builtin / pack-objects.con commit pack-objects: use mru list when iterating over packs (c9af708)
   1#include "builtin.h"
   2#include "cache.h"
   3#include "attr.h"
   4#include "object.h"
   5#include "blob.h"
   6#include "commit.h"
   7#include "tag.h"
   8#include "tree.h"
   9#include "delta.h"
  10#include "pack.h"
  11#include "pack-revindex.h"
  12#include "csum-file.h"
  13#include "tree-walk.h"
  14#include "diff.h"
  15#include "revision.h"
  16#include "list-objects.h"
  17#include "pack-objects.h"
  18#include "progress.h"
  19#include "refs.h"
  20#include "streaming.h"
  21#include "thread-utils.h"
  22#include "pack-bitmap.h"
  23#include "reachable.h"
  24#include "sha1-array.h"
  25#include "argv-array.h"
  26#include "mru.h"
  27
  28static const char *pack_usage[] = {
  29        N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
  30        N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
  31        NULL
  32};
  33
  34/*
  35 * Objects we are going to pack are collected in the `to_pack` structure.
  36 * It contains an array (dynamically expanded) of the object data, and a map
  37 * that can resolve SHA1s to their position in the array.
  38 */
  39static struct packing_data to_pack;
  40
  41static struct pack_idx_entry **written_list;
  42static uint32_t nr_result, nr_written;
  43
  44static int non_empty;
  45static int reuse_delta = 1, reuse_object = 1;
  46static int keep_unreachable, unpack_unreachable, include_tag;
  47static unsigned long unpack_unreachable_expiration;
  48static int pack_loose_unreachable;
  49static int local;
  50static int have_non_local_packs;
  51static int incremental;
  52static int ignore_packed_keep;
  53static int allow_ofs_delta;
  54static struct pack_idx_option pack_idx_opts;
  55static const char *base_name;
  56static int progress = 1;
  57static int window = 10;
  58static unsigned long pack_size_limit;
  59static int depth = 50;
  60static int delta_search_threads;
  61static int pack_to_stdout;
  62static int num_preferred_base;
  63static struct progress *progress_state;
  64static int pack_compression_level = Z_DEFAULT_COMPRESSION;
  65static int pack_compression_seen;
  66
  67static struct packed_git *reuse_packfile;
  68static uint32_t reuse_packfile_objects;
  69static off_t reuse_packfile_offset;
  70
  71static int use_bitmap_index = 1;
  72static int write_bitmap_index;
  73static uint16_t write_bitmap_options;
  74
  75static unsigned long delta_cache_size = 0;
  76static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
  77static unsigned long cache_max_small_delta_size = 1000;
  78
  79static unsigned long window_memory_limit = 0;
  80
  81/*
  82 * stats
  83 */
  84static uint32_t written, written_delta;
  85static uint32_t reused, reused_delta;
  86
  87/*
  88 * Indexed commits
  89 */
  90static struct commit **indexed_commits;
  91static unsigned int indexed_commits_nr;
  92static unsigned int indexed_commits_alloc;
  93
  94static void index_commit_for_bitmap(struct commit *commit)
  95{
  96        if (indexed_commits_nr >= indexed_commits_alloc) {
  97                indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
  98                REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
  99        }
 100
 101        indexed_commits[indexed_commits_nr++] = commit;
 102}
 103
 104static void *get_delta(struct object_entry *entry)
 105{
 106        unsigned long size, base_size, delta_size;
 107        void *buf, *base_buf, *delta_buf;
 108        enum object_type type;
 109
 110        buf = read_sha1_file(entry->idx.sha1, &type, &size);
 111        if (!buf)
 112                die("unable to read %s", sha1_to_hex(entry->idx.sha1));
 113        base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
 114        if (!base_buf)
 115                die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
 116        delta_buf = diff_delta(base_buf, base_size,
 117                               buf, size, &delta_size, 0);
 118        if (!delta_buf || delta_size != entry->delta_size)
 119                die("delta size changed");
 120        free(buf);
 121        free(base_buf);
 122        return delta_buf;
 123}
 124
 125static unsigned long do_compress(void **pptr, unsigned long size)
 126{
 127        git_zstream stream;
 128        void *in, *out;
 129        unsigned long maxsize;
 130
 131        git_deflate_init(&stream, pack_compression_level);
 132        maxsize = git_deflate_bound(&stream, size);
 133
 134        in = *pptr;
 135        out = xmalloc(maxsize);
 136        *pptr = out;
 137
 138        stream.next_in = in;
 139        stream.avail_in = size;
 140        stream.next_out = out;
 141        stream.avail_out = maxsize;
 142        while (git_deflate(&stream, Z_FINISH) == Z_OK)
 143                ; /* nothing */
 144        git_deflate_end(&stream);
 145
 146        free(in);
 147        return stream.total_out;
 148}
 149
 150static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
 151                                           const unsigned char *sha1)
 152{
 153        git_zstream stream;
 154        unsigned char ibuf[1024 * 16];
 155        unsigned char obuf[1024 * 16];
 156        unsigned long olen = 0;
 157
 158        git_deflate_init(&stream, pack_compression_level);
 159
 160        for (;;) {
 161                ssize_t readlen;
 162                int zret = Z_OK;
 163                readlen = read_istream(st, ibuf, sizeof(ibuf));
 164                if (readlen == -1)
 165                        die(_("unable to read %s"), sha1_to_hex(sha1));
 166
 167                stream.next_in = ibuf;
 168                stream.avail_in = readlen;
 169                while ((stream.avail_in || readlen == 0) &&
 170                       (zret == Z_OK || zret == Z_BUF_ERROR)) {
 171                        stream.next_out = obuf;
 172                        stream.avail_out = sizeof(obuf);
 173                        zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
 174                        sha1write(f, obuf, stream.next_out - obuf);
 175                        olen += stream.next_out - obuf;
 176                }
 177                if (stream.avail_in)
 178                        die(_("deflate error (%d)"), zret);
 179                if (readlen == 0) {
 180                        if (zret != Z_STREAM_END)
 181                                die(_("deflate error (%d)"), zret);
 182                        break;
 183                }
 184        }
 185        git_deflate_end(&stream);
 186        return olen;
 187}
 188
 189/*
 190 * we are going to reuse the existing object data as is.  make
 191 * sure it is not corrupt.
 192 */
 193static int check_pack_inflate(struct packed_git *p,
 194                struct pack_window **w_curs,
 195                off_t offset,
 196                off_t len,
 197                unsigned long expect)
 198{
 199        git_zstream stream;
 200        unsigned char fakebuf[4096], *in;
 201        int st;
 202
 203        memset(&stream, 0, sizeof(stream));
 204        git_inflate_init(&stream);
 205        do {
 206                in = use_pack(p, w_curs, offset, &stream.avail_in);
 207                stream.next_in = in;
 208                stream.next_out = fakebuf;
 209                stream.avail_out = sizeof(fakebuf);
 210                st = git_inflate(&stream, Z_FINISH);
 211                offset += stream.next_in - in;
 212        } while (st == Z_OK || st == Z_BUF_ERROR);
 213        git_inflate_end(&stream);
 214        return (st == Z_STREAM_END &&
 215                stream.total_out == expect &&
 216                stream.total_in == len) ? 0 : -1;
 217}
 218
 219static void copy_pack_data(struct sha1file *f,
 220                struct packed_git *p,
 221                struct pack_window **w_curs,
 222                off_t offset,
 223                off_t len)
 224{
 225        unsigned char *in;
 226        unsigned long avail;
 227
 228        while (len) {
 229                in = use_pack(p, w_curs, offset, &avail);
 230                if (avail > len)
 231                        avail = (unsigned long)len;
 232                sha1write(f, in, avail);
 233                offset += avail;
 234                len -= avail;
 235        }
 236}
 237
 238/* Return 0 if we will bust the pack-size limit */
 239static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
 240                                           unsigned long limit, int usable_delta)
 241{
 242        unsigned long size, datalen;
 243        unsigned char header[10], dheader[10];
 244        unsigned hdrlen;
 245        enum object_type type;
 246        void *buf;
 247        struct git_istream *st = NULL;
 248
 249        if (!usable_delta) {
 250                if (entry->type == OBJ_BLOB &&
 251                    entry->size > big_file_threshold &&
 252                    (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
 253                        buf = NULL;
 254                else {
 255                        buf = read_sha1_file(entry->idx.sha1, &type, &size);
 256                        if (!buf)
 257                                die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
 258                }
 259                /*
 260                 * make sure no cached delta data remains from a
 261                 * previous attempt before a pack split occurred.
 262                 */
 263                free(entry->delta_data);
 264                entry->delta_data = NULL;
 265                entry->z_delta_size = 0;
 266        } else if (entry->delta_data) {
 267                size = entry->delta_size;
 268                buf = entry->delta_data;
 269                entry->delta_data = NULL;
 270                type = (allow_ofs_delta && entry->delta->idx.offset) ?
 271                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 272        } else {
 273                buf = get_delta(entry);
 274                size = entry->delta_size;
 275                type = (allow_ofs_delta && entry->delta->idx.offset) ?
 276                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 277        }
 278
 279        if (st) /* large blob case, just assume we don't compress well */
 280                datalen = size;
 281        else if (entry->z_delta_size)
 282                datalen = entry->z_delta_size;
 283        else
 284                datalen = do_compress(&buf, size);
 285
 286        /*
 287         * The object header is a byte of 'type' followed by zero or
 288         * more bytes of length.
 289         */
 290        hdrlen = encode_in_pack_object_header(type, size, header);
 291
 292        if (type == OBJ_OFS_DELTA) {
 293                /*
 294                 * Deltas with relative base contain an additional
 295                 * encoding of the relative offset for the delta
 296                 * base from this object's position in the pack.
 297                 */
 298                off_t ofs = entry->idx.offset - entry->delta->idx.offset;
 299                unsigned pos = sizeof(dheader) - 1;
 300                dheader[pos] = ofs & 127;
 301                while (ofs >>= 7)
 302                        dheader[--pos] = 128 | (--ofs & 127);
 303                if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 304                        if (st)
 305                                close_istream(st);
 306                        free(buf);
 307                        return 0;
 308                }
 309                sha1write(f, header, hdrlen);
 310                sha1write(f, dheader + pos, sizeof(dheader) - pos);
 311                hdrlen += sizeof(dheader) - pos;
 312        } else if (type == OBJ_REF_DELTA) {
 313                /*
 314                 * Deltas with a base reference contain
 315                 * an additional 20 bytes for the base sha1.
 316                 */
 317                if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 318                        if (st)
 319                                close_istream(st);
 320                        free(buf);
 321                        return 0;
 322                }
 323                sha1write(f, header, hdrlen);
 324                sha1write(f, entry->delta->idx.sha1, 20);
 325                hdrlen += 20;
 326        } else {
 327                if (limit && hdrlen + datalen + 20 >= limit) {
 328                        if (st)
 329                                close_istream(st);
 330                        free(buf);
 331                        return 0;
 332                }
 333                sha1write(f, header, hdrlen);
 334        }
 335        if (st) {
 336                datalen = write_large_blob_data(st, f, entry->idx.sha1);
 337                close_istream(st);
 338        } else {
 339                sha1write(f, buf, datalen);
 340                free(buf);
 341        }
 342
 343        return hdrlen + datalen;
 344}
 345
 346/* Return 0 if we will bust the pack-size limit */
 347static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry,
 348                                        unsigned long limit, int usable_delta)
 349{
 350        struct packed_git *p = entry->in_pack;
 351        struct pack_window *w_curs = NULL;
 352        struct revindex_entry *revidx;
 353        off_t offset;
 354        enum object_type type = entry->type;
 355        unsigned long datalen;
 356        unsigned char header[10], dheader[10];
 357        unsigned hdrlen;
 358
 359        if (entry->delta)
 360                type = (allow_ofs_delta && entry->delta->idx.offset) ?
