pack-bitmap.con commit Merge branch 'jc/unpack-trees-plug-leak' (c09988a)
   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
  13/*
  14 * An entry on the bitmap index, representing the bitmap for a given
  15 * commit.
  16 */
  17struct stored_bitmap {
  18        unsigned char sha1[20];
  19        struct ewah_bitmap *root;
  20        struct stored_bitmap *xor;
  21        int flags;
  22};
  23
  24/*
  25 * The currently active bitmap index. By design, repositories only have
  26 * a single bitmap index available (the index for the biggest packfile in
  27 * the repository), since bitmap indexes need full closure.
  28 *
  29 * If there is more than one bitmap index available (e.g. because of alternates),
  30 * the active bitmap index is the largest one.
  31 */
  32static struct bitmap_index {
  33        /* Packfile to which this bitmap index belongs to */
  34        struct packed_git *pack;
  35
  36        /* reverse index for the packfile */
  37        struct pack_revindex *reverse_index;
  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 comits */
  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 int load_bitmap_entries_v1(struct bitmap_index *index)
 201{
 202        static const size_t MAX_XOR_OFFSET = 160;
 203
 204        uint32_t i;
 205        struct stored_bitmap **recent_bitmaps;
 206        struct bitmap_disk_entry *entry;
 207
 208        recent_bitmaps = xcalloc(MAX_XOR_OFFSET, sizeof(struct stored_bitmap));
 209
 210        for (i = 0; i < index->entry_count; ++i) {
 211                int xor_offset, flags;
 212                struct ewah_bitmap *bitmap = NULL;
 213                struct stored_bitmap *xor_bitmap = NULL;
 214                uint32_t commit_idx_pos;
 215                const unsigned char *sha1;
 216
 217                entry = (struct bitmap_disk_entry *)(index->map + index->map_pos);
 218                index->map_pos += sizeof(struct bitmap_disk_entry);
 219
 220                commit_idx_pos = ntohl(entry->object_pos);
 221                sha1 = nth_packed_object_sha1(index->pack, commit_idx_pos);
 222
 223                xor_offset = (int)entry->xor_offset;
 224                flags = (int)entry->flags;
 225
 226                bitmap = read_bitmap_1(index);
 227                if (!bitmap)
 228                        return -1;
 229
 230                if (xor_offset > MAX_XOR_OFFSET || xor_offset > i)
 231                        return error("Corrupted bitmap pack index");
 232
 233                if (xor_offset > 0) {
 234                        xor_bitmap = recent_bitmaps[(i - xor_offset) % MAX_XOR_OFFSET];
 235
 236                        if (xor_bitmap == NULL)
 237                                return error("Invalid XOR offset in bitmap pack index");
 238                }
 239
 240                recent_bitmaps[i % MAX_XOR_OFFSET] = store_bitmap(
 241                        index, bitmap, sha1, xor_bitmap, flags);
 242        }
 243
 244        return 0;
 245}
 246
 247static int open_pack_bitmap_1(struct packed_git *packfile)
 248{
 249        int fd;
 250        struct stat st;
 251        char *idx_name;
 252
 253        if (open_pack_index(packfile))
 254                return -1;
 255
 256        idx_name = pack_bitmap_filename(packfile);
 257        fd = git_open_noatime(idx_name);
 258        free(idx_name);
 259
 260        if (fd < 0)
 261                return -1;
 262
 263        if (fstat(fd, &st)) {
 264                close(fd);
 265                return -1;
 266        }
 267
 268        if (bitmap_git.pack) {
 269                warning("ignoring extra bitmap file: %s", packfile->pack_name);
 270                close(fd);
 271                return -1;
 272        }
 273
 274        bitmap_git.pack = packfile;
