merge-recursive.con commit merge-recursive: apply necessary modifications for directory renames (9c0743f)
   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_DIR,
 184        RENAME_DELETE,
 185        RENAME_ONE_FILE_TO_ONE,
 186        RENAME_ONE_FILE_TO_TWO,
 187        RENAME_TWO_FILES_TO_ONE
 188};
 189
 190struct rename_conflict_info {
 191        enum rename_type rename_type;
 192        struct diff_filepair *pair1;
 193        struct diff_filepair *pair2;
 194        const char *branch1;
 195        const char *branch2;
 196        struct stage_data *dst_entry1;
 197        struct stage_data *dst_entry2;
 198        struct diff_filespec ren1_other;
 199        struct diff_filespec ren2_other;
 200};
 201
 202/*
 203 * Since we want to write the index eventually, we cannot reuse the index
 204 * for these (temporary) data.
 205 */
 206struct stage_data {
 207        struct {
 208                unsigned mode;
 209                struct object_id oid;
 210        } stages[4];
 211        struct rename_conflict_info *rename_conflict_info;
 212        unsigned processed:1;
 213};
 214
 215static inline void setup_rename_conflict_info(enum rename_type rename_type,
 216                                              struct diff_filepair *pair1,
 217                                              struct diff_filepair *pair2,
 218                                              const char *branch1,
 219                                              const char *branch2,
 220                                              struct stage_data *dst_entry1,
 221                                              struct stage_data *dst_entry2,
 222                                              struct merge_options *o,
 223                                              struct stage_data *src_entry1,
 224                                              struct stage_data *src_entry2)
 225{
 226        struct rename_conflict_info *ci = xcalloc(1, sizeof(struct rename_conflict_info));
 227        ci->rename_type = rename_type;
 228        ci->pair1 = pair1;
 229        ci->branch1 = branch1;
 230        ci->branch2 = branch2;
 231
 232        ci->dst_entry1 = dst_entry1;
 233        dst_entry1->rename_conflict_info = ci;
 234        dst_entry1->processed = 0;
 235
 236        assert(!pair2 == !dst_entry2);
 237        if (dst_entry2) {
 238                ci->dst_entry2 = dst_entry2;
 239                ci->pair2 = pair2;
 240                dst_entry2->rename_conflict_info = ci;
 241        }
 242
 243        if (rename_type == RENAME_TWO_FILES_TO_ONE) {
 244                /*
 245                 * For each rename, there could have been
 246                 * modifications on the side of history where that
 247                 * file was not renamed.
 248                 */
 249                int ostage1 = o->branch1 == branch1 ? 3 : 2;
 250                int ostage2 = ostage1 ^ 1;
 251
 252                ci->ren1_other.path = pair1->one->path;
 253                oidcpy(&ci->ren1_other.oid, &src_entry1->stages[ostage1].oid);
 254                ci->ren1_other.mode = src_entry1->stages[ostage1].mode;
 255
 256                ci->ren2_other.path = pair2->one->path;
 257                oidcpy(&ci->ren2_other.oid, &src_entry2->stages[ostage2].oid);
 258                ci->ren2_other.mode = src_entry2->stages[ostage2].mode;
 259        }
 260}
 261
 262static int show(struct merge_options *o, int v)
 263{
 264        return (!o->call_depth && o->verbosity >= v) || o->verbosity >= 5;
 265}
 266
 267__attribute__((format (printf, 3, 4)))
 268static void output(struct merge_options *o, int v, const char *fmt, ...)
 269{
 270        va_list ap;
 271
 272        if (!show(o, v))
 273                return;
 274
 275        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 276
 277        va_start(ap, fmt);
 278        strbuf_vaddf(&o->obuf, fmt, ap);
 279        va_end(ap);
 280
 281        strbuf_addch(&o->obuf, '\n');
 282        if (!o->buffer_output)
 283                flush_output(o);
 284}
 285
 286static void output_commit_title(struct merge_options *o, struct commit *commit)
 287{
 288        strbuf_addchars(&o->obuf, ' ', o->call_depth * 2);
 289        if (commit->util)
 290                strbuf_addf(&o->obuf, "virtual %s\n",
 291                        merge_remote_util(commit)->name);
 292        else {
 293                strbuf_add_unique_abbrev(&o->obuf, &commit->object.oid,
 294                                         DEFAULT_ABBREV);
 295                strbuf_addch(&o->obuf, ' ');
 296                if (parse_commit(commit) != 0)
 297                        strbuf_addstr(&o->obuf, _("(bad commit)\n"));
 298                else {
 299                        const char *title;
 300                        const char *msg = get_commit_buffer(commit, NULL);
 301                        int len = find_commit_subject(msg, &title);
 302                        if (len)
 303                                strbuf_addf(&o->obuf, "%.*s\n", len, title);
 304                        unuse_commit_buffer(commit, msg);
 305                }
 306        }
 307        flush_output(o);
 308}
 309
 310static int add_cacheinfo(struct merge_options *o,
 311                unsigned int mode, const struct object_id *oid,
 312                const char *path, int stage, int refresh, int options)
 313{
 314        struct cache_entry *ce;
 315        int ret;
 316
 317        ce = make_cache_entry(mode, oid ? oid->hash : null_sha1, path, stage, 0);
 318        if (!ce)
 319                return err(o, _("addinfo_cache failed for path '%s'"), path);
 320
 321        ret = add_cache_entry(ce, options);
 322        if (refresh) {
 323                struct cache_entry *nce;
 324
 325                nce = refresh_cache_entry(ce, CE_MATCH_REFRESH | CE_MATCH_IGNORE_MISSING);
 326                if (!nce)
 327                        return err(o, _("addinfo_cache failed for path '%s'"), path);
 328                if (nce != ce)
 329                        ret = add_cache_entry(nce, options);
 330        }
 331        return ret;
 332}
 333
 334static void init_tree_desc_from_tree(struct tree_desc *desc, struct tree *tree)
 335{
 336        parse_tree(tree);
 337        init_tree_desc(desc, tree->buffer, tree->size);
 338}
 339
 340static int git_merge_trees(int index_only,
 341                           struct tree *common,
 342                           struct tree *head,
 343                           struct tree *merge)
 344{
 345        int rc;
 346        struct tree_desc t[3];
 347        struct unpack_trees_options opts;
 348
 349        memset(&opts, 0, sizeof(opts));
 350        if (index_only)
 351                opts.index_only = 1;
 352        else
 353                opts.update = 1;
 354        opts.merge = 1;
 355        opts.head_idx = 2;
 356        opts.fn = threeway_merge;
 357        opts.src_index = &the_index;
 358        opts.dst_index = &the_index;
 359        setup_unpack_trees_porcelain(&opts, "merge");
 360
 361        init_tree_desc_from_tree(t+0, common);
 362        init_tree_desc_from_tree(t+1, head);
 363        init_tree_desc_from_tree(t+2, merge);
 364
 365        rc = unpack_trees(3, t, &opts);
 366        cache_tree_free(&active_cache_tree);
 367        return rc;
 368}
 369
 370struct tree *write_tree_from_memory(struct merge_options *o)
 371{
 372        struct tree *result = NULL;
 373
 374        if (unmerged_cache()) {
 375                int i;
 376                fprintf(stderr, "BUG: There are unmerged index entries:\n");
 377                for (i = 0; i < active_nr; i++) {
 378                        const struct cache_entry *ce = active_cache[i];
 379                        if (ce_stage(ce))
 380                                fprintf(stderr, "BUG: %d %.*s\n", ce_stage(ce),
 381                                        (int)ce_namelen(ce), ce->name);
 382                }
 383                die("BUG: unmerged index entries in merge-recursive.c");
 384        }
 385
 386        if (!active_cache_tree)
 387                active_cache_tree = cache_tree();
 388
 389        if (!cache_tree_fully_valid(active_cache_tree) &&
 390            cache_tree_update(&the_index, 0) < 0) {
 391                err(o, _("error building trees"));
 392                return NULL;
 393        }
 394
 395        result = lookup_tree(&active_cache_tree->oid);
 396
 397        return result;
 398}
 399
 400static int save_files_dirs(const struct object_id *oid,
 401                struct strbuf *base, const char *path,
 402                unsigned int mode, int stage, void *context)
 403{
 404        struct path_hashmap_entry *entry;
 405        int baselen = base->len;
 406        struct merge_options *o = context;
 407
 408        strbuf_addstr(base, path);
 409
 410        FLEX_ALLOC_MEM(entry, path, base->buf, base->len);
 411        hashmap_entry_init(entry, path_hash(entry->path));
 412        hashmap_add(&o->current_file_dir_set, entry);
 413
 414        strbuf_setlen(base, baselen);
 415        return (S_ISDIR(mode) ? READ_TREE_RECURSIVE : 0);
 416}
 417
 418static void get_files_dirs(struct merge_options *o, struct tree *tree)
 419{
 420        struct pathspec match_all;
 421        memset(&match_all, 0, sizeof(match_all));
 422        read_tree_recursive(tree, "", 0, 0, &match_all, save_files_dirs, o);
 423}
 424
 425static int get_tree_entry_if_blob(const struct object_id *tree,
 426                                  const char *path,
 427                                  struct object_id *hashy,
 428                                  unsigned int *mode_o)
 429{
 430        int ret;
 431
 432        ret = get_tree_entry(tree, path, hashy, mode_o);
 433        if (S_ISDIR(*mode_o)) {
 434                oidcpy(hashy, &null_oid);
 435                *mode_o = 0;
 436        }
 437        return ret;
 438}
 439
 440/*
 441 * Returns an index_entry instance which doesn't have to correspond to
 442 * a real cache entry in Git's index.
 443 */
 444static struct stage_data *insert_stage_data(const char *path,
 445                struct tree *o, struct tree *a, struct tree *b,
 446                struct string_list *entries)
 447{
 448        struct string_list_item *item;
 449        struct stage_data *e = xcalloc(1, sizeof(struct stage_data));
 450        get_tree_entry_if_blob(&o->object.oid, path,
 451                               &e->stages[1].oid, &e->stages[1].mode);
 452        get_tree_entry_if_blob(&a->object.oid, path,
 453                               &e->stages[2].oid, &e->stages[2].mode);
 454        get_tree_entry_if_blob(&b->object.oid, path,
 455                               &e->stages[3].oid, &e->stages[3].mode);
 456        item = string_list_insert(entries, path);
 457        item->util = e;
 458        return e;
 459}
 460
 461/*
 462 * Create a dictionary mapping file names to stage_data objects. The
 463 * dictionary contains one entry for every path with a non-zero stage entry.
 464 */
 465static struct string_list *get_unmerged(void)
 466{
 467        struct string_list *unmerged = xcalloc(1, sizeof(struct string_list));
 468        int i;
 469
 470        unmerged->strdup_strings = 1;
 471
 472        for (i = 0; i < active_nr; i++) {
 473                struct string_list_item *item;
 474                struct stage_data *e;
 475                const struct cache_entry *ce = active_cache[i];
 476                if (!ce_stage(ce))
 477                        continue;
 478
 479                item = string_list_lookup(unmerged, ce->name);
 480                if (!item) {
 481                        item = string_list_insert(unmerged, ce->name);
 482                        item->util = xcalloc(1, sizeof(struct stage_data));
 483                }
 484                e = item->util;
 485                e->stages[ce_stage(ce)].mode = ce->ce_mode;
 486                oidcpy(&e->stages[ce_stage(ce)].oid, &ce->oid);
 487        }
 488
 489        return unmerged;
 490}
 491
 492static int string_list_df_name_compare(const char *one, const char *two)
 493{
 494        int onelen = strlen(one);
 495        int twolen = strlen(two);
 496        /*
 497         * Here we only care that entries for D/F conflicts are
 498         * adjacent, in particular with the file of the D/F conflict
 499         * appearing before files below the corresponding directory.
 500         * The order of the rest of the list is irrelevant for us.
 501         *
 502         * To achieve this, we sort with df_name_compare and provide
 503         * the mode S_IFDIR so that D/F conflicts will sort correctly.
 504         * We use the mode S_IFDIR for everything else for simplicity,
 505         * since in other cases any changes in their order due to
 506         * sorting cause no problems for us.
 507         */
 508        int cmp = df_name_compare(one, onelen, S_IFDIR,
 509                                  two, twolen, S_IFDIR);
 510        /*
 511         * Now that 'foo' and 'foo/bar' compare equal, we have to make sure
 512         * that 'foo' comes before 'foo/bar'.
 513         */
 514        if (cmp)
 515                return cmp;
 516        return onelen - twolen;
 517}
 518
 519static void record_df_conflict_files(struct merge_options *o,
 520                                     struct string_list *entries)
 521{
 522        /* If there is a D/F conflict and the file for such a conflict
 523         * currently exist in the working tree, we want to allow it to be
 524         * removed to make room for the corresponding directory if needed.
 525         * The files underneath the directories of such D/F conflicts will
 526         * be processed before the corresponding file involved in the D/F
 527         * conflict.  If the D/F directory ends up being removed by the
 528         * merge, then we won't have to touch the D/F file.  If the D/F
 529         * directory needs to be written to the working copy, then the D/F
 530         * file will simply be removed (in make_room_for_path()) to make
 531         * room for the necessary paths.  Note that if both the directory
 532         * and the file need to be present, then the D/F file will be
 533         * reinstated with a new unique name at the time it is processed.
 534         */
 535        struct string_list df_sorted_entries = STRING_LIST_INIT_NODUP;
 536        const char *last_file = NULL;
 537        int last_len = 0;
 538        int i;
 539
 540        /*
 541         * If we're merging merge-bases, we don't want to bother with
 542         * any working directory changes.
 543         */
 544        if (o->call_depth)
 545                return;
 546
 547        /* Ensure D/F conflicts are adjacent in the entries list. */
 548        for (i = 0; i < entries->nr; i++) {
 549                struct string_list_item *next = &entries->items[i];
 550                string_list_append(&df_sorted_entries, next->string)->util =
 551                                   next->util;
 552        }
 553        df_sorted_entries.cmp = string_list_df_name_compare;
 554        string_list_sort(&df_sorted_entries);
 555
 556        string_list_clear(&o->df_conflict_file_set, 1);
 557        for (i = 0; i < df_sorted_entries.nr; i++) {
 558                const char *path = df_sorted_entries.items[i].string;
 559                int len = strlen(path);
 560                struct stage_data *e = df_sorted_entries.items[i].util;
 561
 562                /*
 563                 * Check if last_file & path correspond to a D/F conflict;
 564                 * i.e. whether path is last_file+'/'+<something>.
 565                 * If so, record that it's okay to remove last_file to make
 566                 * room for path and friends if needed.
