ed77355c6e5515b9aa222a2f4725b129bac0bff7
   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
  27static const char *pack_usage[] = {
  28        N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
  29        N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
  30        NULL
  31};
  32
  33/*
  34 * Objects we are going to pack are collected in the `to_pack` structure.
  35 * It contains an array (dynamically expanded) of the object data, and a map
  36 * that can resolve SHA1s to their position in the array.
  37 */
  38static struct packing_data to_pack;
  39
  40static struct pack_idx_entry **written_list;
  41static uint32_t nr_result, nr_written;
  42
  43static int non_empty;
  44static int reuse_delta = 1, reuse_object = 1;
  45static int keep_unreachable, unpack_unreachable, include_tag;
  46static unsigned long unpack_unreachable_expiration;
  47static int pack_loose_unreachable;
  48static int local;
  49static int have_non_local_packs;
  50static int incremental;
  51static int ignore_packed_keep;
  52static int allow_ofs_delta;
  53static struct pack_idx_option pack_idx_opts;
  54static const char *base_name;
  55static int progress = 1;
  56static int window = 10;
  57static unsigned long pack_size_limit;
  58static int depth = 50;
  59static int delta_search_threads;
  60static int pack_to_stdout;
  61static int num_preferred_base;
  62static struct progress *progress_state;
  63static int pack_compression_level = Z_DEFAULT_COMPRESSION;
  64static int pack_compression_seen;
  65
  66static struct packed_git *reuse_packfile;
  67static uint32_t reuse_packfile_objects;
  68static off_t reuse_packfile_offset;
  69
  70static int use_bitmap_index = 1;
  71static int write_bitmap_index;
  72static uint16_t write_bitmap_options;
  73
  74static unsigned long delta_cache_size = 0;
  75static unsigned long max_delta_cache_size = 256 * 1024 * 1024;
  76static unsigned long cache_max_small_delta_size = 1000;
  77
  78static unsigned long window_memory_limit = 0;
  79
  80/*
  81 * stats
  82 */
  83static uint32_t written, written_delta;
  84static uint32_t reused, reused_delta;
  85
  86/*
  87 * Indexed commits
  88 */
  89static struct commit **indexed_commits;
  90static unsigned int indexed_commits_nr;
  91static unsigned int indexed_commits_alloc;
  92
  93static void index_commit_for_bitmap(struct commit *commit)
  94{
  95        if (indexed_commits_nr >= indexed_commits_alloc) {
  96                indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
  97                REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
  98        }
  99
 100        indexed_commits[indexed_commits_nr++] = commit;
 101}
 102
 103static void *get_delta(struct object_entry *entry)
 104{
 105        unsigned long size, base_size, delta_size;
 106        void *buf, *base_buf, *delta_buf;
 107        enum object_type type;
 108
 109        buf = read_sha1_file(entry->idx.sha1, &type, &size);
 110        if (!buf)
 111                die("unable to read %s", sha1_to_hex(entry->idx.sha1));
 112        base_buf = read_sha1_file(entry->delta->idx.sha1, &type, &base_size);
 113        if (!base_buf)
 114                die("unable to read %s", sha1_to_hex(entry->delta->idx.sha1));
 115        delta_buf = diff_delta(base_buf, base_size,
 116                               buf, size, &delta_size, 0);
 117        if (!delta_buf || delta_size != entry->delta_size)
 118                die("delta size changed");
 119        free(buf);
 120        free(base_buf);
 121        return delta_buf;
 122}
 123
 124static unsigned long do_compress(void **pptr, unsigned long size)
 125{
 126        git_zstream stream;
 127        void *in, *out;
 128        unsigned long maxsize;
 129
 130        git_deflate_init(&stream, pack_compression_level);
 131        maxsize = git_deflate_bound(&stream, size);
 132
 133        in = *pptr;
 134        out = xmalloc(maxsize);
 135        *pptr = out;
 136
 137        stream.next_in = in;
 138        stream.avail_in = size;
 139        stream.next_out = out;
 140        stream.avail_out = maxsize;
 141        while (git_deflate(&stream, Z_FINISH) == Z_OK)
 142                ; /* nothing */
 143        git_deflate_end(&stream);
 144
 145        free(in);
 146        return stream.total_out;
 147}
 148
 149static unsigned long write_large_blob_data(struct git_istream *st, struct sha1file *f,
 150                                           const unsigned char *sha1)
 151{
 152        git_zstream stream;
 153        unsigned char ibuf[1024 * 16];
 154        unsigned char obuf[1024 * 16];
 155        unsigned long olen = 0;
 156
 157        git_deflate_init(&stream, pack_compression_level);
 158
 159        for (;;) {
 160                ssize_t readlen;
 161                int zret = Z_OK;
 162                readlen = read_istream(st, ibuf, sizeof(ibuf));
 163                if (readlen == -1)
 164                        die(_("unable to read %s"), sha1_to_hex(sha1));
 165
 166                stream.next_in = ibuf;
 167                stream.avail_in = readlen;
 168                while ((stream.avail_in || readlen == 0) &&
 169                       (zret == Z_OK || zret == Z_BUF_ERROR)) {
 170                        stream.next_out = obuf;
 171                        stream.avail_out = sizeof(obuf);
 172                        zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
 173                        sha1write(f, obuf, stream.next_out - obuf);
 174                        olen += stream.next_out - obuf;
 175                }
 176                if (stream.avail_in)
 177                        die(_("deflate error (%d)"), zret);
 178                if (readlen == 0) {
 179                        if (zret != Z_STREAM_END)
 180                                die(_("deflate error (%d)"), zret);
 181                        break;
 182                }
 183        }
 184        git_deflate_end(&stream);
 185        return olen;
 186}
 187
 188/*
 189 * we are going to reuse the existing object data as is.  make
 190 * sure it is not corrupt.
 191 */
 192static int check_pack_inflate(struct packed_git *p,
 193                struct pack_window **w_curs,
 194                off_t offset,
 195                off_t len,
 196                unsigned long expect)
 197{
 198        git_zstream stream;
 199        unsigned char fakebuf[4096], *in;
 200        int st;
 201
 202        memset(&stream, 0, sizeof(stream));
 203        git_inflate_init(&stream);
 204        do {
 205                in = use_pack(p, w_curs, offset, &stream.avail_in);
 206                stream.next_in = in;
 207                stream.next_out = fakebuf;
 208                stream.avail_out = sizeof(fakebuf);
 209                st = git_inflate(&stream, Z_FINISH);
 210                offset += stream.next_in - in;
 211        } while (st == Z_OK || st == Z_BUF_ERROR);
 212        git_inflate_end(&stream);
 213        return (st == Z_STREAM_END &&
 214                stream.total_out == expect &&
 215                stream.total_in == len) ? 0 : -1;
 216}
 217
 218static void copy_pack_data(struct sha1file *f,
 219                struct packed_git *p,
 220                struct pack_window **w_curs,
 221                off_t offset,
 222                off_t len)
 223{
 224        unsigned char *in;
 225        unsigned long avail;
 226
 227        while (len) {
 228                in = use_pack(p, w_curs, offset, &avail);
 229                if (avail > len)
 230                        avail = (unsigned long)len;
 231                sha1write(f, in, avail);
 232                offset += avail;
 233                len -= avail;
 234        }
 235}
 236
 237/* Return 0 if we will bust the pack-size limit */
 238static unsigned long write_no_reuse_object(struct sha1file *f, struct object_entry *entry,
 239                                           unsigned long limit, int usable_delta)
 240{
 241        unsigned long size, datalen;
 242        unsigned char header[10], dheader[10];
 243        unsigned hdrlen;
 244        enum object_type type;
 245        void *buf;
 246        struct git_istream *st = NULL;
 247
 248        if (!usable_delta) {
 249                if (entry->type == OBJ_BLOB &&
 250                    entry->size > big_file_threshold &&
 251                    (st = open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL)
 252                        buf = NULL;
 253                else {
 254                        buf = read_sha1_file(entry->idx.sha1, &type, &size);
 255                        if (!buf)
 256                                die(_("unable to read %s"), sha1_to_hex(entry->idx.sha1));
 257                }
 258                /*
 259                 * make sure no cached delta data remains from a
 260                 * previous attempt before a pack split occurred.
 261                 */
 262                free(entry->delta_data);
 263                entry->delta_data = NULL;
 264                entry->z_delta_size = 0;
 265        } else if (entry->delta_data) {
 266                size = entry->delta_size;
 267                buf = entry->delta_data;
 268                entry->delta_data = NULL;
 269                type = (allow_ofs_delta && entry->delta->idx.offset) ?
 270                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 271        } else {
 272                buf = get_delta(entry);
 273                size = entry->delta_size;
 274                type = (allow_ofs_delta && entry->delta->idx.offset) ?
 275                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 276        }
 277
 278        if (st) /* large blob case, just assume we don't compress well */
 279                datalen = size;
 280        else if (entry->z_delta_size)
 281                datalen = entry->z_delta_size;
 282        else
 283                datalen = do_compress(&buf, size);
 284
 285        /*
 286         * The object header is a byte of 'type' followed by zero or
 287         * more bytes of length.
