commit-graph.con commit list-objects-filter: teach tree:# how to handle >0 (c813a7c)
   1#include "cache.h"
   2#include "config.h"
   3#include "dir.h"
   4#include "git-compat-util.h"
   5#include "lockfile.h"
   6#include "pack.h"
   7#include "packfile.h"
   8#include "commit.h"
   9#include "object.h"
  10#include "refs.h"
  11#include "revision.h"
  12#include "sha1-lookup.h"
  13#include "commit-graph.h"
  14#include "object-store.h"
  15#include "alloc.h"
  16#include "hashmap.h"
  17#include "replace-object.h"
  18#include "progress.h"
  19
  20#define GRAPH_SIGNATURE 0x43475048 /* "CGPH" */
  21#define GRAPH_CHUNKID_OIDFANOUT 0x4f494446 /* "OIDF" */
  22#define GRAPH_CHUNKID_OIDLOOKUP 0x4f49444c /* "OIDL" */
  23#define GRAPH_CHUNKID_DATA 0x43444154 /* "CDAT" */
  24#define GRAPH_CHUNKID_LARGEEDGES 0x45444745 /* "EDGE" */
  25
  26#define GRAPH_DATA_WIDTH 36
  27
  28#define GRAPH_VERSION_1 0x1
  29#define GRAPH_VERSION GRAPH_VERSION_1
  30
  31#define GRAPH_OID_VERSION_SHA1 1
  32#define GRAPH_OID_LEN_SHA1 GIT_SHA1_RAWSZ
  33#define GRAPH_OID_VERSION GRAPH_OID_VERSION_SHA1
  34#define GRAPH_OID_LEN GRAPH_OID_LEN_SHA1
  35
  36#define GRAPH_OCTOPUS_EDGES_NEEDED 0x80000000
  37#define GRAPH_PARENT_MISSING 0x7fffffff
  38#define GRAPH_EDGE_LAST_MASK 0x7fffffff
  39#define GRAPH_PARENT_NONE 0x70000000
  40
  41#define GRAPH_LAST_EDGE 0x80000000
  42
  43#define GRAPH_HEADER_SIZE 8
  44#define GRAPH_FANOUT_SIZE (4 * 256)
  45#define GRAPH_CHUNKLOOKUP_WIDTH 12
  46#define GRAPH_MIN_SIZE (GRAPH_HEADER_SIZE + 4 * GRAPH_CHUNKLOOKUP_WIDTH \
  47                        + GRAPH_FANOUT_SIZE + GRAPH_OID_LEN)
  48
  49char *get_commit_graph_filename(const char *obj_dir)
  50{
  51        return xstrfmt("%s/info/commit-graph", obj_dir);
  52}
  53
  54static struct commit_graph *alloc_commit_graph(void)
  55{
  56        struct commit_graph *g = xcalloc(1, sizeof(*g));
  57        g->graph_fd = -1;
  58
  59        return g;
  60}
  61
  62extern int read_replace_refs;
  63
  64static int commit_graph_compatible(struct repository *r)
  65{
  66        if (!r->gitdir)
  67                return 0;
  68
  69        if (read_replace_refs) {
  70                prepare_replace_object(r);
  71                if (hashmap_get_size(&r->objects->replace_map->map))
  72                        return 0;
  73        }
  74
  75        prepare_commit_graft(r);
  76        if (r->parsed_objects && r->parsed_objects->grafts_nr)
  77                return 0;
  78        if (is_repository_shallow(r))
  79                return 0;
  80
  81        return 1;
  82}
  83
  84struct commit_graph *load_commit_graph_one(const char *graph_file)
  85{
  86        void *graph_map;
  87        const unsigned char *data, *chunk_lookup;
  88        size_t graph_size;
  89        struct stat st;
  90        uint32_t i;
  91        struct commit_graph *graph;
  92        int fd = git_open(graph_file);
  93        uint64_t last_chunk_offset;
  94        uint32_t last_chunk_id;
  95        uint32_t graph_signature;
  96        unsigned char graph_version, hash_version;
  97
  98        if (fd < 0)
  99                return NULL;
 100        if (fstat(fd, &st)) {
 101                close(fd);
 102                return NULL;
 103        }
 104        graph_size = xsize_t(st.st_size);
 105
 106        if (graph_size < GRAPH_MIN_SIZE) {
 107                close(fd);
 108                die(_("graph file %s is too small"), graph_file);
 109        }
 110        graph_map = xmmap(NULL, graph_size, PROT_READ, MAP_PRIVATE, fd, 0);
 111        data = (const unsigned char *)graph_map;
 112
 113        graph_signature = get_be32(data);
 114        if (graph_signature != GRAPH_SIGNATURE) {
 115                error(_("graph signature %X does not match signature %X"),
 116                      graph_signature, GRAPH_SIGNATURE);
 117                goto cleanup_fail;
 118        }
 119
 120        graph_version = *(unsigned char*)(data + 4);
 121        if (graph_version != GRAPH_VERSION) {
 122                error(_("graph version %X does not match version %X"),
 123                      graph_version, GRAPH_VERSION);
 124                goto cleanup_fail;
 125        }
 126
 127        hash_version = *(unsigned char*)(data + 5);
 128        if (hash_version != GRAPH_OID_VERSION) {
 129                error(_("hash version %X does not match version %X"),
 130                      hash_version, GRAPH_OID_VERSION);
 131                goto cleanup_fail;
 132        }
 133