 275        bitmap_git.map_size = xsize_t(st.st_size);
 276        bitmap_git.map = xmmap(NULL, bitmap_git.map_size, PROT_READ, MAP_PRIVATE, fd, 0);
 277        bitmap_git.map_pos = 0;
 278        close(fd);
 279
 280        if (load_bitmap_header(&bitmap_git) < 0) {
 281                munmap(bitmap_git.map, bitmap_git.map_size);
 282                bitmap_git.map = NULL;
 283                bitmap_git.map_size = 0;
 284                return -1;
 285        }
 286
 287        return 0;
 288}
 289
 290static int load_pack_bitmap(void)
 291{
 292        assert(bitmap_git.map && !bitmap_git.loaded);
 293
 294        bitmap_git.bitmaps = kh_init_sha1();
 295        bitmap_git.ext_index.positions = kh_init_sha1_pos();
 296        bitmap_git.reverse_index = revindex_for_pack(bitmap_git.pack);
 297
 298        if (!(bitmap_git.commits = read_bitmap_1(&bitmap_git)) ||
 299                !(bitmap_git.trees = read_bitmap_1(&bitmap_git)) ||
 300                !(bitmap_git.blobs = read_bitmap_1(&bitmap_git)) ||
 301                !(bitmap_git.tags = read_bitmap_1(&bitmap_git)))
 302                goto failed;
 303
 304        if (load_bitmap_entries_v1(&bitmap_git) < 0)
 305                goto failed;
 306
 307        bitmap_git.loaded = 1;
 308        return 0;
 309
 310failed:
 311        munmap(bitmap_git.map, bitmap_git.map_size);
 312        bitmap_git.map = NULL;
 313        bitmap_git.map_size = 0;
 314        return -1;
 315}
 316
 317char *pack_bitmap_filename(struct packed_git *p)
 318{
 319        char *idx_name;
 320        int len;
 321
 322        len = strlen(p->pack_name) - strlen(".pack");
 323        idx_name = xmalloc(len + strlen(".bitmap") + 1);
 324
 325        memcpy(idx_name, p->pack_name, len);
 326        memcpy(idx_name + len, ".bitmap", strlen(".bitmap") + 1);
 327
 328        return idx_name;
 329}
 330
 331static int open_pack_bitmap(void)
 332{
 333        struct packed_git *p;
 334        int ret = -1;
 335
 336        assert(!bitmap_git.map && !bitmap_git.loaded);
 337
 338        prepare_packed_git();
 339        for (p = packed_git; p; p = p->next) {
 340                if (open_pack_bitmap_1(p) == 0)
 341                        ret = 0;
 342        }
 343
 344        return ret;
 345}
 346
 347int prepare_bitmap_git(void)
 348{
 349        if (bitmap_git.loaded)
 350                return 0;
 351
 352        if (!open_pack_bitmap())
 353                return load_pack_bitmap();
 354
 355        return -1;
 356}
 357
 358struct include_data {
 359        struct bitmap *base;
 360        struct bitmap *seen;
 361};
 362
 363static inline int bitmap_position_extended(const unsigned char *sha1)
 364{
 365        khash_sha1_pos *positions = bitmap_git.ext_index.positions;
 366        khiter_t pos = kh_get_sha1_pos(positions, sha1);
 367
 368        if (pos < kh_end(positions)) {
 369                int bitmap_pos = kh_value(positions, pos);
 370                return bitmap_pos + bitmap_git.pack->num_objects;
 371        }
 372
 373        return -1;
 374}
 375
 376static inline int bitmap_position_packfile(const unsigned char *sha1)
 377{
 378        off_t offset = find_pack_entry_one(sha1, bitmap_git.pack);
 379        if (!offset)
 380                return -1;
 381
 382        return find_revindex_position(bitmap_git.reverse_index, offset);
 383}
 384
 385static int bitmap_position(const unsigned char *sha1)
 386{
 387        int pos = bitmap_position_packfile(sha1);
 388        return (pos >= 0) ? pos : bitmap_position_extended(sha1);
 389}
 390
 391static int ext_index_add_object(struct object *object, const char *name)
 392{
 393        struct eindex *eindex = &bitmap_git.ext_index;
 394
 395        khiter_t hash_pos;
 396        int hash_ret;
 397        int bitmap_pos;
 398