 288         */
 289        hdrlen = encode_in_pack_object_header(type, size, header);
 290
 291        if (type == OBJ_OFS_DELTA) {
 292                /*
 293                 * Deltas with relative base contain an additional
 294                 * encoding of the relative offset for the delta
 295                 * base from this object's position in the pack.
 296                 */
 297                off_t ofs = entry->idx.offset - entry->delta->idx.offset;
 298                unsigned pos = sizeof(dheader) - 1;
 299                dheader[pos] = ofs & 127;
 300                while (ofs >>= 7)
 301                        dheader[--pos] = 128 | (--ofs & 127);
 302                if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 303                        if (st)
 304                                close_istream(st);
 305                        free(buf);
 306                        return 0;
 307                }
 308                sha1write(f, header, hdrlen);
 309                sha1write(f, dheader + pos, sizeof(dheader) - pos);
 310                hdrlen += sizeof(dheader) - pos;
 311        } else if (type == OBJ_REF_DELTA) {
 312                /*
 313                 * Deltas with a base reference contain
 314                 * an additional 20 bytes for the base sha1.
 315                 */
 316                if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 317                        if (st)
 318                                close_istream(st);
 319                        free(buf);
 320                        return 0;
 321                }
 322                sha1write(f, header, hdrlen);
 323                sha1write(f, entry->delta->idx.sha1, 20);
 324                hdrlen += 20;
 325        } else {
 326                if (limit && hdrlen + datalen + 20 >= limit) {
 327                        if (st)
 328                                close_istream(st);
 329                        free(buf);
 330                        return 0;
 331                }
 332                sha1write(f, header, hdrlen);
 333        }
 334        if (st) {
 335                datalen = write_large_blob_data(st, f, entry->idx.sha1);
 336                close_istream(st);
 337        } else {
 338                sha1write(f, buf, datalen);
 339                free(buf);
 340        }
 341
 342        return hdrlen + datalen;
 343}
 344
 345/* Return 0 if we will bust the pack-size limit */
 346static unsigned long write_reuse_object(struct sha1file *f, struct object_entry *entry,
 347                                        unsigned long limit, int usable_delta)
 348{
 349        struct packed_git *p = entry->in_pack;
 350        struct pack_window *w_curs = NULL;
 351        struct revindex_entry *revidx;
 352        off_t offset;
 353        enum object_type type = entry->type;
 354        unsigned long datalen;
 355        unsigned char header[10], dheader[10];
 356        unsigned hdrlen;
 357
 358        if (entry->delta)
 359                type = (allow_ofs_delta && entry->delta->idx.offset) ?
 360                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 361        hdrlen = encode_in_pack_object_header(type, entry->size, header);
 362
 363        offset = entry->in_pack_offset;
 364        revidx = find_pack_revindex(p, offset);
 365        datalen = revidx[1].offset - offset;
 366        if (!pack_to_stdout && p->index_version > 1 &&
 367            check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
 368                error("bad packed object CRC for %s", sha1_to_hex(entry->idx.sha1));
 369                unuse_pack(&w_curs);
 370                return write_no_reuse_object(f, entry, limit, usable_delta);
 371        }
 372
 373        offset += entry->in_pack_header_size;
 374        datalen -= entry->in_pack_header_size;
 375
 376        if (!pack_to_stdout && p->index_version == 1 &&
 377            check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) {
 378                error("corrupt packed object for %s", sha1_to_hex(entry->idx.sha1));
 379                unuse_pack(&w_curs);
 380                return write_no_reuse_object(f, entry, limit, usable_delta);
 381        }
 382
 383        if (type == OBJ_OFS_DELTA) {
 384                off_t ofs = entry->idx.offset - entry->delta->idx.offset;
 385                unsigned pos = sizeof(dheader) - 1;
 386                dheader[pos] = ofs & 127;
 387                while (ofs >>= 7)
 388                        dheader[--pos] = 128 | (--ofs & 127);
 389                if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 390                        unuse_pack(&w_curs);
 391                        return 0;
 392                }
 393                sha1write(f, header, hdrlen);
 394                sha1write(f, dheader + pos, sizeof(dheader) - pos);
 395                hdrlen += sizeof(dheader) - pos;
 396                reused_delta++;
 397        } else if (type == OBJ_REF_DELTA) {
 398                if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 399                        unuse_pack(&w_curs);
 400                        return 0;
 401                }
 402                sha1write(f, header, hdrlen);
 403                sha1write(f, entry->delta->idx.sha1, 20);
 404                hdrlen += 20;
 405                reused_delta++;
 406        } else {
 407                if (limit && hdrlen + datalen + 20 >= limit) {
 408                        unuse_pack(&w_curs);
 409                        return 0;
 410                }
 411                sha1write(f, header, hdrlen);
 412        }
 413        copy_pack_data(f, p, &w_curs, offset, datalen);
 414        unuse_pack(&w_curs);
 415        reused++;
 416        return hdrlen + datalen;
 417}
 418
 419/* Return 0 if we will bust the pack-size limit */
 420static unsigned long write_object(struct sha1file *f,
 421                                  struct object_entry *entry,
 422                                  off_t write_offset)
 423{
 424        unsigned long limit, len;
 425        int usable_delta, to_reuse;
 426
 427        if (!pack_to_stdout)
 428                crc32_begin(f);
 429
 430        /* apply size limit if limited packsize and not first object */
 431        if (!pack_size_limit || !nr_written)
 432                limit = 0;
 433        else if (pack_size_limit <= write_offset)
 434                /*
 435                 * the earlier object did not fit the limit; avoid
 436                 * mistaking this with unlimited (i.e. limit = 0).
 437                 */
 438                limit = 1;
 439        else
 440                limit = pack_size_limit - write_offset;
 441
 442        if (!entry->delta)
 443                usable_delta = 0;       /* no delta */
 444        else if (!pack_size_limit)
 445               usable_delta = 1;        /* unlimited packfile */
 446        else if (entry->delta->idx.offset == (off_t)-1)
 447                usable_delta = 0;       /* base was written to another pack */
 448        else if (entry->delta->idx.offset)
 449                usable_delta = 1;       /* base already exists in this pack */
 450        else
 451                usable_delta = 0;       /* base could end up in another pack */
 452
 453        if (!reuse_object)
 454                to_reuse = 0;   /* explicit */
 455        else if (!entry->in_pack)
 456                to_reuse = 0;   /* can't reuse what we don't have */
 457        else if (entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA)
 458                                /* check_object() decided it for us ... */
 459                to_reuse = usable_delta;
 460                                /* ... but pack split may override that */
 461        else if (entry->type != entry->in_pack_type)
 462                to_reuse = 0;   /* pack has delta which is unusable */
 463        else if (entry->delta)
 464                to_reuse = 0;   /* we want to pack afresh */
 465        else
 466                to_reuse = 1;   /* we have it in-pack undeltified,
 467                                 * and we do not need to deltify it.
 468                                 */
 469
 470        if (!to_reuse)
 471                len = write_no_reuse_object(f, entry, limit, usable_delta);
 472        else
 473                len = write_reuse_object(f, entry, limit, usable_delta);
 474        if (!len)
 475                return 0;
 476
 477        if (usable_delta)
 478                written_delta++;
 479        written++;
 480        if (!pack_to_stdout)
 481                entry->idx.crc32 = crc32_end(f);
 482        return len;
 483}
 484
 485enum write_one_status {
 486        WRITE_ONE_SKIP = -1, /* already written */
 487        WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
 488        WRITE_ONE_WRITTEN = 1, /* normal */
 489        WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
 490};
 491
 492static enum write_one_status write_one(struct sha1file *f,
 493                                       struct object_entry *e,
 494                                       off_t *offset)
 495{
 496        unsigned long size;
 497        int recursing;
 498
 499        /*
 500         * we set offset to 1 (which is an impossible value) to mark
 501         * the fact that this object is involved in "write its base
 502         * first before writing a deltified object" recursion.