 361                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 362        hdrlen = encode_in_pack_object_header(type, entry->size, header);
 363
 364        offset = entry->in_pack_offset;
 365        revidx = find_pack_revindex(p, offset);
 366        datalen = revidx[1].offset - offset;
 367        if (!pack_to_stdout && p->index_version > 1 &&
 368            check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
 369                error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
 370                unuse_pack(&w_curs);
 371                return write_no_reuse_object(f, entry, limit, usable_delta);
 372        }
 373
 374        offset += entry->in_pack_header_size;
 375        datalen -= entry->in_pack_header_size;
 376
 377        if (!pack_to_stdout && p->index_version == 1 &&
 378            check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
 379                error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
 380                unuse_pack(&w_curs);
 381                return write_no_reuse_object(f, entry, limit, usable_delta);
 382        }
 383
 384        if (type == OBJ_OFS_DELTA) {
 385                off_t ofs = entry->idx.offset - entry->delta->idx.offset;
 386                unsigned pos = sizeof(dheader) - 1;
 387                dheader[pos] = ofs & 127;
 388                while (ofs >>= 7)
 389                        dheader[--pos] = 128 | (--ofs & 127);
 390                if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 391                        unuse_pack(&w_curs);
 392                        return 0;
 393                }
 394                sha1write(f, header, hdrlen);
 395                sha1write(f, dheader + pos, sizeof(dheader) - pos);
 396                hdrlen += sizeof(dheader) - pos;
 397                reused_delta++;
 398        } else if (type == OBJ_REF_DELTA) {
 399                if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 400                        unuse_pack(&w_curs);
 401                        return 0;
 402                }
 403                sha1write(f, header, hdrlen);
 404                sha1write(f, entry->delta->idx.sha1, 20);
 405                hdrlen += 20;
 406                reused_delta++;
 407        } else {
 408                if (limit && hdrlen + datalen + 20 >= limit) {
 409                        unuse_pack(&w_curs);
 410                        return 0;
 411                }
 412                sha1write(f, header, hdrlen);
 413        }
 414        copy_pack_data(f, p, &w_curs, offset, datalen);
 415        unuse_pack(&w_curs);
 416        reused++;
 417        return hdrlen + datalen;
 418}
 419
 420/* Return 0 if we will bust the pack-size limit */
 421static unsigned long write_object(struct sha1file *f,
 422                                  struct object_entry *entry,
 423                                  off_t write_offset)
 424{
 425        unsigned long limit, len;
 426        int usable_delta, to_reuse;
 427
 428        if (!pack_to_stdout)
 429                crc32_begin(f);
 430
 431        /* apply size limit if limited packsize and not first object */
 432        if (!pack_size_limit || !nr_written)
 433                limit = 0;
 434        else if (pack_size_limit <= write_offset)
 435                /*
 436                 * the earlier object did not fit the limit; avoid
 437                 * mistaking this with unlimited (i.e. limit = 0).
 438                 */
 439                limit = 1;
 440        else
 441                limit = pack_size_limit - write_offset;
 442
 443        if (!entry->delta)
 444                usable_delta = 0;       /* no delta */
 445        else if (!pack_size_limit)
 446               usable_delta = 1;        /* unlimited packfile */
 447        else if (entry->delta->idx.offset == (off_t)-1)
 448                usable_delta = 0;       /* base was written to another pack */
 449        else if (entry->delta->idx.offset)
 450                usable_delta = 1;       /* base already exists in this pack */
 451        else
 452                usable_delta = 0;       /* base could end up in another pack */
 453
 454        if (!reuse_object)
 455                to_reuse = 0;   /* explicit */
 456        else if (!entry->in_pack)
 457                to_reuse = 0;   /* can't reuse what we don't have */
 458        else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
 459                                /* check_object() decided it for us ... */
 460                to_reuse = usable_delta;
 461                                /* ... but pack split may override that */
 462        else if (entry->type != entry->in_pack_type)
 463                to_reuse = 0;   /* pack has delta which is unusable */
 464        else if (entry->delta)
 465                to_reuse = 0;   /* we want to pack afresh */
 466        else
 467                to_reuse = 1;   /* we have it in-pack undeltified,
 468                                 * and we do not need to deltify it.
 469                                 */
 470
 471        if (!to_reuse)
 472                len = write_no_reuse_object(f, entry, limit, usable_delta);
 473        else
 474                len = write_reuse_object(f, entry, limit, usable_delta);
 475        if (!len)
 476                return 0;
 477
 478        if (usable_delta)
 479                written_delta++;
 480        written++;
 481        if (!pack_to_stdout)
 482                entry->idx.crc32 = crc32_end(f);
 483        return len;
 484}
 485
 486enum write_one_status {
 487        WRITE_ONE_SKIP = -1, /* already written */
 488        WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
 489        WRITE_ONE_WRITTEN = 1, /* normal */
 490        WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
 491};
 492
 493static enum write_one_status write_one(struct sha1file *f,
 494                                       struct object_entry *e,
 495                                       off_t *offset)
 496{
 497        unsigned long size;
 498        int recursing;
 499
 500        /*
 501         * we set offset to 1 (which is an impossible value) to mark
 502         * the fact that this object is involved in "write its base
 503         * first before writing a deltified object" recursion.
 504         */
 505        recursing = (e->idx.offset == 1);
 506        if (recursing) {
 507                warning("recursive delta detected for object %s",
 508                        sha1_to_hex(e->idx.sha1));
 509                return WRITE_ONE_RECURSIVE;
 510        } else if (e->idx.offset || e->preferred_base) {
 511                /* offset is non zero if object is written already. */
 512                return WRITE_ONE_SKIP;
 513        }
 514
 515        /* if we are deltified, write out base object first. */
 516        if (e->delta) {
 517                e->idx.offset = 1; /* now recurse */
 518                switch (write_one(f, e->delta, offset)) {
 519                case WRITE_ONE_RECURSIVE:
 520                        /* we cannot depend on this one */
 521                        e->delta = NULL;
 522                        break;
 523                default:
 524                        break;
 525                case WRITE_ONE_BREAK:
 526                        e->idx.offset = recursing;
 527                        return WRITE_ONE_BREAK;
 528                }
 529        }
 530
 531        e->idx.offset = *offset;
 532        size = write_object(f, e, *offset);
 533        if (!size) {
 534                e->idx.offset = recursing;
 535                return WRITE_ONE_BREAK;
 536        }
 537        written_list[nr_written++] = &e->idx;
 538
 539        /* make sure off_t is sufficiently large not to wrap */
 540        if (signed_add_overflows(*offset, size))
 541                die("pack too large for current definition of off_t");
 542        *offset += size;
 543        return WRITE_ONE_WRITTEN;
 544}
 545
 546static int mark_tagged(const char *path, const struct object_id *oid, int flag,
 547                       void *cb_data)
 548{
 549        unsigned char peeled[20];
 550        struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
 551
 552        if (entry)
 553                entry->tagged = 1;
 554        if (!peel_ref(path, peeled)) {
 555                entry = packlist_find(&to_pack, peeled, NULL);
 556                if (entry)
 557                        entry->tagged = 1;
 558        }
 559        return 0;
 560}
 561
 562static inline void add_to_write_order(struct object_entry **wo,
 563                               unsigned int *endp,
 564                               struct object_entry *e)
 565{
 566        if (e->filled)
 567                return;
 568        wo[(*endp)++] = e;
 569        e->filled = 1;
 570}
 571
 572static void add_descendants_to_write_order(struct object_entry **wo,
 573                                           unsigned int *endp,
 574                                           struct object_entry *e)
 575{
 576        int add_to_order = 1;
 577        while (e) {
 578                if (add_to_order) {
 579                        struct object_entry *s;
 580                        /* add this node... */
 581                        add_to_write_order(wo, endp, e);
 582                        /* all its siblings... */
 583                        for (s = e->delta_sibling; s; s = s->delta_sibling) {
 584                                add_to_write_order(wo, endp, s);
 585                        }
 586                }
 587                /* drop down a level to add left subtree nodes if possible */
 588                if (e->delta_child) {
 589                        add_to_order = 1;
 590                        e = e->delta_child;
 591                } else {
 592                        add_to_order = 0;
 593                        /* our sibling might have some children, it is next */
 594                        if (e->delta_sibling) {
 595                                e = e->delta_sibling;
 596                                continue;
 597                        }
 598                        /* go back to our parent node */
 599                        e = e->delta;
 600                        while (e && !e->delta_sibling) {
 601                                /* we're on the right side of a subtree, keep
 602                                 * going up until we can go right again */
 603                                e = e->delta;
 604                        }
 605                        if (!e) {
 606                                /* done- we hit our original root node */
 607                                return;
 608                        }
 609                        /* pass it off to sibling at this level */
 610                        e = e->delta_sibling;
 611                }
 612        };
 613}
 614
 615static void add_family_to_write_order(struct object_entry **wo,
 616                                      unsigned int *endp,
 617                                      struct object_entry *e)
 618{
 619        struct object_entry *root;
 620
 621        for (root = e; root->delta; root = root->delta)
 622                ; /* nothing */
 623        add_descendants_to_write_order(wo, endp, root);
 624}
 625
 626static struct object_entry **compute_write_order(void)
 627{
 628        unsigned int i, wo_end, last_untagged;
 629
 630        struct object_entry **wo;
 631        struct object_entry *objects = to_pack.objects;
 632
 633        for (i = 0; i < to_pack.nr_objects; i++) {
 634                objects[i].tagged = 0;
 635                objects[i].filled = 0;
 636                objects[i].delta_child = NULL;
 637                objects[i].delta_sibling = NULL;
 638        }
 639
 640        /*
 641         * Fully connect delta_child/delta_sibling network.
 642         * Make sure delta_sibling is sorted in the original
 643         * recency order.
 644         */
 645        for (i = to_pack.nr_objects; i > 0;) {
 646                struct object_entry *e = &objects[--i];
 647                if (!e->delta)
 648                        continue;
 649                /* Mark me as the first child */
 650                e->delta_sibling = e->delta->delta_child;
 651                e->delta->delta_child = e;
 652        }
 653
 654        /*
 655         * Mark objects that are at the tip of tags.
 656         */
 657        for_each_tag_ref(mark_tagged, NULL);
 658
 659        /*
 660         * Give the objects in the original recency order until
 661         * we see a tagged tip.
 662         */
 663        ALLOC_ARRAY(wo, to_pack.nr_objects);
 664        for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
 665                if (objects[i].tagged)
 666                        break;
 667                add_to_write_order(wo, &wo_end, &objects[i]);
 668        }
 669        last_untagged = i;
 670
 671        /*
 672         * Then fill all the tagged tips.
 673         */
 674        for (; i < to_pack.nr_objects; i++) {
 675                if (objects[i].tagged)
 676                        add_to_write_order(wo, &wo_end, &objects[i]);
 677        }
 678
 679        /*
 680         * And then all remaining commits and tags.
 681         */
 682        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 683                if (objects[i].type != OBJ_COMMIT &&
 684                    objects[i].type != OBJ_TAG)
 685                        continue;
 686                add_to_write_order(wo, &wo_end, &objects[i]);
 687        }
 688
 689        /*
 690         * And then all the trees.
 691         */
 692        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 693                if (objects[i].type != OBJ_TREE)
 694                        continue;
 695                add_to_write_order(wo, &wo_end, &objects[i]);
 696        }
 697
 698        /*
 699         * Finally all the rest in really tight order
 700         */
 701        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 702                if (!objects[i].filled)
 703                        add_family_to_write_order(wo, &wo_end, &objects[i]);
 704        }
 705
 706        if (wo_end != to_pack.nr_objects)
 707                die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
 708
 709        return wo;
 710}
 711
 712static off_t write_reused_pack(struct sha1file *f)
 713{
 714        unsigned char buffer[8192];
 715        off_t to_write, total;
 716        int fd;
 717
 718        if (!is_pack_valid(reuse_packfile))
 719                die("packfile is invalid: %s", reuse_packfile->pack_name);
 720
 721        fd = git_open_noatime(reuse_packfile->pack_name);
 722        if (fd < 0)
 723                die_errno("unable to open packfile for reuse: %s",
 724                          reuse_packfile->pack_name);
 725
 726        if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
 727                die_errno("unable to seek in reused packfile");
 728
 729        if (reuse_packfile_offset < 0)
 730                reuse_packfile_offset = reuse_packfile->pack_size - 20;
 731
 732        total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
 733
 734        while (to_write) {
 735                int read_pack = xread(fd, buffer, sizeof(buffer));
 736
 737                if (read_pack <= 0)
 738                        die_errno("unable to read from reused packfile");
 739
 740                if (read_pack > to_write)
 741                        read_pack = to_write;
 742
 743                sha1write(f, buffer, read_pack);
 744                to_write -= read_pack;
 745
 746                /*
 747                 * We don't know the actual number of objects written,
 748                 * only how many bytes written, how many bytes total, and
 749                 * how many objects total. So we can fake it by pretending all
 750                 * objects we are writing are the same size. This gives us a
 751                 * smooth progress meter, and at the end it matches the true
 752                 * answer.
