pack-bitmap.con commit packfile: allow install_packed_git to handle arbitrary repositories (5babff1)
   1#include "cache.h"
   2#include "commit.h"
   3#include "tag.h"
   4#include "diff.h"
   5#include "revision.h"
   6#include "progress.h"
   7#include "list-objects.h"
   8#include "pack.h"
   9#include "pack-bitmap.h"
  10#include "pack-revindex.h"
  11#include "pack-objects.h"
  12#include "packfile.h"
  13#include "repository.h"
  14#include "object-store.h"
  15
  16/*
  17 * An entry on the bitmap index, representing the bitmap for a given
  18 * commit.
  19 */
  20struct stored_bitmap {
  21        unsigned char sha1[20];
  22        struct ewah_bitmap *root;
  23        struct stored_bitmap *xor;
  24        int flags;
  25};
  26
  27/*
  28 * The currently active bitmap index. By design, repositories only have
  29 * a single bitmap index available (the index for the biggest packfile in
  30 * the repository), since bitmap indexes need full closure.
  31 *
  32 * If there is more than one bitmap index available (e.g. because of alternates),
  33 * the active bitmap index is the largest one.
  34 */
  35static struct bitmap_index {
  36        /* Packfile to which this bitmap index belongs to */
  37        struct packed_git *pack;
  38
  39        /*
  40         * Mark the first `reuse_objects` in the packfile as reused:
  41         * they will be sent as-is without using them for repacking
  42         * calculations
  43         */
  44        uint32_t reuse_objects;
  45
  46        /* mmapped buffer of the whole bitmap index */
  47        unsigned char *map;
  48        size_t map_size; /* size of the mmaped buffer */
  49        size_t map_pos; /* current position when loading the index */
  50
  51        /*
  52         * Type indexes.
  53         *
  54         * Each bitmap marks which objects in the packfile  are of the given
  55         * type. This provides type information when yielding the objects from
  56         * the packfile during a walk, which allows for better delta bases.
  57         */
  58        struct ewah_bitmap *commits;
  59        struct ewah_bitmap *trees;
  60        struct ewah_bitmap *blobs;
  61        struct ewah_bitmap *tags;
  62
  63        /* Map from SHA1 -> `stored_bitmap` for all the bitmapped commits */
  64        khash_sha1 *bitmaps;
  65
  66        /* Number of bitmapped commits */
  67        uint32_t entry_count;
  68
  69        /* Name-hash cache (or NULL if not present). */
  70        uint32_t *hashes;
  71
  72        /*
  73         * Extended index.
  74         *
  75         * When trying to perform bitmap operations with objects that are not
  76         * packed in `pack`, these objects are added to this "fake index" and
  77         * are assumed to appear at the end of the packfile for all operations
  78         */
  79        struct eindex {
  80                struct object **objects;
  81                uint32_t *hashes;
  82                uint32_t count, alloc;
  83                khash_sha1_pos *positions;
  84        } ext_index;
  85
  86        /* Bitmap result of the last performed walk */
  87        struct bitmap *result;
  88
  89        /* Version of the bitmap index */
  90        unsigned int version;
  91
  92        unsigned loaded : 1;
  93
  94} bitmap_git;
  95
  96static struct ewah_bitmap *lookup_stored_bitmap(struct stored_bitmap *st)
  97{
  98        struct ewah_bitmap *parent;
  99        struct ewah_bitmap *composed;
 100
 101        if (st->xor == NULL)
 102                return st->root;
 103
 104        composed = ewah_pool_new();
 105        parent = lookup_stored_bitmap(st->xor);
 106        ewah_xor(st->root, parent, composed);
 107
 108        ewah_pool_free(st->root);
 109        st->root = composed;
 110        st->xor = NULL;
 111
 112        return composed;
 113}
 114
 115/*
 116 * Read a bitmap from the current read position on the mmaped
 117 * index, and increase the read position accordingly
 118 */
 119static struct ewah_bitmap *read_bitmap_1(struct bitmap_index *index)
 120{
 121        struct ewah_bitmap *b = ewah_pool_new();
 122
 123        int bitmap_size = ewah_read_mmap(b,
 124                index->map + index->map_pos,
 125                index->map_size - index->map_pos);
 126
 127        if (bitmap_size < 0) {
 128                error("Failed to load bitmap index (corrupted?)");
 129                ewah_pool_free(b);
 130                return NULL;
 131        }
 132
 133        index->map_pos += bitmap_size;
 134        return b;
 135}
 136
 137static int load_bitmap_header(struct bitmap_index *index)
 138{
 139        struct bitmap_disk_header *header = (void *)index->map;