 567                 */
 568                if (last_file &&
 569                    len > last_len &&
 570                    memcmp(path, last_file, last_len) == 0 &&
 571                    path[last_len] == '/') {
 572                        string_list_insert(&o->df_conflict_file_set, last_file);
 573                }
 574
 575                /*
 576                 * Determine whether path could exist as a file in the
 577                 * working directory as a possible D/F conflict.  This
 578                 * will only occur when it exists in stage 2 as a
 579                 * file.
 580                 */
 581                if (S_ISREG(e->stages[2].mode) || S_ISLNK(e->stages[2].mode)) {
 582                        last_file = path;
 583                        last_len = len;
 584                } else {
 585                        last_file = NULL;
 586                }
 587        }
 588        string_list_clear(&df_sorted_entries, 0);
 589}
 590
 591struct rename {
 592        struct diff_filepair *pair;
 593        /*
 594         * Purpose of src_entry and dst_entry:
 595         *
 596         * If 'before' is renamed to 'after' then src_entry will contain
 597         * the versions of 'before' from the merge_base, HEAD, and MERGE in
 598         * stages 1, 2, and 3; dst_entry will contain the respective
 599         * versions of 'after' in corresponding locations.  Thus, we have a
 600         * total of six modes and oids, though some will be null.  (Stage 0
 601         * is ignored; we're interested in handling conflicts.)
 602         *
 603         * Since we don't turn on break-rewrites by default, neither
 604         * src_entry nor dst_entry can have all three of their stages have
 605         * non-null oids, meaning at most four of the six will be non-null.
 606         * Also, since this is a rename, both src_entry and dst_entry will
 607         * have at least one non-null oid, meaning at least two will be
 608         * non-null.  Of the six oids, a typical rename will have three be
 609         * non-null.  Only two implies a rename/delete, and four implies a
 610         * rename/add.
 611         */
 612        struct stage_data *src_entry;
 613        struct stage_data *dst_entry;
 614        unsigned add_turned_into_rename:1;
 615        unsigned processed:1;
 616};
 617
 618static int update_stages(struct merge_options *opt, const char *path,
 619                         const struct diff_filespec *o,
 620                         const struct diff_filespec *a,
 621                         const struct diff_filespec *b)
 622{
 623
 624        /*
 625         * NOTE: It is usually a bad idea to call update_stages on a path
 626         * before calling update_file on that same path, since it can
 627         * sometimes lead to spurious "refusing to lose untracked file..."
 628         * messages from update_file (via make_room_for path via
 629         * would_lose_untracked).  Instead, reverse the order of the calls
 630         * (executing update_file first and then update_stages).
 631         */
 632        int clear = 1;
 633        int options = ADD_CACHE_OK_TO_ADD | ADD_CACHE_SKIP_DFCHECK;
 634        if (clear)
 635                if (remove_file_from_cache(path))
 636                        return -1;
 637        if (o)
 638                if (add_cacheinfo(opt, o->mode, &o->oid, path, 1, 0, options))
 639                        return -1;
 640        if (a)
 641                if (add_cacheinfo(opt, a->mode, &a->oid, path, 2, 0, options))
 642                        return -1;
 643        if (b)
 644                if (add_cacheinfo(opt, b->mode, &b->oid, path, 3, 0, options))
 645                        return -1;
 646        return 0;
 647}
 648
 649static int update_stages_for_stage_data(struct merge_options *opt,
 650                                        const char *path,
 651                                        const struct stage_data *stage_data)
 652{
 653        struct diff_filespec o, a, b;
 654
 655        o.mode = stage_data->stages[1].mode;
 656        oidcpy(&o.oid, &stage_data->stages[1].oid);
 657
 658        a.mode = stage_data->stages[2].mode;
 659        oidcpy(&a.oid, &stage_data->stages[2].oid);
 660
 661        b.mode = stage_data->stages[3].mode;
 662        oidcpy(&b.oid, &stage_data->stages[3].oid);
 663
 664        return update_stages(opt, path,
 665                             is_null_oid(&o.oid) ? NULL : &o,
 666                             is_null_oid(&a.oid) ? NULL : &a,
 667                             is_null_oid(&b.oid) ? NULL : &b);
 668}
 669
 670static void update_entry(struct stage_data *entry,
 671                         struct diff_filespec *o,
 672                         struct diff_filespec *a,
 673                         struct diff_filespec *b)
 674{
 675        entry->processed = 0;
 676        entry->stages[1].mode = o->mode;
 677        entry->stages[2].mode = a->mode;
 678        entry->stages[3].mode = b->mode;
 679        oidcpy(&entry->stages[1].oid, &o->oid);
 680        oidcpy(&entry->stages[2].oid, &a->oid);
 681        oidcpy(&entry->stages[3].oid, &b->oid);
 682}
 683
 684static int remove_file(struct merge_options *o, int clean,
 685                       const char *path, int no_wd)
 686{
 687        int update_cache = o->call_depth || clean;
 688        int update_working_directory = !o->call_depth && !no_wd;
 689
 690        if (update_cache) {
 691                if (remove_file_from_cache(path))
 692                        return -1;
 693        }
 694        if (update_working_directory) {
 695                if (ignore_case) {
 696                        struct cache_entry *ce;
 697                        ce = cache_file_exists(path, strlen(path), ignore_case);
 698                        if (ce && ce_stage(ce) == 0 && strcmp(path, ce->name))
 699                                return 0;
 700                }
 701                if (remove_path(path))
 702                        return -1;
 703        }
 704        return 0;
 705}
 706
 707/* add a string to a strbuf, but converting "/" to "_" */
 708static void add_flattened_path(struct strbuf *out, const char *s)
 709{
 710        size_t i = out->len;
 711        strbuf_addstr(out, s);
 712        for (; i < out->len; i++)
 713                if (out->buf[i] == '/')
 714                        out->buf[i] = '_';
 715}
 716
 717static char *unique_path(struct merge_options *o, const char *path, const char *branch)
 718{
 719        struct path_hashmap_entry *entry;
 720        struct strbuf newpath = STRBUF_INIT;
 721        int suffix = 0;
 722        size_t base_len;
 723
 724        strbuf_addf(&newpath, "%s~", path);
 725        add_flattened_path(&newpath, branch);
 726
 727        base_len = newpath.len;
 728        while (hashmap_get_from_hash(&o->current_file_dir_set,
 729                                     path_hash(newpath.buf), newpath.buf) ||
 730               (!o->call_depth && file_exists(newpath.buf))) {
 731                strbuf_setlen(&newpath, base_len);
 732                strbuf_addf(&newpath, "_%d", suffix++);
 733        }
 734
 735        FLEX_ALLOC_MEM(entry, path, newpath.buf, newpath.len);
 736        hashmap_entry_init(entry, path_hash(entry->path));
 737        hashmap_add(&o->current_file_dir_set, entry);
 738        return strbuf_detach(&newpath, NULL);
 739}
 740
 741/**
 742 * Check whether a directory in the index is in the way of an incoming
 743 * file.  Return 1 if so.  If check_working_copy is non-zero, also
 744 * check the working directory.  If empty_ok is non-zero, also return
 745 * 0 in the case where the working-tree dir exists but is empty.
 746 */
 747static int dir_in_way(const char *path, int check_working_copy, int empty_ok)
 748{
 749        int pos;
 750        struct strbuf dirpath = STRBUF_INIT;
 751        struct stat st;
 752
 753        strbuf_addstr(&dirpath, path);
 754        strbuf_addch(&dirpath, '/');
 755
 756        pos = cache_name_pos(dirpath.buf, dirpath.len);
 757
 758        if (pos < 0)
 759                pos = -1 - pos;
 760        if (pos < active_nr &&
 761            !strncmp(dirpath.buf, active_cache[pos]->name, dirpath.len)) {
 762                strbuf_release(&dirpath);
 763                return 1;
 764        }
 765
 766        strbuf_release(&dirpath);
 767        return check_working_copy && !lstat(path, &st) && S_ISDIR(st.st_mode) &&
 768                !(empty_ok && is_empty_dir(path));
 769}
 770
 771static int was_tracked(const char *path)
 772{
 773        int pos = cache_name_pos(path, strlen(path));
 774
 775        if (0 <= pos)
 776                /* we have been tracking this path */
 777                return 1;
 778
 779        /*
 780         * Look for an unmerged entry for the path,
 781         * specifically stage #2, which would indicate
 782         * that "our" side before the merge started
 783         * had the path tracked (and resulted in a conflict).
 784         */
 785        for (pos = -1 - pos;
 786             pos < active_nr && !strcmp(path, active_cache[pos]->name);
 787             pos++)
 788                if (ce_stage(active_cache[pos]) == 2)
 789                        return 1;
 790        return 0;
 791}
 792
 793static int would_lose_untracked(const char *path)
 794{
 795        return !was_tracked(path) && file_exists(path);
 796}
 797
 798static int make_room_for_path(struct merge_options *o, const char *path)
 799{
 800        int status, i;
 801        const char *msg = _("failed to create path '%s'%s");
 802
 803        /* Unlink any D/F conflict files that are in the way */
 804        for (i = 0; i < o->df_conflict_file_set.nr; i++) {
 805                const char *df_path = o->df_conflict_file_set.items[i].string;
 806                size_t pathlen = strlen(path);
 807                size_t df_pathlen = strlen(df_path);
 808                if (df_pathlen < pathlen &&
 809                    path[df_pathlen] == '/' &&
 810                    strncmp(path, df_path, df_pathlen) == 0) {
 811                        output(o, 3,
 812                               _("Removing %s to make room for subdirectory\n"),
 813                               df_path);
 814                        unlink(df_path);
 815                        unsorted_string_list_delete_item(&o->df_conflict_file_set,
 816                                                         i, 0);
 817                        break;
 818                }
 819        }
 820
 821        /* Make sure leading directories are created */
 822        status = safe_create_leading_directories_const(path);
 823        if (status) {
 824                if (status == SCLD_EXISTS)
 825                        /* something else exists */
 826                        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 827                return err(o, msg, path, "");
 828        }
 829
 830        /*
 831         * Do not unlink a file in the work tree if we are not
 832         * tracking it.
 833         */
 834        if (would_lose_untracked(path))
 835                return err(o, _("refusing to lose untracked file at '%s'"),
 836                             path);
 837
 838        /* Successful unlink is good.. */
 839        if (!unlink(path))
 840                return 0;
 841        /* .. and so is no existing file */
 842        if (errno == ENOENT)
 843                return 0;
 844        /* .. but not some other error (who really cares what?) */
 845        return err(o, msg, path, _(": perhaps a D/F conflict?"));
 846}
 847
 848static int update_file_flags(struct merge_options *o,
 849                             const struct object_id *oid,
 850                             unsigned mode,
 851                             const char *path,
 852                             int update_cache,
 853                             int update_wd)
 854{
 855        int ret = 0;
 856
 857        if (o->call_depth)
 858                update_wd = 0;
 859
 860        if (update_wd) {
 861                enum object_type type;
 862                void *buf;
 863                unsigned long size;
 864
 865                if (S_ISGITLINK(mode)) {
 866                        /*
 867                         * We may later decide to recursively descend into
 868                         * the submodule directory and update its index
 869                         * and/or work tree, but we do not do that now.