 399        hash_pos = kh_put_sha1_pos(eindex->positions, object->sha1, &hash_ret);
 400        if (hash_ret > 0) {
 401                if (eindex->count >= eindex->alloc) {
 402                        eindex->alloc = (eindex->alloc + 16) * 3 / 2;
 403                        REALLOC_ARRAY(eindex->objects, eindex->alloc);
 404                        REALLOC_ARRAY(eindex->hashes, eindex->alloc);
 405                }
 406
 407                bitmap_pos = eindex->count;
 408                eindex->objects[eindex->count] = object;
 409                eindex->hashes[eindex->count] = pack_name_hash(name);
 410                kh_value(eindex->positions, hash_pos) = bitmap_pos;
 411                eindex->count++;
 412        } else {
 413                bitmap_pos = kh_value(eindex->positions, hash_pos);
 414        }
 415
 416        return bitmap_pos + bitmap_git.pack->num_objects;
 417}
 418
 419static void show_object(struct object *object, const struct name_path *path,
 420                        const char *last, void *data)
 421{
 422        struct bitmap *base = data;
 423        int bitmap_pos;
 424
 425        bitmap_pos = bitmap_position(object->sha1);
 426
 427        if (bitmap_pos < 0) {
 428                char *name = path_name(path, last);
 429                bitmap_pos = ext_index_add_object(object, name);
 430                free(name);
 431        }
 432
 433        bitmap_set(base, bitmap_pos);
 434}
 435
 436static void show_commit(struct commit *commit, void *data)
 437{
 438}
 439
 440static int add_to_include_set(struct include_data *data,
 441                              const unsigned char *sha1,
 442                              int bitmap_pos)
 443{
 444        khiter_t hash_pos;
 445
 446        if (data->seen && bitmap_get(data->seen, bitmap_pos))
 447                return 0;
 448
 449        if (bitmap_get(data->base, bitmap_pos))
 450                return 0;
 451
 452        hash_pos = kh_get_sha1(bitmap_git.bitmaps, sha1);
 453        if (hash_pos < kh_end(bitmap_git.bitmaps)) {
 454                struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, hash_pos);
 455                bitmap_or_ewah(data->base, lookup_stored_bitmap(st));
 456                return 0;
 457        }
 458
 459        bitmap_set(data->base, bitmap_pos);
 460        return 1;
 461}
 462
 463static int should_include(struct commit *commit, void *_data)
 464{
 465        struct include_data *data = _data;
 466        int bitmap_pos;
 467
 468        bitmap_pos = bitmap_position(commit->object.sha1);
 469        if (bitmap_pos < 0)
 470                bitmap_pos = ext_index_add_object((struct object *)commit, NULL);
 471
 472        if (!add_to_include_set(data, commit->object.sha1, bitmap_pos)) {
 473                struct commit_list *parent = commit->parents;
 474
 475                while (parent) {
 476                        parent->item->object.flags |= SEEN;
 477                        parent = parent->next;
 478                }
 479
 480                return 0;
 481        }
 482
 483        return 1;
 484}
 485
 486static struct bitmap *find_objects(struct rev_info *revs,
 487                                   struct object_list *roots,
 488                                   struct bitmap *seen)
 489{
 490        struct bitmap *base = NULL;
 491        int needs_walk = 0;
 492
 493        struct object_list *not_mapped = NULL;
 494
 495        /*
 496         * Go through all the roots for the walk. The ones that have bitmaps
 497         * on the bitmap index will be `or`ed together to form an initial
 498         * global reachability analysis.
 499         *
 500         * The ones without bitmaps in the index will be stored in the
 501         * `not_mapped_list` for further processing.
