         * renamed by the directory rename detection into the same path,
1844         * and we'd get an add/add/.../add conflict, and all those adds
1845         * from *this* side of history.  This is not representable in the
1846         * index, and users aren't going to easily be able to make sense of
1847         * it.  So we need to provide a good warning about what's
1848         * happening, and fall back to no-directory-rename detection
1849         * behavior for those paths.
1850         *
1851         * See testcases 9e and all of section 5 from t6043 for examples.
1852         */
1853        collision_init(collisions);
1854
1855        for (i = 0; i < pairs->nr; ++i) {
1856                struct dir_rename_entry *dir_rename_ent;
1857                struct collision_entry *collision_ent;
1858                char *new_path;
1859                struct diff_filepair *pair = pairs->queue[i];
1860
1861                if (pair->status == 'D')
1862                        continue;
1863                dir_rename_ent = check_dir_renamed(pair->two->path,
1864                                                   dir_renames);
1865                if (!dir_rename_ent)
1866                        continue;
1867
1868                new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
1869                if (!new_path)
1870                        /*
1871                         * dir_rename_ent->non_unique_new_path is true, which
1872                         * means there is no directory rename for us to use,
1873                         * which means it won't cause us any additional
1874                         * collisions.
1875                         */
1876                        continue;
1877                collision_ent = collision_find_entry(collisions, new_path);
1878                if (!collision_ent) {
1879                        collision_ent = xcalloc(1,
1880                                                sizeof(struct collision_entry));
1881                        hashmap_entry_init(collision_ent, strhash(new_path));
1882                        hashmap_put(collisions, collision_ent);
1883                        collision_ent->target_file = new_path;
1884                } else {
1885                        free(new_path);
1886                }
1887                string_list_insert(&collision_ent->source_files,
1888                                   pair->two->path);
1889        }
1890}
1891
1892static char *check_for_directory_rename(struct merge_options *o,
1893                                        const char *path,
1894                                        struct tree *tree,
1895                                        struct hashmap *dir_renames,
1896                                        struct hashmap *dir_rename_exclusions,
1897                                        struct hashmap *collisions,
1898                                        int *clean_merge)
1899{
1900        char *new_path = NULL;
1901        struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
1902        struct dir_rename_entry *oentry = NULL;
1903
1904        if (!entry)
1905                return new_path;
1906
1907        /*
1908         * This next part is a little weird.  We do not want to do an
1909         * implicit rename into a directory we renamed on our side, because
1910         * that will result in a spurious rename/rename(1to2) conflict.  An
1911         * example:
1912         *   Base commit: dumbdir/afile, otherdir/bfile
1913         *   Side 1:      smrtdir/afile, otherdir/bfile
1914         *   Side 2:      dumbdir/afile, dumbdir/bfile
1915         * Here, while working on Side 1, we could notice that otherdir was
1916         * renamed/merged to dumbdir, and change the diff_filepair for
1917         * otherdir/bfile into a rename into dumbdir/bfile.  However, Side
1918         * 2 will notice the rename from dumbdir to smrtdir, and do the
1919         * transitive rename to move it from dumbdir/bfile to
1920         * smrtdir/bfile.  That gives us bfile in dumbdir vs being in
1921         * smrtdir, a rename/rename(1to2) conflict.  We really just want
1922         * the file to end up in smrtdir.  And the way to achieve that is
1923         * to not let Side1 do the rename to dumbdir, since we know that is
1924         * the source of one of our directory renames.
1925         *
1926         * That's why oentry and dir_rename_exclusions is here.
1927         *
1928         * As it turns out, this also prevents N-way transient rename
1929         * confusion; See testcases 9c and 9d of t6043.
1930         */
1931        oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
1932        if (oentry) {
1933                output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
1934                               "to %s, because %s itself was renamed."),
1935                       entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
1936        } else {
1937                new_path = handle_path_level_conflicts(o, path, entry,
1938                                                       collisions, tree);
1939                *clean_merge &= (new_path != NULL);
1940        }
1941
1942        return new_path;
1943}
1944
1945/*
1946 * Get information of all renames which occurred in 'pairs', making use of
1947 * any implicit directory renames inferred from the other side of history.
1948 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
1949 * to be able to associate the correct cache entries with the rename
1950 * information; tree is always equal to either a_tree or b_tree.