 870                         */
 871                        update_wd = 0;
 872                        goto update_index;
 873                }
 874
 875                buf = read_object_file(oid, &type, &size);
 876                if (!buf)
 877                        return err(o, _("cannot read object %s '%s'"), oid_to_hex(oid), path);
 878                if (type != OBJ_BLOB) {
 879                        ret = err(o, _("blob expected for %s '%s'"), oid_to_hex(oid), path);
 880                        goto free_buf;
 881                }
 882                if (S_ISREG(mode)) {
 883                        struct strbuf strbuf = STRBUF_INIT;
 884                        if (convert_to_working_tree(path, buf, size, &strbuf)) {
 885                                free(buf);
 886                                size = strbuf.len;
 887                                buf = strbuf_detach(&strbuf, NULL);
 888                        }
 889                }
 890
 891                if (make_room_for_path(o, path) < 0) {
 892                        update_wd = 0;
 893                        goto free_buf;
 894                }
 895                if (S_ISREG(mode) || (!has_symlinks && S_ISLNK(mode))) {
 896                        int fd;
 897                        if (mode & 0100)
 898                                mode = 0777;
 899                        else
 900                                mode = 0666;
 901                        fd = open(path, O_WRONLY | O_TRUNC | O_CREAT, mode);
 902                        if (fd < 0) {
 903                                ret = err(o, _("failed to open '%s': %s"),
 904                                          path, strerror(errno));
 905                                goto free_buf;
 906                        }
 907                        write_in_full(fd, buf, size);
 908                        close(fd);
 909                } else if (S_ISLNK(mode)) {
 910                        char *lnk = xmemdupz(buf, size);
 911                        safe_create_leading_directories_const(path);
 912                        unlink(path);
 913                        if (symlink(lnk, path))
 914                                ret = err(o, _("failed to symlink '%s': %s"),
 915                                        path, strerror(errno));
 916                        free(lnk);
 917                } else
 918                        ret = err(o,
 919                                  _("do not know what to do with %06o %s '%s'"),
 920                                  mode, oid_to_hex(oid), path);
 921 free_buf:
 922                free(buf);
 923        }
 924 update_index:
 925        if (!ret && update_cache)
 926                add_cacheinfo(o, mode, oid, path, 0, update_wd, ADD_CACHE_OK_TO_ADD);
 927        return ret;
 928}
 929
 930static int update_file(struct merge_options *o,
 931                       int clean,
 932                       const struct object_id *oid,
 933                       unsigned mode,
 934                       const char *path)
 935{
 936        return update_file_flags(o, oid, mode, path, o->call_depth || clean, !o->call_depth);
 937}
 938
 939/* Low level file merging, update and removal */
 940
 941struct merge_file_info {
 942        struct object_id oid;
 943        unsigned mode;
 944        unsigned clean:1,
 945                 merge:1;
 946};
 947
 948static int merge_3way(struct merge_options *o,
 949                      mmbuffer_t *result_buf,
 950                      const struct diff_filespec *one,
 951                      const struct diff_filespec *a,
 952                      const struct diff_filespec *b,
 953                      const char *branch1,
 954                      const char *branch2)
 955{
 956        mmfile_t orig, src1, src2;
 957        struct ll_merge_options ll_opts = {0};
 958        char *base_name, *name1, *name2;
 959        int merge_status;
 960
 961        ll_opts.renormalize = o->renormalize;
 962        ll_opts.xdl_opts = o->xdl_opts;
 963
 964        if (o->call_depth) {
 965                ll_opts.virtual_ancestor = 1;
 966                ll_opts.variant = 0;
 967        } else {
 968                switch (o->recursive_variant) {
 969                case MERGE_RECURSIVE_OURS:
 970                        ll_opts.variant = XDL_MERGE_FAVOR_OURS;
 971                        break;
 972                case MERGE_RECURSIVE_THEIRS:
 973                        ll_opts.variant = XDL_MERGE_FAVOR_THEIRS;
 974                        break;
 975                default:
 976                        ll_opts.variant = 0;
 977                        break;
 978                }
 979        }
 980
 981        if (strcmp(a->path, b->path) ||
 982            (o->ancestor != NULL && strcmp(a->path, one->path) != 0)) {
 983                base_name = o->ancestor == NULL ? NULL :
 984                        mkpathdup("%s:%s", o->ancestor, one->path);
 985                name1 = mkpathdup("%s:%s", branch1, a->path);
 986                name2 = mkpathdup("%s:%s", branch2, b->path);
 987        } else {
 988                base_name = o->ancestor == NULL ? NULL :
 989                        mkpathdup("%s", o->ancestor);
 990                name1 = mkpathdup("%s", branch1);
 991                name2 = mkpathdup("%s", branch2);
 992        }
 993
 994        read_mmblob(&orig, &one->oid);
 995        read_mmblob(&src1, &a->oid);
 996        read_mmblob(&src2, &b->oid);
 997
 998        merge_status = ll_merge(result_buf, a->path, &orig, base_name,
 999                                &src1, name1, &src2, name2, &ll_opts);
1000
1001        free(base_name);
1002        free(name1);
1003        free(name2);
1004        free(orig.ptr);
1005        free(src1.ptr);
1006        free(src2.ptr);
1007        return merge_status;
1008}
1009
1010static int merge_file_1(struct merge_options *o,
1011                                           const struct diff_filespec *one,
1012                                           const struct diff_filespec *a,
1013                                           const struct diff_filespec *b,
1014                                           const char *branch1,
1015                                           const char *branch2,
1016                                           struct merge_file_info *result)
1017{
1018        result->merge = 0;
1019        result->clean = 1;
1020
1021        if ((S_IFMT & a->mode) != (S_IFMT & b->mode)) {
1022                result->clean = 0;
1023                if (S_ISREG(a->mode)) {
1024                        result->mode = a->mode;
1025                        oidcpy(&result->oid, &a->oid);
1026                } else {
1027                        result->mode = b->mode;
1028                        oidcpy(&result->oid, &b->oid);
1029                }
1030        } else {
1031                if (!oid_eq(&a->oid, &one->oid) && !oid_eq(&b->oid, &one->oid))
1032                        result->merge = 1;
1033
1034                /*
1035                 * Merge modes
1036                 */
1037                if (a->mode == b->mode || a->mode == one->mode)
1038                        result->mode = b->mode;
1039                else {
1040                        result->mode = a->mode;
1041                        if (b->mode != one->mode) {
1042                                result->clean = 0;
1043                                result->merge = 1;
1044                        }
1045                }
1046
1047                if (oid_eq(&a->oid, &b->oid) || oid_eq(&a->oid, &one->oid))
1048                        oidcpy(&result->oid, &b->oid);
1049                else if (oid_eq(&b->oid, &one->oid))
1050                        oidcpy(&result->oid, &a->oid);
1051                else if (S_ISREG(a->mode)) {
1052                        mmbuffer_t result_buf;
1053                        int ret = 0, merge_status;
1054
1055                        merge_status = merge_3way(o, &result_buf, one, a, b,
1056                                                  branch1, branch2);
1057
1058                        if ((merge_status < 0) || !result_buf.ptr)
1059                                ret = err(o, _("Failed to execute internal merge"));
1060
1061                        if (!ret &&
1062                            write_object_file(result_buf.ptr, result_buf.size,
1063                                              blob_type, &result->oid))
1064                                ret = err(o, _("Unable to add %s to database"),
1065                                          a->path);
1066
1067                        free(result_buf.ptr);
1068                        if (ret)
1069                                return ret;
1070                        result->clean = (merge_status == 0);
1071                } else if (S_ISGITLINK(a->mode)) {
1072                        result->clean = merge_submodule(&result->oid,
1073                                                       one->path,
1074                                                       &one->oid,
1075                                                       &a->oid,
1076                                                       &b->oid,
1077                                                       !o->call_depth);
1078                } else if (S_ISLNK(a->mode)) {
1079                        switch (o->recursive_variant) {
1080                        case MERGE_RECURSIVE_NORMAL:
1081                                oidcpy(&result->oid, &a->oid);
1082                                if (!oid_eq(&a->oid, &b->oid))
1083                                        result->clean = 0;
1084                                break;
1085                        case MERGE_RECURSIVE_OURS:
1086                                oidcpy(&result->oid, &a->oid);
1087                                break;
1088                        case MERGE_RECURSIVE_THEIRS:
1089                                oidcpy(&result->oid, &b->oid);
1090                                break;
1091                        }
1092                } else
1093                        die("BUG: unsupported object type in the tree");
1094        }
1095
1096        return 0;
1097}
1098
1099static int merge_file_special_markers(struct merge_options *o,
1100                           const struct diff_filespec *one,
1101                           const struct diff_filespec *a,
1102                           const struct diff_filespec *b,
1103                           const char *branch1,
1104                           const char *filename1,
1105                           const char *branch2,
1106                           const char *filename2,
1107                           struct merge_file_info *mfi)
1108{
1109        char *side1 = NULL;
1110        char *side2 = NULL;
1111        int ret;
1112
1113        if (filename1)
1114                side1 = xstrfmt("%s:%s", branch1, filename1);
1115        if (filename2)
1116                side2 = xstrfmt("%s:%s", branch2, filename2);
1117
1118        ret = merge_file_1(o, one, a, b,
1119                           side1 ? side1 : branch1,
1120                           side2 ? side2 : branch2, mfi);
1121        free(side1);
1122        free(side2);
1123        return ret;
1124}
1125
1126static int merge_file_one(struct merge_options *o,
1127                                         const char *path,
1128                                         const struct object_id *o_oid, int o_mode,
1129                                         const struct object_id *a_oid, int a_mode,
1130                                         const struct object_id *b_oid, int b_mode,
1131                                         const char *branch1,
1132                                         const char *branch2,
1133                                         struct merge_file_info *mfi)
1134{
1135        struct diff_filespec one, a, b;
1136
1137        one.path = a.path = b.path = (char *)path;
1138        oidcpy(&one.oid, o_oid);
1139        one.mode = o_mode;
1140        oidcpy(&a.oid, a_oid);
1141        a.mode = a_mode;
1142        oidcpy(&b.oid, b_oid);
1143        b.mode = b_mode;
1144        return merge_file_1(o, &one, &a, &b, branch1, branch2, mfi);
1145}
1146
1147static int conflict_rename_dir(struct merge_options *o,
1148                               struct diff_filepair *pair,
1149                               const char *rename_branch,
1150                               const char *other_branch)
1151{
1152        const struct diff_filespec *dest = pair->two;
1153
1154        if (update_file(o, 1, &dest->oid, dest->mode, dest->path))
1155                return -1;
1156        return 0;
1157}
1158
1159static int handle_change_delete(struct merge_options *o,
1160                                 const char *path, const char *old_path,
1161                                 const struct object_id *o_oid, int o_mode,
1162                                 const struct object_id *changed_oid,
1163                                 int changed_mode,
1164                                 const char *change_branch,
1165                                 const char *delete_branch,
1166                                 const char *change, const char *change_past)
1167{
1168        char *alt_path = NULL;
1169        const char *update_path = path;
1170        int ret = 0;
1171
1172        if (dir_in_way(path, !o->call_depth, 0)) {
1173                update_path = alt_path = unique_path(o, path, change_branch);
1174        }
1175
1176        if (o->call_depth) {
1177                /*
1178                 * We cannot arbitrarily accept either a_sha or b_sha as
1179                 * correct; since there is no true "middle point" between
1180                 * them, simply reuse the base version for virtual merge base.
1181                 */
1182                ret = remove_file_from_cache(path);
1183                if (!ret)
1184                        ret = update_file(o, 0, o_oid, o_mode, update_path);
1185        } else {
1186                if (!alt_path) {
1187                        if (!old_path) {
1188                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1189                                       "and %s in %s. Version %s of %s left in tree."),
1190                                       change, path, delete_branch, change_past,
1191                                       change_branch, change_branch, path);
1192                        } else {
1193                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1194                                       "and %s to %s in %s. Version %s of %s left in tree."),
1195                                       change, old_path, delete_branch, change_past, path,
1196                                       change_branch, change_branch, path);
1197                        }
1198                } else {
1199                        if (!old_path) {
1200                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1201                                       "and %s in %s. Version %s of %s left in tree at %s."),
1202                                       change, path, delete_branch, change_past,
1203                                       change_branch, change_branch, path, alt_path);
1204                        } else {
1205                                output(o, 1, _("CONFLICT (%s/delete): %s deleted in %s "
1206                                       "and %s to %s in %s. Version %s of %s left in tree at %s."),
1207                                       change, old_path, delete_branch, change_past, path,
1208                                       change_branch, change_branch, path, alt_path);
1209                        }
1210                }
1211                /*
1212                 * No need to call update_file() on path when change_branch ==
1213                 * o->branch1 && !alt_path, since that would needlessly touch
1214                 * path.  We could call update_file_flags() with update_cache=0
1215                 * and update_wd=0, but that's a no-op.
1216                 */
1217                if (change_branch != o->branch1 || alt_path)
1218                        ret = update_file(o, 0, changed_oid, changed_mode, update_path);
1219        }
1220        free(alt_path);
1221
1222        return ret;
1223}
1224
1225static int conflict_rename_delete(struct merge_options *o,
1226                                   struct diff_filepair *pair,
1227                                   const char *rename_branch,
1228                                   const char *delete_branch)
1229{
1230        const struct diff_filespec *orig = pair->one;
1231        const struct diff_filespec *dest = pair->two;
1232
1233        if (handle_change_delete(o,
1234                                 o->call_depth ? orig->path : dest->path,
1235                                 o->call_depth ? NULL : orig->path,
1236                                 &orig->oid, orig->mode,
1237                                 &dest->oid, dest->mode,
1238                                 rename_branch, delete_branch,
1239                                 _("rename"), _("renamed")))
1240                return -1;
1241
1242        if (o->call_depth)
1243                return remove_file_from_cache(dest->path);
1244        else
1245                return update_stages(o, dest->path, NULL,
1246                                     rename_branch == o->branch1 ? dest : NULL,
1247                                     rename_branch == o->branch1 ? NULL : dest);
1248}
1249
1250static struct diff_filespec *filespec_from_entry(struct diff_filespec *target,
1251                                                 struct stage_data *entry,
1252                                                 int stage)
1253{
1254        struct object_id *oid = &entry->stages[stage].oid;
1255        unsigned mode = entry->stages[stage].mode;
1256        if (mode == 0 || is_null_oid(oid))
1257                return NULL;
1258        oidcpy(&target->oid, oid);
1259        target->mode = mode;
1260        return target;
1261}
1262
1263static int handle_file(struct merge_options *o,
1264                        struct diff_filespec *rename,
1265                        int stage,
1266                        struct rename_conflict_info *ci)
1267{
1268        char *dst_name = rename->path;
1269        struct stage_data *dst_entry;
1270        const char *cur_branch, *other_branch;
1271        struct diff_filespec other;
1272        struct diff_filespec *add;
1273        int ret;
1274
1275        if (stage == 2) {
1276                dst_entry = ci->dst_entry1;
1277                cur_branch = ci->branch1;
1278                other_branch = ci->branch2;
1279        } else {
1280                dst_entry = ci->dst_entry2;
1281                cur_branch = ci->branch2;
1282                other_branch = ci->branch1;
1283        }
1284
1285        add = filespec_from_entry(&other, dst_entry, stage ^ 1);
1286        if (add) {
1287                char *add_name = unique_path(o, rename->path, other_branch);
1288                if (update_file(o, 0, &add->oid, add->mode, add_name))
1289                        return -1;
1290
1291                remove_file(o, 0, rename->path, 0);
1292                dst_name = unique_path(o, rename->path, cur_branch);
1293        } else {
1294                if (dir_in_way(rename->path, !o->call_depth, 0)) {
1295                        dst_name = unique_path(o, rename->path, cur_branch);
1296                        output(o, 1, _("%s is a directory in %s adding as %s instead"),
1297                               rename->path, other_branch, dst_name);
1298                }
1299        }
1300        if ((ret = update_file(o, 0, &rename->oid, rename->mode, dst_name)))
1301                ; /* fall through, do allow dst_name to be released */
1302        else if (stage == 2)
1303                ret = update_stages(o, rename->path, NULL, rename, add);
1304        else
1305                ret = update_stages(o, rename->path, NULL, add, rename);
1306
1307        if (dst_name != rename->path)
1308                free(dst_name);
1309
1310        return ret;
1311}
1312
1313static int conflict_rename_rename_1to2(struct merge_options *o,
1314                                        struct rename_conflict_info *ci)
1315{
1316        /* One file was renamed in both branches, but to different names. */
1317        struct diff_filespec *one = ci->pair1->one;
1318        struct diff_filespec *a = ci->pair1->two;
1319        struct diff_filespec *b = ci->pair2->two;
1320
1321        output(o, 1, _("CONFLICT (rename/rename): "
1322               "Rename \"%s\"->\"%s\" in branch \"%s\" "
1323               "rename \"%s\"->\"%s\" in \"%s\"%s"),
1324               one->path, a->path, ci->branch1,
1325               one->path, b->path, ci->branch2,
1326               o->call_depth ? _(" (left unresolved)") : "");
1327        if (o->call_depth) {
1328                struct merge_file_info mfi;
1329                struct diff_filespec other;
1330                struct diff_filespec *add;
1331                if (merge_file_one(o, one->path,
1332                                 &one->oid, one->mode,
1333                                 &a->oid, a->mode,
1334                                 &b->oid, b->mode,
1335                                 ci->branch1, ci->branch2, &mfi))
1336                        return -1;
1337
1338                /*
1339                 * FIXME: For rename/add-source conflicts (if we could detect
1340                 * such), this is wrong.  We should instead find a unique
1341                 * pathname and then either rename the add-source file to that
1342                 * unique path, or use that unique path instead of src here.
1343                 */
1344                if (update_file(o, 0, &mfi.oid, mfi.mode, one->path))
1345                        return -1;
1346
1347                /*
1348                 * Above, we put the merged content at the merge-base's
1349                 * path.  Now we usually need to delete both a->path and
1350                 * b->path.  However, the rename on each side of the merge
1351                 * could also be involved in a rename/add conflict.  In
1352                 * such cases, we should keep the added file around,
1353                 * resolving the conflict at that path in its favor.
