index-pack.con commit index-pack: rationalize delta resolution code (9441b61)
   1#include "cache.h"
   2#include "delta.h"
   3#include "pack.h"
   4#include "csum-file.h"
   5#include "blob.h"
   6#include "commit.h"
   7#include "tag.h"
   8#include "tree.h"
   9#include "progress.h"
  10#include "fsck.h"
  11
  12static const char index_pack_usage[] =
  13"git index-pack [-v] [-o <index-file>] [{ ---keep | --keep=<msg> }] [--strict] { <pack-file> | --stdin [--fix-thin] [<pack-file>] }";
  14
  15struct object_entry
  16{
  17        struct pack_idx_entry idx;
  18        unsigned long size;
  19        unsigned int hdr_size;
  20        enum object_type type;
  21        enum object_type real_type;
  22};
  23
  24union delta_base {
  25        unsigned char sha1[20];
  26        off_t offset;
  27};
  28
  29struct base_data {
  30        struct base_data *base;
  31        struct base_data *child;
  32        struct object_entry *obj;
  33        void *data;
  34        unsigned long size;
  35};
  36
  37/*
  38 * Even if sizeof(union delta_base) == 24 on 64-bit archs, we really want
  39 * to memcmp() only the first 20 bytes.
  40 */
  41#define UNION_BASE_SZ   20
  42
  43#define FLAG_LINK (1u<<20)
  44#define FLAG_CHECKED (1u<<21)
  45
  46struct delta_entry
  47{
  48        union delta_base base;
  49        int obj_no;
  50};
  51
  52static struct object_entry *objects;
  53static struct delta_entry *deltas;
  54static struct base_data *base_cache;
  55static size_t base_cache_used;
  56static int nr_objects;
  57static int nr_deltas;
  58static int nr_resolved_deltas;
  59
  60static int from_stdin;
  61static int strict;
  62static int verbose;
  63
  64static struct progress *progress;
  65
  66/* We always read in 4kB chunks. */
  67static unsigned char input_buffer[4096];
  68static unsigned int input_offset, input_len;
  69static off_t consumed_bytes;
  70static git_SHA_CTX input_ctx;
  71static uint32_t input_crc32;
  72static int input_fd, output_fd, pack_fd;
  73
  74static int mark_link(struct object *obj, int type, void *data)
  75{
  76        if (!obj)
  77                return -1;
  78
  79        if (type != OBJ_ANY && obj->type != type)
  80                die("object type mismatch at %s", sha1_to_hex(obj->sha1));
  81
  82        obj->flags |= FLAG_LINK;
  83        return 0;
  84}
  85
  86/* The content of each linked object must have been checked
  87   or it must be already present in the object database */
  88static void check_object(struct object *obj)
  89{
  90        if (!obj)
  91                return;
  92
  93        if (!(obj->flags & FLAG_LINK))
  94                return;
  95
  96        if (!(obj->flags & FLAG_CHECKED)) {
  97                unsigned long size;
  98                int type = sha1_object_info(obj->sha1, &size);
  99                if (type != obj->type || type <= 0)
 100                        die("object of unexpected type");
 101                obj->flags |= FLAG_CHECKED;
 102                return;
 103        }
 104}
 105
 106static void check_objects(void)
 107{
 108        unsigned i, max;
 109
 110        max = get_max_object_index();
 111        for (i = 0; i < max; i++)
 112                check_object(get_indexed_object(i));
 113}
 114
 115
 116/* Discard current buffer used content. */
 117static void flush(void)
 118{
 119        if (input_offset) {
 120                if (output_fd >= 0)
 121                        write_or_die(output_fd, input_buffer, input_offset);
 122                git_SHA1_Update(&input_ctx, input_buffer, input_offset);
 123                memmove(input_buffer, input_buffer + input_offset, input_len);
 124                input_offset = 0;
 125        }
 126}
 127
 128/*
 129 * Make sure at least "min" bytes are available in the buffer, and
 130 * return the pointer to the buffer.