1951 */
1952static struct string_list *get_renames(struct merge_options *o,
1953                                       struct diff_queue_struct *pairs,
1954                                       struct hashmap *dir_renames,
1955                                       struct hashmap *dir_rename_exclusions,
1956                                       struct tree *tree,
1957                                       struct tree *o_tree,
1958                                       struct tree *a_tree,
1959                                       struct tree *b_tree,
1960                                       struct string_list *entries,
1961                                       int *clean_merge)
1962{
1963        int i;
1964        struct hashmap collisions;
1965        struct hashmap_iter iter;
1966        struct collision_entry *e;
1967        struct string_list *renames;
1968
1969        compute_collisions(&collisions, dir_renames, pairs);
1970        renames = xcalloc(1, sizeof(struct string_list));
1971
1972        for (i = 0; i < pairs->nr; ++i) {
1973                struct string_list_item *item;
1974                struct rename *re;
1975                struct diff_filepair *pair = pairs->queue[i];
1976                char *new_path; /* non-NULL only with directory renames */
1977
1978                if (pair->status == 'D') {
1979                        diff_free_filepair(pair);
1980                        continue;
1981                }
1982                new_path = check_for_directory_rename(o, pair->two->path, tree,
1983                                                      dir_renames,
1984                                                      dir_rename_exclusions,
1985                                                      &collisions,
1986                                                      clean_merge);
1987                if (pair->status != 'R' && !new_path) {
1988                        diff_free_filepair(pair);
1989                        continue;
1990                }
1991
1992                re = xmalloc(sizeof(*re));
1993                re->processed = 0;
1994                re->pair = pair;
1995                item = string_list_lookup(entries, re->pair->one->path);
1996                if (!item)
1997                        re->src_entry = insert_stage_data(re->pair->one->path,
1998                                        o_tree, a_tree, b_tree, entries);
1999                else
2000                        re->src_entry = item->util;
2001
2002                item = string_list_lookup(entries, re->pair->two->path);
2003                if (!item)
2004                        re->dst_entry = insert_stage_data(re->pair->two->path,
2005                                        o_tree, a_tree, b_tree, entries);
2006                else
2007                        re->dst_entry = item->util;
2008                item = string_list_insert(renames, pair->one->path);
2009                item->util = re;
2010        }
2011
2012        hashmap_iter_init(&collisions, &iter);
2013        while ((e = hashmap_iter_next(&iter))) {
2014                free(e->target_file);
2015                string_list_clear(&e->source_files, 0);
2016        }
2017        hashmap_free(&collisions, 1);
2018        return renames;
2019}
2020
2021static int process_renames(struct merge_options *o,
2022                           struct string_list *a_renames,
2023                           struct string_list *b_renames)
2024{
2025        int clean_merge = 1, i, j;
2026        struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2027        struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2028        const struct rename *sre;
2029
2030        for (i = 0; i < a_renames->nr; i++) {
2031                sre = a_renames->items[i].util;
2032                string_list_insert(&a_by_dst, sre->pair->two->path)->util
2033                        = (void *)sre;
2034        }
2035        for (i = 0; i < b_renames->nr; i++) {
2036                sre = b_renames->items[i].util;
2037                string_list_insert(&b_by_dst, sre->pair->two->path)->util
2038                        = (void *)sre;
2039        }
2040
2041        for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2042                struct string_list *renames1, *renames2Dst;
2043                struct rename *ren1 = NULL, *ren2 = NULL;
2044                const char *branch1, *branch2;
2045                const char *ren1_src, *ren1_dst;
2046                struct string_list_item *lookup;
2047
2048                if (i >= a_renames->nr) {
2049                        ren2 = b_renames->items[j++].util;
2050                } else if (j >= b_renames->nr) {
2051                        ren1 = a_renames->items[i++].util;
2052                } else {
2053                        int compare = strcmp(a_renames->items[i].string,
2054                                             b_renames->items[j].string);
2055                        if (compare <= 0)
2056                                ren1 = a_renames->items[i++].util;
2057                        if (compare >= 0)
2058                                ren2 = b_renames->items[j++].util;
2059                }
2060
2061                /* TODO: refactor, so that 1/2 are not needed */
2062                if (ren1) {
2063                        renames1 = a_renames;
2064                        renames2Dst = &b_by_dst;
2065                        branch1 = o->branch1;
2066                        branch2 = o->branch2;
2067                } else {
2068                        renames1 = b_renames;
2069                        renames2Dst = &a_by_dst;
2070                        branch1 = o->branch2;
2071                        branch2 = o->branch1;
2072                        SWAP(ren2, ren1);
2073                }
2074
2075                if (ren1->processed)
2076                        continue;
2077                ren1->processed = 1;
2078                ren1->dst_entry->processed = 1;
2079                /* BUG: We should only mark src_entry as processed if we
2080                 * are not dealing with a rename + add-source case.
2081                 */
2082                ren1->src_entry->processed = 1;
2083
2084                ren1_src = ren1->pair->one->path;
2085                ren1_dst = ren1->pair->two->path;
2086
2087                if (ren2) {
2088                        /* One file renamed on both sides */
2089                        const char *ren2_src = ren2->pair->one->path;
2090                        const char *ren2_dst = ren2->pair->two->path;
2091                        enum rename_type rename_type;
2092                        if (strcmp(ren1_src, ren2_src) != 0)
2093                                die("BUG: ren1_src != ren2_src");
2094                        ren2->dst_entry->processed = 1;
2095                        ren2->processed = 1;
2096                        if (strcmp(ren1_dst, ren2_dst) != 0) {
2097                                rename_type = RENAME_ONE_FILE_TO_TWO;
2098                                clean_merge = 0;
2099                        } else {
2100                                rename_type = RENAME_ONE_FILE_TO_ONE;
2101                                /* BUG: We should only remove ren1_src in
2102                                 * the base stage (think of rename +
2103                                 * add-source cases).
2104                                 */
2105                                remove_file(o, 1, ren1_src, 1);
2106                                update_entry(ren1->dst_entry,
2107                                             ren1->pair->one,
2108                                             ren1->pair->two,
2109                                             ren2->pair->two);
2110                        }
2111                        setup_rename_conflict_info(rename_type,
2112                                                   ren1->pair,
2113                                                   ren2->pair,
2114                                                   branch1,
2115                                                   branch2,
2116                                                   ren1->dst_entry,
2117                                                   ren2->dst_entry,
2118                                                   o,
2119                                                   NULL,
2120                                                   NULL);
2121                } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2122                        /* Two different files renamed to the same thing */
2123                        char *ren2_dst;
2124                        ren2 = lookup->util;
2125                        ren2_dst = ren2->pair->two->path;
2126                        if (strcmp(ren1_dst, ren2_dst) != 0)
2127                                die("BUG: ren1_dst != ren2_dst");
2128
2129                        clean_merge = 0;
2130                        ren2->processed = 1;
2131                        /*
2132                         * BUG: We should only mark src_entry as processed
2133                         * if we are not dealing with a rename + add-source
2134                         * case.