 753                 */
 754                written = reuse_packfile_objects *
 755                                (((double)(total - to_write)) / total);
 756                display_progress(progress_state, written);
 757        }
 758
 759        close(fd);
 760        written = reuse_packfile_objects;
 761        display_progress(progress_state, written);
 762        return reuse_packfile_offset - sizeof(struct pack_header);
 763}
 764
 765static const char no_split_warning[] = N_(
 766"disabling bitmap writing, packs are split due to pack.packSizeLimit"
 767);
 768
 769static void write_pack_file(void)
 770{
 771        uint32_t i = 0, j;
 772        struct sha1file *f;
 773        off_t offset;
 774        uint32_t nr_remaining = nr_result;
 775        time_t last_mtime = 0;
 776        struct object_entry **write_order;
 777
 778        if (progress > pack_to_stdout)
 779                progress_state = start_progress(_("Writing objects"), nr_result);
 780        ALLOC_ARRAY(written_list, to_pack.nr_objects);
 781        write_order = compute_write_order();
 782
 783        do {
 784                unsigned char sha1[20];
 785                char *pack_tmp_name = NULL;
 786
 787                if (pack_to_stdout)
 788                        f = sha1fd_throughput(1, "<stdout>", progress_state);
 789                else
 790                        f = create_tmp_packfile(&pack_tmp_name);
 791
 792                offset = write_pack_header(f, nr_remaining);
 793
 794                if (reuse_packfile) {
 795                        off_t packfile_size;
 796                        assert(pack_to_stdout);
 797
 798                        packfile_size = write_reused_pack(f);
 799                        offset += packfile_size;
 800                }
 801
 802                nr_written = 0;
 803                for (; i < to_pack.nr_objects; i++) {
 804                        struct object_entry *e = write_order[i];
 805                        if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
 806                                break;
 807                        display_progress(progress_state, written);
 808                }
 809
 810                /*
 811                 * Did we write the wrong # entries in the header?
 812                 * If so, rewrite it like in fast-import
 813                 */
 814                if (pack_to_stdout) {
 815                        sha1close(f, sha1, CSUM_CLOSE);
 816                } else if (nr_written == nr_remaining) {
 817                        sha1close(f, sha1, CSUM_FSYNC);
 818                } else {
 819                        int fd = sha1close(f, sha1, 0);
 820                        fixup_pack_header_footer(fd, sha1, pack_tmp_name,
 821                                                 nr_written, sha1, offset);
 822                        close(fd);
 823                        if (write_bitmap_index) {
 824                                warning(_(no_split_warning));
 825                                write_bitmap_index = 0;
 826                        }
 827                }
 828
 829                if (!pack_to_stdout) {
 830                        struct stat st;
 831                        struct strbuf tmpname = STRBUF_INIT;
 832
 833                        /*
 834                         * Packs are runtime accessed in their mtime
 835                         * order since newer packs are more likely to contain
 836                         * younger objects.  So if we are creating multiple
 837                         * packs then we should modify the mtime of later ones
 838                         * to preserve this property.
 839                         */
 840                        if (stat(pack_tmp_name, &st) < 0) {
 841                                warning_errno("failed to stat %s", pack_tmp_name);
 842                        } else if (!last_mtime) {
 843                                last_mtime = st.st_mtime;
 844                        } else {
 845                                struct utimbuf utb;
 846                                utb.actime = st.st_atime;
 847                                utb.modtime = --last_mtime;
 848                                if (utime(pack_tmp_name, &utb) < 0)
 849                                        warning_errno("failed utime() on %s", pack_tmp_name);
 850                        }
 851
 852                        strbuf_addf(&tmpname, "%s-", base_name);
 853
 854                        if (write_bitmap_index) {
 855                                bitmap_writer_set_checksum(sha1);
 856                                bitmap_writer_build_type_index(written_list, nr_written);
 857                        }
 858
 859                        finish_tmp_packfile(&tmpname, pack_tmp_name,
 860                                            written_list, nr_written,
 861                                            &pack_idx_opts, sha1);
 862
 863                        if (write_bitmap_index) {
 864                                strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
 865
 866                                stop_progress(&progress_state);
 867
 868                                bitmap_writer_show_progress(progress);
 869                                bitmap_writer_reuse_bitmaps(&to_pack);
 870                                bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
 871                                bitmap_writer_build(&to_pack);
 872                                bitmap_writer_finish(written_list, nr_written,
 873                                                     tmpname.buf, write_bitmap_options);
 874                                write_bitmap_index = 0;
 875                        }
 876
 877                        strbuf_release(&tmpname);
 878                        free(pack_tmp_name);
 879                        puts(sha1_to_hex(sha1));
 880                }
 881
 882                /* mark written objects as written to previous pack */
 883                for (j = 0; j < nr_written; j++) {
 884                        written_list[j]->offset = (off_t)-1;
 885                }
 886                nr_remaining -= nr_written;
 887        } while (nr_remaining && i < to_pack.nr_objects);
 888
 889        free(written_list);
 890        free(write_order);
 891        stop_progress(&progress_state);
 892        if (written != nr_result)
 893                die("wrote %"PRIu32" objects while expecting %"PRIu32,
 894                        written, nr_result);
 895}
 896
 897static void setup_delta_attr_check(struct git_attr_check *check)
 898{
 899        static struct git_attr *attr_delta;
 900
 901        if (!attr_delta)
 902                attr_delta = git_attr("delta");
 903
 904        check[0].attr = attr_delta;
 905}
 906
 907static int no_try_delta(const char *path)
 908{
 909        struct git_attr_check check[1];
 910
 911        setup_delta_attr_check(check);
 912        if (git_check_attr(path, ARRAY_SIZE(check), check))
 913                return 0;
 914        if (ATTR_FALSE(check->value))
 915                return 1;
 916        return 0;
 917}
 918
 919/*
 920 * When adding an object, check whether we have already added it
 921 * to our packing list. If so, we can skip. However, if we are
 922 * being asked to excludei t, but the previous mention was to include
 923 * it, make sure to adjust its flags and tweak our numbers accordingly.
 924 *
 925 * As an optimization, we pass out the index position where we would have
 926 * found the item, since that saves us from having to look it up again a
 927 * few lines later when we want to add the new entry.
 928 */
 929static int have_duplicate_entry(const unsigned char *sha1,
 930                                int exclude,
 931                                uint32_t *index_pos)
 932{
 933        struct object_entry *entry;
 934
 935        entry = packlist_find(&to_pack, sha1, index_pos);
 936        if (!entry)
 937                return 0;
 938
 939        if (exclude) {
 940                if (!entry->preferred_base)
 941                        nr_result--;
 942                entry->preferred_base = 1;
 943        }
 944
 945        return 1;
 946}
 947
 948/*
 949 * Check whether we want the object in the pack (e.g., we do not want
 950 * objects found in non-local stores if the "--local" option was used).
 951 *
 952 * As a side effect of this check, we will find the packed version of this
 953 * object, if any. We therefore pass out the pack information to avoid having
 954 * to look it up again later.
 955 */
 956static int want_object_in_pack(const unsigned char *sha1,
 957                               int exclude,
 958                               struct packed_git **found_pack,
 959                               off_t *found_offset)
 960{
 961        struct mru_entry *entry;
 962
 963        if (!exclude && local && has_loose_object_nonlocal(sha1))
 964                return 0;
 965
 966        *found_pack = NULL;
 967        *found_offset = 0;
 968
 969        for (entry = packed_git_mru->head; entry; entry = entry->next) {
 970                struct packed_git *p = entry->item;
 971                off_t offset = find_pack_entry_one(sha1, p);
 972                if (offset) {
 973                        if (!*found_pack) {
 974                                if (!is_pack_valid(p))
 975                                        continue;
 976                                *found_offset = offset;
 977                                *found_pack = p;
 978                        }
 979                        if (exclude)
 980                                return 1;
 981                        if (incremental)
 982                                return 0;
 983
 984                        /*
 985                         * When asked to do --local (do not include an
 986                         * object that appears in a pack we borrow
 987                         * from elsewhere) or --honor-pack-keep (do not
 988                         * include an object that appears in a pack marked
 989                         * with .keep), we need to make sure no copy of this
 990                         * object come from in _any_ pack that causes us to
 991                         * omit it, and need to complete this loop.  When
 992                         * neither option is in effect, we know the object
 993                         * we just found is going to be packed, so break
 994                         * out of the loop to return 1 now.
 995                         */
 996                        if (!ignore_packed_keep &&
 997                            (!local || !have_non_local_packs)) {
 998                                mru_mark(packed_git_mru, entry);
 999                                break;
1000                        }
1001
1002                        if (local && !p->pack_local)
1003                                return 0;
1004                        if (ignore_packed_keep && p->pack_local && p->pack_keep)
1005                                return 0;
1006                }
1007        }
1008
1009        return 1;
1010}
1011
1012static void create_object_entry(const unsigned char *sha1,
1013                                enum object_type type,
1014                                uint32_t hash,
1015                                int exclude,
1016                                int no_try_delta,
1017                                uint32_t index_pos,
1018                                struct packed_git *found_pack,
1019                                off_t found_offset)
1020{
1021        struct object_entry *entry;
1022
1023        entry = packlist_alloc(&to_pack, sha1, index_pos);
1024        entry->hash = hash;
1025        if (type)
1026                entry->type = type;
1027        if (exclude)
1028                entry->preferred_base = 1;
1029        else
1030                nr_result++;
1031        if (found_pack) {
1032                entry->in_pack = found_pack;
1033                entry->in_pack_offset = found_offset;
1034        }
1035
1036        entry->no_try_delta = no_try_delta;
1037}
1038
1039static const char no_closure_warning[] = N_(
1040"disabling bitmap writing, as some objects are not being packed"
1041);
1042
1043static int add_object_entry(const unsigned char *sha1, enum object_type type,
1044                            const char *name, int exclude)
1045{
1046        struct packed_git *found_pack;
1047        off_t found_offset;
1048        uint32_t index_pos;
1049
1050        if (have_duplicate_entry(sha1, exclude, &index_pos))
1051                return 0;
1052
1053        if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1054                /* The pack is missing an object, so it will not have closure */
1055                if (write_bitmap_index) {
1056                        warning(_(no_closure_warning));
1057                        write_bitmap_index = 0;
1058                }
1059                return 0;
1060        }
1061
1062        create_object_entry(sha1, type, pack_name_hash(name),
1063                            exclude, name && no_try_delta(name),
1064                            index_pos, found_pack, found_offset);
1065
1066        display_progress(progress_state, nr_result);
1067        return 1;
1068}
1069
1070static int add_object_entry_from_bitmap(const unsigned char *sha1,
1071                                        enum object_type type,
1072                                        int flags, uint32_t name_hash,
1073                                        struct packed_git *pack, off_t offset)
1074{
1075        uint32_t index_pos;
1076
1077        if (have_duplicate_entry(sha1, 0, &index_pos))
1078                return 0;
1079
1080        create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1081
1082        display_progress(progress_state, nr_result);
1083        return 1;
1084}
1085
1086struct pbase_tree_cache {
1087        unsigned char sha1[20];
1088        int ref;
1089        int temporary;
1090        void *tree_data;
1091        unsigned long tree_size;
1092};
1093
1094static struct pbase_tree_cache *(pbase_tree_cache[256]);
1095static int pbase_tree_cache_ix(const unsigned char *sha1)
1096{
1097        return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1098}
1099static int pbase_tree_cache_ix_incr(int ix)
1100{
1101        return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1102}
1103
1104static struct pbase_tree {
1105        struct pbase_tree *next;
1106        /* This is a phony "cache" entry; we are not
1107         * going to evict it or find it through _get()
1108         * mechanism -- this is for the toplevel node that
1109         * would almost always change with any commit.