 131 */
 132static void *fill(int min)
 133{
 134        if (min <= input_len)
 135                return input_buffer + input_offset;
 136        if (min > sizeof(input_buffer))
 137                die("cannot fill %d bytes", min);
 138        flush();
 139        do {
 140                ssize_t ret = xread(input_fd, input_buffer + input_len,
 141                                sizeof(input_buffer) - input_len);
 142                if (ret <= 0) {
 143                        if (!ret)
 144                                die("early EOF");
 145                        die("read error on input: %s", strerror(errno));
 146                }
 147                input_len += ret;
 148                if (from_stdin)
 149                        display_throughput(progress, consumed_bytes + input_len);
 150        } while (input_len < min);
 151        return input_buffer;
 152}
 153
 154static void use(int bytes)
 155{
 156        if (bytes > input_len)
 157                die("used more bytes than were available");
 158        input_crc32 = crc32(input_crc32, input_buffer + input_offset, bytes);
 159        input_len -= bytes;
 160        input_offset += bytes;
 161
 162        /* make sure off_t is sufficiently large not to wrap */
 163        if (consumed_bytes > consumed_bytes + bytes)
 164                die("pack too large for current definition of off_t");
 165        consumed_bytes += bytes;
 166}
 167
 168static char *open_pack_file(char *pack_name)
 169{
 170        if (from_stdin) {
 171                input_fd = 0;
 172                if (!pack_name) {
 173                        static char tmpfile[PATH_MAX];
 174                        snprintf(tmpfile, sizeof(tmpfile),
 175                                 "%s/pack/tmp_pack_XXXXXX", get_object_directory());
 176                        output_fd = xmkstemp(tmpfile);
 177                        pack_name = xstrdup(tmpfile);
 178                } else
 179                        output_fd = open(pack_name, O_CREAT|O_EXCL|O_RDWR, 0600);
 180                if (output_fd < 0)
 181                        die("unable to create %s: %s\n", pack_name, strerror(errno));
 182                pack_fd = output_fd;
 183        } else {
 184                input_fd = open(pack_name, O_RDONLY);
 185                if (input_fd < 0)
 186                        die("cannot open packfile '%s': %s",
 187                            pack_name, strerror(errno));
 188                output_fd = -1;
 189                pack_fd = input_fd;
 190        }
 191        git_SHA1_Init(&input_ctx);
 192        return pack_name;
 193}
 194
 195static void parse_pack_header(void)
 196{
 197        struct pack_header *hdr = fill(sizeof(struct pack_header));
 198
 199        /* Header consistency check */
 200        if (hdr->hdr_signature != htonl(PACK_SIGNATURE))
 201                die("pack signature mismatch");
 202        if (!pack_version_ok(hdr->hdr_version))
 203                die("pack version %"PRIu32" unsupported",
 204                        ntohl(hdr->hdr_version));
 205
 206        nr_objects = ntohl(hdr->hdr_entries);
 207        use(sizeof(struct pack_header));
 208}
 209
 210static void bad_object(unsigned long offset, const char *format,
 211                       ...) NORETURN __attribute__((format (printf, 2, 3)));
 212
 213static void bad_object(unsigned long offset, const char *format, ...)
 214{
 215        va_list params;
 216        char buf[1024];
 217
 218        va_start(params, format);
 219        vsnprintf(buf, sizeof(buf), format, params);
 220        va_end(params);
 221        die("pack has bad object at offset %lu: %s", offset, buf);
 222}
 223
 224static void prune_base_data(struct base_data *retain)
 225{
 226        struct base_data *b = base_cache;
 227        for (b = base_cache;
 228             base_cache_used > delta_base_cache_limit && b;
 229             b = b->child) {
 230                if (b->data && b != retain) {
 231                        free(b->data);
 232                        b->data = NULL;
 233                        base_cache_used -= b->size;
 234                }
 235        }
 236}
 237
 238static void link_base_data(struct base_data *base, struct base_data *c)
 239{
 240        if (base)
 241                base->child = c;
 242        else
 243                base_cache = c;
 244
 245        c->base = base;
 246        c->child = NULL;
 247        if (c->data)
 248                base_cache_used += c->size;
 249        prune_base_data(c);
 250}
 251
 252static void unlink_base_data(struct base_data *c)
 253{
 254        struct base_data *base = c->base;
 255        if (base)
 256                base->child = NULL;
 257        else
 258                base_cache = NULL;
 259        if (c->data) {
 260                free(c->data);
 261                base_cache_used -= c->size;
 262        }
 263}
 264
 265static void *unpack_entry_data(unsigned long offset, unsigned long size)
 266{
 267        z_stream stream;
 268        void *buf = xmalloc(size);
 269
 270        memset(&stream, 0, sizeof(stream));
 271        stream.next_out = buf;
 272        stream.avail_out = size;
 273        stream.next_in = fill(1);
 274        stream.avail_in = input_len;
 275        inflateInit(&stream);
 276
 277        for (;;) {
 278                int ret = inflate(&stream, 0);
 279                use(input_len - stream.avail_in);
 280                if (stream.total_out == size && ret == Z_STREAM_END)
 281                        break;
 282                if (ret != Z_OK)
 283                        bad_object(offset, "inflate returned %d", ret);
 284                stream.next_in = fill(1);
 285                stream.avail_in = input_len;
 286        }
 287        inflateEnd(&stream);
 288        return buf;
 289}
 290
 291static void *unpack_raw_entry(struct object_entry *obj, union delta_base *delta_base)
 292{
 293        unsigned char *p, c;
 294        unsigned long size;
 295        off_t base_offset;
 296        unsigned shift;
 297        void *data;
 298
 299        obj->idx.offset = consumed_bytes;
 300        input_crc32 = crc32(0, Z_NULL, 0);
 301
 302        p = fill(1);