 502         */
 503        while (roots) {
 504                struct object *object = roots->item;
 505                roots = roots->next;
 506
 507                if (object->type == OBJ_COMMIT) {
 508                        khiter_t pos = kh_get_sha1(bitmap_git.bitmaps, object->sha1);
 509
 510                        if (pos < kh_end(bitmap_git.bitmaps)) {
 511                                struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
 512                                struct ewah_bitmap *or_with = lookup_stored_bitmap(st);
 513
 514                                if (base == NULL)
 515                                        base = ewah_to_bitmap(or_with);
 516                                else
 517                                        bitmap_or_ewah(base, or_with);
 518
 519                                object->flags |= SEEN;
 520                                continue;
 521                        }
 522                }
 523
 524                object_list_insert(object, &not_mapped);
 525        }
 526
 527        /*
 528         * Best case scenario: We found bitmaps for all the roots,
 529         * so the resulting `or` bitmap has the full reachability analysis
 530         */
 531        if (not_mapped == NULL)
 532                return base;
 533
 534        roots = not_mapped;
 535
 536        /*
 537         * Let's iterate through all the roots that don't have bitmaps to
 538         * check if we can determine them to be reachable from the existing
 539         * global bitmap.
 540         *
 541         * If we cannot find them in the existing global bitmap, we'll need
 542         * to push them to an actual walk and run it until we can confirm
 543         * they are reachable
 544         */
 545        while (roots) {
 546                struct object *object = roots->item;
 547                int pos;
 548
 549                roots = roots->next;
 550                pos = bitmap_position(object->sha1);
 551
 552                if (pos < 0 || base == NULL || !bitmap_get(base, pos)) {
 553                        object->flags &= ~UNINTERESTING;
 554                        add_pending_object(revs, object, "");
 555                        needs_walk = 1;
 556                } else {
 557                        object->flags |= SEEN;
 558                }
 559        }
 560
 561        if (needs_walk) {
 562                struct include_data incdata;
 563
 564                if (base == NULL)
 565                        base = bitmap_new();
 566
 567                incdata.base = base;
 568                incdata.seen = seen;
 569
 570                revs->include_check = should_include;
 571                revs->include_check_data = &incdata;
 572
 573                if (prepare_revision_walk(revs))
 574                        die("revision walk setup failed");
 575
 576                traverse_commit_list(revs, show_commit, show_object, base);
 577        }
 578
 579        return base;
 580}
 581
 582static void show_extended_objects(struct bitmap *objects,
 583                                  show_reachable_fn show_reach)
 584{
 585        struct eindex *eindex = &bitmap_git.ext_index;
 586        uint32_t i;
 587
 588        for (i = 0; i < eindex->count; ++i) {
 589                struct object *obj;
 590
 591                if (!bitmap_get(objects, bitmap_git.pack->num_objects + i))
 592                        continue;
 593
 594                obj = eindex->objects[i];
 595                show_reach(obj->sha1, obj->type, 0, eindex->hashes[i], NULL, 0);
 596        }
 597}
 598
 599static void show_objects_for_type(
 600        struct bitmap *objects,
 601        struct ewah_bitmap *type_filter,
 602        enum object_type object_type,
 603        show_reachable_fn show_reach)
 604{
 605        size_t pos = 0, i = 0;
 606        uint32_t offset;
 607
 608        struct ewah_iterator it;
 609        eword_t filter;
 610
 611        if (bitmap_git.reuse_objects == bitmap_git.pack->num_objects)
 612                return;
 613
 614        ewah_iterator_init(&it, type_filter);
 615
 616        while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