 134        graph = alloc_commit_graph();
 135
 136        graph->hash_len = GRAPH_OID_LEN;
 137        graph->num_chunks = *(unsigned char*)(data + 6);
 138        graph->graph_fd = fd;
 139        graph->data = graph_map;
 140        graph->data_len = graph_size;
 141
 142        last_chunk_id = 0;
 143        last_chunk_offset = 8;
 144        chunk_lookup = data + 8;
 145        for (i = 0; i < graph->num_chunks; i++) {
 146                uint32_t chunk_id = get_be32(chunk_lookup + 0);
 147                uint64_t chunk_offset = get_be64(chunk_lookup + 4);
 148                int chunk_repeated = 0;
 149
 150                chunk_lookup += GRAPH_CHUNKLOOKUP_WIDTH;
 151
 152                if (chunk_offset > graph_size - GIT_MAX_RAWSZ) {
 153                        error(_("improper chunk offset %08x%08x"), (uint32_t)(chunk_offset >> 32),
 154                              (uint32_t)chunk_offset);
 155                        goto cleanup_fail;
 156                }
 157
 158                switch (chunk_id) {
 159                case GRAPH_CHUNKID_OIDFANOUT:
 160                        if (graph->chunk_oid_fanout)
 161                                chunk_repeated = 1;
 162                        else
 163                                graph->chunk_oid_fanout = (uint32_t*)(data + chunk_offset);
 164                        break;
 165
 166                case GRAPH_CHUNKID_OIDLOOKUP:
 167                        if (graph->chunk_oid_lookup)
 168                                chunk_repeated = 1;
 169                        else
 170                                graph->chunk_oid_lookup = data + chunk_offset;
 171                        break;
 172
 173                case GRAPH_CHUNKID_DATA:
 174                        if (graph->chunk_commit_data)
 175                                chunk_repeated = 1;
 176                        else
 177                                graph->chunk_commit_data = data + chunk_offset;
 178                        break;
 179
 180                case GRAPH_CHUNKID_LARGEEDGES:
 181                        if (graph->chunk_large_edges)
 182                                chunk_repeated = 1;
 183                        else
 184                                graph->chunk_large_edges = data + chunk_offset;
 185                        break;
 186                }
 187
 188                if (chunk_repeated) {
 189                        error(_("chunk id %08x appears multiple times"), chunk_id);
 190                        goto cleanup_fail;
 191                }
 192
 193                if (last_chunk_id == GRAPH_CHUNKID_OIDLOOKUP)
 194                {
 195                        graph->num_commits = (chunk_offset - last_chunk_offset)
 196                                             / graph->hash_len;
 197                }
 198
 199                last_chunk_id = chunk_id;
 200                last_chunk_offset = chunk_offset;
 201        }
 202
 203        return graph;
 204
 205cleanup_fail:
 206        munmap(graph_map, graph_size);
 207        close(fd);
 208        exit(1);
 209}
 210
 211static void prepare_commit_graph_one(struct repository *r, const char *obj_dir)
 212{
 213        char *graph_name;
 214
 215        if (r->objects->commit_graph)
 216                return;
 217
 218        graph_name = get_commit_graph_filename(obj_dir);
 219        r->objects->commit_graph =
 220                load_commit_graph_one(graph_name);
 221
 222        FREE_AND_NULL(graph_name);
 223}
 224
 225/*
 226 * Return 1 if commit_graph is non-NULL, and 0 otherwise.
 227 *
 228 * On the first invocation, this function attemps to load the commit
 229 * graph if the_repository is configured to have one.
 230 */
 231static int prepare_commit_graph(struct repository *r)
 232{
 233        struct alternate_object_database *alt;
 234        char *obj_dir;
 235        int config_value;
 236
 237        if (r->objects->commit_graph_attempted)
 238                return !!r->objects->commit_graph;
 239        r->objects->commit_graph_attempted = 1;
 240
 241        if (!git_env_bool(GIT_TEST_COMMIT_GRAPH, 0) &&
 242            (repo_config_get_bool(r, "core.commitgraph", &config_value) ||
 243            !config_value))
 244                /*
 245                 * This repository is not configured to use commit graphs, so
 246                 * do not load one. (But report commit_graph_attempted anyway
 247                 * so that commit graph loading is not attempted again for this
 248                 * repository.)