 303        c = *p;
 304        use(1);
 305        obj->type = (c >> 4) & 7;
 306        size = (c & 15);
 307        shift = 4;
 308        while (c & 0x80) {
 309                p = fill(1);
 310                c = *p;
 311                use(1);
 312                size += (c & 0x7fUL) << shift;
 313                shift += 7;
 314        }
 315        obj->size = size;
 316
 317        switch (obj->type) {
 318        case OBJ_REF_DELTA:
 319                hashcpy(delta_base->sha1, fill(20));
 320                use(20);
 321                break;
 322        case OBJ_OFS_DELTA:
 323                memset(delta_base, 0, sizeof(*delta_base));
 324                p = fill(1);
 325                c = *p;
 326                use(1);
 327                base_offset = c & 127;
 328                while (c & 128) {
 329                        base_offset += 1;
 330                        if (!base_offset || MSB(base_offset, 7))
 331                                bad_object(obj->idx.offset, "offset value overflow for delta base object");
 332                        p = fill(1);
 333                        c = *p;
 334                        use(1);
 335                        base_offset = (base_offset << 7) + (c & 127);
 336                }
 337                delta_base->offset = obj->idx.offset - base_offset;
 338                if (delta_base->offset >= obj->idx.offset)
 339                        bad_object(obj->idx.offset, "delta base offset is out of bound");
 340                break;
 341        case OBJ_COMMIT:
 342        case OBJ_TREE:
 343        case OBJ_BLOB:
 344        case OBJ_TAG:
 345                break;
 346        default:
 347                bad_object(obj->idx.offset, "unknown object type %d", obj->type);
 348        }
 349        obj->hdr_size = consumed_bytes - obj->idx.offset;
 350
 351        data = unpack_entry_data(obj->idx.offset, obj->size);
 352        obj->idx.crc32 = input_crc32;
 353        return data;
 354}
 355
 356static void *get_data_from_pack(struct object_entry *obj)
 357{
 358        off_t from = obj[0].idx.offset + obj[0].hdr_size;
 359        unsigned long len = obj[1].idx.offset - from;
 360        unsigned long rdy = 0;
 361        unsigned char *src, *data;
 362        z_stream stream;
 363        int st;
 364
 365        src = xmalloc(len);
 366        data = src;
 367        do {
 368                ssize_t n = pread(pack_fd, data + rdy, len - rdy, from + rdy);
 369                if (n < 0)
 370                        die("cannot pread pack file: %s", strerror(errno));
 371                if (!n)
 372                        die("premature end of pack file, %lu bytes missing",
 373                            len - rdy);
 374                rdy += n;
 375        } while (rdy < len);
 376        data = xmalloc(obj->size);
 377        memset(&stream, 0, sizeof(stream));
 378        stream.next_out = data;
 379        stream.avail_out = obj->size;
 380        stream.next_in = src;
 381        stream.avail_in = len;
 382        inflateInit(&stream);
 383        while ((st = inflate(&stream, Z_FINISH)) == Z_OK);
 384        inflateEnd(&stream);
 385        if (st != Z_STREAM_END || stream.total_out != obj->size)
 386                die("serious inflate inconsistency");
 387        free(src);
 388        return data;
 389}
 390
 391static int find_delta(const union delta_base *base)
 392{
 393        int first = 0, last = nr_deltas;
 394
 395        while (first < last) {
 396                int next = (first + last) / 2;
 397                struct delta_entry *delta = &deltas[next];
 398                int cmp;
 399
 400                cmp = memcmp(base, &delta->base, UNION_BASE_SZ);
 401                if (!cmp)
 402                        return next;
 403                if (cmp < 0) {
 404                        last = next;
 405                        continue;
 406                }
 407                first = next+1;
 408        }
 409        return -first-1;
 410}
 411
 412static int find_delta_children(const union delta_base *base,
 413                               int *first_index, int *last_index)
 414{
 415        int first = find_delta(base);
 416        int last = first;
 417        int end = nr_deltas - 1;
 418
 419        if (first < 0)
 420                return -1;
 421        while (first > 0 && !memcmp(&deltas[first - 1].base, base, UNION_BASE_SZ))
 422                --first;
 423        while (last < end && !memcmp(&deltas[last + 1].base, base, UNION_BASE_SZ))
 424                ++last;
 425        *first_index = first;
 426        *last_index = last;
 427        return 0;
 428}
 429
 430static void sha1_object(const void *data, unsigned long size,
 431                        enum object_type type, unsigned char *sha1)
 432{
 433        hash_sha1_file(data, size, typename(type), sha1);
 434        if (has_sha1_file(sha1)) {
 435                void *has_data;
 436                enum object_type has_type;
 437                unsigned long has_size;
 438                has_data = read_sha1_file(sha1, &has_type, &has_size);
 439                if (!has_data)
 440                        die("cannot read existing object %s", sha1_to_hex(sha1));
 441                if (size != has_size || type != has_type ||
 442                    memcmp(data, has_data, size) != 0)
 443                        die("SHA1 COLLISION FOUND WITH %s !", sha1_to_hex(sha1));
 444                free(has_data);
 445        }
 446        if (strict) {
 447                if (type == OBJ_BLOB) {
 448                        struct blob *blob = lookup_blob(sha1);
 449                        if (blob)
 450                                blob->object.flags |= FLAG_CHECKED;
 451                        else
 452                                die("invalid blob object %s", sha1_to_hex(sha1));
 453                } else {
 454                        struct object *obj;
 455                        int eaten;
 456                        void *buf = (void *) data;
 457
 458                        /*
 459                         * we do not need to free the memory here, as the
 460                         * buf is deleted by the caller.