 140
 141        if (index->map_size < sizeof(*header) + 20)
 142                return error("Corrupted bitmap index (missing header data)");
 143
 144        if (memcmp(header->magic, BITMAP_IDX_SIGNATURE, sizeof(BITMAP_IDX_SIGNATURE)) != 0)
 145                return error("Corrupted bitmap index file (wrong header)");
 146
 147        index->version = ntohs(header->version);
 148        if (index->version != 1)
 149                return error("Unsupported version for bitmap index file (%d)", index->version);
 150
 151        /* Parse known bitmap format options */
 152        {
 153                uint32_t flags = ntohs(header->options);
 154
 155                if ((flags & BITMAP_OPT_FULL_DAG) == 0)
 156                        return error("Unsupported options for bitmap index file "
 157                                "(Git requires BITMAP_OPT_FULL_DAG)");
 158
 159                if (flags & BITMAP_OPT_HASH_CACHE) {
 160                        unsigned char *end = index->map + index->map_size - 20;
 161                        index->hashes = ((uint32_t *)end) - index->pack->num_objects;
 162                }
 163        }
 164
 165        index->entry_count = ntohl(header->entry_count);
 166        index->map_pos += sizeof(*header);
 167        return 0;
 168}
 169
 170static struct stored_bitmap *store_bitmap(struct bitmap_index *index,
 171                                          struct ewah_bitmap *root,
 172                                          const unsigned char *sha1,
 173                                          struct stored_bitmap *xor_with,
 174                                          int flags)
 175{
 176        struct stored_bitmap *stored;
 177        khiter_t hash_pos;
 178        int ret;
 179
 180        stored = xmalloc(sizeof(struct stored_bitmap));
 181        stored->root = root;
 182        stored->xor = xor_with;
 183        stored->flags = flags;
 184        hashcpy(stored->sha1, sha1);
 185
 186        hash_pos = kh_put_sha1(index->bitmaps, stored->sha1, &ret);
 187
 188        /* a 0 return code means the insertion succeeded with no changes,
 189         * because the SHA1 already existed on the map. this is bad, there
 190         * shouldn't be duplicated commits in the index */
 191        if (ret == 0) {
 192                error("Duplicate entry in bitmap index: %s", sha1_to_hex(sha1));
 193                return NULL;
 194        }
 195
 196        kh_value(index->bitmaps, hash_pos) = stored;
 197        return stored;
 198}
 199
 200static inline uint32_t read_be32(const unsigned char *buffer, size_t *pos)
 201{
 202        uint32_t result = get_be32(buffer + *pos);
 203        (*pos) += sizeof(result);
 204        return result;
 205}
 206
 207static inline uint8_t read_u8(const unsigned char *buffer, size_t *pos)
 208{
 209        return buffer[(*pos)++];
 210}
 211
 212#define MAX_XOR_OFFSET 160
 213
 214static int load_bitmap_entries_v1(struct bitmap_index *index)
 215{
 216        uint32_t i;
 217        struct stored_bitmap *recent_bitmaps[MAX_XOR_OFFSET] = { NULL };
 218
 219        for (i = 0; i < index->entry_count; ++i) {
 220                int xor_offset, flags;
 221                struct ewah_bitmap *bitmap = NULL;
 222                struct stored_bitmap *xor_bitmap = NULL;
 223                uint32_t commit_idx_pos;
 224                const unsigned char *sha1;
 225
 226                commit_idx_pos = read_be32(index->map, &index->map_pos);
 227                xor_offset = read_u8(index->map, &index->map_pos);
 228                flags = read_u8(index->map, &index->map_pos);
 229
 230                sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos);
 231
 232                bitmap = read_bitmap_1(index);
 233                if (!bitmap)
 234                        return -1;
 235
 236                if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
 237                        return error("Corrupted bitmap pack index");
 238
 239                if (xor_offset > 0) {
 240                        xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
 241
 242                        if (xor_bitmap == NULL)
 243                                return error("Invalid XOR offset in bitmap pack index");
 244                }
 245
 246                recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
 247                        index, bitmap, sha1, xor_bitmap, flags);
 248        }
 249
 250        return 0;
 251}
 252
 253static char *pack_bitmap_filename(struct packed_git *p)
 254{
 255        size_t len;
 256
 257        if (!strip_suffix(p->pack_name, ".pack", &len))
 258                die("BUG: pack_name does not end in .pack");