 249                 */
 250                return 0;
 251
 252        if (!commit_graph_compatible(r))
 253                return 0;
 254
 255        obj_dir = r->objects->objectdir;
 256        prepare_commit_graph_one(r, obj_dir);
 257        prepare_alt_odb(r);
 258        for (alt = r->objects->alt_odb_list;
 259             !r->objects->commit_graph && alt;
 260             alt = alt->next)
 261                prepare_commit_graph_one(r, alt->path);
 262        return !!r->objects->commit_graph;
 263}
 264
 265int generation_numbers_enabled(struct repository *r)
 266{
 267        uint32_t first_generation;
 268        struct commit_graph *g;
 269        if (!prepare_commit_graph(r))
 270               return 0;
 271
 272        g = r->objects->commit_graph;
 273
 274        if (!g->num_commits)
 275                return 0;
 276
 277        first_generation = get_be32(g->chunk_commit_data +
 278                                    g->hash_len + 8) >> 2;
 279
 280        return !!first_generation;
 281}
 282
 283void close_commit_graph(struct repository *r)
 284{
 285        free_commit_graph(r->objects->commit_graph);
 286        r->objects->commit_graph = NULL;
 287}
 288
 289static int bsearch_graph(struct commit_graph *g, struct object_id *oid, uint32_t *pos)
 290{
 291        return bsearch_hash(oid->hash, g->chunk_oid_fanout,
 292                            g->chunk_oid_lookup, g->hash_len, pos);
 293}
 294
 295static struct commit_list **insert_parent_or_die(struct repository *r,
 296                                                 struct commit_graph *g,
 297                                                 uint64_t pos,
 298                                                 struct commit_list **pptr)
 299{
 300        struct commit *c;
 301        struct object_id oid;
 302
 303        if (pos >= g->num_commits)
 304                die("invalid parent position %"PRIu64, pos);
 305
 306        hashcpy(oid.hash, g->chunk_oid_lookup + g->hash_len * pos);
 307        c = lookup_commit(r, &oid);
 308        if (!c)
 309                die(_("could not find commit %s"), oid_to_hex(&oid));
 310        c->graph_pos = pos;
 311        return &commit_list_insert(c, pptr)->next;
 312}
 313
 314static void fill_commit_graph_info(struct commit *item, struct commit_graph *g, uint32_t pos)
 315{
 316        const unsigned char *commit_data = g->chunk_commit_data + GRAPH_DATA_WIDTH * pos;
 317        item->graph_pos = pos;
 318        item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
 319}
 320
 321static int fill_commit_in_graph(struct repository *r,
 322                                struct commit *item,
 323                                struct commit_graph *g, uint32_t pos)
 324{
 325        uint32_t edge_value;
 326        uint32_t *parent_data_ptr;
 327        uint64_t date_low, date_high;
 328        struct commit_list **pptr;
 329        const unsigned char *commit_data = g->chunk_commit_data + (g->hash_len + 16) * pos;
 330
 331        item->object.parsed = 1;
 332        item->graph_pos = pos;
 333
 334        item->maybe_tree = NULL;
 335
 336        date_high = get_be32(commit_data + g->hash_len + 8) & 0x3;
 337        date_low = get_be32(commit_data + g->hash_len + 12);
 338        item->date = (timestamp_t)((date_high << 32) | date_low);
 339
 340        item->generation = get_be32(commit_data + g->hash_len + 8) >> 2;
 341
 342        pptr = &item->parents;
 343
 344        edge_value = get_be32(commit_data + g->hash_len);
 345        if (edge_value == GRAPH_PARENT_NONE)
 346                return 1;
 347        pptr = insert_parent_or_die(r, g, edge_value, pptr);
 348
 349        edge_value = get_be32(commit_data + g->hash_len + 4);
 350        if (edge_value == GRAPH_PARENT_NONE)
 351                return 1;
 352        if (!(edge_value & GRAPH_OCTOPUS_EDGES_NEEDED)) {
 353                pptr = insert_parent_or_die(r, g, edge_value, pptr);
 354                return 1;
 355        }
 356
 357        parent_data_ptr = (uint32_t*)(g->chunk_large_edges +
 358                          4 * (uint64_t)(edge_value & GRAPH_EDGE_LAST_MASK));
 359        do {
 360                edge_value = get_be32(parent_data_ptr);
 361                pptr = insert_parent_or_die(r, g,
 362                                            edge_value & GRAPH_EDGE_LAST_MASK,
 363                                            pptr);
 364                parent_data_ptr++;
 365        } while (!(edge_value & GRAPH_LAST_EDGE));
 366
 367        return 1;
 368}
 369
 370static int find_commit_in_graph(struct commit *item, struct commit_graph *g, uint32_t *pos)
 371{
 372        if (item->graph_pos != COMMIT_NOT_FROM_GRAPH) {
 373                *pos = item->graph_pos;
 374                return 1;
 375        } else {
 376                return bsearch_graph(g, &(item->object.oid), pos);
 377        }
 378}
 379
 380static int parse_commit_in_graph_one(struct repository *r,
 381                                     struct commit_graph *g,
 382                                     struct commit *item)
 383{
 384        uint32_t pos;
 385
 386        if (item->object.parsed)
 387                return 1;
 388
 389        if (find_commit_in_graph(item, g, &pos))
 390                return fill_commit_in_graph(r, item, g, pos);
 391
 392        return 0;
 393}
 394
 395int parse_commit_in_graph(struct repository *r, struct commit *item)
 396{
 397        if (!prepare_commit_graph(r))
 398                return 0;
 399        return parse_commit_in_graph_one(r, r->objects->commit_graph, item);
 400}
 401
 402void load_commit_graph_info(struct repository *r, struct commit *item)
 403{
 404        uint32_t pos;
 405        if (!prepare_commit_graph(r))
 406                return;
 407        if (find_commit_in_graph(item, r->objects->commit_graph, &pos))
 408                fill_commit_graph_info(item, r->objects->commit_graph, pos);
 409}
 410
 411static struct tree *load_tree_for_commit(struct repository *r,
 412                                         struct commit_graph *g,
 413                                         struct commit *c)
 414{
 415        struct object_id oid;
 416        const unsigned char *commit_data = g->chunk_commit_data +
 417                                           GRAPH_DATA_WIDTH * (c->graph_pos);
 418
 419        hashcpy(oid.hash, commit_data);
 420        c->maybe_tree = lookup_tree(r, &oid);
 421
 422        return c->maybe_tree;
 423}
 424
 425static struct tree *get_commit_tree_in_graph_one(struct repository *r,
 426                                                 struct commit_graph *g,
 427                                                 const struct commit *c)
 428{
 429        if (c->maybe_tree)
 430                return c->maybe_tree;
 431        if (c->graph_pos == COMMIT_NOT_FROM_GRAPH)
 432                BUG("get_commit_tree_in_graph_one called from non-commit-graph commit");
 433
 434        return load_tree_for_commit(r, g, (struct commit *)c);
 435}
 436
 437struct tree *get_commit_tree_in_graph(struct repository *r, const struct commit *c)
 438{
 439        return get_commit_tree_in_graph_one(r, r->objects->commit_graph, c);
 440}
 441
 442static void write_graph_chunk_fanout(struct hashfile *f,
 443                                     struct commit **commits,
 444                                     int nr_commits)
 445{
 446        int i, count = 0;
 447        struct commit **list = commits;
 448
 449        /*
 450         * Write the first-level table (the list is sorted,
 451         * but we use a 256-entry lookup to be able to avoid
 452         * having to do eight extra binary search iterations).