 503         */
 504        recursing = (e->idx.offset == 1);
 505        if (recursing) {
 506                warning("recursive delta detected for object %s",
 507                        sha1_to_hex(e->idx.sha1));
 508                return WRITE_ONE_RECURSIVE;
 509        } else if (e->idx.offset || e->preferred_base) {
 510                /* offset is non zero if object is written already. */
 511                return WRITE_ONE_SKIP;
 512        }
 513
 514        /* if we are deltified, write out base object first. */
 515        if (e->delta) {
 516                e->idx.offset = 1; /* now recurse */
 517                switch (write_one(f, e->delta, offset)) {
 518                case WRITE_ONE_RECURSIVE:
 519                        /* we cannot depend on this one */
 520                        e->delta = NULL;
 521                        break;
 522                default:
 523                        break;
 524                case WRITE_ONE_BREAK:
 525                        e->idx.offset = recursing;
 526                        return WRITE_ONE_BREAK;
 527                }
 528        }
 529
 530        e->idx.offset = *offset;
 531        size = write_object(f, e, *offset);
 532        if (!size) {
 533                e->idx.offset = recursing;
 534                return WRITE_ONE_BREAK;
 535        }
 536        written_list[nr_written++] = &e->idx;
 537
 538        /* make sure off_t is sufficiently large not to wrap */
 539        if (signed_add_overflows(*offset, size))
 540                die("pack too large for current definition of off_t");
 541        *offset += size;
 542        return WRITE_ONE_WRITTEN;
 543}
 544
 545static int mark_tagged(const char *path, const struct object_id *oid, int flag,
 546                       void *cb_data)
 547{
 548        unsigned char peeled[20];
 549        struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
 550
 551        if (entry)
 552                entry->tagged = 1;
 553        if (!peel_ref(path, peeled)) {
 554                entry = packlist_find(&to_pack, peeled, NULL);
 555                if (entry)
 556                        entry->tagged = 1;
 557        }
 558        return 0;
 559}
 560
 561static inline void add_to_write_order(struct object_entry **wo,
 562                               unsigned int *endp,
 563                               struct object_entry *e)
 564{
 565        if (e->filled)
 566                return;
 567        wo[(*endp)++] = e;
 568        e->filled = 1;
 569}
 570
 571static void add_descendants_to_write_order(struct object_entry **wo,
 572                                           unsigned int *endp,
 573                                           struct object_entry *e)
 574{
 575        int add_to_order = 1;
 576        while (e) {
 577                if (add_to_order) {
 578                        struct object_entry *s;
 579                        /* add this node... */
 580                        add_to_write_order(wo, endp, e);
 581                        /* all its siblings... */
 582                        for (s = e->delta_sibling; s; s = s->delta_sibling) {
 583                                add_to_write_order(wo, endp, s);
 584                        }
 585                }
 586                /* drop down a level to add left subtree nodes if possible */
 587                if (e->delta_child) {
 588                        add_to_order = 1;
 589                        e = e->delta_child;
 590                } else {
 591                        add_to_order = 0;
 592                        /* our sibling might have some children, it is next */
 593                        if (e->delta_sibling) {
 594                                e = e->delta_sibling;
 595                                continue;
 596                        }
 597                        /* go back to our parent node */
 598                        e = e->delta;
 599                        while (e && !e->delta_sibling) {
 600                                /* we're on the right side of a subtree, keep
 601                                 * going up until we can go right again */
 602                                e = e->delta;
 603                        }
 604                        if (!e) {
 605                                /* done- we hit our original root node */
 606                                return;
 607                        }
 608                        /* pass it off to sibling at this level */
 609                        e = e->delta_sibling;
 610                }
 611        };
 612}
 613
 614static void add_family_to_write_order(struct object_entry **wo,
 615                                      unsigned int *endp,
 616                                      struct object_entry *e)
 617{
 618        struct object_entry *root;
 619
 620        for (root = e; root->delta; root = root->delta)
 621                ; /* nothing */
 622        add_descendants_to_write_order(wo, endp, root);
 623}
 624
 625static struct object_entry **compute_write_order(void)
 626{
 627        unsigned int i, wo_end, last_untagged;
 628
 629        struct object_entry **wo;
 630        struct object_entry *objects = to_pack.objects;
 631
 632        for (i = 0; i < to_pack.nr_objects; i++) {
 633                objects[i].tagged = 0;
 634                objects[i].filled = 0;
 635                objects[i].delta_child = NULL;
 636                objects[i].delta_sibling = NULL;
 637        }
 638
 639        /*
 640         * Fully connect delta_child/delta_sibling network.
 641         * Make sure delta_sibling is sorted in the original
 642         * recency order.
 643         */
 644        for (i = to_pack.nr_objects; i > 0;) {
 645                struct object_entry *e = &objects[--i];
 646                if (!e->delta)
 647                        continue;
 648                /* Mark me as the first child */
 649                e->delta_sibling = e->delta->delta_child;
 650                e->delta->delta_child = e;
 651        }
 652
 653        /*
 654         * Mark objects that are at the tip of tags.
 655         */
 656        for_each_tag_ref(mark_tagged, NULL);
 657
 658        /*
 659         * Give the objects in the original recency order until
 660         * we see a tagged tip.
 661         */
 662        ALLOC_ARRAY(wo, to_pack.nr_objects);
 663        for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
 664                if (objects[i].tagged)
 665                        break;
 666                add_to_write_order(wo, &wo_end, &objects[i]);
 667        }
 668        last_untagged = i;
 669
 670        /*
 671         * Then fill all the tagged tips.
 672         */
 673        for (; i < to_pack.nr_objects; i++) {
 674                if (objects[i].tagged)
 675                        add_to_write_order(wo, &wo_end, &objects[i]);
 676        }
 677
 678        /*
 679         * And then all remaining commits and tags.
 680         */
 681        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 682                if (objects[i].type != OBJ_COMMIT &&
 683                    objects[i].type != OBJ_TAG)
 684                        continue;
 685                add_to_write_order(wo, &wo_end, &objects[i]);
 686        }
 687
 688        /*
 689         * And then all the trees.
 690         */
 691        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 692                if (objects[i].type != OBJ_TREE)
 693                        continue;
 694                add_to_write_order(wo, &wo_end, &objects[i]);
 695        }
 696
 697        /*
 698         * Finally all the rest in really tight order
 699         */
 700        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 701                if (!objects[i].filled)
 702                        add_family_to_write_order(wo, &wo_end, &objects[i]);
 703        }
 704
 705        if (wo_end != to_pack.nr_objects)
 706                die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
 707
 708        return wo;
 709}
 710
 711static off_t write_reused_pack(struct sha1file *f)
 712{
 713        unsigned char buffer[8192];
 714        off_t to_write, total;
 715        int fd;
 716
 717        if (!is_pack_valid(reuse_packfile))
 718                die("packfile is invalid: %s", reuse_packfile->pack_name);
 719
 720        fd = git_open_noatime(reuse_packfile->pack_name);
 721        if (fd < 0)
 722                die_errno("unable to open packfile for reuse: %s",
 723                          reuse_packfile->pack_name);
 724
 725        if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
 726                die_errno("unable to seek in reused packfile");
 727
 728        if (reuse_packfile_offset < 0)
 729                reuse_packfile_offset = reuse_packfile->pack_size - 20;
 730
 731        total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
 732
 733        while (to_write) {
 734                int read_pack = xread(fd, buffer, sizeof(buffer));
 735
 736                if (read_pack <= 0)
 737                        die_errno("unable to read from reused packfile");
 738
 739                if (read_pack > to_write)
 740                        read_pack = to_write;
 741
 742                sha1write(f, buffer, read_pack);
 743                to_write -= read_pack;
 744
 745                /*
 746                 * We don't know the actual number of objects written,
 747                 * only how many bytes written, how many bytes total, and
 748                 * how many objects total. So we can fake it by pretending all
 749                 * objects we are writing are the same size. This gives us a
 750                 * smooth progress meter, and at the end it matches the true
 751                 * answer.
 752                 */
 753                written = reuse_packfile_objects *
 754                                (((double)(total - to_write)) / total);
 755                display_progress(progress_state, written);
 756        }
 757
 758        close(fd);
 759        written = reuse_packfile_objects;
 760        display_progress(progress_state, written);
 761        return reuse_packfile_offset - sizeof(struct pack_header);
 762}
 763
 764static const char no_split_warning[] = N_(
 765"disabling bitmap writing, packs are split due to pack.packSizeLimit"
 766);
 767
 768static void write_pack_file(void)
 769{
 770        uint32_t i = 0, j;
 771        struct sha1file *f;
 772        off_t offset;
 773        uint32_t nr_remaining = nr_result;
 774        time_t last_mtime = 0;
 775        struct object_entry **write_order;
 776
 777        if (progress > pack_to_stdout)
 778                progress_state = start_progress(_("Writing objects"), nr_result);
 779        ALLOC_ARRAY(written_list, to_pack.nr_objects);
 780        write_order = compute_write_order();
 781
 782        do {
 783                unsigned char sha1[20];
 784                char *pack_tmp_name = NULL;
 785
 786                if (pack_to_stdout)
 787                        f = sha1fd_throughput(1, "<stdout>", progress_state);
 788                else
 789                        f = create_tmp_packfile(&pack_tmp_name);
 790
 791                offset = write_pack_header(f, nr_remaining);
 792
 793                if (reuse_packfile) {
 794                        off_t packfile_size;
 795                        assert(pack_to_stdout);
 796
 797                        packfile_size = write_reused_pack(f);
 798                        offset += packfile_size;
 799                }
 800
 801                nr_written = 0;
 802                for (; i < to_pack.nr_objects; i++) {
 803                        struct object_entry *e = write_order[i];
 804                        if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
 805                                break;
 806                        display_progress(progress_state, written);
 807                }
 808
 809                /*
 810                 * Did we write the wrong # entries in the header?