 259        return xstrfmt("%.*s.bitmap", (int)len, p->pack_name);
 260}
 261
 262static int open_pack_bitmap_1(struct packed_git *packfile)
 263{
 264        int fd;
 265        struct stat st;
 266        char *idx_name;
 267
 268        if (open_pack_index(packfile))
 269                return -1;
 270
 271        idx_name = pack_bitmap_filename(packfile);
 272        fd = git_open(idx_name);
 273        free(idx_name);
 274
 275        if (fd < 0)
 276                return -1;
 277
 278        if (fstat(fd, &st)) {
 279                close(fd);
 280                return -1;
 281        }
 282
 283        if (bitmap_git.pack) {
 284                warning("ignoring extra bitmap file: %s", packfile->pack_name);
 285                close(fd);
 286                return -1;
 287        }
 288
 289        bitmap_git.pack = packfile;
 290        bitmap_git.map_size = xsize_t(st.st_size);
 291        bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0);
 292        bitmap_git.map_pos = 0;
 293        close(fd);
 294
 295        if (load_bitmap_header(&bitmap_git) < 0) {
 296                munmap(bitmap_git.map, bitmap_git.map_size);
 297                bitmap_git.map = NULL;
 298                bitmap_git.map_size = 0;
 299                return -1;
 300        }
 301
 302        return 0;
 303}
 304
 305static int load_pack_bitmap(void)
 306{
 307        assert(bitmap_git.map && !bitmap_git.loaded);
 308
 309        bitmap_git.bitmaps = kh_init_sha1();
 310        bitmap_git.ext_index.positions = kh_init_sha1_pos();
 311        load_pack_revindex(bitmap_git.pack);
 312
 313        if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
 314                !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
 315                !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
 316                !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
 317                goto failed;
 318
 319        if (load_bitmap_entries_v1(&bitmap_git) < 0)
 320                goto failed;
 321
 322        bitmap_git.loaded = 1;
 323        return 0;
 324
 325failed:
 326        munmap(bitmap_git.map, bitmap_git.map_size);
 327        bitmap_git.map = NULL;
 328        bitmap_git.map_size = 0;
 329        return -1;
 330}
 331
 332static int open_pack_bitmap(void)
 333{
 334        struct packed_git *p;
 335        int ret = -1;
 336
 337        assert(!bitmap_git.map && !bitmap_git.loaded);
 338
 339        prepare_packed_git();
 340        for (p = get_packed_git(the_repository); p; p = p->next) {
 341                if (open_pack_bitmap_1(p) == 0)
 342                        ret = 0;
 343        }
 344
 345        return ret;
 346}
 347
 348int prepare_bitmap_git(void)
 349{
 350        if (bitmap_git.loaded)
 351                return 0;
 352
 353        if (!open_pack_bitmap())
 354                return load_pack_bitmap();
 355
 356        return -1;
 357}
 358
 359struct include_data {
 360        struct bitmap *base;
 361        struct bitmap *seen;
 362};
 363
 364static inline int bitmap_position_extended(const unsigned char *sha1)
 365{
 366        khash_sha1_pos *positions = bitmap_git.ext_index.positions;
 367        khiter_t pos = kh_get_sha1_pos(positions, sha1);
 368
 369        if (pos < kh_end(positions)) {
 370                int bitmap_pos = kh_value(positions, pos);
 371                return bitmap_pos + bitmap_git.pack->num_objects;
 372        }
 373
 374        return -1;
 375}
 376
 377static inline int bitmap_position_packfile(const unsigned char *sha1)
 378{
 379        off_t offset = find_pack_entry_one(sha1, bitmap_git.pack);
 380        if (!offset)
 381                return -1;
 382
 383        return find_revindex_position(bitmap_git.pack, offset);
 384}
 385
 386static int bitmap_position(const unsigned char *sha1)
 387{
 388        int pos = bitmap_position_packfile(sha1);
 389        return (pos >= 0) ? pos : bitmap_position_extended(sha1);
 390}
 391
 392static int ext_index_add_object(struct object *object, const char *name)
 393{
 394        struct eindex *eindex = &bitmap_git.ext_index;
 395
 396        khiter_t hash_pos;
 397        int hash_ret;
 398        int bitmap_pos;
 399
 400        hash_pos = kh_put_sha1_pos(eindex->positions, object->oid.hash, &hash_ret);
 401        if (hash_ret > 0) {
 402                if (eindex->count >= eindex->alloc) {
 403                        eindex->alloc = (eindex->alloc + 16) * 3 / 2;
 404                        REALLOC_ARRAY(eindex->objects, eindex->alloc);
 405                        REALLOC_ARRAY(eindex->hashes, eindex->alloc);
 406                }
 407
 408                bitmap_pos = eindex->count;