 453         */
 454        for (i = 0; i < 256; i++) {
 455                while (count < nr_commits) {
 456                        if ((*list)->object.oid.hash[0] != i)
 457                                break;
 458                        count++;
 459                        list++;
 460                }
 461
 462                hashwrite_be32(f, count);
 463        }
 464}
 465
 466static void write_graph_chunk_oids(struct hashfile *f, int hash_len,
 467                                   struct commit **commits, int nr_commits)
 468{
 469        struct commit **list = commits;
 470        int count;
 471        for (count = 0; count < nr_commits; count++, list++)
 472                hashwrite(f, (*list)->object.oid.hash, (int)hash_len);
 473}
 474
 475static const unsigned char *commit_to_sha1(size_t index, void *table)
 476{
 477        struct commit **commits = table;
 478        return commits[index]->object.oid.hash;
 479}
 480
 481static void write_graph_chunk_data(struct hashfile *f, int hash_len,
 482                                   struct commit **commits, int nr_commits)
 483{
 484        struct commit **list = commits;
 485        struct commit **last = commits + nr_commits;
 486        uint32_t num_extra_edges = 0;
 487
 488        while (list < last) {
 489                struct commit_list *parent;
 490                int edge_value;
 491                uint32_t packedDate[2];
 492
 493                parse_commit(*list);
 494                hashwrite(f, get_commit_tree_oid(*list)->hash, hash_len);
 495
 496                parent = (*list)->parents;
 497
 498                if (!parent)
 499                        edge_value = GRAPH_PARENT_NONE;
 500                else {
 501                        edge_value = sha1_pos(parent->item->object.oid.hash,
 502                                              commits,
 503                                              nr_commits,
 504                                              commit_to_sha1);
 505
 506                        if (edge_value < 0)
 507                                edge_value = GRAPH_PARENT_MISSING;
 508                }
 509
 510                hashwrite_be32(f, edge_value);
 511
 512                if (parent)
 513                        parent = parent->next;
 514
 515                if (!parent)
 516                        edge_value = GRAPH_PARENT_NONE;
 517                else if (parent->next)
 518                        edge_value = GRAPH_OCTOPUS_EDGES_NEEDED | num_extra_edges;
 519                else {
 520                        edge_value = sha1_pos(parent->item->object.oid.hash,
 521                                              commits,
 522                                              nr_commits,
 523                                              commit_to_sha1);
 524                        if (edge_value < 0)
 525                                edge_value = GRAPH_PARENT_MISSING;
 526                }
 527
 528                hashwrite_be32(f, edge_value);
 529
 530                if (edge_value & GRAPH_OCTOPUS_EDGES_NEEDED) {
 531                        do {
 532                                num_extra_edges++;
 533                                parent = parent->next;
 534                        } while (parent);
 535                }
 536
 537                if (sizeof((*list)->date) > 4)
 538                        packedDate[0] = htonl(((*list)->date >> 32) & 0x3);
 539                else
 540                        packedDate[0] = 0;
 541
 542                packedDate[0] |= htonl((*list)->generation << 2);
 543
 544                packedDate[1] = htonl((*list)->date);
 545                hashwrite(f, packedDate, 8);
 546
 547                list++;
 548        }
 549}
 550
 551static void write_graph_chunk_large_edges(struct hashfile *f,
 552                                          struct commit **commits,
 553                                          int nr_commits)
 554{
 555        struct commit **list = commits;
 556        struct commit **last = commits + nr_commits;
 557        struct commit_list *parent;
 558
 559        while (list < last) {
 560                int num_parents = 0;
 561                for (parent = (*list)->parents; num_parents < 3 && parent;
 562                     parent = parent->next)
 563                        num_parents++;
 564
 565                if (num_parents <= 2) {
 566                        list++;
 567                        continue;
 568                }
 569
 570                /* Since num_parents > 2, this initializer is safe. */
 571                for (parent = (*list)->parents->next; parent; parent = parent->next) {
 572                        int edge_value = sha1_pos(parent->item->object.oid.hash,
 573                                                  commits,
 574                                                  nr_commits,
 575                                                  commit_to_sha1);
 576
 577                        if (edge_value < 0)
 578                                edge_value = GRAPH_PARENT_MISSING;
 579                        else if (!parent->next)
 580                                edge_value |= GRAPH_LAST_EDGE;
 581
 582                        hashwrite_be32(f, edge_value);
 583                }
 584
 585                list++;
 586        }
 587}
 588
 589static int commit_compare(const void *_a, const void *_b)
 590{
 591        const struct object_id *a = (const struct object_id *)_a;
 592        const struct object_id *b = (const struct object_id *)_b;
 593        return oidcmp(a, b);
 594}
 595
 596struct packed_commit_list {
 597        struct commit **list;
 598        int nr;
 599        int alloc;
 600};
 601
 602struct packed_oid_list {
 603        struct object_id *list;
 604        int nr;
 605        int alloc;
 606        struct progress *progress;
 607        int progress_done;
 608};
 609
 610static int add_packed_commits(const struct object_id *oid,
 611                              struct packed_git *pack,