1110         */
1111        struct pbase_tree_cache pcache;
1112} *pbase_tree;
1113
1114static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1115{
1116        struct pbase_tree_cache *ent, *nent;
1117        void *data;
1118        unsigned long size;
1119        enum object_type type;
1120        int neigh;
1121        int my_ix = pbase_tree_cache_ix(sha1);
1122        int available_ix = -1;
1123
1124        /* pbase-tree-cache acts as a limited hashtable.
1125         * your object will be found at your index or within a few
1126         * slots after that slot if it is cached.
1127         */
1128        for (neigh = 0; neigh < 8; neigh++) {
1129                ent = pbase_tree_cache[my_ix];
1130                if (ent && !hashcmp(ent->sha1, sha1)) {
1131                        ent->ref++;
1132                        return ent;
1133                }
1134                else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1135                         ((0 <= available_ix) &&
1136                          (!ent && pbase_tree_cache[available_ix])))
1137                        available_ix = my_ix;
1138                if (!ent)
1139                        break;
1140                my_ix = pbase_tree_cache_ix_incr(my_ix);
1141        }
1142
1143        /* Did not find one.  Either we got a bogus request or
1144         * we need to read and perhaps cache.
1145         */
1146        data = read_sha1_file(sha1, &type, &size);
1147        if (!data)
1148                return NULL;
1149        if (type != OBJ_TREE) {
1150                free(data);
1151                return NULL;
1152        }
1153
1154        /* We need to either cache or return a throwaway copy */
1155
1156        if (available_ix < 0)
1157                ent = NULL;
1158        else {
1159                ent = pbase_tree_cache[available_ix];
1160                my_ix = available_ix;
1161        }
1162
1163        if (!ent) {
1164                nent = xmalloc(sizeof(*nent));
1165                nent->temporary = (available_ix < 0);
1166        }
1167        else {
1168                /* evict and reuse */
1169                free(ent->tree_data);
1170                nent = ent;
1171        }
1172        hashcpy(nent->sha1, sha1);
1173        nent->tree_data = data;
1174        nent->tree_size = size;
1175        nent->ref = 1;
1176        if (!nent->temporary)
1177                pbase_tree_cache[my_ix] = nent;
1178        return nent;
1179}
1180
1181static void pbase_tree_put(struct pbase_tree_cache *cache)
1182{
1183        if (!cache->temporary) {
1184                cache->ref--;
1185                return;
1186        }
1187        free(cache->tree_data);
1188        free(cache);
1189}
1190
1191static int name_cmp_len(const char *name)
1192{
1193        int i;
1194        for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1195                ;
1196        return i;
1197}
1198
1199static void add_pbase_object(struct tree_desc *tree,
1200                             const char *name,
1201                             int cmplen,
1202                             const char *fullname)
1203{
1204        struct name_entry entry;
1205        int cmp;
1206
1207        while (tree_entry(tree,&entry)) {
1208                if (S_ISGITLINK(entry.mode))
1209                        continue;
1210                cmp = tree_entry_len(&entry) != cmplen ? 1 :
1211                      memcmp(name, entry.path, cmplen);
1212                if (cmp > 0)
1213                        continue;
1214                if (cmp < 0)
1215                        return;
1216                if (name[cmplen] != '/') {
1217                        add_object_entry(entry.oid->hash,
1218                                         object_type(entry.mode),
1219                                         fullname, 1);
1220                        return;
1221                }
1222                if (S_ISDIR(entry.mode)) {
1223                        struct tree_desc sub;
1224                        struct pbase_tree_cache *tree;
1225                        const char *down = name+cmplen+1;
1226                        int downlen = name_cmp_len(down);
1227
1228                        tree = pbase_tree_get(entry.oid->hash);
1229                        if (!tree)
1230                                return;
1231                        init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1232
1233                        add_pbase_object(&sub, down, downlen, fullname);
1234                        pbase_tree_put(tree);
1235                }
1236        }
1237}
1238
1239static unsigned *done_pbase_paths;
1240static int done_pbase_paths_num;
1241static int done_pbase_paths_alloc;
1242static int done_pbase_path_pos(unsigned hash)
1243{
1244        int lo = 0;
1245        int hi = done_pbase_paths_num;
1246        while (lo < hi) {
1247                int mi = (hi + lo) / 2;
1248                if (done_pbase_paths[mi] == hash)
1249                        return mi;
1250                if (done_pbase_paths[mi] < hash)
1251                        hi = mi;
1252                else
1253                        lo = mi + 1;
1254        }
1255        return -lo-1;
1256}
1257
1258static int check_pbase_path(unsigned hash)
1259{
1260        int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1261        if (0 <= pos)
1262                return 1;
1263        pos = -pos - 1;
1264        ALLOC_GROW(done_pbase_paths,
1265                   done_pbase_paths_num + 1,
1266                   done_pbase_paths_alloc);
1267        done_pbase_paths_num++;
1268        if (pos < done_pbase_paths_num)
1269                memmove(done_pbase_paths + pos + 1,
1270                        done_pbase_paths + pos,
1271                        (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1272        done_pbase_paths[pos] = hash;
1273        return 0;
1274}
1275
1276static void add_preferred_base_object(const char *name)
1277{
1278        struct pbase_tree *it;
1279        int cmplen;
1280        unsigned hash = pack_name_hash(name);
1281
1282        if (!num_preferred_base || check_pbase_path(hash))
1283                return;
1284
1285        cmplen = name_cmp_len(name);
1286        for (it = pbase_tree; it; it = it->next) {
1287                if (cmplen == 0) {
1288                        add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1289                }
1290                else {
1291                        struct tree_desc tree;
1292                        init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1293                        add_pbase_object(&tree, name, cmplen, name);
1294                }
1295        }
1296}
1297
1298static void add_preferred_base(unsigned char *sha1)
1299{
1300        struct pbase_tree *it;
1301        void *data;
1302        unsigned long size;
1303        unsigned char tree_sha1[20];
1304
1305        if (window <= num_preferred_base++)
1306                return;
1307
1308        data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1309        if (!data)
1310                return;
1311
1312        for (it = pbase_tree; it; it = it->next) {
1313                if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1314                        free(data);
1315                        return;
1316                }
1317        }
1318
1319        it = xcalloc(1, sizeof(*it));
1320        it->next = pbase_tree;
1321        pbase_tree = it;
1322
1323        hashcpy(it->pcache.sha1, tree_sha1);
1324        it->pcache.tree_data = data;
1325        it->pcache.tree_size = size;
1326}
1327
1328static void cleanup_preferred_base(void)
1329{
1330        struct pbase_tree *it;
1331        unsigned i;
1332
1333        it = pbase_tree;
1334        pbase_tree = NULL;
1335        while (it) {
1336                struct pbase_tree *this = it;
1337                it = this->next;
1338                free(this->pcache.tree_data);
1339                free(this);
1340        }
1341
1342        for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1343                if (!pbase_tree_cache[i])
1344                        continue;
1345                free(pbase_tree_cache[i]->tree_data);
1346                free(pbase_tree_cache[i]);
1347                pbase_tree_cache[i] = NULL;
1348        }
1349
1350        free(done_pbase_paths);
1351        done_pbase_paths = NULL;
1352        done_pbase_paths_num = done_pbase_paths_alloc = 0;
1353}
1354
1355static void check_object(struct object_entry *entry)
1356{
1357        if (entry->in_pack) {
1358                struct packed_git *p = entry->in_pack;
1359                struct pack_window *w_curs = NULL;
1360                const unsigned char *base_ref = NULL;
1361                struct object_entry *base_entry;
1362                unsigned long used, used_0;
1363                unsigned long avail;
1364                off_t ofs;
1365                unsigned char *buf, c;
1366
1367                buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1368
1369                /*
1370                 * We want in_pack_type even if we do not reuse delta
1371                 * since non-delta representations could still be reused.
1372                 */
1373                used = unpack_object_header_buffer(buf, avail,
1374                                                   &entry->in_pack_type,
1375                                                   &entry->size);
1376                if (used == 0)
1377                        goto give_up;
1378
1379                /*
1380                 * Determine if this is a delta and if so whether we can
1381                 * reuse it or not.  Otherwise let's find out as cheaply as
1382                 * possible what the actual type and size for this object is.
1383                 */
1384                switch (entry->in_pack_type) {
1385                default:
1386                        /* Not a delta hence we've already got all we need. */
1387                        entry->type = entry->in_pack_type;
1388                        entry->in_pack_header_size = used;
1389                        if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1390                                goto give_up;
1391                        unuse_pack(&w_curs);
1392                        return;
1393                case OBJ_REF_DELTA:
1394                        if (reuse_delta && !entry->preferred_base)
1395                                base_ref = use_pack(p, &w_curs,
1396                                                entry->in_pack_offset + used, NULL);
1397                        entry->in_pack_header_size = used + 20;
1398                        break;
1399                case OBJ_OFS_DELTA:
1400                        buf = use_pack(p, &w_curs,
1401                                       entry->in_pack_offset + used, NULL);
1402                        used_0 = 0;
1403                        c = buf[used_0++];
1404                        ofs = c & 127;
1405                        while (c & 128) {
1406                                ofs += 1;
1407                                if (!ofs || MSB(ofs, 7)) {
1408                                        error("delta base offset overflow in pack for %s",
1409                                              sha1_to_hex(entry->idx.sha1));
1410                                        goto give_up;
1411                                }
1412                                c = buf[used_0++];
1413                                ofs = (ofs << 7) + (c & 127);
1414                        }
1415                        ofs = entry->in_pack_offset - ofs;
1416                        if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1417                                error("delta base offset out of bound for %s",
1418                                      sha1_to_hex(entry->idx.sha1));
1419                                goto give_up;
1420                        }
1421                        if (reuse_delta && !entry->preferred_base) {
1422                                struct revindex_entry *revidx;
1423                                revidx = find_pack_revindex(p, ofs);
1424                                if (!revidx)
1425                                        goto give_up;
1426                                base_ref = nth_packed_object_sha1(p, revidx->nr);
1427                        }
1428                        entry->in_pack_header_size = used + used_0;
1429                        break;
1430                }
1431
1432                if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1433                        /*
1434                         * If base_ref was set above that means we wish to
1435                         * reuse delta data, and we even found that base
1436                         * in the list of objects we want to pack. Goodie!
1437                         *
1438                         * Depth value does not matter - find_deltas() will
1439                         * never consider reused delta as the base object to
1440                         * deltify other objects against, in order to avoid
1441                         * circular deltas.
1442                         */
1443                        entry->type = entry->in_pack_type;
1444                        entry->delta = base_entry;
1445                        entry->delta_size = entry->size;
1446                        entry->delta_sibling = base_entry->delta_child;
1447                        base_entry->delta_child = entry;
1448                        unuse_pack(&w_curs);
1449                        return;
1450                }
1451
1452                if (entry->type) {
1453                        /*
1454                         * This must be a delta and we already know what the
1455                         * final object type is.  Let's extract the actual
1456                         * object size from the delta header.
1457                         */
1458                        entry->size = get_size_from_delta(p, &w_curs,
1459                                        entry->in_pack_offset + entry->in_pack_header_size);
1460                        if (entry->size == 0)
1461                                goto give_up;
1462                        unuse_pack(&w_curs);
1463                        return;
1464                }
1465
1466                /*
1467                 * No choice but to fall back to the recursive delta walk
1468                 * with sha1_object_info() to find about the object type
1469                 * at this point...
1470                 */
1471                give_up:
1472                unuse_pack(&w_curs);
1473        }
1474
1475        entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1476        /*
1477         * The error condition is checked in prepare_pack().  This is
1478         * to permit a missing preferred base object to be ignored
1479         * as a preferred base.  Doing so can result in a larger
1480         * pack file, but the transfer will still take place.