 811                 * If so, rewrite it like in fast-import
 812                 */
 813                if (pack_to_stdout) {
 814                        sha1close(f, sha1, CSUM_CLOSE);
 815                } else if (nr_written == nr_remaining) {
 816                        sha1close(f, sha1, CSUM_FSYNC);
 817                } else {
 818                        int fd = sha1close(f, sha1, 0);
 819                        fixup_pack_header_footer(fd, sha1, pack_tmp_name,
 820                                                 nr_written, sha1, offset);
 821                        close(fd);
 822                        if (write_bitmap_index) {
 823                                warning(_(no_split_warning));
 824                                write_bitmap_index = 0;
 825                        }
 826                }
 827
 828                if (!pack_to_stdout) {
 829                        struct stat st;
 830                        struct strbuf tmpname = STRBUF_INIT;
 831
 832                        /*
 833                         * Packs are runtime accessed in their mtime
 834                         * order since newer packs are more likely to contain
 835                         * younger objects.  So if we are creating multiple
 836                         * packs then we should modify the mtime of later ones
 837                         * to preserve this property.
 838                         */
 839                        if (stat(pack_tmp_name, &st) < 0) {
 840                                warning_errno("failed to stat %s", pack_tmp_name);
 841                        } else if (!last_mtime) {
 842                                last_mtime = st.st_mtime;
 843                        } else {
 844                                struct utimbuf utb;
 845                                utb.actime = st.st_atime;
 846                                utb.modtime = --last_mtime;
 847                                if (utime(pack_tmp_name, &utb) < 0)
 848                                        warning_errno("failed utime() on %s", pack_tmp_name);
 849                        }
 850
 851                        strbuf_addf(&tmpname, "%s-", base_name);
 852
 853                        if (write_bitmap_index) {
 854                                bitmap_writer_set_checksum(sha1);
 855                                bitmap_writer_build_type_index(written_list, nr_written);
 856                        }
 857
 858                        finish_tmp_packfile(&tmpname, pack_tmp_name,
 859                                            written_list, nr_written,
 860                                            &pack_idx_opts, sha1);
 861
 862                        if (write_bitmap_index) {
 863                                strbuf_addf(&tmpname, "%s.bitmap", sha1_to_hex(sha1));
 864
 865                                stop_progress(&progress_state);
 866
 867                                bitmap_writer_show_progress(progress);
 868                                bitmap_writer_reuse_bitmaps(&to_pack);
 869                                bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
 870                                bitmap_writer_build(&to_pack);
 871                                bitmap_writer_finish(written_list, nr_written,
 872                                                     tmpname.buf, write_bitmap_options);
 873                                write_bitmap_index = 0;
 874                        }
 875
 876                        strbuf_release(&tmpname);
 877                        free(pack_tmp_name);
 878                        puts(sha1_to_hex(sha1));
 879                }
 880
 881                /* mark written objects as written to previous pack */
 882                for (j = 0; j < nr_written; j++) {
 883                        written_list[j]->offset = (off_t)-1;
 884                }
 885                nr_remaining -= nr_written;
 886        } while (nr_remaining && i < to_pack.nr_objects);
 887
 888        free(written_list);
 889        free(write_order);
 890        stop_progress(&progress_state);
 891        if (written != nr_result)
 892                die("wrote %"PRIu32" objects while expecting %"PRIu32,
 893                        written, nr_result);
 894}
 895
 896static void setup_delta_attr_check(struct git_attr_check *check)
 897{
 898        static struct git_attr *attr_delta;
 899
 900        if (!attr_delta)
 901                attr_delta = git_attr("delta");
 902
 903        check[0].attr = attr_delta;
 904}
 905
 906static int no_try_delta(const char *path)
 907{
 908        struct git_attr_check check[1];
 909
 910        setup_delta_attr_check(check);
 911        if (git_check_attr(path, ARRAY_SIZE(check), check))
 912                return 0;
 913        if (ATTR_FALSE(check->value))
 914                return 1;
 915        return 0;
 916}
 917
 918/*
 919 * When adding an object, check whether we have already added it
 920 * to our packing list. If so, we can skip. However, if we are
 921 * being asked to excludei t, but the previous mention was to include
 922 * it, make sure to adjust its flags and tweak our numbers accordingly.
 923 *
 924 * As an optimization, we pass out the index position where we would have
 925 * found the item, since that saves us from having to look it up again a
 926 * few lines later when we want to add the new entry.
 927 */
 928static int have_duplicate_entry(const unsigned char *sha1,
 929                                int exclude,
 930                                uint32_t *index_pos)
 931{
 932        struct object_entry *entry;
 933
 934        entry = packlist_find(&to_pack, sha1, index_pos);
 935        if (!entry)
 936                return 0;
 937
 938        if (exclude) {
 939                if (!entry->preferred_base)
 940                        nr_result--;
 941                entry->preferred_base = 1;
 942        }
 943
 944        return 1;
 945}
 946
 947/*
 948 * Check whether we want the object in the pack (e.g., we do not want
 949 * objects found in non-local stores if the "--local" option was used).
 950 *
 951 * As a side effect of this check, we will find the packed version of this
 952 * object, if any. We therefore pass out the pack information to avoid having
 953 * to look it up again later.
 954 */
 955static int want_object_in_pack(const unsigned char *sha1,
 956                               int exclude,
 957                               struct packed_git **found_pack,
 958                               off_t *found_offset)
 959{
 960        struct packed_git *p;
 961
 962        if (!exclude && local && has_loose_object_nonlocal(sha1))
 963                return 0;
 964
 965        *found_pack = NULL;
 966        *found_offset = 0;
 967
 968        for (p = packed_git; p; p = p->next) {
 969                off_t offset = find_pack_entry_one(sha1, p);
 970                if (offset) {
 971                        if (!*found_pack) {
 972                                if (!is_pack_valid(p))
 973                                        continue;
 974                                *found_offset = offset;
 975                                *found_pack = p;
 976                        }
 977                        if (exclude)
 978                                return 1;
 979                        if (incremental)
 980                                return 0;
 981
 982                        /*
 983                         * When asked to do --local (do not include an
 984                         * object that appears in a pack we borrow
 985                         * from elsewhere) or --honor-pack-keep (do not
 986                         * include an object that appears in a pack marked
 987                         * with .keep), we need to make sure no copy of this
 988                         * object come from in _any_ pack that causes us to
 989                         * omit it, and need to complete this loop.  When
 990                         * neither option is in effect, we know the object
 991                         * we just found is going to be packed, so break
 992                         * out of the loop to return 1 now.
 993                         */
 994                        if (!ignore_packed_keep &&
 995                            (!local || !have_non_local_packs))
 996                                break;
 997
 998                        if (local && !p->pack_local)
 999                                return 0;
1000                        if (ignore_packed_keep && p->pack_local && p->pack_keep)
1001                                return 0;
1002                }
1003        }
1004
1005        return 1;
1006}
1007
1008static void create_object_entry(const unsigned char *sha1,
1009                                enum object_type type,
1010                                uint32_t hash,
1011                                int exclude,
1012                                int no_try_delta,
1013                                uint32_t index_pos,
1014                                struct packed_git *found_pack,
1015                                off_t found_offset)
1016{
1017        struct object_entry *entry;
1018
1019        entry = packlist_alloc(&to_pack, sha1, index_pos);
1020        entry->hash = hash;
1021        if (type)
1022                entry->type = type;
1023        if (exclude)
1024                entry->preferred_base = 1;
1025        else
1026                nr_result++;
1027        if (found_pack) {
1028                entry->in_pack = found_pack;
1029                entry->in_pack_offset = found_offset;
1030        }
1031
1032        entry->no_try_delta = no_try_delta;
1033}
1034
1035static const char no_closure_warning[] = N_(
1036"disabling bitmap writing, as some objects are not being packed"
1037);
1038
1039static int add_object_entry(const unsigned char *sha1, enum object_type type,
1040                            const char *name, int exclude)
1041{
1042        struct packed_git *found_pack;
1043        off_t found_offset;
1044        uint32_t index_pos;
1045
1046        if (have_duplicate_entry(sha1, exclude, &index_pos))
1047                return 0;
1048
1049        if (!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {
1050                /* The pack is missing an object, so it will not have closure */
1051                if (write_bitmap_index) {
1052                        warning(_(no_closure_warning));
1053                        write_bitmap_index = 0;
1054                }
1055                return 0;
1056        }
1057
1058        create_object_entry(sha1, type, pack_name_hash(name),
1059                            exclude, name && no_try_delta(name),
1060                            index_pos, found_pack, found_offset);
1061
1062        display_progress(progress_state, nr_result);
1063        return 1;
1064}
1065
1066static int add_object_entry_from_bitmap(const unsigned char *sha1,
1067                                        enum object_type type,
1068                                        int flags, uint32_t name_hash,
1069                                        struct packed_git *pack, off_t offset)
1070{
1071        uint32_t index_pos;
1072
1073        if (have_duplicate_entry(sha1, 0, &index_pos))
1074                return 0;
1075
1076        create_object_entry(sha1, type, name_hash, 0, 0, index_pos, pack, offset);
1077
1078        display_progress(progress_state, nr_result);
1079        return 1;
1080}
1081
1082struct pbase_tree_cache {
1083        unsigned char sha1[20];
1084        int ref;
1085        int temporary;
1086        void *tree_data;
1087        unsigned long tree_size;
1088};
1089
1090static struct pbase_tree_cache *(pbase_tree_cache[256]);
1091static int pbase_tree_cache_ix(const unsigned char *sha1)
1092{
1093        return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
1094}
1095static int pbase_tree_cache_ix_incr(int ix)
1096{
1097        return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1098}
1099
1100static struct pbase_tree {
1101        struct pbase_tree *next;
1102        /* This is a phony "cache" entry; we are not
1103         * going to evict it or find it through _get()
1104         * mechanism -- this is for the toplevel node that
1105         * would almost always change with any commit.