 461                         */
 462                        obj = parse_object_buffer(sha1, type, size, buf, &eaten);
 463                        if (!obj)
 464                                die("invalid %s", typename(type));
 465                        if (fsck_object(obj, 1, fsck_error_function))
 466                                die("Error in object");
 467                        if (fsck_walk(obj, mark_link, 0))
 468                                die("Not all child objects of %s are reachable", sha1_to_hex(obj->sha1));
 469
 470                        if (obj->type == OBJ_TREE) {
 471                                struct tree *item = (struct tree *) obj;
 472                                item->buffer = NULL;
 473                        }
 474                        if (obj->type == OBJ_COMMIT) {
 475                                struct commit *commit = (struct commit *) obj;
 476                                commit->buffer = NULL;
 477                        }
 478                        obj->flags |= FLAG_CHECKED;
 479                }
 480        }
 481}
 482
 483static void *get_base_data(struct base_data *c)
 484{
 485        if (!c->data) {
 486                struct object_entry *obj = c->obj;
 487
 488                if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) {
 489                        void *base = get_base_data(c->base);
 490                        void *raw = get_data_from_pack(obj);
 491                        c->data = patch_delta(
 492                                base, c->base->size,
 493                                raw, obj->size,
 494                                &c->size);
 495                        free(raw);
 496                        if (!c->data)
 497                                bad_object(obj->idx.offset, "failed to apply delta");
 498                } else {
 499                        c->data = get_data_from_pack(obj);
 500                        c->size = obj->size;
 501                }
 502
 503                base_cache_used += c->size;
 504                prune_base_data(c);
 505        }
 506        return c->data;
 507}
 508
 509static void resolve_delta(struct object_entry *delta_obj,
 510                          struct base_data *base, struct base_data *result)
 511{
 512        void *delta_data;
 513        unsigned long delta_size;
 514
 515        delta_obj->real_type = base->obj->type;
 516        delta_data = get_data_from_pack(delta_obj);
 517        delta_size = delta_obj->size;
 518        result->obj = delta_obj;
 519        result->data = patch_delta(get_base_data(base), base->obj->size,
 520                                   delta_data, delta_size, &result->size);
 521        free(delta_data);
 522        if (!result->data)
 523                bad_object(delta_obj->idx.offset, "failed to apply delta");
 524        sha1_object(result->data, result->size, delta_obj->real_type,
 525                    delta_obj->idx.sha1);
 526        nr_resolved_deltas++;
 527}
 528
 529static void find_unresolved_deltas(struct base_data *base,
 530                                   struct base_data *prev_base)
 531{
 532        int i, ref, ref_first, ref_last, ofs, ofs_first, ofs_last;
 533
 534        /*
 535         * This is a recursive function. Those brackets should help reducing
 536         * stack usage by limiting the scope of the delta_base union.