1354                 */
1355                add = filespec_from_entry(&other, ci->dst_entry1, 2 ^ 1);
1356                if (add) {
1357                        if (update_file(o, 0, &add->oid, add->mode, a->path))
1358                                return -1;
1359                }
1360                else
1361                        remove_file_from_cache(a->path);
1362                add = filespec_from_entry(&other, ci->dst_entry2, 3 ^ 1);
1363                if (add) {
1364                        if (update_file(o, 0, &add->oid, add->mode, b->path))
1365                                return -1;
1366                }
1367                else
1368                        remove_file_from_cache(b->path);
1369        } else if (handle_file(o, a, 2, ci) || handle_file(o, b, 3, ci))
1370                return -1;
1371
1372        return 0;
1373}
1374
1375static int conflict_rename_rename_2to1(struct merge_options *o,
1376                                        struct rename_conflict_info *ci)
1377{
1378        /* Two files, a & b, were renamed to the same thing, c. */
1379        struct diff_filespec *a = ci->pair1->one;
1380        struct diff_filespec *b = ci->pair2->one;
1381        struct diff_filespec *c1 = ci->pair1->two;
1382        struct diff_filespec *c2 = ci->pair2->two;
1383        char *path = c1->path; /* == c2->path */
1384        struct merge_file_info mfi_c1;
1385        struct merge_file_info mfi_c2;
1386        int ret;
1387
1388        output(o, 1, _("CONFLICT (rename/rename): "
1389               "Rename %s->%s in %s. "
1390               "Rename %s->%s in %s"),
1391               a->path, c1->path, ci->branch1,
1392               b->path, c2->path, ci->branch2);
1393
1394        remove_file(o, 1, a->path, o->call_depth || would_lose_untracked(a->path));
1395        remove_file(o, 1, b->path, o->call_depth || would_lose_untracked(b->path));
1396
1397        if (merge_file_special_markers(o, a, c1, &ci->ren1_other,
1398                                       o->branch1, c1->path,
1399                                       o->branch2, ci->ren1_other.path, &mfi_c1) ||
1400            merge_file_special_markers(o, b, &ci->ren2_other, c2,
1401                                       o->branch1, ci->ren2_other.path,
1402                                       o->branch2, c2->path, &mfi_c2))
1403                return -1;
1404
1405        if (o->call_depth) {
1406                /*
1407                 * If mfi_c1.clean && mfi_c2.clean, then it might make
1408                 * sense to do a two-way merge of those results.  But, I
1409                 * think in all cases, it makes sense to have the virtual
1410                 * merge base just undo the renames; they can be detected
1411                 * again later for the non-recursive merge.
1412                 */
1413                remove_file(o, 0, path, 0);
1414                ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, a->path);
1415                if (!ret)
1416                        ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1417                                          b->path);
1418        } else {
1419                char *new_path1 = unique_path(o, path, ci->branch1);
1420                char *new_path2 = unique_path(o, path, ci->branch2);
1421                output(o, 1, _("Renaming %s to %s and %s to %s instead"),
1422                       a->path, new_path1, b->path, new_path2);
1423                remove_file(o, 0, path, 0);
1424                ret = update_file(o, 0, &mfi_c1.oid, mfi_c1.mode, new_path1);
1425                if (!ret)
1426                        ret = update_file(o, 0, &mfi_c2.oid, mfi_c2.mode,
1427                                          new_path2);
1428                /*
1429                 * unpack_trees() actually populates the index for us for
1430                 * "normal" rename/rename(2to1) situtations so that the
1431                 * correct entries are at the higher stages, which would
1432                 * make the call below to update_stages_for_stage_data
1433                 * unnecessary.  However, if either of the renames came
1434                 * from a directory rename, then unpack_trees() will not
1435                 * have gotten the right data loaded into the index, so we
1436                 * need to do so now.  (While it'd be tempting to move this
1437                 * call to update_stages_for_stage_data() to
1438                 * apply_directory_rename_modifications(), that would break
1439                 * our intermediate calls to would_lose_untracked() since
1440                 * those rely on the current in-memory index.  See also the
1441                 * big "NOTE" in update_stages()).
1442                 */
1443                if (update_stages_for_stage_data(o, path, ci->dst_entry1))
1444                        ret = -1;
1445
1446                free(new_path2);
1447                free(new_path1);
1448        }
1449
1450        return ret;
1451}
1452
1453/*
1454 * Get the diff_filepairs changed between o_tree and tree.
1455 */
1456static struct diff_queue_struct *get_diffpairs(struct merge_options *o,
1457                                               struct tree *o_tree,
1458                                               struct tree *tree)
1459{
1460        struct diff_queue_struct *ret;
1461        struct diff_options opts;
1462
1463        diff_setup(&opts);
1464        opts.flags.recursive = 1;
1465        opts.flags.rename_empty = 0;
1466        opts.detect_rename = DIFF_DETECT_RENAME;
1467        opts.rename_limit = o->merge_rename_limit >= 0 ? o->merge_rename_limit :
1468                            o->diff_rename_limit >= 0 ? o->diff_rename_limit :
1469                            1000;
1470        opts.rename_score = o->rename_score;
1471        opts.show_rename_progress = o->show_rename_progress;
1472        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1473        diff_setup_done(&opts);
1474        diff_tree_oid(&o_tree->object.oid, &tree->object.oid, "", &opts);
1475        diffcore_std(&opts);
1476        if (opts.needed_rename_limit > o->needed_rename_limit)
1477                o->needed_rename_limit = opts.needed_rename_limit;
1478
1479        ret = xmalloc(sizeof(*ret));
1480        *ret = diff_queued_diff;
1481
1482        opts.output_format = DIFF_FORMAT_NO_OUTPUT;
1483        diff_queued_diff.nr = 0;
1484        diff_queued_diff.queue = NULL;
1485        diff_flush(&opts);
1486        return ret;
1487}
1488
1489static int tree_has_path(struct tree *tree, const char *path)
1490{
1491        struct object_id hashy;
1492        unsigned int mode_o;
1493
1494        return !get_tree_entry(&tree->object.oid, path,
1495                               &hashy, &mode_o);
1496}
1497
1498/*
1499 * Return a new string that replaces the beginning portion (which matches
1500 * entry->dir), with entry->new_dir.  In perl-speak:
1501 *   new_path_name = (old_path =~ s/entry->dir/entry->new_dir/);
1502 * NOTE:
1503 *   Caller must ensure that old_path starts with entry->dir + '/'.
1504 */
1505static char *apply_dir_rename(struct dir_rename_entry *entry,
1506                              const char *old_path)
1507{
1508        struct strbuf new_path = STRBUF_INIT;
1509        int oldlen, newlen;
1510
1511        if (entry->non_unique_new_dir)
1512                return NULL;
1513
1514        oldlen = strlen(entry->dir);
1515        newlen = entry->new_dir.len + (strlen(old_path) - oldlen) + 1;
1516        strbuf_grow(&new_path, newlen);
1517        strbuf_addbuf(&new_path, &entry->new_dir);
1518        strbuf_addstr(&new_path, &old_path[oldlen]);
1519
1520        return strbuf_detach(&new_path, NULL);
1521}
1522
1523static void get_renamed_dir_portion(const char *old_path, const char *new_path,
1524                                    char **old_dir, char **new_dir)
1525{
1526        char *end_of_old, *end_of_new;
1527        int old_len, new_len;
1528
1529        *old_dir = NULL;
1530        *new_dir = NULL;
1531
1532        /*
1533         * For
1534         *    "a/b/c/d/e/foo.c" -> "a/b/some/thing/else/e/foo.c"
1535         * the "e/foo.c" part is the same, we just want to know that
1536         *    "a/b/c/d" was renamed to "a/b/some/thing/else"
1537         * so, for this example, this function returns "a/b/c/d" in
1538         * *old_dir and "a/b/some/thing/else" in *new_dir.
1539         *
1540         * Also, if the basename of the file changed, we don't care.  We
1541         * want to know which portion of the directory, if any, changed.
1542         */
1543        end_of_old = strrchr(old_path, '/');
1544        end_of_new = strrchr(new_path, '/');
1545
1546        if (end_of_old == NULL || end_of_new == NULL)
1547                return;
1548        while (*--end_of_new == *--end_of_old &&
1549               end_of_old != old_path &&
1550               end_of_new != new_path)
1551                ; /* Do nothing; all in the while loop */
1552        /*
1553         * We've found the first non-matching character in the directory
1554         * paths.  That means the current directory we were comparing
1555         * represents the rename.  Move end_of_old and end_of_new back
1556         * to the full directory name.
1557         */
1558        if (*end_of_old == '/')
1559                end_of_old++;
1560        if (*end_of_old != '/')
1561                end_of_new++;
1562        end_of_old = strchr(end_of_old, '/');
1563        end_of_new = strchr(end_of_new, '/');
1564
1565        /*
1566         * It may have been the case that old_path and new_path were the same
1567         * directory all along.  Don't claim a rename if they're the same.
1568         */
1569        old_len = end_of_old - old_path;
1570        new_len = end_of_new - new_path;
1571
1572        if (old_len != new_len || strncmp(old_path, new_path, old_len)) {
1573                *old_dir = xstrndup(old_path, old_len);
1574                *new_dir = xstrndup(new_path, new_len);
1575        }
1576}
1577
1578static void remove_hashmap_entries(struct hashmap *dir_renames,
1579                                   struct string_list *items_to_remove)
1580{
1581        int i;
1582        struct dir_rename_entry *entry;
1583
1584        for (i = 0; i < items_to_remove->nr; i++) {
1585                entry = items_to_remove->items[i].util;
1586                hashmap_remove(dir_renames, entry, NULL);
1587        }
1588        string_list_clear(items_to_remove, 0);
1589}
1590
1591/*
1592 * See if there is a directory rename for path, and if there are any file
1593 * level conflicts for the renamed location.  If there is a rename and
1594 * there are no conflicts, return the new name.  Otherwise, return NULL.
1595 */
1596static char *handle_path_level_conflicts(struct merge_options *o,
1597                                         const char *path,
1598                                         struct dir_rename_entry *entry,
1599                                         struct hashmap *collisions,
1600                                         struct tree *tree)
1601{
1602        char *new_path = NULL;
1603        struct collision_entry *collision_ent;
1604        int clean = 1;
1605        struct strbuf collision_paths = STRBUF_INIT;
1606
1607        /*
1608         * entry has the mapping of old directory name to new directory name
1609         * that we want to apply to path.
1610         */
1611        new_path = apply_dir_rename(entry, path);
1612
1613        if (!new_path) {
1614                /* This should only happen when entry->non_unique_new_dir set */
1615                if (!entry->non_unique_new_dir)
1616                        BUG("entry->non_unqiue_dir not set and !new_path");
1617                output(o, 1, _("CONFLICT (directory rename split): "
1618                               "Unclear where to place %s because directory "
1619                               "%s was renamed to multiple other directories, "
1620                               "with no destination getting a majority of the "
1621                               "files."),
1622                       path, entry->dir);
1623                clean = 0;
1624                return NULL;
1625        }
1626
1627        /*
1628         * The caller needs to have ensured that it has pre-populated
1629         * collisions with all paths that map to new_path.  Do a quick check
1630         * to ensure that's the case.
1631         */
1632        collision_ent = collision_find_entry(collisions, new_path);
1633        if (collision_ent == NULL)
1634                BUG("collision_ent is NULL");
1635
1636        /*
1637         * Check for one-sided add/add/.../add conflicts, i.e.
1638         * where implicit renames from the other side doing
1639         * directory rename(s) can affect this side of history
1640         * to put multiple paths into the same location.  Warn
1641         * and bail on directory renames for such paths.
1642         */
1643        if (collision_ent->reported_already) {
1644                clean = 0;
1645        } else if (tree_has_path(tree, new_path)) {
1646                collision_ent->reported_already = 1;
1647                strbuf_add_separated_string_list(&collision_paths, ", ",
1648                                                 &collision_ent->source_files);
1649                output(o, 1, _("CONFLICT (implicit dir rename): Existing "
1650                               "file/dir at %s in the way of implicit "
1651                               "directory rename(s) putting the following "
1652                               "path(s) there: %s."),
1653                       new_path, collision_paths.buf);
1654                clean = 0;
1655        } else if (collision_ent->source_files.nr > 1) {
1656                collision_ent->reported_already = 1;
1657                strbuf_add_separated_string_list(&collision_paths, ", ",
1658                                                 &collision_ent->source_files);
1659                output(o, 1, _("CONFLICT (implicit dir rename): Cannot map "
1660                               "more than one path to %s; implicit directory "
1661                               "renames tried to put these paths there: %s"),
1662                       new_path, collision_paths.buf);
1663                clean = 0;
1664        }
1665
1666        /* Free memory we no longer need */
1667        strbuf_release(&collision_paths);
1668        if (!clean && new_path) {
1669                free(new_path);
1670                return NULL;
1671        }
1672
1673        return new_path;
1674}
1675
1676/*
1677 * There are a couple things we want to do at the directory level:
1678 *   1. Check for both sides renaming to the same thing, in order to avoid
1679 *      implicit renaming of files that should be left in place.  (See
1680 *      testcase 6b in t6043 for details.)
1681 *   2. Prune directory renames if there are still files left in the
1682 *      the original directory.  These represent a partial directory rename,
1683 *      i.e. a rename where only some of the files within the directory
1684 *      were renamed elsewhere.  (Technically, this could be done earlier
1685 *      in get_directory_renames(), except that would prevent us from
1686 *      doing the previous check and thus failing testcase 6b.)
1687 *   3. Check for rename/rename(1to2) conflicts (at the directory level).
1688 *      In the future, we could potentially record this info as well and
1689 *      omit reporting rename/rename(1to2) conflicts for each path within
1690 *      the affected directories, thus cleaning up the merge output.
1691 *   NOTE: We do NOT check for rename/rename(2to1) conflicts at the
1692 *         directory level, because merging directories is fine.  If it
1693 *         causes conflicts for files within those merged directories, then
1694 *         that should be detected at the individual path level.