1106         */
1107        struct pbase_tree_cache pcache;
1108} *pbase_tree;
1109
1110static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1111{
1112        struct pbase_tree_cache *ent, *nent;
1113        void *data;
1114        unsigned long size;
1115        enum object_type type;
1116        int neigh;
1117        int my_ix = pbase_tree_cache_ix(sha1);
1118        int available_ix = -1;
1119
1120        /* pbase-tree-cache acts as a limited hashtable.
1121         * your object will be found at your index or within a few
1122         * slots after that slot if it is cached.
1123         */
1124        for (neigh = 0; neigh < 8; neigh++) {
1125                ent = pbase_tree_cache[my_ix];
1126                if (ent && !hashcmp(ent->sha1, sha1)) {
1127                        ent->ref++;
1128                        return ent;
1129                }
1130                else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1131                         ((0 <= available_ix) &&
1132                          (!ent && pbase_tree_cache[available_ix])))
1133                        available_ix = my_ix;
1134                if (!ent)
1135                        break;
1136                my_ix = pbase_tree_cache_ix_incr(my_ix);
1137        }
1138
1139        /* Did not find one.  Either we got a bogus request or
1140         * we need to read and perhaps cache.
1141         */
1142        data = read_sha1_file(sha1, &type, &size);
1143        if (!data)
1144                return NULL;
1145        if (type != OBJ_TREE) {
1146                free(data);
1147                return NULL;
1148        }
1149
1150        /* We need to either cache or return a throwaway copy */
1151
1152        if (available_ix < 0)
1153                ent = NULL;
1154        else {
1155                ent = pbase_tree_cache[available_ix];
1156                my_ix = available_ix;
1157        }
1158
1159        if (!ent) {
1160                nent = xmalloc(sizeof(*nent));
1161                nent->temporary = (available_ix < 0);
1162        }
1163        else {
1164                /* evict and reuse */
1165                free(ent->tree_data);
1166                nent = ent;
1167        }
1168        hashcpy(nent->sha1, sha1);
1169        nent->tree_data = data;
1170        nent->tree_size = size;
1171        nent->ref = 1;
1172        if (!nent->temporary)
1173                pbase_tree_cache[my_ix] = nent;
1174        return nent;
1175}
1176
1177static void pbase_tree_put(struct pbase_tree_cache *cache)
1178{
1179        if (!cache->temporary) {
1180                cache->ref--;
1181                return;
1182        }
1183        free(cache->tree_data);
1184        free(cache);
1185}
1186
1187static int name_cmp_len(const char *name)
1188{
1189        int i;
1190        for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1191                ;
1192        return i;
1193}
1194
1195static void add_pbase_object(struct tree_desc *tree,
1196                             const char *name,
1197                             int cmplen,
1198                             const char *fullname)
1199{
1200        struct name_entry entry;
1201        int cmp;
1202
1203        while (tree_entry(tree,&entry)) {
1204                if (S_ISGITLINK(entry.mode))
1205                        continue;
1206                cmp = tree_entry_len(&entry) != cmplen ? 1 :
1207                      memcmp(name, entry.path, cmplen);
1208                if (cmp > 0)
1209                        continue;
1210                if (cmp < 0)
1211                        return;
1212                if (name[cmplen] != '/') {
1213                        add_object_entry(entry.oid->hash,
1214                                         object_type(entry.mode),
1215                                         fullname, 1);
1216                        return;
1217                }
1218                if (S_ISDIR(entry.mode)) {
1219                        struct tree_desc sub;
1220                        struct pbase_tree_cache *tree;
1221                        const char *down = name+cmplen+1;
1222                        int downlen = name_cmp_len(down);
1223
1224                        tree = pbase_tree_get(entry.oid->hash);
1225                        if (!tree)
1226                                return;
1227                        init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1228
1229                        add_pbase_object(&sub, down, downlen, fullname);
1230                        pbase_tree_put(tree);
1231                }
1232        }
1233}
1234
1235static unsigned *done_pbase_paths;
1236static int done_pbase_paths_num;
1237static int done_pbase_paths_alloc;
1238static int done_pbase_path_pos(unsigned hash)
1239{
1240        int lo = 0;
1241        int hi = done_pbase_paths_num;
1242        while (lo < hi) {
1243                int mi = (hi + lo) / 2;
1244                if (done_pbase_paths[mi] == hash)
1245                        return mi;
1246                if (done_pbase_paths[mi] < hash)
1247                        hi = mi;
1248                else
1249                        lo = mi + 1;
1250        }
1251        return -lo-1;
1252}
1253
1254static int check_pbase_path(unsigned hash)
1255{
1256        int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1257        if (0 <= pos)
1258                return 1;
1259        pos = -pos - 1;
1260        ALLOC_GROW(done_pbase_paths,
1261                   done_pbase_paths_num + 1,
1262                   done_pbase_paths_alloc);
1263        done_pbase_paths_num++;
1264        if (pos < done_pbase_paths_num)
1265                memmove(done_pbase_paths + pos + 1,
1266                        done_pbase_paths + pos,
1267                        (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1268        done_pbase_paths[pos] = hash;
1269        return 0;
1270}
1271
1272static void add_preferred_base_object(const char *name)
1273{
1274        struct pbase_tree *it;
1275        int cmplen;
1276        unsigned hash = pack_name_hash(name);
1277
1278        if (!num_preferred_base || check_pbase_path(hash))
1279                return;
1280
1281        cmplen = name_cmp_len(name);
1282        for (it = pbase_tree; it; it = it->next) {
1283                if (cmplen == 0) {
1284                        add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1285                }
1286                else {
1287                        struct tree_desc tree;
1288                        init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1289                        add_pbase_object(&tree, name, cmplen, name);
1290                }
1291        }
1292}
1293
1294static void add_preferred_base(unsigned char *sha1)
1295{
1296        struct pbase_tree *it;
1297        void *data;
1298        unsigned long size;
1299        unsigned char tree_sha1[20];
1300
1301        if (window <= num_preferred_base++)
1302                return;
1303
1304        data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1305        if (!data)
1306                return;
1307
1308        for (it = pbase_tree; it; it = it->next) {
1309                if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1310                        free(data);
1311                        return;
1312                }
1313        }
1314
1315        it = xcalloc(1, sizeof(*it));
1316        it->next = pbase_tree;
1317        pbase_tree = it;
1318
1319        hashcpy(it->pcache.sha1, tree_sha1);
1320        it->pcache.tree_data = data;
1321        it->pcache.tree_size = size;
1322}
1323
1324static void cleanup_preferred_base(void)
1325{
1326        struct pbase_tree *it;
1327        unsigned i;
1328
1329        it = pbase_tree;
1330        pbase_tree = NULL;
1331        while (it) {
1332                struct pbase_tree *this = it;
1333                it = this->next;
1334                free(this->pcache.tree_data);
1335                free(this);
1336        }
1337
1338        for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1339                if (!pbase_tree_cache[i])
1340                        continue;
1341                free(pbase_tree_cache[i]->tree_data);
1342                free(pbase_tree_cache[i]);
1343                pbase_tree_cache[i] = NULL;
1344        }
1345
1346        free(done_pbase_paths);
1347        done_pbase_paths = NULL;
1348        done_pbase_paths_num = done_pbase_paths_alloc = 0;
1349}
1350
1351static void check_object(struct object_entry *entry)
1352{
1353        if (entry->in_pack) {
1354                struct packed_git *p = entry->in_pack;
1355                struct pack_window *w_curs = NULL;
1356                const unsigned char *base_ref = NULL;
1357                struct object_entry *base_entry;
1358                unsigned long used, used_0;
1359                unsigned long avail;
1360                off_t ofs;
1361                unsigned char *buf, c;
1362
1363                buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1364
1365                /*
1366                 * We want in_pack_type even if we do not reuse delta
1367                 * since non-delta representations could still be reused.
1368                 */
1369                used = unpack_object_header_buffer(buf, avail,
1370                                                   &entry->in_pack_type,
1371                                                   &entry->size);
1372                if (used == 0)
1373                        goto give_up;
1374
1375                /*
1376                 * Determine if this is a delta and if so whether we can
1377                 * reuse it or not.  Otherwise let's find out as cheaply as
1378                 * possible what the actual type and size for this object is.