 409                eindex->objects[eindex->count] = object;
 410                eindex->hashes[eindex->count] = pack_name_hash(name);
 411                kh_value(eindex->positions, hash_pos) = bitmap_pos;
 412                eindex->count++;
 413        } else {
 414                bitmap_pos = kh_value(eindex->positions, hash_pos);
 415        }
 416
 417        return bitmap_pos + bitmap_git.pack->num_objects;
 418}
 419
 420static void show_object(struct object *object, const char *name, void *data)
 421{
 422        struct bitmap *base = data;
 423        int bitmap_pos;
 424
 425        bitmap_pos = bitmap_position(object->oid.hash);
 426
 427        if (bitmap_pos < 0)
 428                bitmap_pos = ext_index_add_object(object, name);
 429
 430        bitmap_set(base, bitmap_pos);
 431}
 432
 433static void show_commit(struct commit *commit, void *data)
 434{
 435}
 436
 437static int add_to_include_set(struct include_data *data,
 438                              const unsigned char *sha1,
 439                              int bitmap_pos)
 440{
 441        khiter_t hash_pos;
 442
 443        if (data->seen && bitmap_get(data->seen, bitmap_pos))
 444                return 0;
 445
 446        if (bitmap_get(data->base, bitmap_pos))
 447                return 0;
 448
 449        hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
 450        if (hash_pos < kh_end(bitmap_git.bitmaps)) {
 451                struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
 452                bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
 453                return 0;
 454        }
 455
 456        bitmap_set(data->base, bitmap_pos);
 457        return 1;
 458}
 459
 460static int should_include(struct commit *commit, void *_data)
 461{
 462        struct include_data *data = _data;
 463        int bitmap_pos;
 464
 465        bitmap_pos = bitmap_position(commit->object.oid.hash);
 466        if (bitmap_pos < 0)
 467                bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
 468
 469        if (!add_to_include_set(data, commit->object.oid.hash, bitmap_pos)) {
 470                struct commit_list *parent = commit->parents;
 471
 472                while (parent) {
 473                        parent->item->object.flags |= SEEN;
 474                        parent = parent->next;
 475                }
 476
 477                return 0;
 478        }
 479
 480        return 1;
 481}
 482
 483static struct bitmap *find_objects(struct rev_info *revs,
 484                                   struct object_list *roots,
 485                                   struct bitmap *seen)
 486{
 487        struct bitmap *base = NULL;
 488        int needs_walk = 0;
 489
 490        struct object_list *not_mapped = NULL;
 491
 492        /*
 493         * Go through all the roots for the walk. The ones that have bitmaps
 494         * on the bitmap index will be `or`ed together to form an initial
 495         * global reachability analysis.
 496         *
 497         * The ones without bitmaps in the index will be stored in the
 498         * `not_mapped_list` for further processing.
 499         */
 500        while (roots) {
 501                struct object *object = roots->item;
 502                roots = roots->next;
 503
 504                if (object->type == OBJ_COMMIT) {
 505                        khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->oid.hash);
 506
 507                        if (pos < kh_end(bitmap_git.bitmaps)) {
 508                                struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
 509                                struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
 510
 511                                if (base == NULL)
 512                                        base = ewah_to_bitmap(or_with);
 513                                else
 514                                        bitmap_or_ewah(base, or_with);
 515
 516                                object->flags |= SEEN;
 517                                continue;
 518                        }
 519                }
 520
 521                object_list_insert(object, &not_mapped);
 522        }
 523
 524        /*
 525         * Best case scenario: We found bitmaps for all the roots,
 526         * so the resulting `or` bitmap has the full reachability analysis
 527         */
 528        if (not_mapped == NULL)
 529                return base;
 530
 531        roots = not_mapped;
 532
 533        /*
 534         * Let's iterate through all the roots that don't have bitmaps to
 535         * check if we can determine them to be reachable from the existing
 536         * global bitmap.