1695 */
1696static void handle_directory_level_conflicts(struct merge_options *o,
1697                                             struct hashmap *dir_re_head,
1698                                             struct tree *head,
1699                                             struct hashmap *dir_re_merge,
1700                                             struct tree *merge)
1701{
1702        struct hashmap_iter iter;
1703        struct dir_rename_entry *head_ent;
1704        struct dir_rename_entry *merge_ent;
1705
1706        struct string_list remove_from_head = STRING_LIST_INIT_NODUP;
1707        struct string_list remove_from_merge = STRING_LIST_INIT_NODUP;
1708
1709        hashmap_iter_init(dir_re_head, &iter);
1710        while ((head_ent = hashmap_iter_next(&iter))) {
1711                merge_ent = dir_rename_find_entry(dir_re_merge, head_ent->dir);
1712                if (merge_ent &&
1713                    !head_ent->non_unique_new_dir &&
1714                    !merge_ent->non_unique_new_dir &&
1715                    !strbuf_cmp(&head_ent->new_dir, &merge_ent->new_dir)) {
1716                        /* 1. Renamed identically; remove it from both sides */
1717                        string_list_append(&remove_from_head,
1718                                           head_ent->dir)->util = head_ent;
1719                        strbuf_release(&head_ent->new_dir);
1720                        string_list_append(&remove_from_merge,
1721                                           merge_ent->dir)->util = merge_ent;
1722                        strbuf_release(&merge_ent->new_dir);
1723                } else if (tree_has_path(head, head_ent->dir)) {
1724                        /* 2. This wasn't a directory rename after all */
1725                        string_list_append(&remove_from_head,
1726                                           head_ent->dir)->util = head_ent;
1727                        strbuf_release(&head_ent->new_dir);
1728                }
1729        }
1730
1731        remove_hashmap_entries(dir_re_head, &remove_from_head);
1732        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1733
1734        hashmap_iter_init(dir_re_merge, &iter);
1735        while ((merge_ent = hashmap_iter_next(&iter))) {
1736                head_ent = dir_rename_find_entry(dir_re_head, merge_ent->dir);
1737                if (tree_has_path(merge, merge_ent->dir)) {
1738                        /* 2. This wasn't a directory rename after all */
1739                        string_list_append(&remove_from_merge,
1740                                           merge_ent->dir)->util = merge_ent;
1741                } else if (head_ent &&
1742                           !head_ent->non_unique_new_dir &&
1743                           !merge_ent->non_unique_new_dir) {
1744                        /* 3. rename/rename(1to2) */
1745                        /*
1746                         * We can assume it's not rename/rename(1to1) because
1747                         * that was case (1), already checked above.  So we
1748                         * know that head_ent->new_dir and merge_ent->new_dir
1749                         * are different strings.
1750                         */
1751                        output(o, 1, _("CONFLICT (rename/rename): "
1752                                       "Rename directory %s->%s in %s. "
1753                                       "Rename directory %s->%s in %s"),
1754                               head_ent->dir, head_ent->new_dir.buf, o->branch1,
1755                               head_ent->dir, merge_ent->new_dir.buf, o->branch2);
1756                        string_list_append(&remove_from_head,
1757                                           head_ent->dir)->util = head_ent;
1758                        strbuf_release(&head_ent->new_dir);
1759                        string_list_append(&remove_from_merge,
1760                                           merge_ent->dir)->util = merge_ent;
1761                        strbuf_release(&merge_ent->new_dir);
1762                }
1763        }
1764
1765        remove_hashmap_entries(dir_re_head, &remove_from_head);
1766        remove_hashmap_entries(dir_re_merge, &remove_from_merge);
1767}
1768
1769static struct hashmap *get_directory_renames(struct diff_queue_struct *pairs,
1770                                             struct tree *tree)
1771{
1772        struct hashmap *dir_renames;
1773        struct hashmap_iter iter;
1774        struct dir_rename_entry *entry;
1775        int i;
1776
1777        /*
1778         * Typically, we think of a directory rename as all files from a
1779         * certain directory being moved to a target directory.  However,
1780         * what if someone first moved two files from the original
1781         * directory in one commit, and then renamed the directory
1782         * somewhere else in a later commit?  At merge time, we just know
1783         * that files from the original directory went to two different
1784         * places, and that the bulk of them ended up in the same place.
1785         * We want each directory rename to represent where the bulk of the
1786         * files from that directory end up; this function exists to find
1787         * where the bulk of the files went.
1788         *
1789         * The first loop below simply iterates through the list of file
1790         * renames, finding out how often each directory rename pair
1791         * possibility occurs.
1792         */
1793        dir_renames = xmalloc(sizeof(*dir_renames));
1794        dir_rename_init(dir_renames);
1795        for (i = 0; i < pairs->nr; ++i) {
1796                struct string_list_item *item;
1797                int *count;
1798                struct diff_filepair *pair = pairs->queue[i];
1799                char *old_dir, *new_dir;
1800
1801                /* File not part of directory rename if it wasn't renamed */
1802                if (pair->status != 'R')
1803                        continue;
1804
1805                get_renamed_dir_portion(pair->one->path, pair->two->path,
1806                                        &old_dir,        &new_dir);
1807                if (!old_dir)
1808                        /* Directory didn't change at all; ignore this one. */
1809                        continue;
1810
1811                entry = dir_rename_find_entry(dir_renames, old_dir);
1812                if (!entry) {
1813                        entry = xmalloc(sizeof(*entry));
1814                        dir_rename_entry_init(entry, old_dir);
1815                        hashmap_put(dir_renames, entry);
1816                } else {
1817                        free(old_dir);
1818                }
1819                item = string_list_lookup(&entry->possible_new_dirs, new_dir);
1820                if (!item) {
1821                        item = string_list_insert(&entry->possible_new_dirs,
1822                                                  new_dir);
1823                        item->util = xcalloc(1, sizeof(int));
1824                } else {
1825                        free(new_dir);
1826                }
1827                count = item->util;
1828                *count += 1;
1829        }
1830
1831        /*
1832         * For each directory with files moved out of it, we find out which
1833         * target directory received the most files so we can declare it to
1834         * be the "winning" target location for the directory rename.  This
1835         * winner gets recorded in new_dir.  If there is no winner
1836         * (multiple target directories received the same number of files),
1837         * we set non_unique_new_dir.  Once we've determined the winner (or
1838         * that there is no winner), we no longer need possible_new_dirs.
1839         */
1840        hashmap_iter_init(dir_renames, &iter);
1841        while ((entry = hashmap_iter_next(&iter))) {
1842                int max = 0;
1843                int bad_max = 0;
1844                char *best = NULL;
1845
1846                for (i = 0; i < entry->possible_new_dirs.nr; i++) {
1847                        int *count = entry->possible_new_dirs.items[i].util;
1848
1849                        if (*count == max)
1850                                bad_max = max;
1851                        else if (*count > max) {
1852                                max = *count;
1853                                best = entry->possible_new_dirs.items[i].string;
1854                        }
1855                }
1856                if (bad_max == max)
1857                        entry->non_unique_new_dir = 1;
1858                else {
1859                        assert(entry->new_dir.len == 0);
1860                        strbuf_addstr(&entry->new_dir, best);
1861                }
1862                /*
1863                 * The relevant directory sub-portion of the original full
1864                 * filepaths were xstrndup'ed before inserting into
1865                 * possible_new_dirs, and instead of manually iterating the
1866                 * list and free'ing each, just lie and tell
1867                 * possible_new_dirs that it did the strdup'ing so that it
1868                 * will free them for us.
1869                 */
1870                entry->possible_new_dirs.strdup_strings = 1;
1871                string_list_clear(&entry->possible_new_dirs, 1);
1872        }
1873
1874        return dir_renames;
1875}
1876
1877static struct dir_rename_entry *check_dir_renamed(const char *path,
1878                                                  struct hashmap *dir_renames)
1879{
1880        char temp[PATH_MAX];
1881        char *end;
1882        struct dir_rename_entry *entry;
1883
1884        strcpy(temp, path);
1885        while ((end = strrchr(temp, '/'))) {
1886                *end = '\0';
1887                entry = dir_rename_find_entry(dir_renames, temp);
1888                if (entry)
1889                        return entry;
1890        }
1891        return NULL;
1892}
1893
1894static void compute_collisions(struct hashmap *collisions,
1895                               struct hashmap *dir_renames,
1896                               struct diff_queue_struct *pairs)
1897{
1898        int i;
1899
1900        /*
1901         * Multiple files can be mapped to the same path due to directory
1902         * renames done by the other side of history.  Since that other
1903         * side of history could have merged multiple directories into one,
1904         * if our side of history added the same file basename to each of
1905         * those directories, then all N of them would get implicitly
1906         * renamed by the directory rename detection into the same path,
1907         * and we'd get an add/add/.../add conflict, and all those adds
1908         * from *this* side of history.  This is not representable in the
1909         * index, and users aren't going to easily be able to make sense of
1910         * it.  So we need to provide a good warning about what's
1911         * happening, and fall back to no-directory-rename detection
1912         * behavior for those paths.
1913         *
1914         * See testcases 9e and all of section 5 from t6043 for examples.
1915         */
1916        collision_init(collisions);
1917
1918        for (i = 0; i < pairs->nr; ++i) {
1919                struct dir_rename_entry *dir_rename_ent;
1920                struct collision_entry *collision_ent;
1921                char *new_path;
1922                struct diff_filepair *pair = pairs->queue[i];
1923
1924                if (pair->status == 'D')
1925                        continue;
1926                dir_rename_ent = check_dir_renamed(pair->two->path,
1927                                                   dir_renames);
1928                if (!dir_rename_ent)
1929                        continue;
1930
1931                new_path = apply_dir_rename(dir_rename_ent, pair->two->path);
1932                if (!new_path)
1933                        /*
1934                         * dir_rename_ent->non_unique_new_path is true, which
1935                         * means there is no directory rename for us to use,
1936                         * which means it won't cause us any additional
1937                         * collisions.
1938                         */
1939                        continue;
1940                collision_ent = collision_find_entry(collisions, new_path);
1941                if (!collision_ent) {
1942                        collision_ent = xcalloc(1,
1943                                                sizeof(struct collision_entry));
1944                        hashmap_entry_init(collision_ent, strhash(new_path));
1945                        hashmap_put(collisions, collision_ent);
1946                        collision_ent->target_file = new_path;
1947                } else {
1948                        free(new_path);
1949                }
1950                string_list_insert(&collision_ent->source_files,
1951                                   pair->two->path);
1952        }
1953}
1954
1955static char *check_for_directory_rename(struct merge_options *o,
1956                                        const char *path,
1957                                        struct tree *tree,
1958                                        struct hashmap *dir_renames,
1959                                        struct hashmap *dir_rename_exclusions,
1960                                        struct hashmap *collisions,
1961                                        int *clean_merge)
1962{
1963        char *new_path = NULL;
1964        struct dir_rename_entry *entry = check_dir_renamed(path, dir_renames);
1965        struct dir_rename_entry *oentry = NULL;
1966
1967        if (!entry)
1968                return new_path;
1969
1970        /*
1971         * This next part is a little weird.  We do not want to do an
1972         * implicit rename into a directory we renamed on our side, because
1973         * that will result in a spurious rename/rename(1to2) conflict.  An
1974         * example:
1975         *   Base commit: dumbdir/afile, otherdir/bfile
1976         *   Side 1:      smrtdir/afile, otherdir/bfile
1977         *   Side 2:      dumbdir/afile, dumbdir/bfile
1978         * Here, while working on Side 1, we could notice that otherdir was
1979         * renamed/merged to dumbdir, and change the diff_filepair for
1980         * otherdir/bfile into a rename into dumbdir/bfile.  However, Side
1981         * 2 will notice the rename from dumbdir to smrtdir, and do the
1982         * transitive rename to move it from dumbdir/bfile to
1983         * smrtdir/bfile.  That gives us bfile in dumbdir vs being in
1984         * smrtdir, a rename/rename(1to2) conflict.  We really just want
1985         * the file to end up in smrtdir.  And the way to achieve that is
1986         * to not let Side1 do the rename to dumbdir, since we know that is
1987         * the source of one of our directory renames.
1988         *
1989         * That's why oentry and dir_rename_exclusions is here.
1990         *
1991         * As it turns out, this also prevents N-way transient rename
1992         * confusion; See testcases 9c and 9d of t6043.
1993         */
1994        oentry = dir_rename_find_entry(dir_rename_exclusions, entry->new_dir.buf);
1995        if (oentry) {
1996                output(o, 1, _("WARNING: Avoiding applying %s -> %s rename "
1997                               "to %s, because %s itself was renamed."),
1998                       entry->dir, entry->new_dir.buf, path, entry->new_dir.buf);
1999        } else {
2000                new_path = handle_path_level_conflicts(o, path, entry,
2001                                                       collisions, tree);
2002                *clean_merge &= (new_path != NULL);
2003        }
2004
2005        return new_path;
2006}
2007
2008static void apply_directory_rename_modifications(struct merge_options *o,
2009                                                 struct diff_filepair *pair,
2010                                                 char *new_path,
2011                                                 struct rename *re,
2012                                                 struct tree *tree,
2013                                                 struct tree *o_tree,
2014                                                 struct tree *a_tree,
2015                                                 struct tree *b_tree,
2016                                                 struct string_list *entries,
2017                                                 int *clean)
2018{
2019        struct string_list_item *item;
2020        int stage = (tree == a_tree ? 2 : 3);
2021
2022        /*
2023         * In all cases where we can do directory rename detection,
2024         * unpack_trees() will have read pair->two->path into the
2025         * index and the working copy.  We need to remove it so that
2026         * we can instead place it at new_path.  It is guaranteed to
2027         * not be untracked (unpack_trees() would have errored out
2028         * saying the file would have been overwritten), but it might
2029         * be dirty, though.
2030         */
2031        remove_file(o, 1, pair->two->path, 0 /* no_wd */);
2032
2033        /* Find or create a new re->dst_entry */
2034        item = string_list_lookup(entries, new_path);
2035        if (item) {
2036                /*
2037                 * Since we're renaming on this side of history, and it's
2038                 * due to a directory rename on the other side of history
2039                 * (which we only allow when the directory in question no
2040                 * longer exists on the other side of history), the
2041                 * original entry for re->dst_entry is no longer
2042                 * necessary...
2043                 */
2044                re->dst_entry->processed = 1;
2045
2046                /*
2047                 * ...because we'll be using this new one.
2048                 */
2049                re->dst_entry = item->util;
2050        } else {
2051                /*
2052                 * re->dst_entry is for the before-dir-rename path, and we
2053                 * need it to hold information for the after-dir-rename
2054                 * path.  Before creating a new entry, we need to mark the
2055                 * old one as unnecessary (...unless it is shared by
2056                 * src_entry, i.e. this didn't use to be a rename, in which
2057                 * case we can just allow the normal processing to happen
2058                 * for it).
2059                 */
2060                if (pair->status == 'R')
2061                        re->dst_entry->processed = 1;
2062
2063                re->dst_entry = insert_stage_data(new_path,
2064                                                  o_tree, a_tree, b_tree,
2065                                                  entries);
2066                item = string_list_insert(entries, new_path);
2067                item->util = re->dst_entry;
2068        }
2069
2070        /*
2071         * Update the stage_data with the information about the path we are
2072         * moving into place.  That slot will be empty and available for us
2073         * to write to because of the collision checks in
2074         * handle_path_level_conflicts().  In other words,
2075         * re->dst_entry->stages[stage].oid will be the null_oid, so it's
2076         * open for us to write to.
2077         *
2078         * It may be tempting to actually update the index at this point as
2079         * well, using update_stages_for_stage_data(), but as per the big
2080         * "NOTE" in update_stages(), doing so will modify the current
2081         * in-memory index which will break calls to would_lose_untracked()
2082         * that we need to make.  Instead, we need to just make sure that
2083         * the various conflict_rename_*() functions update the index
2084         * explicitly rather than relying on unpack_trees() to have done it.