 617                eword_t word = objects->words[i] & filter;
 618
 619                for (offset = 0; offset < BITS_IN_WORD; ++offset) {
 620                        const unsigned char *sha1;
 621                        struct revindex_entry *entry;
 622                        uint32_t hash = 0;
 623
 624                        if ((word >> offset) == 0)
 625                                break;
 626
 627                        offset += ewah_bit_ctz64(word >> offset);
 628
 629                        if (pos + offset < bitmap_git.reuse_objects)
 630                                continue;
 631
 632                        entry = &bitmap_git.reverse_index->revindex[pos + offset];
 633                        sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
 634
 635                        if (bitmap_git.hashes)
 636                                hash = ntohl(bitmap_git.hashes[entry->nr]);
 637
 638                        show_reach(sha1, object_type, 0, hash, bitmap_git.pack, entry->offset);
 639                }
 640
 641                pos += BITS_IN_WORD;
 642                i++;
 643        }
 644}
 645
 646static int in_bitmapped_pack(struct object_list *roots)
 647{
 648        while (roots) {
 649                struct object *object = roots->item;
 650                roots = roots->next;
 651
 652                if (find_pack_entry_one(object->sha1, bitmap_git.pack) > 0)
 653                        return 1;
 654        }
 655
 656        return 0;
 657}
 658
 659int prepare_bitmap_walk(struct rev_info *revs)
 660{
 661        unsigned int i;
 662        unsigned int pending_nr = revs->pending.nr;
 663        struct object_array_entry *pending_e = revs->pending.objects;
 664
 665        struct object_list *wants = NULL;
 666        struct object_list *haves = NULL;
 667
 668        struct bitmap *wants_bitmap = NULL;
 669        struct bitmap *haves_bitmap = NULL;
 670
 671        if (!bitmap_git.loaded) {
 672                /* try to open a bitmapped pack, but don't parse it yet
 673                 * because we may not need to use it */
 674                if (open_pack_bitmap() < 0)
 675                        return -1;
 676        }
 677
 678        for (i = 0; i < pending_nr; ++i) {
 679                struct object *object = pending_e[i].item;
 680
 681                if (object->type == OBJ_NONE)
 682                        parse_object_or_die(object->sha1, NULL);
 683
 684                while (object->type == OBJ_TAG) {
 685                        struct tag *tag = (struct tag *) object;
 686
 687                        if (object->flags & UNINTERESTING)
 688                                object_list_insert(object, &haves);
 689                        else
 690                                object_list_insert(object, &wants);
 691
 692                        if (!tag->tagged)
 693                                die("bad tag");
 694                        object = parse_object_or_die(tag->tagged->sha1, NULL);
 695                }
 696
 697                if (object->flags & UNINTERESTING)
 698                        object_list_insert(object, &haves);
 699                else
 700                        object_list_insert(object, &wants);
 701        }
 702
 703        /*
 704         * if we have a HAVES list, but none of those haves is contained
 705         * in the packfile that has a bitmap, we don't have anything to
 706         * optimize here
 707         */
 708        if (haves && !in_bitmapped_pack(haves))
 709                return -1;
 710
 711        /* if we don't want anything, we're done here */
 712        if (!wants)
 713                return -1;
 714
 715        /*
 716         * now we're going to use bitmaps, so load the actual bitmap entries
 717         * from disk. this is the point of no return; after this the rev_list
 718         * becomes invalidated and we must perform the revwalk through bitmaps
 719         */
 720        if (!bitmap_git.loaded && load_pack_bitmap() < 0)
 721                return -1;
 722
 723        revs->pending.nr = 0;
 724        revs->pending.alloc = 0;