2135                         */
2136                        ren2->src_entry->processed = 1;
2137
2138                        setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2139                                                   ren1->pair,
2140                                                   ren2->pair,
2141                                                   branch1,
2142                                                   branch2,
2143                                                   ren1->dst_entry,
2144                                                   ren2->dst_entry,
2145                                                   o,
2146                                                   ren1->src_entry,
2147                                                   ren2->src_entry);
2148
2149                } else {
2150                        /* Renamed in 1, maybe changed in 2 */
2151                        /* we only use sha1 and mode of these */
2152                        struct diff_filespec src_other, dst_other;
2153                        int try_merge;
2154
2155                        /*
2156                         * unpack_trees loads entries from common-commit
2157                         * into stage 1, from head-commit into stage 2, and
2158                         * from merge-commit into stage 3.  We keep track
2159                         * of which side corresponds to the rename.
2160                         */
2161                        int renamed_stage = a_renames == renames1 ? 2 : 3;
2162                        int other_stage =   a_renames == renames1 ? 3 : 2;
2163
2164                        /* BUG: We should only remove ren1_src in the base
2165                         * stage and in other_stage (think of rename +
2166                         * add-source case).
2167                         */
2168                        remove_file(o, 1, ren1_src,
2169                                    renamed_stage == 2 || !was_tracked(ren1_src));
2170
2171                        oidcpy(&src_other.oid,
2172                               &ren1->src_entry->stages[other_stage].oid);
2173                        src_other.mode = ren1->src_entry->stages[other_stage].mode;
2174                        oidcpy(&dst_other.oid,
2175                               &ren1->dst_entry->stages[other_stage].oid);
2176                        dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2177                        try_merge = 0;
2178
2179                        if (oid_eq(&src_other.oid, &null_oid)) {
2180                                setup_rename_conflict_info(RENAME_DELETE,
2181                                                           ren1->pair,
2182                                                           NULL,
2183                                                           branch1,
2184                                                           branch2,
2185                                                           ren1->dst_entry,
2186                                                           NULL,
2187                                                           o,
2188                                                           NULL,
2189                                                           NULL);
2190                        } else if ((dst_other.mode == ren1->pair->two->mode) &&
2191                                   oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2192                                /*
2193                                 * Added file on the other side identical to
2194                                 * the file being renamed: clean merge.
2195                                 * Also, there is no need to overwrite the
2196                                 * file already in the working copy, so call
2197                                 * update_file_flags() instead of
2198                                 * update_file().
2199                                 */
2200                                if (update_file_flags(o,
2201                                                      &ren1->pair->two->oid,
2202                                                      ren1->pair->two->mode,
2203                                                      ren1_dst,
2204                                                      1, /* update_cache */
2205                                                      0  /* update_wd    */))
2206                                        clean_merge = -1;
2207                        } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2208                                clean_merge = 0;
2209                                try_merge = 1;
2210                                output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2211                                       "%s added in %s"),
2212                                       ren1_src, ren1_dst, branch1,
2213                                       ren1_dst, branch2);
2214                                if (o->call_depth) {
2215                                        struct merge_file_info mfi;
2216                                        if (merge_file_one(o, ren1_dst, &null_oid, 0,
2217                                                           &ren1->pair->two->oid,
2218                                                           ren1->pair->two->mode,
2219                                                           &dst_other.oid,
2220                                                           dst_other.mode,
2221                                                           branch1, branch2, &mfi)) {
2222                                                clean_merge = -1;
2223                                                goto cleanup_and_return;
2224                                        }
2225                                        output(o, 1, _("Adding merged %s"), ren1_dst);
2226                                        if (update_file(o, 0, &mfi.oid,
2227                                                        mfi.mode, ren1_dst))
2228                                                clean_merge = -1;
2229                                        try_merge = 0;
2230                                } else {
2231                                        char *new_path = unique_path(o, ren1_dst, branch2);
2232                                        output(o, 1, _("Adding as %s instead"), new_path);
2233                                        if (update_file(o, 0, &dst_other.oid,
2234                                                        dst_other.mode, new_path))
2235                                                clean_merge = -1;
2236                                        free(new_path);
2237                                }
2238                        } else
2239                                try_merge = 1;
2240
2241                        if (clean_merge < 0)
2242                                goto cleanup_and_return;
2243                        if (try_merge) {
2244                                struct diff_filespec *one, *a, *b;