2085         */
2086        get_tree_entry(&tree->object.oid,
2087                       pair->two->path,
2088                       &re->dst_entry->stages[stage].oid,
2089                       &re->dst_entry->stages[stage].mode);
2090
2091        /* Update pair status */
2092        if (pair->status == 'A') {
2093                /*
2094                 * Recording rename information for this add makes it look
2095                 * like a rename/delete conflict.  Make sure we can
2096                 * correctly handle this as an add that was moved to a new
2097                 * directory instead of reporting a rename/delete conflict.
2098                 */
2099                re->add_turned_into_rename = 1;
2100        }
2101        /*
2102         * We don't actually look at pair->status again, but it seems
2103         * pedagogically correct to adjust it.
2104         */
2105        pair->status = 'R';
2106
2107        /*
2108         * Finally, record the new location.
2109         */
2110        pair->two->path = new_path;
2111}
2112
2113/*
2114 * Get information of all renames which occurred in 'pairs', making use of
2115 * any implicit directory renames inferred from the other side of history.
2116 * We need the three trees in the merge ('o_tree', 'a_tree' and 'b_tree')
2117 * to be able to associate the correct cache entries with the rename
2118 * information; tree is always equal to either a_tree or b_tree.
2119 */
2120static struct string_list *get_renames(struct merge_options *o,
2121                                       struct diff_queue_struct *pairs,
2122                                       struct hashmap *dir_renames,
2123                                       struct hashmap *dir_rename_exclusions,
2124                                       struct tree *tree,
2125                                       struct tree *o_tree,
2126                                       struct tree *a_tree,
2127                                       struct tree *b_tree,
2128                                       struct string_list *entries,
2129                                       int *clean_merge)
2130{
2131        int i;
2132        struct hashmap collisions;
2133        struct hashmap_iter iter;
2134        struct collision_entry *e;
2135        struct string_list *renames;
2136
2137        compute_collisions(&collisions, dir_renames, pairs);
2138        renames = xcalloc(1, sizeof(struct string_list));
2139
2140        for (i = 0; i < pairs->nr; ++i) {
2141                struct string_list_item *item;
2142                struct rename *re;
2143                struct diff_filepair *pair = pairs->queue[i];
2144                char *new_path; /* non-NULL only with directory renames */
2145
2146                if (pair->status == 'D') {
2147                        diff_free_filepair(pair);
2148                        continue;
2149                }
2150                new_path = check_for_directory_rename(o, pair->two->path, tree,
2151                                                      dir_renames,
2152                                                      dir_rename_exclusions,
2153                                                      &collisions,
2154                                                      clean_merge);
2155                if (pair->status != 'R' && !new_path) {
2156                        diff_free_filepair(pair);
2157                        continue;
2158                }
2159
2160                re = xmalloc(sizeof(*re));
2161                re->processed = 0;
2162                re->add_turned_into_rename = 0;
2163                re->pair = pair;
2164                item = string_list_lookup(entries, re->pair->one->path);
2165                if (!item)
2166                        re->src_entry = insert_stage_data(re->pair->one->path,
2167                                        o_tree, a_tree, b_tree, entries);
2168                else
2169                        re->src_entry = item->util;
2170
2171                item = string_list_lookup(entries, re->pair->two->path);
2172                if (!item)
2173                        re->dst_entry = insert_stage_data(re->pair->two->path,
2174                                        o_tree, a_tree, b_tree, entries);
2175                else
2176                        re->dst_entry = item->util;
2177                item = string_list_insert(renames, pair->one->path);
2178                item->util = re;
2179                if (new_path)
2180                        apply_directory_rename_modifications(o, pair, new_path,
2181                                                             re, tree, o_tree,
2182                                                             a_tree, b_tree,
2183                                                             entries,
2184                                                             clean_merge);
2185        }
2186
2187        hashmap_iter_init(&collisions, &iter);
2188        while ((e = hashmap_iter_next(&iter))) {
2189                free(e->target_file);
2190                string_list_clear(&e->source_files, 0);
2191        }
2192        hashmap_free(&collisions, 1);
2193        return renames;
2194}
2195
2196static int process_renames(struct merge_options *o,
2197                           struct string_list *a_renames,
2198                           struct string_list *b_renames)
2199{
2200        int clean_merge = 1, i, j;
2201        struct string_list a_by_dst = STRING_LIST_INIT_NODUP;
2202        struct string_list b_by_dst = STRING_LIST_INIT_NODUP;
2203        const struct rename *sre;
2204
2205        for (i = 0; i < a_renames->nr; i++) {
2206                sre = a_renames->items[i].util;
2207                string_list_insert(&a_by_dst, sre->pair->two->path)->util
2208                        = (void *)sre;
2209        }
2210        for (i = 0; i < b_renames->nr; i++) {
2211                sre = b_renames->items[i].util;
2212                string_list_insert(&b_by_dst, sre->pair->two->path)->util
2213                        = (void *)sre;
2214        }
2215
2216        for (i = 0, j = 0; i < a_renames->nr || j < b_renames->nr;) {
2217                struct string_list *renames1, *renames2Dst;
2218                struct rename *ren1 = NULL, *ren2 = NULL;
2219                const char *branch1, *branch2;
2220                const char *ren1_src, *ren1_dst;
2221                struct string_list_item *lookup;
2222
2223                if (i >= a_renames->nr) {
2224                        ren2 = b_renames->items[j++].util;
2225                } else if (j >= b_renames->nr) {
2226                        ren1 = a_renames->items[i++].util;
2227                } else {
2228                        int compare = strcmp(a_renames->items[i].string,
2229                                             b_renames->items[j].string);
2230                        if (compare <= 0)
2231                                ren1 = a_renames->items[i++].util;
2232                        if (compare >= 0)
2233                                ren2 = b_renames->items[j++].util;
2234                }
2235
2236                /* TODO: refactor, so that 1/2 are not needed */
2237                if (ren1) {
2238                        renames1 = a_renames;
2239                        renames2Dst = &b_by_dst;
2240                        branch1 = o->branch1;
2241                        branch2 = o->branch2;
2242                } else {
2243                        renames1 = b_renames;
2244                        renames2Dst = &a_by_dst;
2245                        branch1 = o->branch2;
2246                        branch2 = o->branch1;
2247                        SWAP(ren2, ren1);
2248                }
2249
2250                if (ren1->processed)
2251                        continue;
2252                ren1->processed = 1;
2253                ren1->dst_entry->processed = 1;
2254                /* BUG: We should only mark src_entry as processed if we
2255                 * are not dealing with a rename + add-source case.
2256                 */
2257                ren1->src_entry->processed = 1;
2258
2259                ren1_src = ren1->pair->one->path;
2260                ren1_dst = ren1->pair->two->path;
2261
2262                if (ren2) {
2263                        /* One file renamed on both sides */
2264                        const char *ren2_src = ren2->pair->one->path;
2265                        const char *ren2_dst = ren2->pair->two->path;
2266                        enum rename_type rename_type;
2267                        if (strcmp(ren1_src, ren2_src) != 0)
2268                                die("BUG: ren1_src != ren2_src");
2269                        ren2->dst_entry->processed = 1;
2270                        ren2->processed = 1;
2271                        if (strcmp(ren1_dst, ren2_dst) != 0) {
2272                                rename_type = RENAME_ONE_FILE_TO_TWO;
2273                                clean_merge = 0;
2274                        } else {
2275                                rename_type = RENAME_ONE_FILE_TO_ONE;
2276                                /* BUG: We should only remove ren1_src in
2277                                 * the base stage (think of rename +
2278                                 * add-source cases).
2279                                 */
2280                                remove_file(o, 1, ren1_src, 1);
2281                                update_entry(ren1->dst_entry,
2282                                             ren1->pair->one,
2283                                             ren1->pair->two,
2284                                             ren2->pair->two);
2285                        }
2286                        setup_rename_conflict_info(rename_type,
2287                                                   ren1->pair,
2288                                                   ren2->pair,
2289                                                   branch1,
2290                                                   branch2,
2291                                                   ren1->dst_entry,
2292                                                   ren2->dst_entry,
2293                                                   o,
2294                                                   NULL,
2295                                                   NULL);
2296                } else if ((lookup = string_list_lookup(renames2Dst, ren1_dst))) {
2297                        /* Two different files renamed to the same thing */
2298                        char *ren2_dst;
2299                        ren2 = lookup->util;
2300                        ren2_dst = ren2->pair->two->path;
2301                        if (strcmp(ren1_dst, ren2_dst) != 0)
2302                                die("BUG: ren1_dst != ren2_dst");
2303
2304                        clean_merge = 0;
2305                        ren2->processed = 1;
2306                        /*
2307                         * BUG: We should only mark src_entry as processed
2308                         * if we are not dealing with a rename + add-source
2309                         * case.
2310                         */
2311                        ren2->src_entry->processed = 1;
2312
2313                        setup_rename_conflict_info(RENAME_TWO_FILES_TO_ONE,
2314                                                   ren1->pair,
2315                                                   ren2->pair,
2316                                                   branch1,
2317                                                   branch2,
2318                                                   ren1->dst_entry,
2319                                                   ren2->dst_entry,
2320                                                   o,
2321                                                   ren1->src_entry,
2322                                                   ren2->src_entry);
2323
2324                } else {
2325                        /* Renamed in 1, maybe changed in 2 */
2326                        /* we only use sha1 and mode of these */
2327                        struct diff_filespec src_other, dst_other;
2328                        int try_merge;
2329
2330                        /*
2331                         * unpack_trees loads entries from common-commit
2332                         * into stage 1, from head-commit into stage 2, and
2333                         * from merge-commit into stage 3.  We keep track
2334                         * of which side corresponds to the rename.
2335                         */
2336                        int renamed_stage = a_renames == renames1 ? 2 : 3;
2337                        int other_stage =   a_renames == renames1 ? 3 : 2;
2338
2339                        /* BUG: We should only remove ren1_src in the base
2340                         * stage and in other_stage (think of rename +
2341                         * add-source case).
2342                         */
2343                        remove_file(o, 1, ren1_src,
2344                                    renamed_stage == 2 || !was_tracked(ren1_src));
2345
2346                        oidcpy(&src_other.oid,
2347                               &ren1->src_entry->stages[other_stage].oid);
2348                        src_other.mode = ren1->src_entry->stages[other_stage].mode;
2349                        oidcpy(&dst_other.oid,
2350                               &ren1->dst_entry->stages[other_stage].oid);
2351                        dst_other.mode = ren1->dst_entry->stages[other_stage].mode;
2352                        try_merge = 0;
2353
2354                        if (oid_eq(&src_other.oid, &null_oid) &&
2355                            ren1->add_turned_into_rename) {
2356                                setup_rename_conflict_info(RENAME_DIR,
2357                                                           ren1->pair,
2358                                                           NULL,
2359                                                           branch1,
2360                                                           branch2,
2361                                                           ren1->dst_entry,
2362                                                           NULL,
2363                                                           o,
2364                                                           NULL,
2365                                                           NULL);
2366                        } else if (oid_eq(&src_other.oid, &null_oid)) {
2367                                setup_rename_conflict_info(RENAME_DELETE,
2368                                                           ren1->pair,
2369                                                           NULL,
2370                                                           branch1,
2371                                                           branch2,
2372                                                           ren1->dst_entry,
2373                                                           NULL,
2374                                                           o,
2375                                                           NULL,
2376                                                           NULL);
2377                        } else if ((dst_other.mode == ren1->pair->two->mode) &&
2378                                   oid_eq(&dst_other.oid, &ren1->pair->two->oid)) {
2379                                /*
2380                                 * Added file on the other side identical to
2381                                 * the file being renamed: clean merge.
2382                                 * Also, there is no need to overwrite the
2383                                 * file already in the working copy, so call
2384                                 * update_file_flags() instead of
2385                                 * update_file().
2386                                 */
2387                                if (update_file_flags(o,
2388                                                      &ren1->pair->two->oid,
2389                                                      ren1->pair->two->mode,
2390                                                      ren1_dst,
2391                                                      1, /* update_cache */
2392                                                      0  /* update_wd    */))
2393                                        clean_merge = -1;
2394                        } else if (!oid_eq(&dst_other.oid, &null_oid)) {
2395                                clean_merge = 0;
2396                                try_merge = 1;
2397                                output(o, 1, _("CONFLICT (rename/add): Rename %s->%s in %s. "
2398                                       "%s added in %s"),
2399                                       ren1_src, ren1_dst, branch1,
2400                                       ren1_dst, branch2);
2401                                if (o->call_depth) {
2402                                        struct merge_file_info mfi;
2403                                        if (merge_file_one(o, ren1_dst, &null_oid, 0,
2404                                                           &ren1->pair->two->oid,
2405                                                           ren1->pair->two->mode,
2406                                                           &dst_other.oid,
2407                                                           dst_other.mode,
2408                                                           branch1, branch2, &mfi)) {
2409                                                clean_merge = -1;
2410                                                goto cleanup_and_return;
2411                                        }
2412                                        output(o, 1, _("Adding merged %s"), ren1_dst);
2413                                        if (update_file(o, 0, &mfi.oid,
2414                                                        mfi.mode, ren1_dst))
2415                                                clean_merge = -1;
2416                                        try_merge = 0;
2417                                } else {
2418                                        char *new_path = unique_path(o, ren1_dst, branch2);
2419                                        output(o, 1, _("Adding as %s instead"), new_path);
2420                                        if (update_file(o, 0, &dst_other.oid,
2421                                                        dst_other.mode, new_path))
2422                                                clean_merge = -1;
2423                                        free(new_path);
2424                                }
2425                        } else
2426                                try_merge = 1;
2427
2428                        if (clean_merge < 0)
2429                                goto cleanup_and_return;
2430                        if (try_merge) {
2431                                struct diff_filespec *one, *a, *b;
2432                                src_other.path = (char *)ren1_src;
2433
2434                                one = ren1->pair->one;
2435                                if (a_renames == renames1) {
2436                                        a = ren1->pair->two;
2437                                        b = &src_other;
2438                                } else {
2439                                        b = ren1->pair->two;
2440                                        a = &src_other;
2441                                }
2442                                update_entry(ren1->dst_entry, one, a, b);
2443                                setup_rename_conflict_info(RENAME_NORMAL,
2444                                                           ren1->pair,
2445                                                           NULL,
2446                                                           branch1,
2447                                                           NULL,
2448                                                           ren1->dst_entry,
2449                                                           NULL,
2450                                                           o,
2451                                                           NULL,
2452                                                           NULL);
2453                        }
2454                }
2455        }
2456cleanup_and_return:
2457        string_list_clear(&a_by_dst, 0);
2458        string_list_clear(&b_by_dst, 0);
2459
2460        return clean_merge;
2461}
2462
2463struct rename_info {
2464        struct string_list *head_renames;
2465        struct string_list *merge_renames;
2466};
2467
2468static void initial_cleanup_rename(struct diff_queue_struct *pairs,
2469                                   struct hashmap *dir_renames)
2470{
2471        struct hashmap_iter iter;
2472        struct dir_rename_entry *e;
2473
2474        hashmap_iter_init(dir_renames, &iter);
2475        while ((e = hashmap_iter_next(&iter))) {
2476                free(e->dir);
2477                strbuf_release(&e->new_dir);
2478                /* possible_new_dirs already cleared in get_directory_renames */
2479        }
2480        hashmap_free(dir_renames, 1);
2481        free(dir_renames);
2482
2483        free(pairs->queue);
2484        free(pairs);
2485}
2486
2487static int handle_renames(struct merge_options *o,
2488                          struct tree *common,
2489                          struct tree *head,
2490                          struct tree *merge,
2491                          struct string_list *entries,
2492                          struct rename_info *ri)
2493{
2494        struct diff_queue_struct *head_pairs, *merge_pairs;
2495        struct hashmap *dir_re_head, *dir_re_merge;
2496        int clean = 1;
2497
2498        ri->head_renames = NULL;
2499        ri->merge_renames = NULL;
2500
2501        if (!o->detect_rename)
2502                return 1;
2503
2504        head_pairs = get_diffpairs(o, common, head);
2505        merge_pairs = get_diffpairs(o, common, merge);
2506
2507        dir_re_head = get_directory_renames(head_pairs, head);
2508        dir_re_merge = get_directory_renames(merge_pairs, merge);
2509
2510        handle_directory_level_conflicts(o,
2511                                         dir_re_head, head,
2512                                         dir_re_merge, merge);
2513
2514        ri->head_renames  = get_renames(o, head_pairs,
2515                                        dir_re_merge, dir_re_head, head,
2516                                        common, head, merge, entries,
2517                                        &clean);
2518        if (clean < 0)
2519                goto cleanup;
2520        ri->merge_renames = get_renames(o, merge_pairs,
2521                                        dir_re_head, dir_re_merge, merge,
2522                                        common, head, merge, entries,
2523                                        &clean);
2524        if (clean < 0)
2525                goto cleanup;
2526        clean &= process_renames(o, ri->head_renames, ri->merge_renames);
2527
2528cleanup:
2529        /*
2530         * Some cleanup is deferred until cleanup_renames() because the
2531         * data structures are still needed and referenced in
2532         * process_entry().  But there are a few things we can free now.