 612                              uint32_t pos,
 613                              void *data)
 614{
 615        struct packed_oid_list *list = (struct packed_oid_list*)data;
 616        enum object_type type;
 617        off_t offset = nth_packed_object_offset(pack, pos);
 618        struct object_info oi = OBJECT_INFO_INIT;
 619
 620        if (list->progress)
 621                display_progress(list->progress, ++list->progress_done);
 622
 623        oi.typep = &type;
 624        if (packed_object_info(the_repository, pack, offset, &oi) < 0)
 625                die(_("unable to get type of object %s"), oid_to_hex(oid));
 626
 627        if (type != OBJ_COMMIT)
 628                return 0;
 629
 630        ALLOC_GROW(list->list, list->nr + 1, list->alloc);
 631        oidcpy(&(list->list[list->nr]), oid);
 632        list->nr++;
 633
 634        return 0;
 635}
 636
 637static void add_missing_parents(struct packed_oid_list *oids, struct commit *commit)
 638{
 639        struct commit_list *parent;
 640        for (parent = commit->parents; parent; parent = parent->next) {
 641                if (!(parent->item->object.flags & UNINTERESTING)) {
 642                        ALLOC_GROW(oids->list, oids->nr + 1, oids->alloc);
 643                        oidcpy(&oids->list[oids->nr], &(parent->item->object.oid));
 644                        oids->nr++;
 645                        parent->item->object.flags |= UNINTERESTING;
 646                }
 647        }
 648}
 649
 650static void close_reachable(struct packed_oid_list *oids, int report_progress)
 651{
 652        int i;
 653        struct commit *commit;
 654        struct progress *progress = NULL;
 655        int j = 0;
 656
 657        if (report_progress)
 658                progress = start_delayed_progress(
 659                        _("Annotating commits in commit graph"), 0);
 660        for (i = 0; i < oids->nr; i++) {
 661                display_progress(progress, ++j);
 662                commit = lookup_commit(the_repository, &oids->list[i]);
 663                if (commit)
 664                        commit->object.flags |= UNINTERESTING;
 665        }
 666
 667        /*
 668         * As this loop runs, oids->nr may grow, but not more
 669         * than the number of missing commits in the reachable
 670         * closure.
 671         */
 672        for (i = 0; i < oids->nr; i++) {
 673                display_progress(progress, ++j);
 674                commit = lookup_commit(the_repository, &oids->list[i]);
 675
 676                if (commit && !parse_commit(commit))
 677                        add_missing_parents(oids, commit);
 678        }
 679
 680        for (i = 0; i < oids->nr; i++) {
 681                display_progress(progress, ++j);
 682                commit = lookup_commit(the_repository, &oids->list[i]);
 683
 684                if (commit)
 685                        commit->object.flags &= ~UNINTERESTING;
 686        }
 687        stop_progress(&progress);
 688}
 689
 690static void compute_generation_numbers(struct packed_commit_list* commits,
 691                                       int report_progress)
 692{
 693        int i;
 694        struct commit_list *list = NULL;
 695        struct progress *progress = NULL;
 696
 697        if (report_progress)
 698                progress = start_progress(
 699                        _("Computing commit graph generation numbers"),
 700                        commits->nr);
 701        for (i = 0; i < commits->nr; i++) {
 702                display_progress(progress, i + 1);
 703                if (commits->list[i]->generation != GENERATION_NUMBER_INFINITY &&
 704                    commits->list[i]->generation != GENERATION_NUMBER_ZERO)
 705                        continue;
 706
 707                commit_list_insert(commits->list[i], &list);
 708                while (list) {
 709                        struct commit *current = list->item;
 710                        struct commit_list *parent;
 711                        int all_parents_computed = 1;
 712                        uint32_t max_generation = 0;
 713
 714                        for (parent = current->parents; parent; parent = parent->next) {
 715                                if (parent->item->generation == GENERATION_NUMBER_INFINITY ||
 716                                    parent->item->generation == GENERATION_NUMBER_ZERO) {
 717                                        all_parents_computed = 0;
 718                                        commit_list_insert(parent->item, &list);
 719                                        break;
 720                                } else if (parent->item->generation > max_generation) {
 721                                        max_generation = parent->item->generation;
 722                                }
 723                        }
 724
 725                        if (all_parents_computed) {
 726                                current->generation = max_generation + 1;
 727                                pop_commit(&list);
 728
 729                                if (current->generation > GENERATION_NUMBER_MAX)
 730                                        current->generation = GENERATION_NUMBER_MAX;
 731                        }
 732                }
 733        }
 734        stop_progress(&progress);
 735}
 736
 737static int add_ref_to_list(const char *refname,
 738                           const struct object_id *oid,
 739                           int flags, void *cb_data)
 740{
 741        struct string_list *list = (struct string_list *)cb_data;
 742
 743        string_list_append(list, oid_to_hex(oid));
 744        return 0;
 745}
 746
 747void write_commit_graph_reachable(const char *obj_dir, int append,
 748                                  int report_progress)
 749{
 750        struct string_list list = STRING_LIST_INIT_DUP;
 751
 752        for_each_ref(add_ref_to_list, &list);