2245                                src_other.path = (char *)ren1_src;
2246
2247                                one = ren1->pair->one;
2248                                if (a_renames == renames1) {
2249                                        a = ren1->pair->two;
2250                                        b = &src_other;
2251                                } else {
2252                                        b = ren1->pair->two;
2253                                        a = &src_other;
2254                                }
2255                                update_entry(ren1->dst_entry, one, a, b);
2256                                setup_rename_conflict_info(RENAME_NORMAL,
2257                                                           ren1->pair,
2258                                                           NULL,
2259                                                           branch1,
2260                                                           NULL,
2261                                                           ren1->dst_entry,
2262                                                           NULL,
2263                                                           o,
2264                                                           NULL,
2265                                                           NULL);
2266                        }
2267                }
2268        }
2269cleanup_and_return:
2270        string_list_clear(&a_by_dst, 0);
2271        string_list_clear(&b_by_dst, 0);
2272
2273        return clean_merge;
2274}
2275
2276struct rename_info {
2277        struct string_list *head_renames;
2278        struct string_list *merge_renames;
2279};
2280
2281static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2282                                   struct hashmap *dir_renames)
2283{
2284        struct hashmap_iter iter;
2285        struct dir_rename_entry *e;
2286
2287        hashmap_iter_init(dir_renames, &iter);
2288        while ((e = hashmap_iter_next(&iter))) {
2289                free(e->dir);
2290                strbuf_release(&e->new_dir);
2291                /* possible_new_dirs already cleared in get_directory_renames */
2292        }
2293        hashmap_free(dir_renames, 1);
2294        free(dir_renames);
2295
2296        free(pairs->queue);
2297        free(pairs);
2298}
2299
2300static int handle_renames(struct merge_options *o,
2301                          struct tree *common,
2302                          struct tree *head,
2303                          struct tree *merge,
2304                          struct string_list *entries,
2305                          struct rename_info *ri)
2306{
2307        struct diff_queue_struct *head_pairs, *merge_pairs;
2308        struct hashmap *dir_re_head, *dir_re_merge;
2309        int clean = 1;
2310
2311        ri->head_renames = NULL;
2312        ri->merge_renames = NULL;
2313
2314        if (!o->detect_rename)
2315                return 1;
2316
2317        head_pairs = get_diffpairs(o, common, head);
2318        merge_pairs = get_diffpairs(o, common, merge);
2319
2320        dir_re_head = get_directory_renames(head_pairs, head);
2321        dir_re_merge = get_directory_renames(merge_pairs, merge);
2322
2323        handle_directory_level_conflicts(o,
2324                                         dir_re_head, head,
2325                                         dir_re_merge, merge);
2326
2327        ri->head_renames  = get_renames(o, head_pairs,
2328                                        dir_re_merge, dir_re_head, head,
2329                                        common, head, merge, entries,
2330                                        &clean);
2331        if (clean < 0)
2332                goto cleanup;
2333        ri->merge_renames = get_renames(o, merge_pairs,
2334                                        dir_re_head, dir_re_merge, merge,
2335                                        common, head, merge, entries,
2336                                        &clean);
2337        if (clean < 0)
2338                goto cleanup;
2339        clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2340
2341cleanup:
2342        /*
2343         * Some cleanup is deferred until cleanup_renames() because the
2344         * data structures are still needed and referenced in
2345         * process_entry().  But there are a few things we can free now.
2346         */
2347        initial_cleanup_rename(head_pairs, dir_re_head);
2348        initial_cleanup_rename(merge_pairs, dir_re_merge);
2349
2350        return clean;
2351}
2352
2353static void final_cleanup_rename(struct string_list *rename)
2354{
2355        const struct rename *re;
2356        int i;
2357
2358        if (rename == NULL)
2359                return;
2360
2361        for (i = 0; i < rename->nr; i++) {
2362                re = rename->items[i].util;
2363                diff_free_filepair(re->pair);
2364        }
2365        string_list_clear(rename, 1);
2366        free(rename);
2367}
2368
2369static void final_cleanup_renames(struct rename_info *re_info)
2370{
2371        final_cleanup_rename(re_info->head_renames);
2372        final_cleanup_rename(re_info->merge_renames);
2373}
2374
2375static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2376{
2377        return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2378}
2379
2380static int read_oid_strbuf(struct merge_options *o,
2381        const struct object_id *oid, struct strbuf *dst)
2382{
2383        void *buf;
2384        enum object_type type;
2385        unsigned long size;
2386        buf = read_sha1_file(oid->hash, &type, &size);
2387        if (!buf)
2388                return err(o, _("cannot read object %s"), oid_to_hex(oid));
2389        if (type != OBJ_BLOB) {
2390                free(buf);
2391                return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2392        }
2393        strbuf_attach(dst, buf, size, size + 1);
2394        return 0;
2395}
2396
2397static int blob_unchanged(struct merge_options *opt,
2398                          const struct object_id *o_oid,
2399                          unsigned o_mode,
2400                          const struct object_id *a_oid,
2401                          unsigned a_mode,
2402                          int renormalize, const char *path)
2403{
2404        struct strbuf o = STRBUF_INIT;
2405        struct strbuf a = STRBUF_INIT;
2406        int ret = 0; /* assume changed for safety */
2407
2408        if (a_mode != o_mode)
2409                return 0;
2410        if (oid_eq(o_oid, a_oid))
2411                return 1;
2412        if (!renormalize)
2413                return 0;
2414
2415        assert(o_oid && a_oid);
2416        if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2417                goto error_return;
2418        /*
2419         * Note: binary | is used so that both renormalizations are
2420         * performed.  Comparison can be skipped if both files are
2421         * unchanged since their sha1s have already been compared.