1379                 */
1380                switch (entry->in_pack_type) {
1381                default:
1382                        /* Not a delta hence we've already got all we need. */
1383                        entry->type = entry->in_pack_type;
1384                        entry->in_pack_header_size = used;
1385                        if (entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)
1386                                goto give_up;
1387                        unuse_pack(&w_curs);
1388                        return;
1389                case OBJ_REF_DELTA:
1390                        if (reuse_delta && !entry->preferred_base)
1391                                base_ref = use_pack(p, &w_curs,
1392                                                entry->in_pack_offset + used, NULL);
1393                        entry->in_pack_header_size = used + 20;
1394                        break;
1395                case OBJ_OFS_DELTA:
1396                        buf = use_pack(p, &w_curs,
1397                                       entry->in_pack_offset + used, NULL);
1398                        used_0 = 0;
1399                        c = buf[used_0++];
1400                        ofs = c & 127;
1401                        while (c & 128) {
1402                                ofs += 1;
1403                                if (!ofs || MSB(ofs, 7)) {
1404                                        error("delta base offset overflow in pack for %s",
1405                                              sha1_to_hex(entry->idx.sha1));
1406                                        goto give_up;
1407                                }
1408                                c = buf[used_0++];
1409                                ofs = (ofs << 7) + (c & 127);
1410                        }
1411                        ofs = entry->in_pack_offset - ofs;
1412                        if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1413                                error("delta base offset out of bound for %s",
1414                                      sha1_to_hex(entry->idx.sha1));
1415                                goto give_up;
1416                        }
1417                        if (reuse_delta && !entry->preferred_base) {
1418                                struct revindex_entry *revidx;
1419                                revidx = find_pack_revindex(p, ofs);
1420                                if (!revidx)
1421                                        goto give_up;
1422                                base_ref = nth_packed_object_sha1(p, revidx->nr);
1423                        }
1424                        entry->in_pack_header_size = used + used_0;
1425                        break;
1426                }
1427
1428                if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1429                        /*
1430                         * If base_ref was set above that means we wish to
1431                         * reuse delta data, and we even found that base
1432                         * in the list of objects we want to pack. Goodie!
1433                         *
1434                         * Depth value does not matter - find_deltas() will
1435                         * never consider reused delta as the base object to
1436                         * deltify other objects against, in order to avoid
1437                         * circular deltas.
1438                         */
1439                        entry->type = entry->in_pack_type;
1440                        entry->delta = base_entry;
1441                        entry->delta_size = entry->size;
1442                        entry->delta_sibling = base_entry->delta_child;
1443                        base_entry->delta_child = entry;
1444                        unuse_pack(&w_curs);
1445                        return;
1446                }
1447
1448                if (entry->type) {
1449                        /*
1450                         * This must be a delta and we already know what the
1451                         * final object type is.  Let's extract the actual
1452                         * object size from the delta header.
1453                         */
1454                        entry->size = get_size_from_delta(p, &w_curs,
1455                                        entry->in_pack_offset + entry->in_pack_header_size);
1456                        if (entry->size == 0)
1457                                goto give_up;
1458                        unuse_pack(&w_curs);
1459                        return;
1460                }
1461
1462                /*
1463                 * No choice but to fall back to the recursive delta walk
1464                 * with sha1_object_info() to find about the object type
1465                 * at this point...
1466                 */
1467                give_up:
1468                unuse_pack(&w_curs);
1469        }
1470
1471        entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1472        /*
1473         * The error condition is checked in prepare_pack().  This is
1474         * to permit a missing preferred base object to be ignored
1475         * as a preferred base.  Doing so can result in a larger
1476         * pack file, but the transfer will still take place.
1477         */
1478}
1479
1480static int pack_offset_sort(const void *_a, const void *_b)
1481{
1482        const struct object_entry *a = *(struct object_entry **)_a;
1483        const struct object_entry *b = *(struct object_entry **)_b;
1484
1485        /* avoid filesystem trashing with loose objects */
1486        if (!a->in_pack && !b->in_pack)
1487                return hashcmp(a->idx.sha1, b->idx.sha1);
1488
1489        if (a->in_pack < b->in_pack)
1490                return -1;
1491        if (a->in_pack > b->in_pack)
1492                return 1;
1493        return a->in_pack_offset < b->in_pack_offset ? -1 :
1494                        (a->in_pack_offset > b->in_pack_offset);
1495}
1496
1497/*
1498 * Drop an on-disk delta we were planning to reuse. Naively, this would
1499 * just involve blanking out the "delta" field, but we have to deal
1500 * with some extra book-keeping:
1501 *
1502 *   1. Removing ourselves from the delta_sibling linked list.
1503 *
1504 *   2. Updating our size/type to the non-delta representation. These were
1505 *      either not recorded initially (size) or overwritten with the delta type
1506 *      (type) when check_object() decided to reuse the delta.
1507 */
1508static void drop_reused_delta(struct object_entry *entry)
1509{
1510        struct object_entry **p = &entry->delta->delta_child;
1511        struct object_info oi = OBJECT_INFO_INIT;
1512
1513        while (*p) {
1514                if (*p == entry)
1515                        *p = (*p)->delta_sibling;
1516                else
1517                        p = &(*p)->delta_sibling;
1518        }
1519        entry->delta = NULL;
1520
1521        oi.sizep = &entry->size;
1522        oi.typep = &entry->type;
1523        if (packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) < 0) {
1524                /*
1525                 * We failed to get the info from this pack for some reason;
1526                 * fall back to sha1_object_info, which may find another copy.
1527                 * And if that fails, the error will be recorded in entry->type
1528                 * and dealt with in prepare_pack().
1529                 */
1530                entry->type = sha1_object_info(entry->idx.sha1, &entry->size);
1531        }
1532}
1533
1534/*
1535 * Follow the chain of deltas from this entry onward, throwing away any links
1536 * that cause us to hit a cycle (as determined by the DFS state flags in
1537 * the entries).
1538 */
1539static void break_delta_chains(struct object_entry *entry)
1540{
1541        /* If it's not a delta, it can't be part of a cycle. */
1542        if (!entry->delta) {
1543                entry->dfs_state = DFS_DONE;
1544                return;
1545        }
1546
1547        switch (entry->dfs_state) {
1548        case DFS_NONE:
1549                /*
1550                 * This is the first time we've seen the object. We mark it as
1551                 * part of the active potential cycle and recurse.
1552                 */
1553                entry->dfs_state = DFS_ACTIVE;
1554                break_delta_chains(entry->delta);
1555                entry->dfs_state = DFS_DONE;
1556                break;
1557
1558        case DFS_DONE:
1559                /* object already examined, and not part of a cycle */
1560                break;
1561
1562        case DFS_ACTIVE:
1563                /*
1564                 * We found a cycle that needs broken. It would be correct to
1565                 * break any link in the chain, but it's convenient to
1566                 * break this one.
1567                 */
1568                drop_reused_delta(entry);
1569                entry->dfs_state = DFS_DONE;
1570                break;
1571        }
1572}
1573
1574static void get_object_details(void)
1575{
1576        uint32_t i;
1577        struct object_entry **sorted_by_offset;
1578
1579        sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1580        for (i = 0; i < to_pack.nr_objects; i++)
1581                sorted_by_offset[i] = to_pack.objects + i;
1582        qsort(sorted_by_offset, to_pack.nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1583
1584        for (i = 0; i < to_pack.nr_objects; i++) {
1585                struct object_entry *entry = sorted_by_offset[i];
1586                check_object(entry);
1587                if (big_file_threshold < entry->size)
1588                        entry->no_try_delta = 1;
1589        }
1590
1591        /*
1592         * This must happen in a second pass, since we rely on the delta
1593         * information for the whole list being completed.
1594         */
1595        for (i = 0; i < to_pack.nr_objects; i++)
1596                break_delta_chains(&to_pack.objects[i]);
1597
1598        free(sorted_by_offset);
1599}
1600
1601/*
1602 * We search for deltas in a list sorted by type, by filename hash, and then
1603 * by size, so that we see progressively smaller and smaller files.
1604 * That's because we prefer deltas to be from the bigger file
1605 * to the smaller -- deletes are potentially cheaper, but perhaps
1606 * more importantly, the bigger file is likely the more recent
1607 * one.  The deepest deltas are therefore the oldest objects which are
1608 * less susceptible to be accessed often.