1481         */
1482}
1483
1484static int pack_offset_sort(const void *_a, const void *_b)
1485{
1486        const struct object_entry *a = *(struct object_entry **)_a;
1487        const struct object_entry *b = *(struct object_entry **)_b;
1488
1489        /* avoid filesystem trashing with loose objects */
1490        if (!a->in_pack && !b->in_pack)
1491                return hashcmp(a->idx.sha1, b->idx.sha1);
1492
1493        if (a->in_pack < b->in_pack)
1494                return -1;
1495        if (a->in_pack > b->in_pack)
1496                return 1;
1497        return a->in_pack_offset < b->in_pack_offset ? -1 :
1498                        (a->in_pack_offset > b->in_pack_offset);
1499}
1500
1501/*
1502 * Drop an on-disk delta we were planning to reuse. Naively, this would
1503 * just involve blanking out the "delta" field, but we have to deal
1504 * with some extra book-keeping:
1505 *
1506 *   1. Removing ourselves from the delta_sibling linked list.
1507 *
1508 *   2. Updating our size/type to the non-delta representation. These were
1509 *      either not recorded initially (size) or overwritten with the delta type
1510 *      (type) when check_object() decided to reuse the delta.
1511 */
1512static void drop_reused_delta(struct object_entry *entry)
1513{
1514        struct object_entry **p = &entry->delta->delta_child;
1515        struct object_info oi = OBJECT_INFO_INIT;
1516
1517        while (*p) {
1518                if (*p == entry)
1519                        *p = (*p)->delta_sibling;
1520                else
1521                        p = &(*p)->delta_sibling;
1522        }
1523        entry->delta = NULL;
1524
1525        oi.sizep = &entry->size;
1526        oi.typep = &entry->type;
1527        if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1528                /*
1529                 * We failed to get the info from this pack for some reason;
1530                 * fall back to sha1_object_info, which may find another copy.
1531                 * And if that fails, the error will be recorded in entry->type
1532                 * and dealt with in prepare_pack().
1533                 */
1534                entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1535        }
1536}
1537
1538/*
1539 * Follow the chain of deltas from this entry onward, throwing away any links
1540 * that cause us to hit a cycle (as determined by the DFS state flags in
1541 * the entries).
1542 */
1543static void break_delta_chains(struct object_entry *entry)
1544{
1545        /* If it's not a delta, it can't be part of a cycle. */
1546        if (!entry->delta) {
1547                entry->dfs_state = DFS_DONE;
1548                return;
1549        }
1550
1551        switch (entry->dfs_state) {
1552        case DFS_NONE:
1553                /*
1554                 * This is the first time we've seen the object. We mark it as
1555                 * part of the active potential cycle and recurse.
1556                 */
1557                entry->dfs_state = DFS_ACTIVE;
1558                break_delta_chains(entry->delta);
1559                entry->dfs_state = DFS_DONE;
1560                break;
1561
1562        case DFS_DONE:
1563                /* object already examined, and not part of a cycle */
1564                break;
1565
1566        case DFS_ACTIVE:
1567                /*
1568                 * We found a cycle that needs broken. It would be correct to
1569                 * break any link in the chain, but it's convenient to
1570                 * break this one.
1571                 */
1572                drop_reused_delta(entry);
1573                entry->dfs_state = DFS_DONE;
1574                break;
1575        }
1576}
1577
1578static void get_object_details(void)
1579{
1580        uint32_t i;
1581        struct object_entry **sorted_by_offset;
1582
1583        sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1584        for (i = 0; i < to_pack.nr_objects; i++)
1585                sorted_by_offset[i] = to_pack.objects + i;
1586        qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1587
1588        for (i = 0; i < to_pack.nr_objects; i++) {
1589                struct object_entry *entry = sorted_by_offset[i];
1590                check_object(entry);
1591                if (big_file_threshold < entry->size)
1592                        entry->no_try_delta = 1;
1593        }
1594
1595        /*
1596         * This must happen in a second pass, since we rely on the delta
1597         * information for the whole list being completed.
1598         */
1599        for (i = 0; i < to_pack.nr_objects; i++)
1600                break_delta_chains(&to_pack.objects[i]);
1601
1602        free(sorted_by_offset);
1603}
1604
1605/*
1606 * We search for deltas in a list sorted by type, by filename hash, and then
1607 * by size, so that we see progressively smaller and smaller files.
1608 * That's because we prefer deltas to be from the bigger file
1609 * to the smaller -- deletes are potentially cheaper, but perhaps
1610 * more importantly, the bigger file is likely the more recent
1611 * one.  The deepest deltas are therefore the oldest objects which are
1612 * less susceptible to be accessed often.
1613 */
1614static int type_size_sort(const void *_a, const void *_b)
1615{
1616        const struct object_entry *a = *(struct object_entry **)_a;
1617        const struct object_entry *b = *(struct object_entry **)_b;
1618
1619        if (a->type > b->type)
1620                return -1;
1621        if (a->type < b->type)
1622                return 1;
1623        if (a->hash > b->hash)
1624                return -1;
1625        if (a->hash < b->hash)
1626                return 1;
1627        if (a->preferred_base > b->preferred_base)
1628                return -1;
1629        if (a->preferred_base < b->preferred_base)
1630                return 1;
1631        if (a->size > b->size)
1632                return -1;
1633        if (a->size < b->size)
1634                return 1;
1635        return a < b ? -1 : (a > b);  /* newest first */
1636}
1637
1638struct unpacked {
1639        struct object_entry *entry;
1640        void *data;
1641        struct delta_index *index;
1642        unsigned depth;
1643};
1644
1645static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1646                           unsigned long delta_size)
1647{
1648        if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1649                return 0;
1650
1651        if (delta_size < cache_max_small_delta_size)
1652                return 1;
1653
1654        /* cache delta, if objects are large enough compared to delta size */
1655        if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1656                return 1;
1657
1658        return 0;
1659}
1660
1661#ifndef NO_PTHREADS
1662
1663static pthread_mutex_t read_mutex;
1664#define read_lock()             pthread_mutex_lock(&read_mutex)
1665#define read_unlock()           pthread_mutex_unlock(&read_mutex)
1666
1667static pthread_mutex_t cache_mutex;
1668#define cache_lock()            pthread_mutex_lock(&cache_mutex)
1669#define cache_unlock()          pthread_mutex_unlock(&cache_mutex)
1670
1671static pthread_mutex_t progress_mutex;
1672#define progress_lock()         pthread_mutex_lock(&progress_mutex)
1673#define progress_unlock()       pthread_mutex_unlock(&progress_mutex)
1674
1675#else
1676
1677#define read_lock()             (void)0
1678#define read_unlock()           (void)0
1679#define cache_lock()            (void)0
1680#define cache_unlock()          (void)0
1681#define progress_lock()         (void)0
1682#define progress_unlock()       (void)0
1683
1684#endif
1685
1686static int try_delta(struct unpacked *trg, struct unpacked *src,
1687                     unsigned max_depth, unsigned long *mem_usage)
1688{
1689        struct object_entry *trg_entry = trg->entry;
1690        struct object_entry *src_entry = src->entry;
1691        unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1692        unsigned ref_depth;
1693        enum object_type type;
1694        void *delta_buf;
1695
1696        /* Don't bother doing diffs between different types */
1697        if (trg_entry->type != src_entry->type)
1698                return -1;
1699
1700        /*
1701         * We do not bother to try a delta that we discarded on an
1702         * earlier try, but only when reusing delta data.  Note that
1703         * src_entry that is marked as the preferred_base should always
1704         * be considered, as even if we produce a suboptimal delta against
1705         * it, we will still save the transfer cost, as we already know
1706         * the other side has it and we won't send src_entry at all.
1707         */
1708        if (reuse_delta && trg_entry->in_pack &&
1709            trg_entry->in_pack == src_entry->in_pack &&
1710            !src_entry->preferred_base &&
1711            trg_entry->in_pack_type != OBJ_REF_DELTA &&
1712            trg_entry->in_pack_type != OBJ_OFS_DELTA)
1713                return 0;
1714
1715        /* Let's not bust the allowed depth. */
1716        if (src->depth >= max_depth)
1717                return 0;
1718
1719        /* Now some size filtering heuristics. */
1720        trg_size = trg_entry->size;
1721        if (!trg_entry->delta) {
1722                max_size = trg_size/2 - 20;
1723                ref_depth = 1;
1724        } else {
1725                max_size = trg_entry->delta_size;
1726                ref_depth = trg->depth;
1727        }
1728        max_size = (uint64_t)max_size * (max_depth - src->depth) /
1729                                                (max_depth - ref_depth + 1);
1730        if (max_size == 0)
1731                return 0;
1732        src_size = src_entry->size;
1733        sizediff = src_size < trg_size ? trg_size - src_size : 0;
1734        if (sizediff >= max_size)
1735                return 0;
1736        if (trg_size < src_size / 32)
1737                return 0;
1738
1739        /* Load data if not already done */
1740        if (!trg->data) {
1741                read_lock();
1742                trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1743                read_unlock();
1744                if (!trg->data)
1745                        die("object %s cannot be read",
1746                            sha1_to_hex(trg_entry->idx.sha1));
1747                if (sz != trg_size)
1748                        die("object %s inconsistent object length (%lu vs %lu)",
1749                            sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1750                *mem_usage += sz;
1751        }
1752        if (!src->data) {
1753                read_lock();
1754                src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1755                read_unlock();
1756                if (!src->data) {
1757                        if (src_entry->preferred_base) {
1758                                static int warned = 0;
1759                                if (!warned++)
1760                                        warning("object %s cannot be read",
1761                                                sha1_to_hex(src_entry->idx.sha1));
1762                                /*
1763                                 * Those objects are not included in the
1764                                 * resulting pack.  Be resilient and ignore
1765                                 * them if they can't be read, in case the
1766                                 * pack could be created nevertheless.
1767                                 */
1768                                return 0;
1769                        }
1770                        die("object %s cannot be read",
1771                            sha1_to_hex(src_entry->idx.sha1));
1772                }
1773                if (sz != src_size)
1774                        die("object %s inconsistent object length (%lu vs %lu)",
1775                            sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1776                *mem_usage += sz;
1777        }
1778        if (!src->index) {
1779                src->index = create_delta_index(src->data, src_size);
1780                if (!src->index) {
1781                        static int warned = 0;
1782                        if (!warned++)
1783                                warning("suboptimal pack - out of memory");
1784                        return 0;
1785                }
1786                *mem_usage += sizeof_delta_index(src->index);
1787        }
1788
1789        delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1790        if (!delta_buf)
1791                return 0;
1792
1793        if (trg_entry->delta) {
1794                /* Prefer only shallower same-sized deltas. */
1795                if (delta_size == trg_entry->delta_size &&
1796                    src->depth + 1 >= trg->depth) {
1797                        free(delta_buf);
1798                        return 0;
1799                }
1800        }
1801
1802        /*
1803         * Handle memory allocation outside of the cache
1804         * accounting lock.  Compiler will optimize the strangeness
1805         * away when NO_PTHREADS is defined.