2422         */
2423        if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2424            renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2425                ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2426
2427error_return:
2428        strbuf_release(&o);
2429        strbuf_release(&a);
2430        return ret;
2431}
2432
2433static int handle_modify_delete(struct merge_options *o,
2434                                 const char *path,
2435                                 struct object_id *o_oid, int o_mode,
2436                                 struct object_id *a_oid, int a_mode,
2437                                 struct object_id *b_oid, int b_mode)
2438{
2439        const char *modify_branch, *delete_branch;
2440        struct object_id *changed_oid;
2441        int changed_mode;
2442
2443        if (a_oid) {
2444                modify_branch = o->branch1;
2445                delete_branch = o->branch2;
2446                changed_oid = a_oid;
2447                changed_mode = a_mode;
2448        } else {
2449                modify_branch = o->branch2;
2450                delete_branch = o->branch1;
2451                changed_oid = b_oid;
2452                changed_mode = b_mode;
2453        }
2454
2455        return handle_change_delete(o,
2456                                    path, NULL,
2457                                    o_oid, o_mode,
2458                                    changed_oid, changed_mode,
2459                                    modify_branch, delete_branch,
2460                                    _("modify"), _("modified"));
2461}
2462
2463static int merge_content(struct merge_options *o,
2464                         const char *path,
2465                         struct object_id *o_oid, int o_mode,
2466                         struct object_id *a_oid, int a_mode,
2467                         struct object_id *b_oid, int b_mode,
2468                         struct rename_conflict_info *rename_conflict_info)
2469{
2470        const char *reason = _("content");
2471        const char *path1 = NULL, *path2 = NULL;
2472        struct merge_file_info mfi;
2473        struct diff_filespec one, a, b;
2474        unsigned df_conflict_remains = 0;
2475
2476        if (!o_oid) {
2477                reason = _("add/add");
2478                o_oid = (struct object_id *)&null_oid;
2479        }
2480        one.path = a.path = b.path = (char *)path;
2481        oidcpy(&one.oid, o_oid);
2482        one.mode = o_mode;
2483        oidcpy(&a.oid, a_oid);
2484        a.mode = a_mode;
2485        oidcpy(&b.oid, b_oid);
2486        b.mode = b_mode;
2487
2488        if (rename_conflict_info) {
2489                struct diff_filepair *pair1 = rename_conflict_info->pair1;
2490
2491                path1 = (o->branch1 == rename_conflict_info->branch1) ?
2492                        pair1->two->path : pair1->one->path;
2493                /* If rename_conflict_info->pair2 != NULL, we are in
2494                 * RENAME_ONE_FILE_TO_ONE case.  Otherwise, we have a
2495                 * normal rename.
2496                 */
2497                path2 = (rename_conflict_info->pair2 ||
2498                         o->branch2 == rename_conflict_info->branch1) ?
2499                        pair1->two->path : pair1->one->path;
2500
2501                if (dir_in_way(path, !o->call_depth,
2502                               S_ISGITLINK(pair1->two->mode)))
2503                        df_conflict_remains = 1;
2504        }
2505        if (merge_file_special_markers(o, &one, &a, &b,
2506                                       o->branch1, path1,
2507                                       o->branch2, path2, &mfi))
2508                return -1;
2509
2510        if (mfi.clean && !df_conflict_remains &&
2511            oid_eq(&mfi.oid, a_oid) && mfi.mode == a_mode) {
2512                int path_renamed_outside_HEAD;
2513                output(o, 3, _("Skipped %s (merged same as existing)"), path);
2514                /*
2515                 * The content merge resulted in the same file contents we
2516                 * already had.  We can return early if those file contents
2517                 * are recorded at the correct path (which may not be true
2518                 * if the merge involves a rename).
2519                 */
2520                path_renamed_outside_HEAD = !path2 || !strcmp(path, path2);
2521                if (!path_renamed_outside_HEAD) {
2522                        add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2523                                      0, (!o->call_depth), 0);
2524                        return mfi.clean;
2525                }
2526        } else
2527                output(o, 2, _("Auto-merging %s"), path);
2528
2529        if (!mfi.clean) {
2530                if (S_ISGITLINK(mfi.mode))
2531                        reason = _("submodule");
2532                output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2533                                reason, path);
2534                if (rename_conflict_info && !df_conflict_remains)
2535                        if (update_stages(o, path, &one, &a, &b))
2536                                return -1;
2537        }
2538
2539        if (df_conflict_remains) {
2540                char *new_path;
2541                if (o->call_depth) {
2542                        remove_file_from_cache(path);
2543                } else {
2544                        if (!mfi.clean) {
2545                                if (update_stages(o, path, &one, &a, &b))
2546                                        return -1;
2547                        } else {
2548                                int file_from_stage2 = was_tracked(path);
2549                                struct diff_filespec merged;
2550                                oidcpy(&merged.oid, &mfi.oid);
2551                                merged.mode = mfi.mode;
2552
2553                                if (update_stages(o, path, NULL,
2554                                                  file_from_stage2 ? &merged : NULL,
2555                                                  file_from_stage2 ? NULL : &merged))
2556                                        return -1;
2557                        }
2558
2559                }
2560                new_path = unique_path(o, path, rename_conflict_info->branch1);
2561                output(o, 1, _("Adding as %s instead"), new_path);
2562                if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2563                        free(new_path);
2564                        return -1;
2565                }
2566                free(new_path);
2567                mfi.clean = 0;
2568        } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2569                return -1;
2570        return mfi.clean;
2571}
2572
2573/* Per entry merge function */
2574static int process_entry(struct merge_options *o,
2575                         const char *path, struct stage_data *entry)
2576{
2577        int clean_merge = 1;
2578        int normalize = o->renormalize;
2579        unsigned o_mode = entry->stages[1].mode;
2580        unsigned a_mode = entry->stages[2].mode;
2581        unsigned b_mode = entry->stages[3].mode;
2582        struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2583        struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2584        struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2585
2586        entry->processed = 1;
2587        if (entry->rename_conflict_info) {
2588                struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2589                switch (conflict_info->rename_type) {