 537         *
 538         * If we cannot find them in the existing global bitmap, we'll need
 539         * to push them to an actual walk and run it until we can confirm
 540         * they are reachable
 541         */
 542        while (roots) {
 543                struct object *object = roots->item;
 544                int pos;
 545
 546                roots = roots->next;
 547                pos = bitmap_position(object->oid.hash);
 548
 549                if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
 550                        object->flags &= ~UNINTERESTING;
 551                        add_pending_object(revs, object, "");
 552                        needs_walk = 1;
 553                } else {
 554                        object->flags |= SEEN;
 555                }
 556        }
 557
 558        if (needs_walk) {
 559                struct include_data incdata;
 560
 561                if (base == NULL)
 562                        base = bitmap_new();
 563
 564                incdata.base = base;
 565                incdata.seen = seen;
 566
 567                revs->include_check = should_include;
 568                revs->include_check_data = &incdata;
 569
 570                if (prepare_revision_walk(revs))
 571                        die("revision walk setup failed");
 572
 573                traverse_commit_list(revs, show_commit, show_object, base);
 574        }
 575
 576        return base;
 577}
 578
 579static void show_extended_objects(struct bitmap *objects,
 580                                  show_reachable_fn show_reach)
 581{
 582        struct eindex *eindex = &bitmap_git.ext_index;
 583        uint32_t i;
 584
 585        for (i = 0; i < eindex->count; ++i) {
 586                struct object *obj;
 587
 588                if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
 589                        continue;
 590
 591                obj = eindex->objects[i];
 592                show_reach(&obj->oid, obj->type, 0, eindex->hashes[i], NULL, 0);
 593        }
 594}
 595
 596static void show_objects_for_type(
 597        struct bitmap *objects,
 598        struct ewah_bitmap *type_filter,
 599        enum object_type object_type,
 600        show_reachable_fn show_reach)
 601{
 602        size_t pos = 0, i = 0;
 603        uint32_t offset;
 604
 605        struct ewah_iterator it;
 606        eword_t filter;
 607
 608        if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
 609                return;
 610
 611        ewah_iterator_init(&it, type_filter);
 612
 613        while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
 614                eword_t word = objects->words[i] & filter;
 615
 616                for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
 617                        struct object_id oid;
 618                        struct revindex_entry *entry;
 619                        uint32_t hash = 0;
 620
 621                        if ((word >> offset) == 0)
 622                                break;
 623
 624                        offset += ewah_bit_ctz64(word >> offset);
 625
 626                        if (pos + offset < bitmap_git.reuse_objects)
 627                                continue;
 628
 629                        entry = &bitmap_git.pack->revindex[pos + offset];
 630                        nth_packed_object_oid(&oid, bitmap_git.pack, entry->nr);
 631
 632                        if (bitmap_git.hashes)
 633                                hash = get_be32(bitmap_git.hashes + entry->nr);
 634
 635                        show_reach(&oid, object_type, 0, hash, bitmap_git.pack, entry->offset);
 636                }
 637
 638                pos += BITS_IN_EWORD;
 639                i++;
 640        }
 641}
 642
 643static int in_bitmapped_pack(struct object_list *roots)
 644{
 645        while (roots) {
 646                struct object *object = roots->item;
 647                roots = roots->next;
 648
 649                if (find_pack_entry_one(object->oid.hash, bitmap_git.pack) > 0)
 650                        return 1;
 651        }
 652
 653        return 0;
 654}
 655
 656int prepare_bitmap_walk(struct rev_info *revs)
 657{
 658        unsigned int i;
 659
 660        struct object_list *wants = NULL;
 661        struct object_list *haves = NULL;
 662
 663        struct bitmap *wants_bitmap = NULL;
 664        struct bitmap *haves_bitmap = NULL;
 665
 666        if (!bitmap_git.loaded) {
 667                /* try to open a bitmapped pack, but don't parse it yet
 668                 * because we may not need to use it */
 669                if (open_pack_bitmap() < 0)
 670                        return -1;
 671        }
 672