1609 */
1610static int type_size_sort(const void *_a, const void *_b)
1611{
1612        const struct object_entry *a = *(struct object_entry **)_a;
1613        const struct object_entry *b = *(struct object_entry **)_b;
1614
1615        if (a->type > b->type)
1616                return -1;
1617        if (a->type < b->type)
1618                return 1;
1619        if (a->hash > b->hash)
1620                return -1;
1621        if (a->hash < b->hash)
1622                return 1;
1623        if (a->preferred_base > b->preferred_base)
1624                return -1;
1625        if (a->preferred_base < b->preferred_base)
1626                return 1;
1627        if (a->size > b->size)
1628                return -1;
1629        if (a->size < b->size)
1630                return 1;
1631        return a < b ? -1 : (a > b);  /* newest first */
1632}
1633
1634struct unpacked {
1635        struct object_entry *entry;
1636        void *data;
1637        struct delta_index *index;
1638        unsigned depth;
1639};
1640
1641static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1642                           unsigned long delta_size)
1643{
1644        if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1645                return 0;
1646
1647        if (delta_size < cache_max_small_delta_size)
1648                return 1;
1649
1650        /* cache delta, if objects are large enough compared to delta size */
1651        if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1652                return 1;
1653
1654        return 0;
1655}
1656
1657#ifndef NO_PTHREADS
1658
1659static pthread_mutex_t read_mutex;
1660#define read_lock()             pthread_mutex_lock(&read_mutex)
1661#define read_unlock()           pthread_mutex_unlock(&read_mutex)
1662
1663static pthread_mutex_t cache_mutex;
1664#define cache_lock()            pthread_mutex_lock(&cache_mutex)
1665#define cache_unlock()          pthread_mutex_unlock(&cache_mutex)
1666
1667static pthread_mutex_t progress_mutex;
1668#define progress_lock()         pthread_mutex_lock(&progress_mutex)
1669#define progress_unlock()       pthread_mutex_unlock(&progress_mutex)
1670
1671#else
1672
1673#define read_lock()             (void)0
1674#define read_unlock()           (void)0
1675#define cache_lock()            (void)0
1676#define cache_unlock()          (void)0
1677#define progress_lock()         (void)0
1678#define progress_unlock()       (void)0
1679
1680#endif
1681
1682static int try_delta(struct unpacked *trg, struct unpacked *src,
1683                     unsigned max_depth, unsigned long *mem_usage)
1684{
1685        struct object_entry *trg_entry = trg->entry;
1686        struct object_entry *src_entry = src->entry;
1687        unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1688        unsigned ref_depth;
1689        enum object_type type;
1690        void *delta_buf;
1691
1692        /* Don't bother doing diffs between different types */
1693        if (trg_entry->type != src_entry->type)
1694                return -1;
1695
1696        /*
1697         * We do not bother to try a delta that we discarded on an
1698         * earlier try, but only when reusing delta data.  Note that
1699         * src_entry that is marked as the preferred_base should always
1700         * be considered, as even if we produce a suboptimal delta against
1701         * it, we will still save the transfer cost, as we already know
1702         * the other side has it and we won't send src_entry at all.
1703         */
1704        if (reuse_delta && trg_entry->in_pack &&
1705            trg_entry->in_pack == src_entry->in_pack &&
1706            !src_entry->preferred_base &&
1707            trg_entry->in_pack_type != OBJ_REF_DELTA &&
1708            trg_entry->in_pack_type != OBJ_OFS_DELTA)
1709                return 0;
1710
1711        /* Let's not bust the allowed depth. */
1712        if (src->depth >= max_depth)
1713                return 0;
1714
1715        /* Now some size filtering heuristics. */
1716        trg_size = trg_entry->size;
1717        if (!trg_entry->delta) {
1718                max_size = trg_size/2 - 20;
1719                ref_depth = 1;
1720        } else {
1721                max_size = trg_entry->delta_size;
1722                ref_depth = trg->depth;
1723        }
1724        max_size = (uint64_t)max_size * (max_depth - src->depth) /
1725                                                (max_depth - ref_depth + 1);
1726        if (max_size == 0)
1727                return 0;
1728        src_size = src_entry->size;
1729        sizediff = src_size < trg_size ? trg_size - src_size : 0;
1730        if (sizediff >= max_size)
1731                return 0;
1732        if (trg_size < src_size / 32)
1733                return 0;
1734
1735        /* Load data if not already done */
1736        if (!trg->data) {
1737                read_lock();
1738                trg->data = read_sha1_file(trg_entry->idx.sha1, &type, &sz);
1739                read_unlock();
1740                if (!trg->data)
1741                        die("object %s cannot be read",
1742                            sha1_to_hex(trg_entry->idx.sha1));
1743                if (sz != trg_size)
1744                        die("object %s inconsistent object length (%lu vs %lu)",
1745                            sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);
1746                *mem_usage += sz;
1747        }
1748        if (!src->data) {
1749                read_lock();
1750                src->data = read_sha1_file(src_entry->idx.sha1, &type, &sz);
1751                read_unlock();
1752                if (!src->data) {
1753                        if (src_entry->preferred_base) {
1754                                static int warned = 0;
1755                                if (!warned++)
1756                                        warning("object %s cannot be read",
1757                                                sha1_to_hex(src_entry->idx.sha1));
1758                                /*
1759                                 * Those objects are not included in the
1760                                 * resulting pack.  Be resilient and ignore
1761                                 * them if they can't be read, in case the
1762                                 * pack could be created nevertheless.
1763                                 */
1764                                return 0;
1765                        }
1766                        die("object %s cannot be read",
1767                            sha1_to_hex(src_entry->idx.sha1));
1768                }
1769                if (sz != src_size)
1770                        die("object %s inconsistent object length (%lu vs %lu)",
1771                            sha1_to_hex(src_entry->idx.sha1), sz, src_size);
1772                *mem_usage += sz;
1773        }
1774        if (!src->index) {
1775                src->index = create_delta_index(src->data, src_size);
1776                if (!src->index) {
1777                        static int warned = 0;
1778                        if (!warned++)
1779                                warning("suboptimal pack - out of memory");
1780                        return 0;
1781                }
1782                *mem_usage += sizeof_delta_index(src->index);
1783        }
1784
1785        delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1786        if (!delta_buf)
1787                return 0;
1788
1789        if (trg_entry->delta) {
1790                /* Prefer only shallower same-sized deltas. */
1791                if (delta_size == trg_entry->delta_size &&
1792                    src->depth + 1 >= trg->depth) {
1793                        free(delta_buf);
1794                        return 0;
1795                }
1796        }
1797
1798        /*
1799         * Handle memory allocation outside of the cache
1800         * accounting lock.  Compiler will optimize the strangeness
1801         * away when NO_PTHREADS is defined.
1802         */
1803        free(trg_entry->delta_data);
1804        cache_lock();
1805        if (trg_entry->delta_data) {
1806                delta_cache_size -= trg_entry->delta_size;
1807                trg_entry->delta_data = NULL;
1808        }
1809        if (delta_cacheable(src_size, trg_size, delta_size)) {
1810                delta_cache_size += delta_size;
1811                cache_unlock();
1812                trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1813        } else {
1814                cache_unlock();
1815                free(delta_buf);
1816        }
1817
1818        trg_entry->delta = src_entry;
1819        trg_entry->delta_size = delta_size;
1820        trg->depth = src->depth + 1;
1821
1822        return 1;
1823}
1824
1825static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1826{
1827        struct object_entry *child = me->delta_child;
1828        unsigned int m = n;
1829        while (child) {
1830                unsigned int c = check_delta_limit(child, n + 1);
1831                if (m < c)
1832                        m = c;
1833                child = child->delta_sibling;
1834        }
1835        return m;
1836}
1837
1838static unsigned long free_unpacked(struct unpacked *n)
1839{
1840        unsigned long freed_mem = sizeof_delta_index(n->index);
1841        free_delta_index(n->index);
1842        n->index = NULL;
1843        if (n->data) {
1844                freed_mem += n->entry->size;
1845                free(n->data);
1846                n->data = NULL;
1847        }
1848        n->entry = NULL;
1849        n->depth = 0;
1850        return freed_mem;
1851}
1852
1853static void find_deltas(struct object_entry **list, unsigned *list_size,
1854                        int window, int depth, unsigned *processed)
1855{
1856        uint32_t i, idx = 0, count = 0;
1857        struct unpacked *array;
1858        unsigned long mem_usage = 0;
1859
1860        array = xcalloc(window, sizeof(struct unpacked));
1861
1862        for (;;) {
1863                struct object_entry *entry;
1864                struct unpacked *n = array + idx;
1865                int j, max_depth, best_base = -1;
1866
1867                progress_lock();
1868                if (!*list_size) {
1869                        progress_unlock();
1870                        break;
1871                }
1872                entry = *list++;
1873                (*list_size)--;
1874                if (!entry->preferred_base) {
1875                        (*processed)++;
1876                        display_progress(progress_state, *processed);
1877                }
1878                progress_unlock();
1879
1880                mem_usage -= free_unpacked(n);
1881                n->entry = entry;
1882
1883                while (window_memory_limit &&
1884                       mem_usage > window_memory_limit &&
1885                       count > 1) {
1886                        uint32_t tail = (idx + window - count) % window;
1887                        mem_usage -= free_unpacked(array + tail);
1888                        count--;
1889                }
1890
1891                /* We do not compute delta to *create* objects we are not
1892                 * going to pack.
1893                 */
1894                if (entry->preferred_base)
1895                        goto next;
1896
1897                /*
1898                 * If the current object is at pack edge, take the depth the
1899                 * objects that depend on the current object into account
1900                 * otherwise they would become too deep.