2590                case RENAME_NORMAL:
2591                case RENAME_ONE_FILE_TO_ONE:
2592                        clean_merge = merge_content(o, path,
2593                                                    o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2594                                                    conflict_info);
2595                        break;
2596                case RENAME_DELETE:
2597                        clean_merge = 0;
2598                        if (conflict_rename_delete(o,
2599                                                   conflict_info->pair1,
2600                                                   conflict_info->branch1,
2601                                                   conflict_info->branch2))
2602                                clean_merge = -1;
2603                        break;
2604                case RENAME_ONE_FILE_TO_TWO:
2605                        clean_merge = 0;
2606                        if (conflict_rename_rename_1to2(o, conflict_info))
2607                                clean_merge = -1;
2608                        break;
2609                case RENAME_TWO_FILES_TO_ONE:
2610                        clean_merge = 0;
2611                        if (conflict_rename_rename_2to1(o, conflict_info))
2612                                clean_merge = -1;
2613                        break;
2614                default:
2615                        entry->processed = 0;
2616                        break;
2617                }
2618        } else if (o_oid && (!a_oid || !b_oid)) {
2619                /* Case A: Deleted in one */
2620                if ((!a_oid && !b_oid) ||
2621                    (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2622                    (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2623                        /* Deleted in both or deleted in one and
2624                         * unchanged in the other */
2625                        if (a_oid)
2626                                output(o, 2, _("Removing %s"), path);
2627                        /* do not touch working file if it did not exist */
2628                        remove_file(o, 1, path, !a_oid);
2629                } else {
2630                        /* Modify/delete; deleted side may have put a directory in the way */
2631                        clean_merge = 0;
2632                        if (handle_modify_delete(o, path, o_oid, o_mode,
2633                                                 a_oid, a_mode, b_oid, b_mode))
2634                                clean_merge = -1;
2635                }
2636        } else if ((!o_oid && a_oid && !b_oid) ||
2637                   (!o_oid && !a_oid && b_oid)) {
2638                /* Case B: Added in one. */
2639                /* [nothing|directory] -> ([nothing|directory], file) */
2640
2641                const char *add_branch;
2642                const char *other_branch;
2643                unsigned mode;
2644                const struct object_id *oid;
2645                const char *conf;
2646
2647                if (a_oid) {
2648                        add_branch = o->branch1;
2649                        other_branch = o->branch2;
2650                        mode = a_mode;
2651                        oid = a_oid;
2652                        conf = _("file/directory");
2653                } else {
2654                        add_branch = o->branch2;
2655                        other_branch = o->branch1;
2656                        mode = b_mode;
2657                        oid = b_oid;
2658                        conf = _("directory/file");
2659                }
2660                if (dir_in_way(path,
2661                               !o->call_depth && !S_ISGITLINK(a_mode),
2662                               0)) {
2663                        char *new_path = unique_path(o, path, add_branch);
2664                        clean_merge = 0;
2665                        output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
2666                               "Adding %s as %s"),
2667                               conf, path, other_branch, path, new_path);
2668                        if (update_file(o, 0, oid, mode, new_path))
2669                                clean_merge = -1;
2670                        else if (o->call_depth)
2671                                remove_file_from_cache(path);
2672                        free(new_path);
2673                } else {
2674                        output(o, 2, _("Adding %s"), path);
2675                        /* do not overwrite file if already present */
2676                        if (update_file_flags(o, oid, mode, path, 1, !a_oid))
2677                                clean_merge = -1;
2678                }
2679        } else if (a_oid && b_oid) {
2680                /* Case C: Added in both (check for same permissions) and */
2681                /* case D: Modified in both, but differently. */
2682                clean_merge = merge_content(o, path,
2683                                            o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2684                                            NULL);
2685        } else if (!o_oid && !a_oid && !b_oid) {
2686                /*
2687                 * this entry was deleted altogether. a_mode == 0 means
2688                 * we had that path and want to actively remove it.
2689                 */
2690                remove_file(o, 1, path, !a_mode);
2691        } else
2692                die("BUG: fatal merge failure, shouldn't happen.");
2693
2694        return clean_merge;
2695}
2696
2697int merge_trees(struct merge_options *o,
2698                struct tree *head,
2699                struct tree *merge,
2700                struct tree *common,
2701                struct tree **result)
2702{
2703        int code, clean;
2704
2705        if (o->subtree_shift) {
2706                merge = shift_tree_object(head, merge, o->subtree_shift);
2707                common = shift_tree_object(head, common, o->subtree_shift);
2708        }
2709
2710        if (oid_eq(&common->object.oid, &merge->object.oid)) {
2711                struct strbuf sb = STRBUF_INIT;
2712
2713                if (!o->call_depth && index_has_changes(&sb)) {
2714                        err(o, _("Dirty index: cannot merge (dirty: %s)"),
2715                            sb.buf);
2716                        return 0;
2717                }
2718                output(o, 0, _("Already up to date!"));
2719                *result = head;
2720                return 1;
2721        }
2722
2723        code = git_merge_trees(o->call_depth, common, head, merge);
2724
2725        if (code != 0) {
2726                if (show(o, 4) || o->call_depth)
2727                        err(o, _("merging of trees %s and %s failed"),
2728                            oid_to_hex(&head->object.oid),
2729                            oid_to_hex(&merge->object.oid));
2730                return -1;
2731        }
2732
2733        if (unmerged_cache()) {
2734                struct string_list *entries;
2735                struct rename_info re_info;
2736                int i;
2737                /*
2738                 * Only need the hashmap while processing entries, so
2739                 * initialize it here and free it when we are done running
2740                 * through the entries. Keeping it in the merge_options as
2741                 * opposed to decaring a local hashmap is for convenience
2742                 * so that we don't have to pass it to around.