1806         */
1807        free(trg_entry->delta_data);
1808        cache_lock();
1809        if (trg_entry->delta_data) {
1810                delta_cache_size -= trg_entry->delta_size;
1811                trg_entry->delta_data = NULL;
1812        }
1813        if (delta_cacheable(src_size, trg_size, delta_size)) {
1814                delta_cache_size += delta_size;
1815                cache_unlock();
1816                trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1817        } else {
1818                cache_unlock();
1819                free(delta_buf);
1820        }
1821
1822        trg_entry->delta = src_entry;
1823        trg_entry->delta_size = delta_size;
1824        trg->depth = src->depth + 1;
1825
1826        return 1;
1827}
1828
1829static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1830{
1831        struct object_entry *child = me->delta_child;
1832        unsigned int m = n;
1833        while (child) {
1834                unsigned int c = check_delta_limit(child, n + 1);
1835                if (m < c)
1836                        m = c;
1837                child = child->delta_sibling;
1838        }
1839        return m;
1840}
1841
1842static unsigned long free_unpacked(struct unpacked *n)
1843{
1844        unsigned long freed_mem = sizeof_delta_index(n->index);
1845        free_delta_index(n->index);
1846        n->index = NULL;
1847        if (n->data) {
1848                freed_mem += n->entry->size;
1849                free(n->data);
1850                n->data = NULL;
1851        }
1852        n->entry = NULL;
1853        n->depth = 0;
1854        return freed_mem;
1855}
1856
1857static void find_deltas(struct object_entry **list, unsigned *list_size,
1858                        int window, int depth, unsigned *processed)
1859{
1860        uint32_t i, idx = 0, count = 0;
1861        struct unpacked *array;
1862        unsigned long mem_usage = 0;
1863
1864        array = xcalloc(window, sizeof(struct unpacked));
1865
1866        for (;;) {
1867                struct object_entry *entry;
1868                struct unpacked *n = array + idx;
1869                int j, max_depth, best_base = -1;
1870
1871                progress_lock();
1872                if (!*list_size) {
1873                        progress_unlock();
1874                        break;
1875                }
1876                entry = *list++;
1877                (*list_size)--;
1878                if (!entry->preferred_base) {
1879                        (*processed)++;
1880                        display_progress(progress_state, *processed);
1881                }
1882                progress_unlock();
1883
1884                mem_usage -= free_unpacked(n);
1885                n->entry = entry;
1886
1887                while (window_memory_limit &&
1888                       mem_usage > window_memory_limit &&
1889                       count > 1) {
1890                        uint32_t tail = (idx + window - count) % window;
1891                        mem_usage -= free_unpacked(array + tail);
1892                        count--;
1893                }
1894
1895                /* We do not compute delta to *create* objects we are not
1896                 * going to pack.
1897                 */
1898                if (entry->preferred_base)
1899                        goto next;
1900
1901                /*
1902                 * If the current object is at pack edge, take the depth the
1903                 * objects that depend on the current object into account
1904                 * otherwise they would become too deep.
1905                 */
1906                max_depth = depth;
1907                if (entry->delta_child) {
1908                        max_depth -= check_delta_limit(entry, 0);
1909                        if (max_depth <= 0)
1910                                goto next;
1911                }
1912
1913                j = window;
1914                while (--j > 0) {
1915                        int ret;
1916                        uint32_t other_idx = idx + j;
1917                        struct unpacked *m;
1918                        if (other_idx >= window)
1919                                other_idx -= window;
1920                        m = array + other_idx;
1921                        if (!m->entry)
1922                                break;
1923                        ret = try_delta(n, m, max_depth, &mem_usage);
1924                        if (ret < 0)
1925                                break;
1926                        else if (ret > 0)
1927                                best_base = other_idx;
1928                }
1929
1930                /*
1931                 * If we decided to cache the delta data, then it is best
1932                 * to compress it right away.  First because we have to do
1933                 * it anyway, and doing it here while we're threaded will
1934                 * save a lot of time in the non threaded write phase,
1935                 * as well as allow for caching more deltas within
1936                 * the same cache size limit.
1937                 * ...
1938                 * But only if not writing to stdout, since in that case
1939                 * the network is most likely throttling writes anyway,
1940                 * and therefore it is best to go to the write phase ASAP
1941                 * instead, as we can afford spending more time compressing
1942                 * between writes at that moment.
1943                 */
1944                if (entry->delta_data && !pack_to_stdout) {
1945                        entry->z_delta_size = do_compress(&entry->delta_data,
1946                                                          entry->delta_size);
1947                        cache_lock();
1948                        delta_cache_size -= entry->delta_size;
1949                        delta_cache_size += entry->z_delta_size;
1950                        cache_unlock();
1951                }
1952
1953                /* if we made n a delta, and if n is already at max
1954                 * depth, leaving it in the window is pointless.  we
1955                 * should evict it first.
1956                 */
1957                if (entry->delta && max_depth <= n->depth)
1958                        continue;
1959
1960                /*
1961                 * Move the best delta base up in the window, after the
1962                 * currently deltified object, to keep it longer.  It will
1963                 * be the first base object to be attempted next.
1964                 */
1965                if (entry->delta) {
1966                        struct unpacked swap = array[best_base];
1967                        int dist = (window + idx - best_base) % window;
1968                        int dst = best_base;
1969                        while (dist--) {
1970                                int src = (dst + 1) % window;
1971                                array[dst] = array[src];
1972                                dst = src;
1973                        }
1974                        array[dst] = swap;
1975                }
1976
1977                next:
1978                idx++;
1979                if (count + 1 < window)
1980                        count++;
1981                if (idx >= window)
1982                        idx = 0;
1983        }
1984
1985        for (i = 0; i < window; ++i) {
1986                free_delta_index(array[i].index);
1987                free(array[i].data);
1988        }
1989        free(array);
1990}
1991
1992#ifndef NO_PTHREADS
1993
1994static void try_to_free_from_threads(size_t size)
1995{
1996        read_lock();
1997        release_pack_memory(size);
1998        read_unlock();
1999}
2000
2001static try_to_free_t old_try_to_free_routine;
2002
2003/*
2004 * The main thread waits on the condition that (at least) one of the workers
2005 * has stopped working (which is indicated in the .working member of
2006 * struct thread_params).
2007 * When a work thread has completed its work, it sets .working to 0 and
2008 * signals the main thread and waits on the condition that .data_ready
2009 * becomes 1.
2010 */
2011
2012struct thread_params {
2013        pthread_t thread;
2014        struct object_entry **list;
2015        unsigned list_size;
2016        unsigned remaining;
2017        int window;
2018        int depth;
2019        int working;
2020        int data_ready;
2021        pthread_mutex_t mutex;
2022        pthread_cond_t cond;
2023        unsigned *processed;
2024};
2025
2026static pthread_cond_t progress_cond;
2027
2028/*
2029 * Mutex and conditional variable can't be statically-initialized on Windows.
2030 */
2031static void init_threaded_search(void)
2032{
2033        init_recursive_mutex(&read_mutex);
2034        pthread_mutex_init(&cache_mutex, NULL);
2035        pthread_mutex_init(&progress_mutex, NULL);
2036        pthread_cond_init(&progress_cond, NULL);
2037        old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2038}
2039
2040static void cleanup_threaded_search(void)
2041{
2042        set_try_to_free_routine(old_try_to_free_routine);
2043        pthread_cond_destroy(&progress_cond);
2044        pthread_mutex_destroy(&read_mutex);
2045        pthread_mutex_destroy(&cache_mutex);
2046        pthread_mutex_destroy(&progress_mutex);
2047}
2048
2049static void *threaded_find_deltas(void *arg)
2050{
2051        struct thread_params *me = arg;
2052
2053        while (me->remaining) {
2054                find_deltas(me->list, &me->remaining,
2055                            me->window, me->depth, me->processed);
2056
2057                progress_lock();
2058                me->working = 0;
2059                pthread_cond_signal(&progress_cond);
2060                progress_unlock();
2061
2062                /*
2063                 * We must not set ->data_ready before we wait on the
2064                 * condition because the main thread may have set it to 1
2065                 * before we get here. In order to be sure that new
2066                 * work is available if we see 1 in ->data_ready, it
2067                 * was initialized to 0 before this thread was spawned
2068                 * and we reset it to 0 right away.
2069                 */
2070                pthread_mutex_lock(&me->mutex);
2071                while (!me->data_ready)
2072                        pthread_cond_wait(&me->cond, &me->mutex);
2073                me->data_ready = 0;
2074                pthread_mutex_unlock(&me->mutex);
2075        }
2076        /* leave ->working 1 so that this doesn't get more work assigned */
2077        return NULL;
2078}
2079
2080static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2081                           int window, int depth, unsigned *processed)
2082{
2083        struct thread_params *p;
2084        int i, ret, active_threads = 0;
2085
2086        init_threaded_search();
2087
2088        if (delta_search_threads <= 1) {
2089                find_deltas(list, &list_size, window, depth, processed);
2090                cleanup_threaded_search();
2091                return;
2092        }
2093        if (progress > pack_to_stdout)
2094                fprintf(stderr, "Delta compression using up to %d threads.\n",
2095                                delta_search_threads);
2096        p = xcalloc(delta_search_threads, sizeof(*p));
2097
2098        /* Partition the work amongst work threads. */
2099        for (i = 0; i < delta_search_threads; i++) {
2100                unsigned sub_size = list_size / (delta_search_threads - i);
2101
2102                /* don't use too small segments or no deltas will be found */
2103                if (sub_size < 2*window && i+1 < delta_search_threads)
2104                        sub_size = 0;
2105
2106                p[i].window = window;
2107                p[i].depth = depth;
2108                p[i].processed = processed;
2109                p[i].working = 1;
2110                p[i].data_ready = 0;
2111
2112                /* try to split chunks on "path" boundaries */
2113                while (sub_size && sub_size < list_size &&
2114                       list[sub_size]->hash &&
2115                       list[sub_size]->hash == list[sub_size-1]->hash)
2116                        sub_size++;
2117
2118                p[i].list = list;
2119                p[i].list_size = sub_size;
2120                p[i].remaining = sub_size;
2121
2122                list += sub_size;
2123                list_size -= sub_size;
2124        }
2125
2126        /* Start work threads. */
2127        for (i = 0; i < delta_search_threads; i++) {
2128                if (!p[i].list_size)
2129                        continue;
2130                pthread_mutex_init(&p[i].mutex, NULL);
2131                pthread_cond_init(&p[i].cond, NULL);
2132                ret = pthread_create(&p[i].thread, NULL,
2133                                     threaded_find_deltas, &p[i]);
2134                if (ret)
2135                        die("unable to create thread: %s", strerror(ret));
2136                active_threads++;
2137        }
2138
2139        /*
2140         * Now let's wait for work completion.  Each time a thread is done
2141         * with its work, we steal half of the remaining work from the
2142         * thread with the largest number of unprocessed objects and give
2143         * it to that newly idle thread.  This ensure good load balancing
2144         * until the remaining object list segments are simply too short
2145         * to be worth splitting anymore.
2146         */
2147        while (active_threads) {
2148                struct thread_params *target = NULL;
2149                struct thread_params *victim = NULL;
2150                unsigned sub_size = 0;
2151
2152                progress_lock();
2153                for (;;) {
2154                        for (i = 0; !target && i < delta_search_threads; i++)
2155                                if (!p[i].working)
2156                                        target = &p[i];
2157                        if (target)
2158                                break;
2159                        pthread_cond_wait(&progress_cond, &progress_mutex);
2160                }
2161
2162                for (i = 0; i < delta_search_threads; i++)
2163                        if (p[i].remaining > 2*window &&
2164                            (!victim || victim->remaining < p[i].remaining))
2165                                victim = &p[i];
2166                if (victim) {
2167                        sub_size = victim->remaining / 2;
2168                        list = victim->list + victim->list_size - sub_size;
2169                        while (sub_size && list[0]->hash &&
2170                               list[0]->hash == list[-1]->hash) {
2171                                list++;
2172                                sub_size--;
2173                        }
2174                        if (!sub_size) {
2175                                /*
2176                                 * It is possible for some "paths" to have
2177                                 * so many objects that no hash boundary
2178                                 * might be found.  Let's just steal the
2179                                 * exact half in that case.
2180                                 */
2181                                sub_size = victim->remaining / 2;
2182                                list -= sub_size;
2183                        }
2184                        target->list = list;
2185                        victim->list_size -= sub_size;
2186                        victim->remaining -= sub_size;
2187                }
2188                target->list_size = sub_size;
2189                target->remaining = sub_size;
2190                target->working = 1;
2191                progress_unlock();
2192
2193                pthread_mutex_lock(&target->mutex);
2194                target->data_ready = 1;
2195                pthread_cond_signal(&target->cond);
2196                pthread_mutex_unlock(&target->mutex);
2197
2198                if (!sub_size) {
2199                        pthread_join(target->thread, NULL);
2200                        pthread_cond_destroy(&target->cond);
2201                        pthread_mutex_destroy(&target->mutex);
2202                        active_threads--;
2203                }
2204        }
2205        cleanup_threaded_search();
2206        free(p);
2207}
2208
2209#else
2210#define ll_find_deltas(l, s, w, d, p)   find_deltas(l, &s, w, d, p)
2211#endif
2212
2213static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2214{
2215        struct object_id peeled;
2216
2217        if (starts_with(path, "refs/tags/") && /* is a tag? */
2218            !peel_ref(path, peeled.hash)    && /* peelable? */
2219            packlist_find(&to_pack, peeled.hash, NULL))      /* object packed? */
2220                add_object_entry(oid->hash, OBJ_TAG, NULL, 0);
2221        return 0;
2222}
2223
2224static void prepare_pack(int window, int depth)
2225{
2226        struct object_entry **delta_list;
2227        uint32_t i, nr_deltas;
2228        unsigned n;
2229
2230        get_object_details();
2231
2232        /*
2233         * If we're locally repacking then we need to be doubly careful
2234         * from now on in order to make sure no stealth corruption gets
2235         * propagated to the new pack.  Clients receiving streamed packs
2236         * should validate everything they get anyway so no need to incur
2237         * the additional cost here in that case.