 753        write_commit_graph(obj_dir, NULL, &list, append, report_progress);
 754
 755        string_list_clear(&list, 0);
 756}
 757
 758void write_commit_graph(const char *obj_dir,
 759                        struct string_list *pack_indexes,
 760                        struct string_list *commit_hex,
 761                        int append, int report_progress)
 762{
 763        struct packed_oid_list oids;
 764        struct packed_commit_list commits;
 765        struct hashfile *f;
 766        uint32_t i, count_distinct = 0;
 767        char *graph_name;
 768        struct lock_file lk = LOCK_INIT;
 769        uint32_t chunk_ids[5];
 770        uint64_t chunk_offsets[5];
 771        int num_chunks;
 772        int num_extra_edges;
 773        struct commit_list *parent;
 774        struct progress *progress = NULL;
 775
 776        if (!commit_graph_compatible(the_repository))
 777                return;
 778
 779        oids.nr = 0;
 780        oids.alloc = approximate_object_count() / 32;
 781        oids.progress = NULL;
 782        oids.progress_done = 0;
 783
 784        if (append) {
 785                prepare_commit_graph_one(the_repository, obj_dir);
 786                if (the_repository->objects->commit_graph)
 787                        oids.alloc += the_repository->objects->commit_graph->num_commits;
 788        }
 789
 790        if (oids.alloc < 1024)
 791                oids.alloc = 1024;
 792        ALLOC_ARRAY(oids.list, oids.alloc);
 793
 794        if (append && the_repository->objects->commit_graph) {
 795                struct commit_graph *commit_graph =
 796                        the_repository->objects->commit_graph;
 797                for (i = 0; i < commit_graph->num_commits; i++) {
 798                        const unsigned char *hash = commit_graph->chunk_oid_lookup +
 799                                commit_graph->hash_len * i;
 800                        hashcpy(oids.list[oids.nr++].hash, hash);
 801                }
 802        }
 803
 804        if (pack_indexes) {
 805                struct strbuf packname = STRBUF_INIT;
 806                int dirlen;
 807                strbuf_addf(&packname, "%s/pack/", obj_dir);
 808                dirlen = packname.len;
 809                if (report_progress) {
 810                        oids.progress = start_delayed_progress(
 811                                _("Finding commits for commit graph"), 0);
 812                        oids.progress_done = 0;
 813                }
 814                for (i = 0; i < pack_indexes->nr; i++) {
 815                        struct packed_git *p;
 816                        strbuf_setlen(&packname, dirlen);
 817                        strbuf_addstr(&packname, pack_indexes->items[i].string);
 818                        p = add_packed_git(packname.buf, packname.len, 1);
 819                        if (!p)
 820                                die(_("error adding pack %s"), packname.buf);
 821                        if (open_pack_index(p))
 822                                die(_("error opening index for %s"), packname.buf);
 823                        for_each_object_in_pack(p, add_packed_commits, &oids, 0);
 824                        close_pack(p);
 825                        free(p);
 826                }
 827                stop_progress(&oids.progress);
 828                strbuf_release(&packname);
 829        }
 830
 831        if (commit_hex) {
 832                if (report_progress)
 833                        progress = start_delayed_progress(
 834                                _("Finding commits for commit graph"),
 835                                commit_hex->nr);
 836                for (i = 0; i < commit_hex->nr; i++) {
 837                        const char *end;
 838                        struct object_id oid;
 839                        struct commit *result;
 840
 841                        display_progress(progress, i + 1);
 842                        if (commit_hex->items[i].string &&
 843                            parse_oid_hex(commit_hex->items[i].string, &oid, &end))
 844                                continue;
 845
 846                        result = lookup_commit_reference_gently(the_repository, &oid, 1);
 847
 848                        if (result) {
 849                                ALLOC_GROW(oids.list, oids.nr + 1, oids.alloc);
 850                                oidcpy(&oids.list[oids.nr], &(result->object.oid));
 851                                oids.nr++;
 852                        }
 853                }
 854                stop_progress(&progress);
 855        }
 856
 857        if (!pack_indexes && !commit_hex) {
 858                if (report_progress)
 859                        oids.progress = start_delayed_progress(
 860                                _("Finding commits for commit graph"), 0);
 861                for_each_packed_object(add_packed_commits, &oids, 0);
 862                stop_progress(&oids.progress);
 863        }
 864
 865        close_reachable(&oids, report_progress);
 866
 867        QSORT(oids.list, oids.nr, commit_compare);
 868
 869        count_distinct = 1;
 870        for (i = 1; i < oids.nr; i++) {
 871                if (!oideq(&oids.list[i - 1], &oids.list[i]))
 872                        count_distinct++;
 873        }
 874
 875        if (count_distinct >= GRAPH_PARENT_MISSING)
 876                die(_("the commit graph format cannot write %d commits"), count_distinct);
 877
 878        commits.nr = 0;
 879        commits.alloc = count_distinct;
 880        ALLOC_ARRAY(commits.list, commits.alloc);
 881
 882        num_extra_edges = 0;
 883        for (i = 0; i < oids.nr; i++) {
 884                int num_parents = 0;
 885                if (i > 0 && oideq(&oids.list[i - 1], &oids.list[i]))
 886                        continue;
 887
 888                commits.list[commits.nr] = lookup_commit(the_repository, &oids.list[i]);
 889                parse_commit(commits.list[commits.nr]);
 890