 725        revs->pending.objects = NULL;
 726
 727        if (haves) {
 728                revs->ignore_missing_links = 1;
 729                haves_bitmap = find_objects(revs, haves, NULL);
 730                reset_revision_walk();
 731                revs->ignore_missing_links = 0;
 732
 733                if (haves_bitmap == NULL)
 734                        die("BUG: failed to perform bitmap walk");
 735        }
 736
 737        wants_bitmap = find_objects(revs, wants, haves_bitmap);
 738
 739        if (!wants_bitmap)
 740                die("BUG: failed to perform bitmap walk");
 741
 742        if (haves_bitmap)
 743                bitmap_and_not(wants_bitmap, haves_bitmap);
 744
 745        bitmap_git.result = wants_bitmap;
 746
 747        bitmap_free(haves_bitmap);
 748        return 0;
 749}
 750
 751int reuse_partial_packfile_from_bitmap(struct packed_git **packfile,
 752                                       uint32_t *entries,
 753                                       off_t *up_to)
 754{
 755        /*
 756         * Reuse the packfile content if we need more than
 757         * 90% of its objects
 758         */
 759        static const double REUSE_PERCENT = 0.9;
 760
 761        struct bitmap *result = bitmap_git.result;
 762        uint32_t reuse_threshold;
 763        uint32_t i, reuse_objects = 0;
 764
 765        assert(result);
 766
 767        for (i = 0; i < result->word_alloc; ++i) {
 768                if (result->words[i] != (eword_t)~0) {
 769                        reuse_objects += ewah_bit_ctz64(~result->words[i]);
 770                        break;
 771                }
 772
 773                reuse_objects += BITS_IN_WORD;
 774        }
 775
 776#ifdef GIT_BITMAP_DEBUG
 777        {
 778                const unsigned char *sha1;
 779                struct revindex_entry *entry;
 780
 781                entry = &bitmap_git.reverse_index->revindex[reuse_objects];
 782                sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
 783
 784                fprintf(stderr, "Failed to reuse at %d (%016llx)\n",
 785                        reuse_objects, result->words[i]);
 786                fprintf(stderr, " %s\n", sha1_to_hex(sha1));
 787        }
 788#endif
 789
 790        if (!reuse_objects)
 791                return -1;
 792
 793        if (reuse_objects >= bitmap_git.pack->num_objects) {
 794                bitmap_git.reuse_objects = *entries = bitmap_git.pack->num_objects;
 795                *up_to = -1; /* reuse the full pack */
 796                *packfile = bitmap_git.pack;
 797                return 0;
 798        }
 799
 800        reuse_threshold = bitmap_popcount(bitmap_git.result) * REUSE_PERCENT;
 801
 802        if (reuse_objects < reuse_threshold)
 803                return -1;
 804
 805        bitmap_git.reuse_objects = *entries = reuse_objects;
 806        *up_to = bitmap_git.reverse_index->revindex[reuse_objects].offset;
 807        *packfile = bitmap_git.pack;
 808
 809        return 0;
 810}
 811
 812void traverse_bitmap_commit_list(show_reachable_fn show_reachable)
 813{
 814        assert(bitmap_git.result);
 815
 816        show_objects_for_type(bitmap_git.result, bitmap_git.commits,
 817                OBJ_COMMIT, show_reachable);
 818        show_objects_for_type(bitmap_git.result, bitmap_git.trees,
 819                OBJ_TREE, show_reachable);
 820        show_objects_for_type(bitmap_git.result, bitmap_git.blobs,
 821                OBJ_BLOB, show_reachable);
 822        show_objects_for_type(bitmap_git.result, bitmap_git.tags,
 823                OBJ_TAG, show_reachable);
 824
 825        show_extended_objects(bitmap_git.result, show_reachable);
 826
 827        bitmap_free(bitmap_git.result);
 828        bitmap_git.result = NULL;
 829}
 830
 831static uint32_t count_object_type(struct bitmap *objects,
 832                                  enum object_type type)
 833{
 834        struct eindex *eindex = &bitmap_git.ext_index;
 835
 836        uint32_t i = 0, count = 0;
 837        struct ewah_iterator it;
 838        eword_t filter;
 839
 840        switch (type) {
 841        case OBJ_COMMIT:
 842                ewah_iterator_init(&it, bitmap_git.commits);
 843                break;
 844
 845        case OBJ_TREE:
 846                ewah_iterator_init(&it, bitmap_git.trees);
 847                break;
 848
 849        case OBJ_BLOB:
 850                ewah_iterator_init(&it, bitmap_git.blobs);
 851                break;
 852
 853        case OBJ_TAG:
 854                ewah_iterator_init(&it, bitmap_git.tags);
 855                break;
 856
 857        default:
 858                return 0;
 859        }
 860
 861        while (i < objects->word_alloc && ewah_iterator_next(&filter, &it)) {
 862                eword_t word = objects->words[i++] & filter;
 863                count += ewah_bit_popcount64(word);
 864        }
 865
 866        for (i = 0; i < eindex->count; ++i) {
 867                if (eindex->objects[i]->type == type &&
 868                        bitmap_get(objects, bitmap_git.pack->num_objects + i))
 869                        count++;
 870        }
 871
 872        return count;
 873}
 874
 875void count_bitmap_commit_list(uint32_t *commits, uint32_t *trees,
 876                              uint32_t *blobs, uint32_t *tags)
 877{
 878        assert(bitmap_git.result);
 879
 880        if (commits)
 881                *commits = count_object_type(bitmap_git.result, OBJ_COMMIT);
 882
 883        if (trees)
 884                *trees = count_object_type(bitmap_git.result, OBJ_TREE);
 885
 886        if (blobs)
 887                *blobs = count_object_type(bitmap_git.result, OBJ_BLOB);
 888
 889        if (tags)
 890                *tags = count_object_type(bitmap_git.result, OBJ_TAG);
 891}
 892
 893struct bitmap_test_data {
 894        struct bitmap *base;
 895        struct progress *prg;
 896        size_t seen;
 897};
 898
 899static void test_show_object(struct object *object,
 900                             const struct name_path *path,
 901                             const char *last, void *data)
 902{
 903        struct bitmap_test_data *tdata = data;
 904        int bitmap_pos;
 905
 906        bitmap_pos = bitmap_position(object->sha1);
 907        if (bitmap_pos < 0)
 908                die("Object not in bitmap: %s\n", sha1_to_hex(object->sha1));
 909
 910        bitmap_set(tdata->base, bitmap_pos);
 911        display_progress(tdata->prg, ++tdata->seen);
 912}
 913
 914static void test_show_commit(struct commit *commit, void *data)
 915{
 916        struct bitmap_test_data *tdata = data;
 917        int bitmap_pos;
 918
 919        bitmap_pos = bitmap_position(commit->object.sha1);
 920        if (bitmap_pos < 0)
 921                die("Object not in bitmap: %s\n", sha1_to_hex(commit->object.sha1));
 922
 923        bitmap_set(tdata->base, bitmap_pos);
 924        display_progress(tdata->prg, ++tdata->seen);
 925}
 926
 927void test_bitmap_walk(struct rev_info *revs)
 928{
 929        struct object *root;
 930        struct bitmap *result = NULL;
 931        khiter_t pos;
 932        size_t result_popcnt;
 933        struct bitmap_test_data tdata;
 934
 935        if (prepare_bitmap_git())
 936                die("failed to load bitmap indexes");
 937
 938        if (revs->pending.nr != 1)
 939                die("you must specify exactly one commit to test");
 940
 941        fprintf(stderr, "Bitmap v%d test (%d entries loaded)\n",
 942                bitmap_git.version, bitmap_git.entry_count);
 943
 944        root = revs->pending.objects[0].item;
 945        pos = kh_get_sha1(bitmap_git.bitmaps, root->sha1);
 946
 947        if (pos < kh_end(bitmap_git.bitmaps)) {
 948                struct stored_bitmap *st = kh_value(bitmap_git.bitmaps, pos);
 949                struct ewah_bitmap *bm = lookup_stored_bitmap(st);
 950
 951                fprintf(stderr, "Found bitmap for %s. %d bits / %08x checksum\n",
 952                        sha1_to_hex(root->sha1), (int)bm->bit_size, ewah_checksum(bm));
 953
 954                result = ewah_to_bitmap(bm);
 955        }
 956
 957        if (result == NULL)
 958                die("Commit %s doesn't have an indexed bitmap", sha1_to_hex(root->sha1));
 959