 673        for (i = 0; i < revs->pending.nr; ++i) {
 674                struct object *object = revs->pending.objects[i].item;
 675
 676                if (object->type == OBJ_NONE)
 677                        parse_object_or_die(&object->oid, NULL);
 678
 679                while (object->type == OBJ_TAG) {
 680                        struct tag *tag = (struct tag *) object;
 681
 682                        if (object->flags & UNINTERESTING)
 683                                object_list_insert(object, &haves);
 684                        else
 685                                object_list_insert(object, &wants);
 686
 687                        if (!tag->tagged)
 688                                die("bad tag");
 689                        object = parse_object_or_die(&tag->tagged->oid, NULL);
 690                }
 691
 692                if (object->flags & UNINTERESTING)
 693                        object_list_insert(object, &haves);
 694                else
 695                        object_list_insert(object, &wants);
 696        }
 697
 698        /*
 699         * if we have a HAVES list, but none of those haves is contained
 700         * in the packfile that has a bitmap, we don't have anything to
 701         * optimize here
 702         */
 703        if (haves && !in_bitmapped_pack(haves))
 704                return -1;
 705
 706        /* if we don't want anything, we're done here */
 707        if (!wants)
 708                return -1;
 709
 710        /*
 711         * now we're going to use bitmaps, so load the actual bitmap entries
 712         * from disk. this is the point of no return; after this the rev_list
 713         * becomes invalidated and we must perform the revwalk through bitmaps
 714         */
 715        if (!bitmap_git.loaded && load_pack_bitmap() < 0)
 716                return -1;
 717
 718        object_array_clear(&revs->pending);
 719
 720        if (haves) {
 721                revs->ignore_missing_links = 1;
 722                haves_bitmap = find_objects(revs, haves, NULL);
 723                reset_revision_walk();
 724                revs->ignore_missing_links = 0;
 725
 726                if (haves_bitmap == NULL)
 727                        die("BUG: failed to perform bitmap walk");
 728        }
 729
 730        wants_bitmap = find_objects(revs, wants, haves_bitmap);
 731
 732        if (!wants_bitmap)
 733                die("BUG: failed to perform bitmap walk");
 734
 735        if (haves_bitmap)
 736                bitmap_and_not(wants_bitmap, haves_bitmap);
 737
 738        bitmap_git.result = wants_bitmap;
 739
 740        bitmap_free(haves_bitmap);
 741        return 0;
 742}
 743
 744int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
 745                                       uint32_t *entries,
 746                                       off_t *up_to)
 747{
 748        /*
 749         * Reuse the packfile content if we need more than
 750         * 90% of its objects
 751         */
 752        static const double REUSE_PERCENT = 0.9;
 753
 754        struct bitmap *result = bitmap_git.result;
 755        uint32_t reuse_threshold;
 756        uint32_t i, reuse_objects = 0;
 757
 758        assert(result);
 759
 760        for (i = 0; i < result->word_alloc; ++i) {
 761                if (result->words[i] != (eword_t)~0) {
 762                        reuse_objects += ewah_bit_ctz64(~result->words[i]);
 763                        break;
 764                }
 765
 766                reuse_objects += BITS_IN_EWORD;
 767        }
 768
 769#ifdef GIT_BITMAP_DEBUG
 770        {
 771                const unsigned char *sha1;
 772                struct revindex_entry *entry;
 773
 774                entry = &bitmap_git.reverse_index->revindex[reuse_objects];
 775                sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
 776
 777                fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
 778                        reuse_objects, result->words[i]);
 779                fprintf(stderr, " %s\n", sha1_to_hex(sha1));
 780        }
 781#endif
 782
 783        if (!reuse_objects)
 784                return -1;
 785
 786        if (reuse_objects >= bitmap_git.pack->num_objects) {
 787                bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
 788                *up_to = -1; /* reuse the full pack */
 789                *packfile = bitmap_git.pack;
 790                return 0;
 791        }
 792
 793        reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
 794
 795        if (reuse_objects < reuse_threshold)
 796                return -1;
 797
 798        bitmap_git.reuse_objects = *entries = reuse_objects;