2238         */
2239        if (!pack_to_stdout)
2240                do_check_packed_object_crc = 1;
2241
2242        if (!to_pack.nr_objects || !window || !depth)
2243                return;
2244
2245        ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2246        nr_deltas = n = 0;
2247
2248        for (i = 0; i < to_pack.nr_objects; i++) {
2249                struct object_entry *entry = to_pack.objects + i;
2250
2251                if (entry->delta)
2252                        /* This happens if we decided to reuse existing
2253                         * delta from a pack.  "reuse_delta &&" is implied.
2254                         */
2255                        continue;
2256
2257                if (entry->size < 50)
2258                        continue;
2259
2260                if (entry->no_try_delta)
2261                        continue;
2262
2263                if (!entry->preferred_base) {
2264                        nr_deltas++;
2265                        if (entry->type < 0)
2266                                die("unable to get type of object %s",
2267                                    sha1_to_hex(entry->idx.sha1));
2268                } else {
2269                        if (entry->type < 0) {
2270                                /*
2271                                 * This object is not found, but we
2272                                 * don't have to include it anyway.
2273                                 */
2274                                continue;
2275                        }
2276                }
2277
2278                delta_list[n++] = entry;
2279        }
2280
2281        if (nr_deltas && n > 1) {
2282                unsigned nr_done = 0;
2283                if (progress)
2284                        progress_state = start_progress(_("Compressing objects"),
2285                                                        nr_deltas);
2286                qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2287                ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2288                stop_progress(&progress_state);
2289                if (nr_done != nr_deltas)
2290                        die("inconsistency with delta count");
2291        }
2292        free(delta_list);
2293}
2294
2295static int git_pack_config(const char *k, const char *v, void *cb)
2296{
2297        if (!strcmp(k, "pack.window")) {
2298                window = git_config_int(k, v);
2299                return 0;
2300        }
2301        if (!strcmp(k, "pack.windowmemory")) {
2302                window_memory_limit = git_config_ulong(k, v);
2303                return 0;
2304        }
2305        if (!strcmp(k, "pack.depth")) {
2306                depth = git_config_int(k, v);
2307                return 0;
2308        }
2309        if (!strcmp(k, "pack.compression")) {
2310                int level = git_config_int(k, v);
2311                if (level == -1)
2312                        level = Z_DEFAULT_COMPRESSION;
2313                else if (level < 0 || level > Z_BEST_COMPRESSION)
2314                        die("bad pack compression level %d", level);
2315                pack_compression_level = level;
2316                pack_compression_seen = 1;
2317                return 0;
2318        }
2319        if (!strcmp(k, "pack.deltacachesize")) {
2320                max_delta_cache_size = git_config_int(k, v);
2321                return 0;
2322        }
2323        if (!strcmp(k, "pack.deltacachelimit")) {
2324                cache_max_small_delta_size = git_config_int(k, v);
2325                return 0;
2326        }
2327        if (!strcmp(k, "pack.writebitmaphashcache")) {
2328                if (git_config_bool(k, v))
2329                        write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2330                else
2331                        write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2332        }
2333        if (!strcmp(k, "pack.usebitmaps")) {
2334                use_bitmap_index = git_config_bool(k, v);
2335                return 0;
2336        }
2337        if (!strcmp(k, "pack.threads")) {
2338                delta_search_threads = git_config_int(k, v);
2339                if (delta_search_threads < 0)
2340                        die("invalid number of threads specified (%d)",
2341                            delta_search_threads);
2342#ifdef NO_PTHREADS
2343                if (delta_search_threads != 1)
2344                        warning("no threads support, ignoring %s", k);
2345#endif
2346                return 0;
2347        }
2348        if (!strcmp(k, "pack.indexversion")) {
2349                pack_idx_opts.version = git_config_int(k, v);
2350                if (pack_idx_opts.version > 2)
2351                        die("bad pack.indexversion=%"PRIu32,
2352                            pack_idx_opts.version);
2353                return 0;
2354        }
2355        return git_default_config(k, v, cb);
2356}
2357
2358static void read_object_list_from_stdin(void)
2359{
2360        char line[40 + 1 + PATH_MAX + 2];
2361        unsigned char sha1[20];
2362
2363        for (;;) {
2364                if (!fgets(line, sizeof(line), stdin)) {
2365                        if (feof(stdin))
2366                                break;
2367                        if (!ferror(stdin))
2368                                die("fgets returned NULL, not EOF, not error!");
2369                        if (errno != EINTR)
2370                                die_errno("fgets");
2371                        clearerr(stdin);
2372                        continue;
2373                }
2374                if (line[0] == '-') {
2375                        if (get_sha1_hex(line+1, sha1))
2376                                die("expected edge sha1, got garbage:\n %s",
2377                                    line);
2378                        add_preferred_base(sha1);
2379                        continue;
2380                }
2381                if (get_sha1_hex(line, sha1))
2382                        die("expected sha1, got garbage:\n %s", line);
2383
2384                add_preferred_base_object(line+41);
2385                add_object_entry(sha1, 0, line+41, 0);
2386        }
2387}
2388
2389#define OBJECT_ADDED (1u<<20)
2390
2391static void show_commit(struct commit *commit, void *data)
2392{
2393        add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2394        commit->object.flags |= OBJECT_ADDED;
2395
2396        if (write_bitmap_index)
2397                index_commit_for_bitmap(commit);
2398}
2399
2400static void show_object(struct object *obj, const char *name, void *data)
2401{
2402        add_preferred_base_object(name);
2403        add_object_entry(obj->oid.hash, obj->type, name, 0);
2404        obj->flags |= OBJECT_ADDED;
2405}
2406
2407static void show_edge(struct commit *commit)
2408{
2409        add_preferred_base(commit->object.oid.hash);
2410}
2411
2412struct in_pack_object {
2413        off_t offset;
2414        struct object *object;
2415};
2416
2417struct in_pack {
2418        int alloc;
2419        int nr;
2420        struct in_pack_object *array;
2421};
2422
2423static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2424{
2425        in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2426        in_pack->array[in_pack->nr].object = object;
2427        in_pack->nr++;
2428}
2429
2430/*
2431 * Compare the objects in the offset order, in order to emulate the
2432 * "git rev-list --objects" output that produced the pack originally.
2433 */
2434static int ofscmp(const void *a_, const void *b_)
2435{
2436        struct in_pack_object *a = (struct in_pack_object *)a_;
2437        struct in_pack_object *b = (struct in_pack_object *)b_;
2438
2439        if (a->offset < b->offset)
2440                return -1;
2441        else if (a->offset > b->offset)
2442                return 1;
2443        else
2444                return oidcmp(&a->object->oid, &b->object->oid);
2445}
2446
2447static void add_objects_in_unpacked_packs(struct rev_info *revs)
2448{
2449        struct packed_git *p;
2450        struct in_pack in_pack;
2451        uint32_t i;
2452
2453        memset(&in_pack, 0, sizeof(in_pack));
2454
2455        for (p = packed_git; p; p = p->next) {
2456                const unsigned char *sha1;
2457                struct object *o;
2458
2459                if (!p->pack_local || p->pack_keep)
2460                        continue;
2461                if (open_pack_index(p))
2462                        die("cannot open pack index");
2463
2464                ALLOC_GROW(in_pack.array,
2465                           in_pack.nr + p->num_objects,
2466                           in_pack.alloc);
2467
2468                for (i = 0; i < p->num_objects; i++) {
2469                        sha1 = nth_packed_object_sha1(p, i);
2470                        o = lookup_unknown_object(sha1);
2471                        if (!(o->flags & OBJECT_ADDED))
2472                                mark_in_pack_object(o, p, &in_pack);
2473                        o->flags |= OBJECT_ADDED;
2474                }
2475        }
2476
2477        if (in_pack.nr) {
2478                qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2479                      ofscmp);
2480                for (i = 0; i < in_pack.nr; i++) {
2481                        struct object *o = in_pack.array[i].object;
2482                        add_object_entry(o->oid.hash, o->type, "", 0);
2483                }
2484        }
2485        free(in_pack.array);
2486}
2487
2488static int add_loose_object(const unsigned char *sha1, const char *path,
2489                            void *data)
2490{
2491        enum object_type type = sha1_object_info(sha1, NULL);
2492
2493        if (type < 0) {
2494                warning("loose object at %s could not be examined", path);
2495                return 0;
2496        }
2497
2498        add_object_entry(sha1, type, "", 0);
2499        return 0;
2500}
2501
2502/*
2503 * We actually don't even have to worry about reachability here.
2504 * add_object_entry will weed out duplicates, so we just add every
2505 * loose object we find.
2506 */
2507static void add_unreachable_loose_objects(void)
2508{
2509        for_each_loose_file_in_objdir(get_object_directory(),
2510                                      add_loose_object,
2511                                      NULL, NULL, NULL);
2512}
2513
2514static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2515{
2516        static struct packed_git *last_found = (void *)1;
2517        struct packed_git *p;
2518
2519        p = (last_found != (void *)1) ? last_found : packed_git;
2520
2521        while (p) {
2522                if ((!p->pack_local || p->pack_keep) &&
2523                        find_pack_entry_one(sha1, p)) {
2524                        last_found = p;
2525                        return 1;
2526                }
2527                if (p == last_found)
2528                        p = packed_git;
2529                else
2530                        p = p->next;
2531                if (p == last_found)
2532                        p = p->next;
2533        }
2534        return 0;
2535}
2536
2537/*
2538 * Store a list of sha1s that are should not be discarded
2539 * because they are either written too recently, or are
2540 * reachable from another object that was.
2541 *
2542 * This is filled by get_object_list.
2543 */
2544static struct sha1_array recent_objects;
2545
2546static int loosened_object_can_be_discarded(const unsigned char *sha1,
2547                                            unsigned long mtime)
2548{
2549        if (!unpack_unreachable_expiration)
2550                return 0;
2551        if (mtime > unpack_unreachable_expiration)
2552                return 0;
2553        if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2554                return 0;
2555        return 1;
2556}
2557
2558static void loosen_unused_packed_objects(struct rev_info *revs)
2559{
2560        struct packed_git *p;
2561        uint32_t i;
2562        const unsigned char *sha1;
2563
2564        for (p = packed_git; p; p = p->next) {
2565                if (!p->pack_local || p->pack_keep)
2566                        continue;
2567
2568                if (open_pack_index(p))
2569                        die("cannot open pack index");
2570
2571                for (i = 0; i < p->num_objects; i++) {
2572                        sha1 = nth_packed_object_sha1(p, i);
2573                        if (!packlist_find(&to_pack, sha1, NULL) &&
2574                            !has_sha1_pack_kept_or_nonlocal(sha1) &&
2575                            !loosened_object_can_be_discarded(sha1, p->mtime))
2576                                if (force_object_loose(sha1, p->mtime))
2577                                        die("unable to force loose object");
2578                }
2579        }
2580}
2581
2582/*
2583 * This tracks any options which a reader of the pack might
2584 * not understand, and which would therefore prevent blind reuse
2585 * of what we have on disk.