 891                for (parent = commits.list[commits.nr]->parents;
 892                     parent; parent = parent->next)
 893                        num_parents++;
 894
 895                if (num_parents > 2)
 896                        num_extra_edges += num_parents - 1;
 897
 898                commits.nr++;
 899        }
 900        num_chunks = num_extra_edges ? 4 : 3;
 901
 902        if (commits.nr >= GRAPH_PARENT_MISSING)
 903                die(_("too many commits to write graph"));
 904
 905        compute_generation_numbers(&commits, report_progress);
 906
 907        graph_name = get_commit_graph_filename(obj_dir);
 908        if (safe_create_leading_directories(graph_name)) {
 909                UNLEAK(graph_name);
 910                die_errno(_("unable to create leading directories of %s"),
 911                          graph_name);
 912        }
 913
 914        hold_lock_file_for_update(&lk, graph_name, LOCK_DIE_ON_ERROR);
 915        f = hashfd(lk.tempfile->fd, lk.tempfile->filename.buf);
 916
 917        hashwrite_be32(f, GRAPH_SIGNATURE);
 918
 919        hashwrite_u8(f, GRAPH_VERSION);
 920        hashwrite_u8(f, GRAPH_OID_VERSION);
 921        hashwrite_u8(f, num_chunks);
 922        hashwrite_u8(f, 0); /* unused padding byte */
 923
 924        chunk_ids[0] = GRAPH_CHUNKID_OIDFANOUT;
 925        chunk_ids[1] = GRAPH_CHUNKID_OIDLOOKUP;
 926        chunk_ids[2] = GRAPH_CHUNKID_DATA;
 927        if (num_extra_edges)
 928                chunk_ids[3] = GRAPH_CHUNKID_LARGEEDGES;
 929        else
 930                chunk_ids[3] = 0;
 931        chunk_ids[4] = 0;
 932
 933        chunk_offsets[0] = 8 + (num_chunks + 1) * GRAPH_CHUNKLOOKUP_WIDTH;
 934        chunk_offsets[1] = chunk_offsets[0] + GRAPH_FANOUT_SIZE;
 935        chunk_offsets[2] = chunk_offsets[1] + GRAPH_OID_LEN * commits.nr;
 936        chunk_offsets[3] = chunk_offsets[2] + (GRAPH_OID_LEN + 16) * commits.nr;
 937        chunk_offsets[4] = chunk_offsets[3] + 4 * num_extra_edges;
 938
 939        for (i = 0; i <= num_chunks; i++) {
 940                uint32_t chunk_write[3];
 941
 942                chunk_write[0] = htonl(chunk_ids[i]);
 943                chunk_write[1] = htonl(chunk_offsets[i] >> 32);
 944                chunk_write[2] = htonl(chunk_offsets[i] & 0xffffffff);
 945                hashwrite(f, chunk_write, 12);
 946        }
 947
 948        write_graph_chunk_fanout(f, commits.list, commits.nr);
 949        write_graph_chunk_oids(f, GRAPH_OID_LEN, commits.list, commits.nr);
 950        write_graph_chunk_data(f, GRAPH_OID_LEN, commits.list, commits.nr);
 951        write_graph_chunk_large_edges(f, commits.list, commits.nr);
 952
 953        close_commit_graph(the_repository);
 954        finalize_hashfile(f, NULL, CSUM_HASH_IN_STREAM | CSUM_FSYNC);
 955        commit_lock_file(&lk);
 956
 957        free(graph_name);
 958        free(commits.list);
 959        free(oids.list);
 960}
 961
 962#define VERIFY_COMMIT_GRAPH_ERROR_HASH 2
 963static int verify_commit_graph_error;
 964
 965static void graph_report(const char *fmt, ...)
 966{
 967        va_list ap;
 968
 969        verify_commit_graph_error = 1;
 970        va_start(ap, fmt);
 971        vfprintf(stderr, fmt, ap);
 972        fprintf(stderr, "\n");
 973        va_end(ap);
 974}
 975
 976#define GENERATION_ZERO_EXISTS 1
 977#define GENERATION_NUMBER_EXISTS 2
 978
 979int verify_commit_graph(struct repository *r, struct commit_graph *g)
 980{
 981        uint32_t i, cur_fanout_pos = 0;
 982        struct object_id prev_oid, cur_oid, checksum;
 983        int generation_zero = 0;
 984        struct hashfile *f;
 985        int devnull;
 986        struct progress *progress = NULL;
 987
 988        if (!g) {
 989                graph_report("no commit-graph file loaded");
 990                return 1;
 991        }
 992
 993        verify_commit_graph_error = 0;
 994
 995        if (!g->chunk_oid_fanout)
 996                graph_report("commit-graph is missing the OID Fanout chunk");
 997        if (!g->chunk_oid_lookup)
 998                graph_report("commit-graph is missing the OID Lookup chunk");
 999        if (!g->chunk_commit_data)
1000                graph_report("commit-graph is missing the Commit Data chunk");
1001
1002        if (verify_commit_graph_error)
1003                return verify_commit_graph_error;
1004
1005        devnull = open("/dev/null", O_WRONLY);
1006        f = hashfd(devnull, NULL);
1007        hashwrite(f, g->data, g->data_len - g->hash_len);
1008        finalize_hashfile(f, checksum.hash, CSUM_CLOSE);
1009        if (!hasheq(checksum.hash, g->data + g->data_len - g->hash_len)) {
1010                graph_report(_("the commit-graph file has incorrect checksum and is likely corrupt"));
1011                verify_commit_graph_error = VERIFY_COMMIT_GRAPH_ERROR_HASH;
1012        }
1013
1014        for (i = 0; i < g->num_commits; i++) {
1015                struct commit *graph_commit;
1016
1017                hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
1018
1019                if (i && oidcmp(&prev_oid, &cur_oid) >= 0)
1020                        graph_report("commit-graph has incorrect OID order: %s then %s",
1021                                     oid_to_hex(&prev_oid),
1022                                     oid_to_hex(&cur_oid));
1023
1024                oidcpy(&prev_oid, &cur_oid);
1025
1026                while (cur_oid.hash[0] > cur_fanout_pos) {
1027                        uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
1028
1029                        if (i != fanout_value)
1030                                graph_report("commit-graph has incorrect fanout value: fanout[%d] = %u != %u",
1031                                             cur_fanout_pos, fanout_value, i);
1032                        cur_fanout_pos++;
1033                }
1034
1035                graph_commit = lookup_commit(r, &cur_oid);
1036                if (!parse_commit_in_graph_one(r, g, graph_commit))