 799        *up_to = bitmap_git.pack->revindex[reuse_objects].offset;
 800        *packfile = bitmap_git.pack;
 801
 802        return 0;
 803}
 804
 805void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
 806{
 807        assert(bitmap_git.result);
 808
 809        show_objects_for_type(bitmap_git.result, bitmap_git.commits,
 810                OBJ_COMMIT, show_reachable);
 811        show_objects_for_type(bitmap_git.result, bitmap_git.trees,
 812                OBJ_TREE, show_reachable);
 813        show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
 814                OBJ_BLOB, show_reachable);
 815        show_objects_for_type(bitmap_git.result, bitmap_git.tags,
 816                OBJ_TAG, show_reachable);
 817
 818        show_extended_objects(bitmap_git.result, show_reachable);
 819
 820        bitmap_free(bitmap_git.result);
 821        bitmap_git.result = NULL;
 822}
 823
 824static uint32_t count_object_type(struct bitmap *objects,
 825                                  enum object_type type)
 826{
 827        struct eindex *eindex = &bitmap_git.ext_index;
 828
 829        uint32_t i = 0, count = 0;
 830        struct ewah_iterator it;
 831        eword_t filter;
 832
 833        switch (type) {
 834        case OBJ_COMMIT:
 835                ewah_iterator_init(&it, bitmap_git.commits);
 836                break;
 837
 838        case OBJ_TREE:
 839                ewah_iterator_init(&it, bitmap_git.trees);
 840                break;
 841
 842        case OBJ_BLOB:
 843                ewah_iterator_init(&it, bitmap_git.blobs);
 844                break;
 845
 846        case OBJ_TAG:
 847                ewah_iterator_init(&it, bitmap_git.tags);
 848                break;
 849
 850        default:
 851                return 0;
 852        }
 853
 854        while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
 855                eword_t word = objects->words[i++] & filter;
 856                count += ewah_bit_popcount64(word);
 857        }
 858
 859        for (i = 0; i < eindex->count; ++i) {
 860                if (eindex->objects[i]->type == type &&
 861                        bitmap_get(objects, bitmap_git.pack->num_objects + i))
 862                        count++;
 863        }
 864
 865        return count;
 866}
 867
 868void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
 869                              uint32_t *blobs, uint32_t *tags)
 870{
 871        assert(bitmap_git.result);
 872
 873        if (commits)
 874                *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
 875
 876        if (trees)
 877                *trees = count_object_type(bitmap_git.result, OBJ_TREE);
 878
 879        if (blobs)
 880                *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
 881
 882        if (tags)
 883                *tags = count_object_type(bitmap_git.result, OBJ_TAG);
 884}
 885
 886struct bitmap_test_data {
 887        struct bitmap *base;
 888        struct progress *prg;
 889        size_t seen;
 890};
 891
 892static void test_show_object(struct object *object, const char *name,
 893                             void *data)
 894{
 895        struct bitmap_test_data *tdata = data;
 896        int bitmap_pos;
 897
 898        bitmap_pos = bitmap_position(object->oid.hash);
 899        if (bitmap_pos < 0)
 900                die("Object not in bitmap: %s\n", oid_to_hex(&object->oid));
 901
 902        bitmap_set(tdata->base, bitmap_pos);
 903        display_progress(tdata->prg, ++tdata->seen);
 904}
 905
 906static void test_show_commit(struct commit *commit, void *data)
 907{
 908        struct bitmap_test_data *tdata = data;
 909        int bitmap_pos;
 910
 911        bitmap_pos = bitmap_position(commit->object.oid.hash);
 912        if (bitmap_pos < 0)
 913                die("Object not in bitmap: %s\n", oid_to_hex(&commit->object.oid));
 914
 915        bitmap_set(tdata->base, bitmap_pos);
 916        display_progress(tdata->prg, ++tdata->seen);
 917}
 918
 919void test_bitmap_walk(struct rev_info *revs)
 920{
 921        struct object *root;
 922        struct bitmap *result = NULL;
 923        khiter_t pos;
 924        size_t result_popcnt;
 925        struct bitmap_test_data tdata;
 926
 927        if (prepare_bitmap_git())
 928                die("failed to load bitmap indexes");
 929
 930        if (revs->pending.nr != 1)
 931                die("you must specify exactly one commit to test");
 932
 933        fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
 934                bitmap_git.version, bitmap_git.entry_count);
 935
 936        root = revs->pending.objects[0].item;
 937        pos = kh_get_sha1(bitmap_git.bitmaps, root->oid.hash);
 938