2533         */
2534        initial_cleanup_rename(head_pairs, dir_re_head);
2535        initial_cleanup_rename(merge_pairs, dir_re_merge);
2536
2537        return clean;
2538}
2539
2540static void final_cleanup_rename(struct string_list *rename)
2541{
2542        const struct rename *re;
2543        int i;
2544
2545        if (rename == NULL)
2546                return;
2547
2548        for (i = 0; i < rename->nr; i++) {
2549                re = rename->items[i].util;
2550                diff_free_filepair(re->pair);
2551        }
2552        string_list_clear(rename, 1);
2553        free(rename);
2554}
2555
2556static void final_cleanup_renames(struct rename_info *re_info)
2557{
2558        final_cleanup_rename(re_info->head_renames);
2559        final_cleanup_rename(re_info->merge_renames);
2560}
2561
2562static struct object_id *stage_oid(const struct object_id *oid, unsigned mode)
2563{
2564        return (is_null_oid(oid) || mode == 0) ? NULL: (struct object_id *)oid;
2565}
2566
2567static int read_oid_strbuf(struct merge_options *o,
2568        const struct object_id *oid, struct strbuf *dst)
2569{
2570        void *buf;
2571        enum object_type type;
2572        unsigned long size;
2573        buf = read_object_file(oid, &type, &size);
2574        if (!buf)
2575                return err(o, _("cannot read object %s"), oid_to_hex(oid));
2576        if (type != OBJ_BLOB) {
2577                free(buf);
2578                return err(o, _("object %s is not a blob"), oid_to_hex(oid));
2579        }
2580        strbuf_attach(dst, buf, size, size + 1);
2581        return 0;
2582}
2583
2584static int blob_unchanged(struct merge_options *opt,
2585                          const struct object_id *o_oid,
2586                          unsigned o_mode,
2587                          const struct object_id *a_oid,
2588                          unsigned a_mode,
2589                          int renormalize, const char *path)
2590{
2591        struct strbuf o = STRBUF_INIT;
2592        struct strbuf a = STRBUF_INIT;
2593        int ret = 0; /* assume changed for safety */
2594
2595        if (a_mode != o_mode)
2596                return 0;
2597        if (oid_eq(o_oid, a_oid))
2598                return 1;
2599        if (!renormalize)
2600                return 0;
2601
2602        assert(o_oid && a_oid);
2603        if (read_oid_strbuf(opt, o_oid, &o) || read_oid_strbuf(opt, a_oid, &a))
2604                goto error_return;
2605        /*
2606         * Note: binary | is used so that both renormalizations are
2607         * performed.  Comparison can be skipped if both files are
2608         * unchanged since their sha1s have already been compared.
2609         */
2610        if (renormalize_buffer(&the_index, path, o.buf, o.len, &o) |
2611            renormalize_buffer(&the_index, path, a.buf, a.len, &a))
2612                ret = (o.len == a.len && !memcmp(o.buf, a.buf, o.len));
2613
2614error_return:
2615        strbuf_release(&o);
2616        strbuf_release(&a);
2617        return ret;
2618}
2619
2620static int handle_modify_delete(struct merge_options *o,
2621                                 const char *path,
2622                                 struct object_id *o_oid, int o_mode,
2623                                 struct object_id *a_oid, int a_mode,
2624                                 struct object_id *b_oid, int b_mode)
2625{
2626        const char *modify_branch, *delete_branch;
2627        struct object_id *changed_oid;
2628        int changed_mode;
2629
2630        if (a_oid) {
2631                modify_branch = o->branch1;
2632                delete_branch = o->branch2;
2633                changed_oid = a_oid;
2634                changed_mode = a_mode;
2635        } else {
2636                modify_branch = o->branch2;
2637                delete_branch = o->branch1;
2638                changed_oid = b_oid;
2639                changed_mode = b_mode;
2640        }
2641
2642        return handle_change_delete(o,
2643                                    path, NULL,
2644                                    o_oid, o_mode,
2645                                    changed_oid, changed_mode,
2646                                    modify_branch, delete_branch,
2647                                    _("modify"), _("modified"));
2648}
2649
2650static int merge_content(struct merge_options *o,
2651                         const char *path,
2652                         struct object_id *o_oid, int o_mode,
2653                         struct object_id *a_oid, int a_mode,
2654                         struct object_id *b_oid, int b_mode,
2655                         struct rename_conflict_info *rename_conflict_info)
2656{
2657        const char *reason = _("content");
2658        const char *path1 = NULL, *path2 = NULL;
2659        struct merge_file_info mfi;
2660        struct diff_filespec one, a, b;
2661        unsigned df_conflict_remains = 0;
2662
2663        if (!o_oid) {
2664                reason = _("add/add");
2665                o_oid = (struct object_id *)&null_oid;
2666        }
2667        one.path = a.path = b.path = (char *)path;
2668        oidcpy(&one.oid, o_oid);
2669        one.mode = o_mode;
2670        oidcpy(&a.oid, a_oid);
2671        a.mode = a_mode;
2672        oidcpy(&b.oid, b_oid);
2673        b.mode = b_mode;
2674
2675        if (rename_conflict_info) {
2676                struct diff_filepair *pair1 = rename_conflict_info->pair1;
2677
2678                path1 = (o->branch1 == rename_conflict_info->branch1) ?
2679                        pair1->two->path : pair1->one->path;
2680                /* If rename_conflict_info->pair2 != NULL, we are in
2681                 * RENAME_ONE_FILE_TO_ONE case.  Otherwise, we have a
2682                 * normal rename.
2683                 */
2684                path2 = (rename_conflict_info->pair2 ||
2685                         o->branch2 == rename_conflict_info->branch1) ?
2686                        pair1->two->path : pair1->one->path;
2687
2688                if (dir_in_way(path, !o->call_depth,
2689                               S_ISGITLINK(pair1->two->mode)))
2690                        df_conflict_remains = 1;
2691        }
2692        if (merge_file_special_markers(o, &one, &a, &b,
2693                                       o->branch1, path1,
2694                                       o->branch2, path2, &mfi))
2695                return -1;
2696
2697        if (mfi.clean && !df_conflict_remains &&
2698            oid_eq(&mfi.oid, a_oid) && mfi.mode == a_mode) {
2699                int path_renamed_outside_HEAD;
2700                output(o, 3, _("Skipped %s (merged same as existing)"), path);
2701                /*
2702                 * The content merge resulted in the same file contents we
2703                 * already had.  We can return early if those file contents
2704                 * are recorded at the correct path (which may not be true
2705                 * if the merge involves a rename).
2706                 */
2707                path_renamed_outside_HEAD = !path2 || !strcmp(path, path2);
2708                if (!path_renamed_outside_HEAD) {
2709                        add_cacheinfo(o, mfi.mode, &mfi.oid, path,
2710                                      0, (!o->call_depth), 0);
2711                        return mfi.clean;
2712                }
2713        } else
2714                output(o, 2, _("Auto-merging %s"), path);
2715
2716        if (!mfi.clean) {
2717                if (S_ISGITLINK(mfi.mode))
2718                        reason = _("submodule");
2719                output(o, 1, _("CONFLICT (%s): Merge conflict in %s"),
2720                                reason, path);
2721                if (rename_conflict_info && !df_conflict_remains)
2722                        if (update_stages(o, path, &one, &a, &b))
2723                                return -1;
2724        }
2725
2726        if (df_conflict_remains) {
2727                char *new_path;
2728                if (o->call_depth) {
2729                        remove_file_from_cache(path);
2730                } else {
2731                        if (!mfi.clean) {
2732                                if (update_stages(o, path, &one, &a, &b))
2733                                        return -1;
2734                        } else {
2735                                int file_from_stage2 = was_tracked(path);
2736                                struct diff_filespec merged;
2737                                oidcpy(&merged.oid, &mfi.oid);
2738                                merged.mode = mfi.mode;
2739
2740                                if (update_stages(o, path, NULL,
2741                                                  file_from_stage2 ? &merged : NULL,
2742                                                  file_from_stage2 ? NULL : &merged))
2743                                        return -1;
2744                        }
2745
2746                }
2747                new_path = unique_path(o, path, rename_conflict_info->branch1);
2748                output(o, 1, _("Adding as %s instead"), new_path);
2749                if (update_file(o, 0, &mfi.oid, mfi.mode, new_path)) {
2750                        free(new_path);
2751                        return -1;
2752                }
2753                free(new_path);
2754                mfi.clean = 0;
2755        } else if (update_file(o, mfi.clean, &mfi.oid, mfi.mode, path))
2756                return -1;
2757        return mfi.clean;
2758}
2759
2760/* Per entry merge function */
2761static int process_entry(struct merge_options *o,
2762                         const char *path, struct stage_data *entry)
2763{
2764        int clean_merge = 1;
2765        int normalize = o->renormalize;
2766        unsigned o_mode = entry->stages[1].mode;
2767        unsigned a_mode = entry->stages[2].mode;
2768        unsigned b_mode = entry->stages[3].mode;
2769        struct object_id *o_oid = stage_oid(&entry->stages[1].oid, o_mode);
2770        struct object_id *a_oid = stage_oid(&entry->stages[2].oid, a_mode);
2771        struct object_id *b_oid = stage_oid(&entry->stages[3].oid, b_mode);
2772
2773        entry->processed = 1;
2774        if (entry->rename_conflict_info) {
2775                struct rename_conflict_info *conflict_info = entry->rename_conflict_info;
2776                switch (conflict_info->rename_type) {
2777                case RENAME_NORMAL:
2778                case RENAME_ONE_FILE_TO_ONE:
2779                        clean_merge = merge_content(o, path,
2780                                                    o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2781                                                    conflict_info);
2782                        break;
2783                case RENAME_DIR:
2784                        clean_merge = 1;
2785                        if (conflict_rename_dir(o,
2786                                                conflict_info->pair1,
2787                                                conflict_info->branch1,
2788                                                conflict_info->branch2))
2789                                clean_merge = -1;
2790                        break;
2791                case RENAME_DELETE:
2792                        clean_merge = 0;
2793                        if (conflict_rename_delete(o,
2794                                                   conflict_info->pair1,
2795                                                   conflict_info->branch1,
2796                                                   conflict_info->branch2))
2797                                clean_merge = -1;
2798                        break;
2799                case RENAME_ONE_FILE_TO_TWO:
2800                        clean_merge = 0;
2801                        if (conflict_rename_rename_1to2(o, conflict_info))
2802                                clean_merge = -1;
2803                        break;
2804                case RENAME_TWO_FILES_TO_ONE:
2805                        clean_merge = 0;
2806                        if (conflict_rename_rename_2to1(o, conflict_info))
2807                                clean_merge = -1;
2808                        break;
2809                default:
2810                        entry->processed = 0;
2811                        break;
2812                }
2813        } else if (o_oid && (!a_oid || !b_oid)) {
2814                /* Case A: Deleted in one */
2815                if ((!a_oid && !b_oid) ||
2816                    (!b_oid && blob_unchanged(o, o_oid, o_mode, a_oid, a_mode, normalize, path)) ||
2817                    (!a_oid && blob_unchanged(o, o_oid, o_mode, b_oid, b_mode, normalize, path))) {
2818                        /* Deleted in both or deleted in one and
2819                         * unchanged in the other */
2820                        if (a_oid)
2821                                output(o, 2, _("Removing %s"), path);
2822                        /* do not touch working file if it did not exist */
2823                        remove_file(o, 1, path, !a_oid);
2824                } else {
2825                        /* Modify/delete; deleted side may have put a directory in the way */
2826                        clean_merge = 0;
2827                        if (handle_modify_delete(o, path, o_oid, o_mode,
2828                                                 a_oid, a_mode, b_oid, b_mode))
2829                                clean_merge = -1;
2830                }
2831        } else if ((!o_oid && a_oid && !b_oid) ||
2832                   (!o_oid && !a_oid && b_oid)) {
2833                /* Case B: Added in one. */
2834                /* [nothing|directory] -> ([nothing|directory], file) */
2835
2836                const char *add_branch;
2837                const char *other_branch;
2838                unsigned mode;
2839                const struct object_id *oid;
2840                const char *conf;
2841
2842                if (a_oid) {
2843                        add_branch = o->branch1;
2844                        other_branch = o->branch2;
2845                        mode = a_mode;
2846                        oid = a_oid;
2847                        conf = _("file/directory");
2848                } else {
2849                        add_branch = o->branch2;
2850                        other_branch = o->branch1;
2851                        mode = b_mode;
2852                        oid = b_oid;
2853                        conf = _("directory/file");
2854                }
2855                if (dir_in_way(path,
2856                               !o->call_depth && !S_ISGITLINK(a_mode),
2857                               0)) {
2858                        char *new_path = unique_path(o, path, add_branch);
2859                        clean_merge = 0;
2860                        output(o, 1, _("CONFLICT (%s): There is a directory with name %s in %s. "
2861                               "Adding %s as %s"),
2862                               conf, path, other_branch, path, new_path);
2863                        if (update_file(o, 0, oid, mode, new_path))
2864                                clean_merge = -1;
2865                        else if (o->call_depth)
2866                                remove_file_from_cache(path);
2867                        free(new_path);
2868                } else {
2869                        output(o, 2, _("Adding %s"), path);
2870                        /* do not overwrite file if already present */
2871                        if (update_file_flags(o, oid, mode, path, 1, !a_oid))
2872                                clean_merge = -1;
2873                }
2874        } else if (a_oid && b_oid) {
2875                /* Case C: Added in both (check for same permissions) and */
2876                /* case D: Modified in both, but differently. */
2877                clean_merge = merge_content(o, path,
2878                                            o_oid, o_mode, a_oid, a_mode, b_oid, b_mode,
2879                                            NULL);
2880        } else if (!o_oid && !a_oid && !b_oid) {
2881                /*
2882                 * this entry was deleted altogether. a_mode == 0 means
2883                 * we had that path and want to actively remove it.