2743                 */
2744                hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
2745                get_files_dirs(o, head);
2746                get_files_dirs(o, merge);
2747
2748                entries = get_unmerged();
2749                clean = handle_renames(o, common, head, merge, entries,
2750                                       &re_info);
2751                record_df_conflict_files(o, entries);
2752                if (clean < 0)
2753                        goto cleanup;
2754                for (i = entries->nr-1; 0 <= i; i--) {
2755                        const char *path = entries->items[i].string;
2756                        struct stage_data *e = entries->items[i].util;
2757                        if (!e->processed) {
2758                                int ret = process_entry(o, path, e);
2759                                if (!ret)
2760                                        clean = 0;
2761                                else if (ret < 0) {
2762                                        clean = ret;
2763                                        goto cleanup;
2764                                }
2765                        }
2766                }
2767                for (i = 0; i < entries->nr; i++) {
2768                        struct stage_data *e = entries->items[i].util;
2769                        if (!e->processed)
2770                                die("BUG: unprocessed path??? %s",
2771                                    entries->items[i].string);
2772                }
2773
2774cleanup:
2775                final_cleanup_renames(&re_info);
2776
2777                string_list_clear(entries, 1);
2778                free(entries);
2779
2780                hashmap_free(&o->current_file_dir_set, 1);
2781
2782                if (clean < 0)
2783                        return clean;
2784        }
2785        else
2786                clean = 1;
2787
2788        if (o->call_depth && !(*result = write_tree_from_memory(o)))
2789                return -1;
2790
2791        return clean;
2792}
2793
2794static struct commit_list *reverse_commit_list(struct commit_list *list)
2795{
2796        struct commit_list *next = NULL, *current, *backup;
2797        for (current = list; current; current = backup) {
2798                backup = current->next;
2799                current->next = next;
2800                next = current;
2801        }
2802        return next;
2803}
2804
2805/*
2806 * Merge the commits h1 and h2, return the resulting virtual
2807 * commit object and a flag indicating the cleanness of the merge.
2808 */
2809int merge_recursive(struct merge_options *o,
2810                    struct commit *h1,
2811                    struct commit *h2,
2812                    struct commit_list *ca,
2813                    struct commit **result)
2814{
2815        struct commit_list *iter;
2816        struct commit *merged_common_ancestors;
2817        struct tree *mrtree = mrtree;
2818        int clean;
2819
2820        if (show(o, 4)) {
2821                output(o, 4, _("Merging:"));
2822                output_commit_title(o, h1);
2823                output_commit_title(o, h2);
2824        }
2825
2826        if (!ca) {
2827                ca = get_merge_bases(h1, h2);
2828                ca = reverse_commit_list(ca);
2829        }
2830
2831        if (show(o, 5)) {
2832                unsigned cnt = commit_list_count(ca);
2833
2834                output(o, 5, Q_("found %u common ancestor:",
2835                                "found %u common ancestors:", cnt), cnt);
2836                for (iter = ca; iter; iter = iter->next)
2837                        output_commit_title(o, iter->item);
2838        }
2839
2840        merged_common_ancestors = pop_commit(&ca);
2841        if (merged_common_ancestors == NULL) {
2842                /* if there is no common ancestor, use an empty tree */
2843                struct tree *tree;
2844
2845                tree = lookup_tree(the_hash_algo->empty_tree);
2846                merged_common_ancestors = make_virtual_commit(tree, "ancestor");
2847        }
2848
2849        for (iter = ca; iter; iter = iter->next) {
2850                const char *saved_b1, *saved_b2;
2851                o->call_depth++;
2852                /*
2853                 * When the merge fails, the result contains files
2854                 * with conflict markers. The cleanness flag is
2855                 * ignored (unless indicating an error), it was never
2856                 * actually used, as result of merge_trees has always
2857                 * overwritten it: the committed "conflicts" were
2858                 * already resolved.