 960        revs->tag_objects = 1;
 961        revs->tree_objects = 1;
 962        revs->blob_objects = 1;
 963
 964        result_popcnt = bitmap_popcount(result);
 965
 966        if (prepare_revision_walk(revs))
 967                die("revision walk setup failed");
 968
 969        tdata.base = bitmap_new();
 970        tdata.prg = start_progress("Verifying bitmap entries", result_popcnt);
 971        tdata.seen = 0;
 972
 973        traverse_commit_list(revs, &test_show_commit, &test_show_object, &tdata);
 974
 975        stop_progress(&tdata.prg);
 976
 977        if (bitmap_equals(result, tdata.base))
 978                fprintf(stderr, "OK!\n");
 979        else
 980                fprintf(stderr, "Mismatch!\n");
 981}
 982
 983static int rebuild_bitmap(uint32_t *reposition,
 984                          struct ewah_bitmap *source,
 985                          struct bitmap *dest)
 986{
 987        uint32_t pos = 0;
 988        struct ewah_iterator it;
 989        eword_t word;
 990
 991        ewah_iterator_init(&it, source);
 992
 993        while (ewah_iterator_next(&word, &it)) {
 994                uint32_t offset, bit_pos;
 995
 996                for (offset = 0; offset < BITS_IN_WORD; ++offset) {
 997                        if ((word >> offset) == 0)
 998                                break;
 999
1000                        offset += ewah_bit_ctz64(word >> offset);
1001
1002                        bit_pos = reposition[pos + offset];
1003                        if (bit_pos > 0)
1004                                bitmap_set(dest, bit_pos - 1);
1005                        else /* can't reuse, we don't have the object */
1006                                return -1;
1007                }
1008
1009                pos += BITS_IN_WORD;
1010        }
1011        return 0;
1012}
1013
1014int rebuild_existing_bitmaps(struct packing_data *mapping,
1015                             khash_sha1 *reused_bitmaps,
1016                             int show_progress)
1017{
1018        uint32_t i, num_objects;
1019        uint32_t *reposition;
1020        struct bitmap *rebuild;
1021        struct stored_bitmap *stored;
1022        struct progress *progress = NULL;
1023
1024        khiter_t hash_pos;
1025        int hash_ret;
1026
1027        if (prepare_bitmap_git() < 0)
1028                return -1;
1029
1030        num_objects = bitmap_git.pack->num_objects;
1031        reposition = xcalloc(num_objects, sizeof(uint32_t));
1032
1033        for (i = 0; i < num_objects; ++i) {
1034                const unsigned char *sha1;
1035                struct revindex_entry *entry;
1036                struct object_entry *oe;
1037
1038                entry = &bitmap_git.reverse_index->revindex[i];
1039                sha1 = nth_packed_object_sha1(bitmap_git.pack, entry->nr);
1040                oe = packlist_find(mapping, sha1, NULL);
1041
1042                if (oe)
1043                        reposition[i] = oe->in_pack_pos + 1;
1044        }
1045
1046        rebuild = bitmap_new();
1047        i = 0;
1048
1049        if (show_progress)
1050                progress = start_progress("Reusing bitmaps", 0);
1051
1052        kh_foreach_value(bitmap_git.bitmaps, stored, {
1053                if (stored->flags & BITMAP_FLAG_REUSE) {
1054                        if (!rebuild_bitmap(reposition,
1055                                            lookup_stored_bitmap(stored),
1056                                            rebuild)) {
1057                                hash_pos = kh_put_sha1(reused_bitmaps,
1058                                                       stored->sha1,
1059                                                       &hash_ret);
1060                                kh_value(reused_bitmaps, hash_pos) =
1061                                        bitmap_to_ewah(rebuild);
1062                        }
1063                        bitmap_reset(rebuild);
1064                        display_progress(progress, ++i);
1065                }
1066        });
1067
1068        stop_progress(&progress);
1069
1070        free(reposition);
1071        bitmap_free(rebuild);
1072        return 0;
1073}