1037                        graph_report("failed to parse %s from commit-graph",
1038                                     oid_to_hex(&cur_oid));
1039        }
1040
1041        while (cur_fanout_pos < 256) {
1042                uint32_t fanout_value = get_be32(g->chunk_oid_fanout + cur_fanout_pos);
1043
1044                if (g->num_commits != fanout_value)
1045                        graph_report("commit-graph has incorrect fanout value: fanout[%d] = %u != %u",
1046                                     cur_fanout_pos, fanout_value, i);
1047
1048                cur_fanout_pos++;
1049        }
1050
1051        if (verify_commit_graph_error & ~VERIFY_COMMIT_GRAPH_ERROR_HASH)
1052                return verify_commit_graph_error;
1053
1054        progress = start_progress(_("Verifying commits in commit graph"),
1055                                  g->num_commits);
1056        for (i = 0; i < g->num_commits; i++) {
1057                struct commit *graph_commit, *odb_commit;
1058                struct commit_list *graph_parents, *odb_parents;
1059                uint32_t max_generation = 0;
1060
1061                display_progress(progress, i + 1);
1062                hashcpy(cur_oid.hash, g->chunk_oid_lookup + g->hash_len * i);
1063
1064                graph_commit = lookup_commit(r, &cur_oid);
1065                odb_commit = (struct commit *)create_object(r, cur_oid.hash, alloc_commit_node(r));
1066                if (parse_commit_internal(odb_commit, 0, 0)) {
1067                        graph_report("failed to parse %s from object database",
1068                                     oid_to_hex(&cur_oid));
1069                        continue;
1070                }
1071
1072                if (!oideq(&get_commit_tree_in_graph_one(r, g, graph_commit)->object.oid,
1073                           get_commit_tree_oid(odb_commit)))
1074                        graph_report("root tree OID for commit %s in commit-graph is %s != %s",
1075                                     oid_to_hex(&cur_oid),
1076                                     oid_to_hex(get_commit_tree_oid(graph_commit)),
1077                                     oid_to_hex(get_commit_tree_oid(odb_commit)));
1078
1079                graph_parents = graph_commit->parents;
1080                odb_parents = odb_commit->parents;
1081
1082                while (graph_parents) {
1083                        if (odb_parents == NULL) {
1084                                graph_report("commit-graph parent list for commit %s is too long",
1085                                             oid_to_hex(&cur_oid));
1086                                break;
1087                        }
1088
1089                        if (!oideq(&graph_parents->item->object.oid, &odb_parents->item->object.oid))
1090                                graph_report("commit-graph parent for %s is %s != %s",
1091                                             oid_to_hex(&cur_oid),
1092                                             oid_to_hex(&graph_parents->item->object.oid),
1093                                             oid_to_hex(&odb_parents->item->object.oid));
1094
1095                        if (graph_parents->item->generation > max_generation)
1096                                max_generation = graph_parents->item->generation;
1097
1098                        graph_parents = graph_parents->next;
1099                        odb_parents = odb_parents->next;
1100                }
1101
1102                if (odb_parents != NULL)
1103                        graph_report("commit-graph parent list for commit %s terminates early",
1104                                     oid_to_hex(&cur_oid));
1105
1106                if (!graph_commit->generation) {
1107                        if (generation_zero == GENERATION_NUMBER_EXISTS)
1108                                graph_report("commit-graph has generation number zero for commit %s, but non-zero elsewhere",
1109                                             oid_to_hex(&cur_oid));
1110                        generation_zero = GENERATION_ZERO_EXISTS;
1111                } else if (generation_zero == GENERATION_ZERO_EXISTS)
1112                        graph_report("commit-graph has non-zero generation number for commit %s, but zero elsewhere",
1113                                     oid_to_hex(&cur_oid));
1114
1115                if (generation_zero == GENERATION_ZERO_EXISTS)
1116                        continue;
1117
1118                /*
1119                 * If one of our parents has generation GENERATION_NUMBER_MAX, then
1120                 * our generation is also GENERATION_NUMBER_MAX. Decrement to avoid
1121                 * extra logic in the following condition.
1122                 */
1123                if (max_generation == GENERATION_NUMBER_MAX)
1124                        max_generation--;
1125
1126                if (graph_commit->generation != max_generation + 1)
1127                        graph_report("commit-graph generation for commit %s is %u != %u",
1128                                     oid_to_hex(&cur_oid),
1129                                     graph_commit->generation,
1130                                     max_generation + 1);
1131
1132                if (graph_commit->date != odb_commit->date)
1133                        graph_report("commit date for commit %s in commit-graph is %"PRItime" != %"PRItime,
1134                                     oid_to_hex(&cur_oid),
1135                                     graph_commit->date,
1136                                     odb_commit->date);
1137        }
1138        stop_progress(&progress);
1139
1140        return verify_commit_graph_error;
1141}
1142
1143void free_commit_graph(struct commit_graph *g)
1144{
1145        if (!g)
1146                return;
1147        if (g->graph_fd >= 0) {
1148                munmap((void *)g->data, g->data_len);
1149                g->data = NULL;
1150                close(g->graph_fd);
1151        }
1152        free(g);
1153}