2884                 */
2885                remove_file(o, 1, path, !a_mode);
2886        } else
2887                die("BUG: fatal merge failure, shouldn't happen.");
2888
2889        return clean_merge;
2890}
2891
2892int merge_trees(struct merge_options *o,
2893                struct tree *head,
2894                struct tree *merge,
2895                struct tree *common,
2896                struct tree **result)
2897{
2898        int code, clean;
2899
2900        if (o->subtree_shift) {
2901                merge = shift_tree_object(head, merge, o->subtree_shift);
2902                common = shift_tree_object(head, common, o->subtree_shift);
2903        }
2904
2905        if (oid_eq(&common->object.oid, &merge->object.oid)) {
2906                struct strbuf sb = STRBUF_INIT;
2907
2908                if (!o->call_depth && index_has_changes(&sb)) {
2909                        err(o, _("Dirty index: cannot merge (dirty: %s)"),
2910                            sb.buf);
2911                        return 0;
2912                }
2913                output(o, 0, _("Already up to date!"));
2914                *result = head;
2915                return 1;
2916        }
2917
2918        code = git_merge_trees(o->call_depth, common, head, merge);
2919
2920        if (code != 0) {
2921                if (show(o, 4) || o->call_depth)
2922                        err(o, _("merging of trees %s and %s failed"),
2923                            oid_to_hex(&head->object.oid),
2924                            oid_to_hex(&merge->object.oid));
2925                return -1;
2926        }
2927
2928        if (unmerged_cache()) {
2929                struct string_list *entries;
2930                struct rename_info re_info;
2931                int i;
2932                /*
2933                 * Only need the hashmap while processing entries, so
2934                 * initialize it here and free it when we are done running
2935                 * through the entries. Keeping it in the merge_options as
2936                 * opposed to decaring a local hashmap is for convenience
2937                 * so that we don't have to pass it to around.
2938                 */
2939                hashmap_init(&o->current_file_dir_set, path_hashmap_cmp, NULL, 512);
2940                get_files_dirs(o, head);
2941                get_files_dirs(o, merge);
2942
2943                entries = get_unmerged();
2944                clean = handle_renames(o, common, head, merge, entries,
2945                                       &re_info);
2946                record_df_conflict_files(o, entries);
2947                if (clean < 0)
2948                        goto cleanup;
2949                for (i = entries->nr-1; 0 <= i; i--) {
2950                        const char *path = entries->items[i].string;
2951                        struct stage_data *e = entries->items[i].util;
2952                        if (!e->processed) {
2953                                int ret = process_entry(o, path, e);
2954                                if (!ret)
2955                                        clean = 0;
2956                                else if (ret < 0) {
2957                                        clean = ret;
2958                                        goto cleanup;
2959                                }
2960                        }
2961                }
2962                for (i = 0; i < entries->nr; i++) {
2963                        struct stage_data *e = entries->items[i].util;
2964                        if (!e->processed)
2965                                die("BUG: unprocessed path??? %s",
2966                                    entries->items[i].string);
2967                }
2968
2969cleanup:
2970                final_cleanup_renames(&re_info);
2971
2972                string_list_clear(entries, 1);
2973                free(entries);
2974
2975                hashmap_free(&o->current_file_dir_set, 1);
2976
2977                if (clean < 0)
2978                        return clean;
2979        }
2980        else
2981                clean = 1;
2982
2983        if (o->call_depth && !(*result = write_tree_from_memory(o)))
2984                return -1;
2985
2986        return clean;
2987}
2988
2989static struct commit_list *reverse_commit_list(struct commit_list *list)
2990{
2991        struct commit_list *next = NULL, *current, *backup;
2992        for (current = list; current; current = backup) {
2993                backup = current->next;
2994                current->next = next;
2995                next = current;
2996        }
2997        return next;
2998}
2999
3000/*
3001 * Merge the commits h1 and h2, return the resulting virtual
3002 * commit object and a flag indicating the cleanness of the merge.
3003 */
3004int merge_recursive(struct merge_options *o,
3005                    struct commit *h1,
3006                    struct commit *h2,
3007                    struct commit_list *ca,
3008                    struct commit **result)
3009{
3010        struct commit_list *iter;
3011        struct commit *merged_common_ancestors;
3012        struct tree *mrtree;
3013        int clean;
3014
3015        if (show(o, 4)) {
3016                output(o, 4, _("Merging:"));
3017                output_commit_title(o, h1);
3018                output_commit_title(o, h2);
3019        }
3020
3021        if (!ca) {
3022                ca = get_merge_bases(h1, h2);
3023                ca = reverse_commit_list(ca);
3024        }
3025
3026        if (show(o, 5)) {
3027                unsigned cnt = commit_list_count(ca);
3028
3029                output(o, 5, Q_("found %u common ancestor:",
3030                                "found %u common ancestors:", cnt), cnt);
3031                for (iter = ca; iter; iter = iter->next)
3032                        output_commit_title(o, iter->item);
3033        }
3034
3035        merged_common_ancestors = pop_commit(&ca);
3036        if (merged_common_ancestors == NULL) {
3037                /* if there is no common ancestor, use an empty tree */
3038                struct tree *tree;
3039
3040                tree = lookup_tree(the_hash_algo->empty_tree);
3041                merged_common_ancestors = make_virtual_commit(tree, "ancestor");
3042        }
3043
3044        for (iter = ca; iter; iter = iter->next) {
3045                const char *saved_b1, *saved_b2;
3046                o->call_depth++;
3047                /*
3048                 * When the merge fails, the result contains files
3049                 * with conflict markers. The cleanness flag is
3050                 * ignored (unless indicating an error), it was never
3051                 * actually used, as result of merge_trees has always
3052                 * overwritten it: the committed "conflicts" were
3053                 * already resolved.
3054                 */
3055                discard_cache();
3056                saved_b1 = o->branch1;
3057                saved_b2 = o->branch2;
3058                o->branch1 = "Temporary merge branch 1";
3059                o->branch2 = "Temporary merge branch 2";
3060                if (merge_recursive(o, merged_common_ancestors, iter->item,
3061                                    NULL, &merged_common_ancestors) < 0)
3062                        return -1;
3063                o->branch1 = saved_b1;
3064                o->branch2 = saved_b2;
3065                o->call_depth--;
3066
3067                if (!merged_common_ancestors)
3068                        return err(o, _("merge returned no commit"));
3069        }
3070
3071        discard_cache();
3072        if (!o->call_depth)
3073                read_cache();
3074
3075        o->ancestor = "merged common ancestors";
3076        clean = merge_trees(o, h1->tree, h2->tree, merged_common_ancestors->tree,
3077                            &mrtree);
3078        if (clean < 0) {
3079                flush_output(o);
3080                return clean;
3081        }
3082
3083        if (o->call_depth) {
3084                *result = make_virtual_commit(mrtree, "merged tree");
3085                commit_list_insert(h1, &(*result)->parents);
3086                commit_list_insert(h2, &(*result)->parents->next);
3087        }
3088        flush_output(o);
3089        if (!o->call_depth && o->buffer_output < 2)
3090                strbuf_release(&o->obuf);
3091        if (show(o, 2))
3092                diff_warn_rename_limit("merge.renamelimit",
3093                                       o->needed_rename_limit, 0);
3094        return clean;
3095}
3096
3097static struct commit *get_ref(const struct object_id *oid, const char *name)
3098{
3099        struct object *object;
3100
3101        object = deref_tag(parse_object(oid), name, strlen(name));
3102        if (!object)
3103                return NULL;
3104        if (object->type == OBJ_TREE)
3105                return make_virtual_commit((struct tree*)object, name);
3106        if (object->type != OBJ_COMMIT)
3107                return NULL;
3108        if (parse_commit((struct commit *)object))
3109                return NULL;
3110        return (struct commit *)object;
3111}
3112
3113int merge_recursive_generic(struct merge_options *o,
3114                            const struct object_id *head,
3115                            const struct object_id *merge,
3116                            int num_base_list,
3117                            const struct object_id **base_list,
3118                            struct commit **result)
3119{
3120        int clean;
3121        struct lock_file lock = LOCK_INIT;
3122        struct commit *head_commit = get_ref(head, o->branch1);
3123        struct commit *next_commit = get_ref(merge, o->branch2);
3124        struct commit_list *ca = NULL;
3125
3126        if (base_list) {
3127                int i;
3128                for (i = 0; i < num_base_list; ++i) {
3129                        struct commit *base;
3130                        if (!(base = get_ref(base_list[i], oid_to_hex(base_list[i]))))
3131                                return err(o, _("Could not parse object '%s'"),
3132                                        oid_to_hex(base_list[i]));
3133                        commit_list_insert(base, &ca);
3134                }
3135        }
3136
3137        hold_locked_index(&lock, LOCK_DIE_ON_ERROR);
3138        clean = merge_recursive(o, head_commit, next_commit, ca,
3139                        result);
3140        if (clean < 0) {
3141                rollback_lock_file(&lock);
3142                return clean;
3143        }
3144
3145        if (write_locked_index(&the_index, &lock,
3146                               COMMIT_LOCK | SKIP_IF_UNCHANGED))
3147                return err(o, _("Unable to write index."));
3148
3149        return clean ? 0 : 1;
3150}
3151
3152static void merge_recursive_config(struct merge_options *o)
3153{
3154        git_config_get_int("merge.verbosity", &o->verbosity);
3155        git_config_get_int("diff.renamelimit", &o->diff_rename_limit);
3156        git_config_get_int("merge.renamelimit", &o->merge_rename_limit);
3157        git_config(git_xmerge_config, NULL);
3158}
3159
3160void init_merge_options(struct merge_options *o)
3161{
3162        const char *merge_verbosity;
3163        memset(o, 0, sizeof(struct merge_options));
3164        o->verbosity = 2;
3165        o->buffer_output = 1;
3166        o->diff_rename_limit = -1;
3167        o->merge_rename_limit = -1;
3168        o->renormalize = 0;
3169        o->detect_rename = 1;
3170        merge_recursive_config(o);
3171        merge_verbosity = getenv("GIT_MERGE_VERBOSITY");
3172        if (merge_verbosity)
3173                o->verbosity = strtol(merge_verbosity, NULL, 10);
3174        if (o->verbosity >= 5)
3175                o->buffer_output = 0;
3176        strbuf_init(&o->obuf, 0);
3177        string_list_init(&o->df_conflict_file_set, 1);
3178}
3179
3180int parse_merge_opt(struct merge_options *o, const char *s)
3181{
3182        const char *arg;
3183
3184        if (!s || !*s)
3185                return -1;
3186        if (!strcmp(s, "ours"))
3187                o->recursive_variant = MERGE_RECURSIVE_OURS;
3188        else if (!strcmp(s, "theirs"))
3189                o->recursive_variant = MERGE_RECURSIVE_THEIRS;
3190        else if (!strcmp(s, "subtree"))
3191                o->subtree_shift = "";
3192        else if (skip_prefix(s, "subtree=", &arg))
3193                o->subtree_shift = arg;
3194        else if (!strcmp(s, "patience"))
3195                o->xdl_opts = DIFF_WITH_ALG(o, PATIENCE_DIFF);
3196        else if (!strcmp(s, "histogram"))
3197                o->xdl_opts = DIFF_WITH_ALG(o, HISTOGRAM_DIFF);
3198        else if (skip_prefix(s, "diff-algorithm=", &arg)) {
3199                long value = parse_algorithm_value(arg);
3200                if (value < 0)
3201                        return -1;
3202                /* clear out previous settings */
3203                DIFF_XDL_CLR(o, NEED_MINIMAL);
3204                o->xdl_opts &= ~XDF_DIFF_ALGORITHM_MASK;
3205                o->xdl_opts |= value;
3206        }
3207        else if (!strcmp(s, "ignore-space-change"))
3208                DIFF_XDL_SET(o, IGNORE_WHITESPACE_CHANGE);
3209        else if (!strcmp(s, "ignore-all-space"))
3210                DIFF_XDL_SET(o, IGNORE_WHITESPACE);
3211        else if (!strcmp(s, "ignore-space-at-eol"))
3212                DIFF_XDL_SET(o, IGNORE_WHITESPACE_AT_EOL);
3213        else if (!strcmp(s, "ignore-cr-at-eol"))
3214                DIFF_XDL_SET(o, IGNORE_CR_AT_EOL);
3215        else if (!strcmp(s, "renormalize"))
3216                o->renormalize = 1;
3217        else if (!strcmp(s, "no-renormalize"))
3218                o->renormalize = 0;
3219        else if (!strcmp(s, "no-renames"))
3220                o->detect_rename = 0;
3221        else if (!strcmp(s, "find-renames")) {
3222                o->detect_rename = 1;
3223                o->rename_score = 0;
3224        }
3225        else if (skip_prefix(s, "find-renames=", &arg) ||
3226                 skip_prefix(s, "rename-threshold=", &arg)) {
3227                if ((o->rename_score = parse_rename_score(&arg)) == -1 || *arg != 0)
3228                        return -1;
3229                o->detect_rename = 1;
3230        }
3231        else
3232                return -1;
3233        return 0;
3234}