 939        if (pos < kh_end(bitmap_git.bitmaps)) {
 940                struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
 941                struct ewah_bitmap *bm = lookup_stored_bitmap(st);
 942
 943                fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
 944                        oid_to_hex(&root->oid), (int)bm->bit_size, ewah_checksum(bm));
 945
 946                result = ewah_to_bitmap(bm);
 947        }
 948
 949        if (result == NULL)
 950                die("Commit %s doesn't have an indexed bitmap", oid_to_hex(&root->oid));
 951
 952        revs->tag_objects = 1;
 953        revs->tree_objects = 1;
 954        revs->blob_objects = 1;
 955
 956        result_popcnt = bitmap_popcount(result);
 957
 958        if (prepare_revision_walk(revs))
 959                die("revision walk setup failed");
 960
 961        tdata.base = bitmap_new();
 962        tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
 963        tdata.seen = 0;
 964
 965        traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
 966
 967        stop_progress(&tdata.prg);
 968
 969        if (bitmap_equals(result, tdata.base))
 970                fprintf(stderr, "OK!\n");
 971        else
 972                fprintf(stderr, "Mismatch!\n");
 973
 974        bitmap_free(result);
 975}
 976
 977static int rebuild_bitmap(uint32_t *reposition,
 978                          struct ewah_bitmap *source,
 979                          struct bitmap *dest)
 980{
 981        uint32_t pos = 0;
 982        struct ewah_iterator it;
 983        eword_t word;
 984
 985        ewah_iterator_init(&it, source);
 986
 987        while (ewah_iterator_next(&word, &it)) {
 988                uint32_t offset, bit_pos;
 989
 990                for (offset = 0; offset < BITS_IN_EWORD; ++offset) {
 991                        if ((word >> offset) == 0)
 992                                break;
 993
 994                        offset += ewah_bit_ctz64(word >> offset);
 995
 996                        bit_pos = reposition[pos + offset];
 997                        if (bit_pos > 0)
 998                                bitmap_set(dest, bit_pos - 1);
 999                        else /* can't reuse, we don't have the object */
1000                                return -1;
1001                }
1002
1003                pos += BITS_IN_EWORD;
1004        }
1005        return 0;
1006}
1007
1008int rebuild_existing_bitmaps(struct packing_data *mapping,
1009                             khash_sha1 *reused_bitmaps,
1010                             int show_progress)
1011{
1012        uint32_t i, num_objects;
1013        uint32_t *reposition;
1014        struct bitmap *rebuild;
1015        struct stored_bitmap *stored;
1016        struct progress *progress = NULL;
1017
1018        khiter_t hash_pos;
1019        int hash_ret;
1020
1021        if (prepare_bitmap_git() < 0)
1022                return -1;
1023
1024        num_objects = bitmap_git.pack->num_objects;
1025        reposition = xcalloc(num_objects, sizeof(uint32_t));
1026
1027        for (i = 0; i < num_objects; ++i) {
1028                const unsigned char *sha1;
1029                struct revindex_entry *entry;
1030                struct object_entry *oe;
1031
1032                entry = &bitmap_git.pack->revindex[i];
1033                sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
1034                oe = packlist_find(mapping, sha1, NULL);
1035
1036                if (oe)
1037                        reposition[i] = oe->in_pack_pos + 1;
1038        }
1039
1040        rebuild = bitmap_new();
1041        i = 0;
1042
1043        if (show_progress)
1044                progress = start_progress("Reusing bitmaps", 0);
1045
1046        kh_foreach_value(bitmap_git.bitmaps, stored, {
1047                if (stored->flags & BITMAP_FLAG_REUSE) {
1048                        if (!rebuild_bitmap(reposition,
1049                                            lookup_stored_bitmap(stored),
1050                                            rebuild)) {
1051                                hash_pos = kh_put_sha1(reused_bitmaps,
1052                                                       stored->sha1,
1053                                                       &hash_ret);
1054                                kh_value(reused_bitmaps, hash_pos) =
1055                                        bitmap_to_ewah(rebuild);
1056                        }
1057                        bitmap_reset(rebuild);
1058                        display_progress(progress, ++i);
1059                }
1060        });
1061
1062        stop_progress(&progress);
1063
1064        free(reposition);
1065        bitmap_free(rebuild);
1066        return 0;
1067}