 537         */
 538        {
 539                union delta_base base_spec;
 540
 541                hashcpy(base_spec.sha1, base->obj->idx.sha1);
 542                ref = !find_delta_children(&base_spec, &ref_first, &ref_last);
 543
 544                memset(&base_spec, 0, sizeof(base_spec));
 545                base_spec.offset = base->obj->idx.offset;
 546                ofs = !find_delta_children(&base_spec, &ofs_first, &ofs_last);
 547        }
 548
 549        if (!ref && !ofs)
 550                return;
 551
 552        link_base_data(prev_base, base);
 553
 554        if (ref) {
 555                for (i = ref_first; i <= ref_last; i++) {
 556                        struct object_entry *child = objects + deltas[i].obj_no;
 557                        if (child->real_type == OBJ_REF_DELTA) {
 558                                struct base_data result;
 559                                resolve_delta(child, base, &result);
 560                                find_unresolved_deltas(&result, base);
 561                        }
 562                }
 563        }
 564
 565        if (ofs) {
 566                for (i = ofs_first; i <= ofs_last; i++) {
 567                        struct object_entry *child = objects + deltas[i].obj_no;
 568                        if (child->real_type == OBJ_OFS_DELTA) {
 569                                struct base_data result;
 570                                resolve_delta(child, base, &result);
 571                                find_unresolved_deltas(&result, base);
 572                        }
 573                }
 574        }
 575
 576        unlink_base_data(base);
 577}
 578
 579static int compare_delta_entry(const void *a, const void *b)
 580{
 581        const struct delta_entry *delta_a = a;
 582        const struct delta_entry *delta_b = b;
 583        return memcmp(&delta_a->base, &delta_b->base, UNION_BASE_SZ);
 584}
 585
 586/* Parse all objects and return the pack content SHA1 hash */
 587static void parse_pack_objects(unsigned char *sha1)
 588{
 589        int i;
 590        struct delta_entry *delta = deltas;
 591        struct stat st;
 592
 593        /*
 594         * First pass:
 595         * - find locations of all objects;
 596         * - calculate SHA1 of all non-delta objects;
 597         * - remember base (SHA1 or offset) for all deltas.
 598         */
 599        if (verbose)
 600                progress = start_progress(
 601                                from_stdin ? "Receiving objects" : "Indexing objects",
 602                                nr_objects);
 603        for (i = 0; i < nr_objects; i++) {
 604                struct object_entry *obj = &objects[i];
 605                void *data = unpack_raw_entry(obj, &delta->base);
 606                obj->real_type = obj->type;
 607                if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA) {
 608                        nr_deltas++;
 609                        delta->obj_no = i;
 610                        delta++;
 611                } else
 612                        sha1_object(data, obj->size, obj->type, obj->idx.sha1);
 613                free(data);
 614                display_progress(progress, i+1);
 615        }
 616        objects[i].idx.offset = consumed_bytes;
 617        stop_progress(&progress);
 618
 619        /* Check pack integrity */
 620        flush();
 621        git_SHA1_Final(sha1, &input_ctx);
 622        if (hashcmp(fill(20), sha1))
 623                die("pack is corrupted (SHA1 mismatch)");
 624        use(20);
 625
 626        /* If input_fd is a file, we should have reached its end now. */
 627        if (fstat(input_fd, &st))
 628                die("cannot fstat packfile: %s", strerror(errno));
 629        if (S_ISREG(st.st_mode) &&
 630                        lseek(input_fd, 0, SEEK_CUR) - input_len != st.st_size)
 631                die("pack has junk at the end");
 632
 633        if (!nr_deltas)
 634                return;
 635
 636        /* Sort deltas by base SHA1/offset for fast searching */
 637        qsort(deltas, nr_deltas, sizeof(struct delta_entry),
 638              compare_delta_entry);
 639
 640        /*
 641         * Second pass:
 642         * - for all non-delta objects, look if it is used as a base for
 643         *   deltas;
 644         * - if used as a base, uncompress the object and apply all deltas,
 645         *   recursively checking if the resulting object is used as a base
 646         *   for some more deltas.