2859                 */
2860                discard_cache();
2861                saved_b1 = o->branch1;
2862                saved_b2 = o->branch2;
2863                o->branch1 = "Temporary merge branch 1";
2864                o->branch2 = "Temporary merge branch 2";
2865                if (merge_recursive(o, merged_common_ancestors, iter->item,
2866                                    NULL, &merged_common_ancestors) < 0)
2867                        return -1;
2868                o->branch1 = saved_b1;
2869                o->branch2 = saved_b2;
2870                o->call_depth--;
2871
2872                if (!merged_common_ancestors)
2873                        return err(o, _("merge returned no commit"));
2874        }
2875
2876        discard_cache();
2877        if (!o->call_depth)
2878                read_cache();
2879
2880        o->ancestor = "merged common ancestors";
2881        clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,
2882                            &mrtree);
2883        if (clean < 0) {
2884                flush_output(o);
2885                return clean;
2886        }
2887
2888        if (o->call_depth) {
2889                *result = make_virtual_commit(mrtree, "merged tree");
2890                commit_list_insert(h1, &(*result)->parents);
2891                commit_list_insert(h2, &(*result)->parents->next);
2892        }
2893        flush_output(o);
2894        if (!o->call_depth && o->buffer_output < 2)
2895                strbuf_release(&o->obuf);
2896        if (show(o, 2))
2897                diff_warn_rename_limit("merge.renamelimit",
2898                                       o->needed_rename_limit, 0);
2899        return clean;
2900}
2901
2902static struct commit *get_ref(const struct object_id *oid, const char *name)
2903{
2904        struct object *object;
2905
2906        object = deref_tag(parse_object(oid), name, strlen(name));
2907        if (!object)
2908                return NULL;
2909        if (object->type == OBJ_TREE)
2910                return make_virtual_commit((struct tree*)object, name);
2911        if (object->type != OBJ_COMMIT)
2912                return NULL;
2913        if (parse_commit((struct commit *)object))
2914                return NULL;
2915        return (struct commit *)object;
2916}
2917
2918int merge_recursive_generic(struct merge_options *o,
2919                            const struct object_id *head,
2920                            const struct object_id *merge,
2921                            int num_base_list,
2922                            const struct object_id **base_list,
2923                            struct commit **result)
2924{
2925        int clean;
2926        struct lock_file lock = LOCK_INIT;
2927        struct commit *head_commit = get_ref(head, o->branch1);
2928        struct commit *next_commit = get_ref(merge, o->branch2);
2929        struct commit_list *ca = NULL;
2930
2931        if (base_list) {
2932                int i;
2933                for (i = 0; i < num_base_list; ++i) {
2934                        struct commit *base;
2935                        if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
2936                                return err(o, _("Could not parse object '%s'"),
2937                                        oid_to_hex(base_list[i]));
2938                        commit_list_insert(base, &ca);
2939                }
2940        }
2941
2942        hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
2943        clean = merge_recursive(o, head_commit, next_commit, ca,
2944                        result);
2945        if (clean < 0)
2946                return clean;
2947
2948        if (active_cache_changed &&
2949            write_locked_index(&the_index, &lock, COMMIT_LOCK))
2950                return err(o, _("Unable to write index."));
2951
2952        return clean ? 0 : 1;
2953}
2954
2955static void merge_recursive_config(struct merge_options *o)
2956{
2957        git_config_get_int("merge.verbosity", &o->verbosity);
2958        git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
2959        git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
2960        git_config(git_xmerge_config, NULL);
2961}
2962
2963void init_merge_options(struct merge_options *o)
2964{
2965        const char *merge_verbosity;
2966        memset(o, 0, sizeof(struct merge_options));
2967        o->verbosity = 2;
2968        o->buffer_output = 1;
2969        o->diff_rename_limit = -1;
2970        o->merge_rename_limit = -1;
2971        o->renormalize = 0;
2972        o->detect_rename = 1;
2973        merge_recursive_config(o);
2974        merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
2975        if (merge_verbosity)
2976                o->verbosity = strtol(merge_verbosity, NULL, 10);
2977        if (o->verbosity >= 5)
2978                o->buffer_output = 0;
2979        strbuf_init(&o->obuf, 0);
2980        string_list_init(&o->df_conflict_file_set, 1);
2981}
2982
2983int parse_merge_opt(struct merge_options *o, const char *s)
2984{
2985        const char *arg;
2986
2987        if (!s || !*s)
2988                return -1;
2989        if (!strcmp(s, "ours"))
2990                o->recursive_variant = MERGE_RECURSIVE_OURS;
2991        else if (!strcmp(s, "theirs"))
2992                o->recursive_variant = MERGE_RECURSIVE_THEIRS;
2993        else if (!strcmp(s, "subtree"))
2994                o->subtree_shift = "";
2995        else if (skip_prefix(s, "subtree=", &arg))
2996                o->subtree_shift = arg;
2997        else if (!strcmp(s, "patience"))
2998                o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
2999        else if (!strcmp(s, "histogram"))
3000                o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3001        else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3002                long value = parse_algorithm_value(arg);
3003                if (value < 0)
3004                        return -1;
3005                /* clear out previous settings */
3006                DIFF_XDL_CLR(o, NEED_MINIMAL);
3007                o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3008                o->xdl_opts |= value;
3009        }
3010        else if (!strcmp(s, "ignore-space-change"))
3011                DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3012        else if (!strcmp(s, "ignore-all-space"))
3013                DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3014        else if (!strcmp(s, "ignore-space-at-eol"))
3015                DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3016        else if (!strcmp(s, "ignore-cr-at-eol"))
3017                DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3018        else if (!strcmp(s, "renormalize"))
3019                o->renormalize = 1;
3020        else if (!strcmp(s, "no-renormalize"))
3021                o->renormalize = 0;
3022        else if (!strcmp(s, "no-renames"))
3023                o->detect_rename = 0;
3024        else if (!strcmp(s, "find-renames")) {
3025                o->detect_rename = 1;
3026                o->rename_score = 0;
3027        }
3028        else if (skip_prefix(s, "find-renames=", &arg) ||
3029                 skip_prefix(s, "rename-threshold=", &arg)) {
3030                if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3031                        return -1;
3032                o->detect_rename = 1;
3033        }
3034        else
3035                return -1;
3036        return 0;
3037}