2586 */
2587static int pack_options_allow_reuse(void)
2588{
2589        return allow_ofs_delta;
2590}
2591
2592static int get_object_list_from_bitmap(struct rev_info *revs)
2593{
2594        if (prepare_bitmap_walk(revs) < 0)
2595                return -1;
2596
2597        if (pack_options_allow_reuse() &&
2598            !reuse_partial_packfile_from_bitmap(
2599                        &reuse_packfile,
2600                        &reuse_packfile_objects,
2601                        &reuse_packfile_offset)) {
2602                assert(reuse_packfile_objects);
2603                nr_result += reuse_packfile_objects;
2604                display_progress(progress_state, nr_result);
2605        }
2606
2607        traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2608        return 0;
2609}
2610
2611static void record_recent_object(struct object *obj,
2612                                 const char *name,
2613                                 void *data)
2614{
2615        sha1_array_append(&recent_objects, obj->oid.hash);
2616}
2617
2618static void record_recent_commit(struct commit *commit, void *data)
2619{
2620        sha1_array_append(&recent_objects, commit->object.oid.hash);
2621}
2622
2623static void get_object_list(int ac, const char **av)
2624{
2625        struct rev_info revs;
2626        char line[1000];
2627        int flags = 0;
2628
2629        init_revisions(&revs, NULL);
2630        save_commit_buffer = 0;
2631        setup_revisions(ac, av, &revs, NULL);
2632
2633        /* make sure shallows are read */
2634        is_repository_shallow();
2635
2636        while (fgets(line, sizeof(line), stdin) != NULL) {
2637                int len = strlen(line);
2638                if (len && line[len - 1] == '\n')
2639                        line[--len] = 0;
2640                if (!len)
2641                        break;
2642                if (*line == '-') {
2643                        if (!strcmp(line, "--not")) {
2644                                flags ^= UNINTERESTING;
2645                                write_bitmap_index = 0;
2646                                continue;
2647                        }
2648                        if (starts_with(line, "--shallow ")) {
2649                                unsigned char sha1[20];
2650                                if (get_sha1_hex(line + 10, sha1))
2651                                        die("not an SHA-1 '%s'", line + 10);
2652                                register_shallow(sha1);
2653                                use_bitmap_index = 0;
2654                                continue;
2655                        }
2656                        die("not a rev '%s'", line);
2657                }
2658                if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2659                        die("bad revision '%s'", line);
2660        }
2661
2662        if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2663                return;
2664
2665        if (prepare_revision_walk(&revs))
2666                die("revision walk setup failed");
2667        mark_edges_uninteresting(&revs, show_edge);
2668        traverse_commit_list(&revs, show_commit, show_object, NULL);
2669
2670        if (unpack_unreachable_expiration) {
2671                revs.ignore_missing_links = 1;
2672                if (add_unseen_recent_objects_to_traversal(&revs,
2673                                unpack_unreachable_expiration))
2674                        die("unable to add recent objects");
2675                if (prepare_revision_walk(&revs))
2676                        die("revision walk setup failed");
2677                traverse_commit_list(&revs, record_recent_commit,
2678                                     record_recent_object, NULL);
2679        }
2680
2681        if (keep_unreachable)
2682                add_objects_in_unpacked_packs(&revs);
2683        if (pack_loose_unreachable)
2684                add_unreachable_loose_objects();
2685        if (unpack_unreachable)
2686                loosen_unused_packed_objects(&revs);
2687
2688        sha1_array_clear(&recent_objects);
2689}
2690
2691static int option_parse_index_version(const struct option *opt,
2692                                      const char *arg, int unset)
2693{
2694        char *c;
2695        const char *val = arg;
2696        pack_idx_opts.version = strtoul(val, &c, 10);
2697        if (pack_idx_opts.version > 2)
2698                die(_("unsupported index version %s"), val);
2699        if (*c == ',' && c[1])
2700                pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2701        if (*c || pack_idx_opts.off32_limit & 0x80000000)
2702                die(_("bad index version '%s'"), val);
2703        return 0;
2704}
2705
2706static int option_parse_unpack_unreachable(const struct option *opt,
2707                                           const char *arg, int unset)
2708{
2709        if (unset) {
2710                unpack_unreachable = 0;
2711                unpack_unreachable_expiration = 0;
2712        }
2713        else {
2714                unpack_unreachable = 1;
2715                if (arg)
2716                        unpack_unreachable_expiration = approxidate(arg);
2717        }
2718        return 0;
2719}
2720
2721int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2722{
2723        int use_internal_rev_list = 0;
2724        int thin = 0;
2725        int shallow = 0;
2726        int all_progress_implied = 0;
2727        struct argv_array rp = ARGV_ARRAY_INIT;
2728        int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2729        int rev_list_index = 0;
2730        struct option pack_objects_options[] = {
2731                OPT_SET_INT('q', "quiet", &progress,
2732                            N_("do not show progress meter"), 0),
2733                OPT_SET_INT(0, "progress", &progress,
2734                            N_("show progress meter"), 1),
2735                OPT_SET_INT(0, "all-progress", &progress,
2736                            N_("show progress meter during object writing phase"), 2),
2737                OPT_BOOL(0, "all-progress-implied",
2738                         &all_progress_implied,
2739                         N_("similar to --all-progress when progress meter is shown")),
2740                { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2741                  N_("write the pack index file in the specified idx format version"),
2742                  0, option_parse_index_version },
2743                OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2744                              N_("maximum size of each output pack file")),
2745                OPT_BOOL(0, "local", &local,
2746                         N_("ignore borrowed objects from alternate object store")),
2747                OPT_BOOL(0, "incremental", &incremental,
2748                         N_("ignore packed objects")),
2749                OPT_INTEGER(0, "window", &window,
2750                            N_("limit pack window by objects")),
2751                OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2752                              N_("limit pack window by memory in addition to object limit")),
2753                OPT_INTEGER(0, "depth", &depth,
2754                            N_("maximum length of delta chain allowed in the resulting pack")),
2755                OPT_BOOL(0, "reuse-delta", &reuse_delta,
2756                         N_("reuse existing deltas")),
2757                OPT_BOOL(0, "reuse-object", &reuse_object,
2758                         N_("reuse existing objects")),
2759                OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2760                         N_("use OFS_DELTA objects")),
2761                OPT_INTEGER(0, "threads", &delta_search_threads,
2762                            N_("use threads when searching for best delta matches")),
2763                OPT_BOOL(0, "non-empty", &non_empty,
2764                         N_("do not create an empty pack output")),
2765                OPT_BOOL(0, "revs", &use_internal_rev_list,
2766                         N_("read revision arguments from standard input")),
2767                { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2768                  N_("limit the objects to those that are not yet packed"),
2769                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2770                { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2771                  N_("include objects reachable from any reference"),
2772                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2773                { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2774                  N_("include objects referred by reflog entries"),
2775                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2776                { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2777                  N_("include objects referred to by the index"),
2778                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2779                OPT_BOOL(0, "stdout", &pack_to_stdout,
2780                         N_("output pack to stdout")),
2781                OPT_BOOL(0, "include-tag", &include_tag,
2782                         N_("include tag objects that refer to objects to be packed")),
2783                OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2784                         N_("keep unreachable objects")),
2785                OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2786                         N_("pack loose unreachable objects")),
2787                { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2788                  N_("unpack unreachable objects newer than <time>"),
2789                  PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2790                OPT_BOOL(0, "thin", &thin,
2791                         N_("create thin packs")),
2792                OPT_BOOL(0, "shallow", &shallow,
2793                         N_("create packs suitable for shallow fetches")),
2794                OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2795                         N_("ignore packs that have companion .keep file")),
2796                OPT_INTEGER(0, "compression", &pack_compression_level,
2797                            N_("pack compression level")),
2798                OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2799                            N_("do not hide commits by grafts"), 0),
2800                OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2801                         N_("use a bitmap index if available to speed up counting objects")),
2802                OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2803                         N_("write a bitmap index together with the pack index")),
2804                OPT_END(),
2805        };
2806
2807        check_replace_refs = 0;
2808
2809        reset_pack_idx_option(&pack_idx_opts);
2810        git_config(git_pack_config, NULL);
2811        if (!pack_compression_seen && core_compression_seen)
2812                pack_compression_level = core_compression_level;
2813
2814        progress = isatty(2);
2815        argc = parse_options(argc, argv, prefix, pack_objects_options,
2816                             pack_usage, 0);
2817
2818        if (argc) {
2819                base_name = argv[0];
2820                argc--;
2821        }
2822        if (pack_to_stdout != !base_name || argc)
2823                usage_with_options(pack_usage, pack_objects_options);
2824
2825        argv_array_push(&rp, "pack-objects");
2826        if (thin) {
2827                use_internal_rev_list = 1;
2828                argv_array_push(&rp, shallow
2829                                ? "--objects-edge-aggressive"
2830                                : "--objects-edge");
2831        } else
2832                argv_array_push(&rp, "--objects");
2833
2834        if (rev_list_all) {
2835                use_internal_rev_list = 1;
2836                argv_array_push(&rp, "--all");
2837        }
2838        if (rev_list_reflog) {
2839                use_internal_rev_list = 1;
2840                argv_array_push(&rp, "--reflog");
2841        }
2842        if (rev_list_index) {
2843                use_internal_rev_list = 1;
2844                argv_array_push(&rp, "--indexed-objects");
2845        }
2846        if (rev_list_unpacked) {
2847                use_internal_rev_list = 1;
2848                argv_array_push(&rp, "--unpacked");
2849        }
2850
2851        if (!reuse_object)
2852                reuse_delta = 0;
2853        if (pack_compression_level == -1)
2854                pack_compression_level = Z_DEFAULT_COMPRESSION;
2855        else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2856                die("bad pack compression level %d", pack_compression_level);
2857
2858        if (!delta_search_threads)      /* --threads=0 means autodetect */
2859                delta_search_threads = online_cpus();
2860
2861#ifdef NO_PTHREADS
2862        if (delta_search_threads != 1)
2863                warning("no threads support, ignoring --threads");
2864#endif
2865        if (!pack_to_stdout && !pack_size_limit)
2866                pack_size_limit = pack_size_limit_cfg;
2867        if (pack_to_stdout && pack_size_limit)
2868                die("--max-pack-size cannot be used to build a pack for transfer.");
2869        if (pack_size_limit && pack_size_limit < 1024*1024) {
2870                warning("minimum pack size limit is 1 MiB");
2871                pack_size_limit = 1024*1024;
2872        }
2873
2874        if (!pack_to_stdout && thin)
2875                die("--thin cannot be used to build an indexable pack.");
2876
2877        if (keep_unreachable && unpack_unreachable)
2878                die("--keep-unreachable and --unpack-unreachable are incompatible.");
2879        if (!rev_list_all || !rev_list_reflog || !rev_list_index)
2880                unpack_unreachable_expiration = 0;
2881
2882        if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2883                use_bitmap_index = 0;
2884
2885        if (pack_to_stdout || !rev_list_all)
2886                write_bitmap_index = 0;
2887
2888        if (progress && all_progress_implied)
2889                progress = 2;
2890
2891        prepare_packed_git();
2892        if (ignore_packed_keep) {
2893                struct packed_git *p;
2894                for (p = packed_git; p; p = p->next)
2895                        if (p->pack_local && p->pack_keep)
2896                                break;
2897                if (!p) /* no keep-able packs found */
2898                        ignore_packed_keep = 0;
2899        }
2900        if (local) {
2901                /*
2902                 * unlike ignore_packed_keep above, we do not want to
2903                 * unset "local" based on looking at packs, as it
2904                 * also covers non-local objects
2905                 */
2906                struct packed_git *p;
2907                for (p = packed_git; p; p = p->next) {
2908                        if (!p->pack_local) {
2909                                have_non_local_packs = 1;
2910                                break;
2911                        }
2912                }
2913        }
2914
2915        if (progress)
2916                progress_state = start_progress(_("Counting objects"), 0);
2917        if (!use_internal_rev_list)
2918                read_object_list_from_stdin();
2919        else {
2920                get_object_list(rp.argc, rp.argv);
2921                argv_array_clear(&rp);
2922        }
2923        cleanup_preferred_base();
2924        if (include_tag && nr_result)
2925                for_each_ref(add_ref_tag, NULL);
2926        stop_progress(&progress_state);
2927
2928        if (non_empty && !nr_result)
2929                return 0;
2930        if (nr_result)
2931                prepare_pack(window, depth);
2932        write_pack_file();
2933        if (progress)
2934                fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2935                        " reused %"PRIu32" (delta %"PRIu32")\n",
2936                        written, written_delta, reused, reused_delta);
2937        return 0;
2938}