 647         */
 648        if (verbose)
 649                progress = start_progress("Resolving deltas", nr_deltas);
 650        for (i = 0; i < nr_objects; i++) {
 651                struct object_entry *obj = &objects[i];
 652                struct base_data base_obj;
 653
 654                if (obj->type == OBJ_REF_DELTA || obj->type == OBJ_OFS_DELTA)
 655                        continue;
 656                base_obj.obj = obj;
 657                base_obj.data = NULL;
 658                find_unresolved_deltas(&base_obj, NULL);
 659                display_progress(progress, nr_resolved_deltas);
 660        }
 661}
 662
 663static int write_compressed(struct sha1file *f, void *in, unsigned int size)
 664{
 665        z_stream stream;
 666        unsigned long maxsize;
 667        void *out;
 668
 669        memset(&stream, 0, sizeof(stream));
 670        deflateInit(&stream, zlib_compression_level);
 671        maxsize = deflateBound(&stream, size);
 672        out = xmalloc(maxsize);
 673
 674        /* Compress it */
 675        stream.next_in = in;
 676        stream.avail_in = size;
 677        stream.next_out = out;
 678        stream.avail_out = maxsize;
 679        while (deflate(&stream, Z_FINISH) == Z_OK);
 680        deflateEnd(&stream);
 681
 682        size = stream.total_out;
 683        sha1write(f, out, size);
 684        free(out);
 685        return size;
 686}
 687
 688static struct object_entry *append_obj_to_pack(struct sha1file *f,
 689                               const unsigned char *sha1, void *buf,
 690                               unsigned long size, enum object_type type)
 691{
 692        struct object_entry *obj = &objects[nr_objects++];
 693        unsigned char header[10];
 694        unsigned long s = size;
 695        int n = 0;
 696        unsigned char c = (type << 4) | (s & 15);
 697        s >>= 4;
 698        while (s) {
 699                header[n++] = c | 0x80;
 700                c = s & 0x7f;
 701                s >>= 7;
 702        }
 703        header[n++] = c;
 704        crc32_begin(f);
 705        sha1write(f, header, n);
 706        obj[0].size = size;
 707        obj[0].hdr_size = n;
 708        obj[0].type = type;
 709        obj[0].real_type = type;
 710        obj[1].idx.offset = obj[0].idx.offset + n;
 711        obj[1].idx.offset += write_compressed(f, buf, size);
 712        obj[0].idx.crc32 = crc32_end(f);
 713        sha1flush(f);
 714        hashcpy(obj->idx.sha1, sha1);
 715        return obj;
 716}
 717
 718static int delta_pos_compare(const void *_a, const void *_b)
 719{
 720        struct delta_entry *a = *(struct delta_entry **)_a;
 721        struct delta_entry *b = *(struct delta_entry **)_b;
 722        return a->obj_no - b->obj_no;
 723}
 724
 725static void fix_unresolved_deltas(struct sha1file *f, int nr_unresolved)
 726{
 727        struct delta_entry **sorted_by_pos;
 728        int i, n = 0;
 729
 730        /*
 731         * Since many unresolved deltas may well be themselves base objects
 732         * for more unresolved deltas, we really want to include the
 733         * smallest number of base objects that would cover as much delta
 734         * as possible by picking the
 735         * trunc deltas first, allowing for other deltas to resolve without
 736         * additional base objects.  Since most base objects are to be found
 737         * before deltas depending on them, a good heuristic is to start
 738         * resolving deltas in the same order as their position in the pack.
 739         */
 740        sorted_by_pos = xmalloc(nr_unresolved * sizeof(*sorted_by_pos));
 741        for (i = 0; i < nr_deltas; i++) {
 742                if (objects[deltas[i].obj_no].real_type != OBJ_REF_DELTA)
 743                        continue;
 744                sorted_by_pos[n++] = &deltas[i];
 745        }
 746        qsort(sorted_by_pos, n, sizeof(*sorted_by_pos), delta_pos_compare);
 747
 748        for (i = 0; i < n; i++) {
 749                struct delta_entry *d = sorted_by_pos[i];
 750                enum object_type type;
 751                struct base_data base_obj;
 752
 753                if (objects[d->obj_no].real_type != OBJ_REF_DELTA)
 754                        continue;
 755                base_obj.data = read_sha1_file(d->base.sha1, &type, &base_obj.size);
 756                if (!base_obj.data)
 757                        continue;
 758
 759                if (check_sha1_signature(d->base.sha1, base_obj.data,
 760                                base_obj.size, typename(type)))
 761                        die("local object %s is corrupt", sha1_to_hex(d->base.sha1));
 762                base_obj.obj = append_obj_to_pack(f, d->base.sha1,
 763                                        base_obj.data, base_obj.size, type);
 764                find_unresolved_deltas(&base_obj, NULL);
 765                display_progress(progress, nr_resolved_deltas);
 766        }
 767        free(sorted_by_pos);
 768}
 769
 770static void final(const char *final_pack_name, const char *curr_pack_name,
 771                  const char *final_index_name, const char *curr_index_name,
 772                  const char *keep_name, const char *keep_msg,
 773                  unsigned char *sha1)
 774{
 775        const char *report = "pack";
 776        char name[PATH_MAX];
 777        int err;
 778
 779        if (!from_stdin) {
 780                close(input_fd);
 781        } else {
 782                fsync_or_die(output_fd, curr_pack_name);
 783                err = close(output_fd);
 784                if (err)
 785                        die("error while closing pack file: %s", strerror(errno));
 786                chmod(curr_pack_name, 0444);
 787        }
 788
 789        if (keep_msg) {
 790                int keep_fd, keep_msg_len = strlen(keep_msg);
 791                if (!keep_name) {
 792                        snprintf(name, sizeof(name), "%s/pack/pack-%s.keep",
 793                                 get_object_directory(), sha1_to_hex(sha1));
 794                        keep_name = name;
 795                }
 796                keep_fd = open(keep_name, O_RDWR|O_CREAT|O_EXCL, 0600);
 797                if (keep_fd < 0) {
 798                        if (errno != EEXIST)
 799                                die("cannot write keep file");
 800                } else {
 801                        if (keep_msg_len > 0) {
 802                                write_or_die(keep_fd, keep_msg, keep_msg_len);
 803                                write_or_die(keep_fd, "\n", 1);
 804                        }
 805                        if (close(keep_fd) != 0)
 806                                die("cannot write keep file");
 807                        report = "keep";
 808                }
 809        }
 810
 811        if (final_pack_name != curr_pack_name) {
 812                if (!final_pack_name) {
 813                        snprintf(name, sizeof(name), "%s/pack/pack-%s.pack",
 814                                 get_object_directory(), sha1_to_hex(sha1));
 815                        final_pack_name = name;
 816                }
 817                if (move_temp_to_file(curr_pack_name, final_pack_name))
 818                        die("cannot store pack file");
 819        }
 820
 821        chmod(curr_index_name, 0444);
 822        if (final_index_name != curr_index_name) {
 823                if (!final_index_name) {
 824                        snprintf(name, sizeof(name), "%s/pack/pack-%s.idx",
 825                                 get_object_directory(), sha1_to_hex(sha1));
 826                        final_index_name = name;
 827                }
 828                if (move_temp_to_file(curr_index_name, final_index_name))
 829                        die("cannot store index file");
 830        }
 831
 832        if (!from_stdin) {
 833                printf("%s\n", sha1_to_hex(sha1));
 834        } else {
 835                char buf[48];
 836                int len = snprintf(buf, sizeof(buf), "%s\t%s\n",
 837                                   report, sha1_to_hex(sha1));
 838                write_or_die(1, buf, len);
 839
 840                /*
 841                 * Let's just mimic git-unpack-objects here and write
 842                 * the last part of the input buffer to stdout.
 843                 */
 844                while (input_len) {
 845                        err = xwrite(1, input_buffer + input_offset, input_len);
 846                        if (err <= 0)
 847                                break;
 848                        input_len -= err;
 849                        input_offset += err;
 850                }
 851        }
 852}
 853
 854static int git_index_pack_config(const char *k, const char *v, void *cb)
 855{
 856        if (!strcmp(k, "pack.indexversion")) {
 857                pack_idx_default_version = git_config_int(k, v);
 858                if (pack_idx_default_version > 2)
 859                        die("bad pack.indexversion=%"PRIu32,
 860                                pack_idx_default_version);
 861                return 0;
 862        }
 863        return git_default_config(k, v, cb);
 864}
 865
 866int main(int argc, char **argv)
 867{
 868        int i, fix_thin_pack = 0;
 869        char *curr_pack, *pack_name = NULL;
 870        char *curr_index, *index_name = NULL;
 871        const char *keep_name = NULL, *keep_msg = NULL;
 872        char *index_name_buf = NULL, *keep_name_buf = NULL;
 873        struct pack_idx_entry **idx_objects;
 874        unsigned char pack_sha1[20];
 875        int nongit = 0;
 876
 877        setup_git_directory_gently(&nongit);
 878        git_config(git_index_pack_config, NULL);
 879
 880        for (i = 1; i < argc; i++) {
 881                char *arg = argv[i];
 882
 883                if (*arg == '-') {
 884                        if (!strcmp(arg, "--stdin")) {
 885                                from_stdin = 1;
 886                        } else if (!strcmp(arg, "--fix-thin")) {
 887                                fix_thin_pack = 1;
 888                        } else if (!strcmp(arg, "--strict")) {
 889                                strict = 1;
 890                        } else if (!strcmp(arg, "--keep")) {
 891                                keep_msg = "";
 892                        } else if (!prefixcmp(arg, "--keep=")) {
 893                                keep_msg = arg + 7;
 894                        } else if (!prefixcmp(arg, "--pack_header=")) {
 895                                struct pack_header *hdr;
 896                                char *c;
 897
 898                                hdr = (struct pack_header *)input_buffer;
 899                                hdr->hdr_signature = htonl(PACK_SIGNATURE);
 900                                hdr->hdr_version = htonl(strtoul(arg + 14, &c, 10));
 901                                if (*c != ',')
 902                                        die("bad %s", arg);
 903                                hdr->hdr_entries = htonl(strtoul(c + 1, &c, 10));
 904                                if (*c)
 905                                        die("bad %s", arg);
 906                                input_len = sizeof(*hdr);
 907                        } else if (!strcmp(arg, "-v")) {
 908                                verbose = 1;
 909                        } else if (!strcmp(arg, "-o")) {
 910                                if (index_name || (i+1) >= argc)
 911                                        usage(index_pack_usage);
 912                                index_name = argv[++i];
 913                        } else if (!prefixcmp(arg, "--index-version=")) {
 914                                char *c;
 915                                pack_idx_default_version = strtoul(arg + 16, &c, 10);
 916                                if (pack_idx_default_version > 2)
 917                                        die("bad %s", arg);
 918                                if (*c == ',')
 919                                        pack_idx_off32_limit = strtoul(c+1, &c, 0);
 920                                if (*c || pack_idx_off32_limit & 0x80000000)
 921                                        die("bad %s", arg);
 922                        } else
 923                                usage(index_pack_usage);
 924                        continue;
 925                }
 926
 927                if (pack_name)
 928                        usage(index_pack_usage);
 929                pack_name = arg;
 930        }
 931
 932        if (!pack_name && !from_stdin)
 933                usage(index_pack_usage);
 934        if (fix_thin_pack && !from_stdin)
 935                die("--fix-thin cannot be used without --stdin");
 936        if (!index_name && pack_name) {
 937                int len = strlen(pack_name);
 938                if (!has_extension(pack_name, ".pack"))
 939                        die("packfile name '%s' does not end with '.pack'",
 940                            pack_name);
 941                index_name_buf = xmalloc(len);
 942                memcpy(index_name_buf, pack_name, len - 5);
 943                strcpy(index_name_buf + len - 5, ".idx");
 944                index_name = index_name_buf;
 945        }
 946        if (keep_msg && !keep_name && pack_name) {
 947                int len = strlen(pack_name);
 948                if (!has_extension(pack_name, ".pack"))
 949                        die("packfile name '%s' does not end with '.pack'",
 950                            pack_name);
 951                keep_name_buf = xmalloc(len);
 952                memcpy(keep_name_buf, pack_name, len - 5);
 953                strcpy(keep_name_buf + len - 5, ".keep");
 954                keep_name = keep_name_buf;
 955        }
 956
 957        curr_pack = open_pack_file(pack_name);
 958        parse_pack_header();
 959        objects = xmalloc((nr_objects + 1) * sizeof(struct object_entry));
 960        deltas = xmalloc(nr_objects * sizeof(struct delta_entry));
 961        parse_pack_objects(pack_sha1);
 962        if (nr_deltas == nr_resolved_deltas) {
 963                stop_progress(&progress);
 964                /* Flush remaining pack final 20-byte SHA1. */
 965                flush();
 966        } else {
 967                if (fix_thin_pack) {
 968                        struct sha1file *f;
 969                        unsigned char read_sha1[20], tail_sha1[20];
 970                        char msg[48];
 971                        int nr_unresolved = nr_deltas - nr_resolved_deltas;
 972                        int nr_objects_initial = nr_objects;
 973                        if (nr_unresolved <= 0)
 974                                die("confusion beyond insanity");
 975                        objects = xrealloc(objects,
 976                                           (nr_objects + nr_unresolved + 1)
 977                                           * sizeof(*objects));
 978                        f = sha1fd(output_fd, curr_pack);
 979                        fix_unresolved_deltas(f, nr_unresolved);
 980                        sprintf(msg, "completed with %d local objects",
 981                                nr_objects - nr_objects_initial);
 982                        stop_progress_msg(&progress, msg);
 983                        sha1close(f, tail_sha1, 0);
 984                        hashcpy(read_sha1, pack_sha1);
 985                        fixup_pack_header_footer(output_fd, pack_sha1,
 986                                                 curr_pack, nr_objects,
 987                                                 read_sha1, consumed_bytes-20);
 988                        if (hashcmp(read_sha1, tail_sha1) != 0)
 989                                die("Unexpected tail checksum for %s "
 990                                    "(disk corruption?)", curr_pack);
 991                }
 992                if (nr_deltas != nr_resolved_deltas)
 993                        die("pack has %d unresolved deltas",
 994                            nr_deltas - nr_resolved_deltas);
 995        }
 996        free(deltas);
 997        if (strict)
 998                check_objects();
 999
1000        idx_objects = xmalloc((nr_objects) * sizeof(struct pack_idx_entry *));
1001        for (i = 0; i < nr_objects; i++)
1002                idx_objects[i] = &objects[i].idx;
1003        curr_index = write_idx_file(index_name, idx_objects, nr_objects, pack_sha1);
1004        free(idx_objects);
1005
1006        final(pack_name, curr_pack,
1007                index_name, curr_index,
1008                keep_name, keep_msg,
1009                pack_sha1);
1010        free(objects);
1011        free(index_name_buf);
1012        free(keep_name_buf);
1013        if (pack_name == NULL)
1014                free(curr_pack);
1015        if (index_name == NULL)
1016                free(curr_index);
1017
1018        return 0;
1019}