1901                 */
1902                max_depth = depth;
1903                if (entry->delta_child) {
1904                        max_depth -= check_delta_limit(entry, 0);
1905                        if (max_depth <= 0)
1906                                goto next;
1907                }
1908
1909                j = window;
1910                while (--j > 0) {
1911                        int ret;
1912                        uint32_t other_idx = idx + j;
1913                        struct unpacked *m;
1914                        if (other_idx >= window)
1915                                other_idx -= window;
1916                        m = array + other_idx;
1917                        if (!m->entry)
1918                                break;
1919                        ret = try_delta(n, m, max_depth, &mem_usage);
1920                        if (ret < 0)
1921                                break;
1922                        else if (ret > 0)
1923                                best_base = other_idx;
1924                }
1925
1926                /*
1927                 * If we decided to cache the delta data, then it is best
1928                 * to compress it right away.  First because we have to do
1929                 * it anyway, and doing it here while we're threaded will
1930                 * save a lot of time in the non threaded write phase,
1931                 * as well as allow for caching more deltas within
1932                 * the same cache size limit.
1933                 * ...
1934                 * But only if not writing to stdout, since in that case
1935                 * the network is most likely throttling writes anyway,
1936                 * and therefore it is best to go to the write phase ASAP
1937                 * instead, as we can afford spending more time compressing
1938                 * between writes at that moment.
1939                 */
1940                if (entry->delta_data && !pack_to_stdout) {
1941                        entry->z_delta_size = do_compress(&entry->delta_data,
1942                                                          entry->delta_size);
1943                        cache_lock();
1944                        delta_cache_size -= entry->delta_size;
1945                        delta_cache_size += entry->z_delta_size;
1946                        cache_unlock();
1947                }
1948
1949                /* if we made n a delta, and if n is already at max
1950                 * depth, leaving it in the window is pointless.  we
1951                 * should evict it first.
1952                 */
1953                if (entry->delta && max_depth <= n->depth)
1954                        continue;
1955
1956                /*
1957                 * Move the best delta base up in the window, after the
1958                 * currently deltified object, to keep it longer.  It will
1959                 * be the first base object to be attempted next.
1960                 */
1961                if (entry->delta) {
1962                        struct unpacked swap = array[best_base];
1963                        int dist = (window + idx - best_base) % window;
1964                        int dst = best_base;
1965                        while (dist--) {
1966                                int src = (dst + 1) % window;
1967                                array[dst] = array[src];
1968                                dst = src;
1969                        }
1970                        array[dst] = swap;
1971                }
1972
1973                next:
1974                idx++;
1975                if (count + 1 < window)
1976                        count++;
1977                if (idx >= window)
1978                        idx = 0;
1979        }
1980
1981        for (i = 0; i < window; ++i) {
1982                free_delta_index(array[i].index);
1983                free(array[i].data);
1984        }
1985        free(array);
1986}
1987
1988#ifndef NO_PTHREADS
1989
1990static void try_to_free_from_threads(size_t size)
1991{
1992        read_lock();
1993        release_pack_memory(size);
1994        read_unlock();
1995}
1996
1997static try_to_free_t old_try_to_free_routine;
1998
1999/*
2000 * The main thread waits on the condition that (at least) one of the workers
2001 * has stopped working (which is indicated in the .working member of
2002 * struct thread_params).
2003 * When a work thread has completed its work, it sets .working to 0 and
2004 * signals the main thread and waits on the condition that .data_ready
2005 * becomes 1.
2006 */
2007
2008struct thread_params {
2009        pthread_t thread;
2010        struct object_entry **list;
2011        unsigned list_size;
2012        unsigned remaining;
2013        int window;
2014        int depth;
2015        int working;
2016        int data_ready;
2017        pthread_mutex_t mutex;
2018        pthread_cond_t cond;
2019        unsigned *processed;
2020};
2021
2022static pthread_cond_t progress_cond;
2023
2024/*
2025 * Mutex and conditional variable can't be statically-initialized on Windows.
2026 */
2027static void init_threaded_search(void)
2028{
2029        init_recursive_mutex(&read_mutex);
2030        pthread_mutex_init(&cache_mutex, NULL);
2031        pthread_mutex_init(&progress_mutex, NULL);
2032        pthread_cond_init(&progress_cond, NULL);
2033        old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2034}
2035
2036static void cleanup_threaded_search(void)
2037{
2038        set_try_to_free_routine(old_try_to_free_routine);
2039        pthread_cond_destroy(&progress_cond);
2040        pthread_mutex_destroy(&read_mutex);
2041        pthread_mutex_destroy(&cache_mutex);
2042        pthread_mutex_destroy(&progress_mutex);
2043}
2044
2045static void *threaded_find_deltas(void *arg)
2046{
2047        struct thread_params *me = arg;
2048
2049        while (me->remaining) {
2050                find_deltas(me->list, &me->remaining,
2051                            me->window, me->depth, me->processed);
2052
2053                progress_lock();
2054                me->working = 0;
2055                pthread_cond_signal(&progress_cond);
2056                progress_unlock();
2057
2058                /*
2059                 * We must not set ->data_ready before we wait on the
2060                 * condition because the main thread may have set it to 1
2061                 * before we get here. In order to be sure that new
2062                 * work is available if we see 1 in ->data_ready, it
2063                 * was initialized to 0 before this thread was spawned
2064                 * and we reset it to 0 right away.
2065                 */
2066                pthread_mutex_lock(&me->mutex);
2067                while (!me->data_ready)
2068                        pthread_cond_wait(&me->cond, &me->mutex);
2069                me->data_ready = 0;
2070                pthread_mutex_unlock(&me->mutex);
2071        }
2072        /* leave ->working 1 so that this doesn't get more work assigned */
2073        return NULL;
2074}
2075
2076static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2077                           int window, int depth, unsigned *processed)
2078{
2079        struct thread_params *p;
2080        int i, ret, active_threads = 0;
2081
2082        init_threaded_search();
2083
2084        if (delta_search_threads <= 1) {
2085                find_deltas(list, &list_size, window, depth, processed);
2086                cleanup_threaded_search();
2087                return;
2088        }
2089        if (progress > pack_to_stdout)
2090                fprintf(stderr, "Delta compression using up to %d threads.\n",
2091                                delta_search_threads);
2092        p = xcalloc(delta_search_threads, sizeof(*p));
2093
2094        /* Partition the work amongst work threads. */
2095        for (i = 0; i < delta_search_threads; i++) {
2096                unsigned sub_size = list_size / (delta_search_threads - i);
2097
2098                /* don't use too small segments or no deltas will be found */
2099                if (sub_size < 2*window && i+1 < delta_search_threads)
2100                        sub_size = 0;
2101
2102                p[i].window = window;
2103                p[i].depth = depth;
2104                p[i].processed = processed;
2105                p[i].working = 1;
2106                p[i].data_ready = 0;
2107
2108                /* try to split chunks on "path" boundaries */
2109                while (sub_size && sub_size < list_size &&
2110                       list[sub_size]->hash &&
2111                       list[sub_size]->hash == list[sub_size-1]->hash)
2112                        sub_size++;
2113
2114                p[i].list = list;
2115                p[i].list_size = sub_size;
2116                p[i].remaining = sub_size;
2117
2118                list += sub_size;
2119                list_size -= sub_size;
2120        }
2121
2122        /* Start work threads. */
2123        for (i = 0; i < delta_search_threads; i++) {
2124                if (!p[i].list_size)
2125                        continue;
2126                pthread_mutex_init(&p[i].mutex, NULL);
2127                pthread_cond_init(&p[i].cond, NULL);
2128                ret = pthread_create(&p[i].thread, NULL,
2129                                     threaded_find_deltas, &p[i]);
2130                if (ret)
2131                        die("unable to create thread: %s", strerror(ret));
2132                active_threads++;
2133        }
2134
2135        /*
2136         * Now let's wait for work completion.  Each time a thread is done
2137         * with its work, we steal half of the remaining work from the
2138         * thread with the largest number of unprocessed objects and give
2139         * it to that newly idle thread.  This ensure good load balancing
2140         * until the remaining object list segments are simply too short
2141         * to be worth splitting anymore.
2142         */
2143        while (active_threads) {
2144                struct thread_params *target = NULL;
2145                struct thread_params *victim = NULL;
2146                unsigned sub_size = 0;
2147
2148                progress_lock();
2149                for (;;) {
2150                        for (i = 0; !target && i < delta_search_threads; i++)
2151                                if (!p[i].working)
2152                                        target = &p[i];
2153                        if (target)
2154                                break;
2155                        pthread_cond_wait(&progress_cond, &progress_mutex);
2156                }
2157
2158                for (i = 0; i < delta_search_threads; i++)
2159                        if (p[i].remaining > 2*window &&
2160                            (!victim || victim->remaining < p[i].remaining))
2161                                victim = &p[i];
2162                if (victim) {
2163                        sub_size = victim->remaining / 2;
2164                        list = victim->list + victim->list_size - sub_size;
2165                        while (sub_size && list[0]->hash &&
2166                               list[0]->hash == list[-1]->hash) {
2167                                list++;
2168                                sub_size--;
2169                        }
2170                        if (!sub_size) {
2171                                /*
2172                                 * It is possible for some "paths" to have
2173                                 * so many objects that no hash boundary
2174                                 * might be found.  Let's just steal the
2175                                 * exact half in that case.
2176                                 */
2177                                sub_size = victim->remaining / 2;
2178                                list -= sub_size;
2179                        }
2180                        target->list = list;
2181                        victim->list_size -= sub_size;
2182                        victim->remaining -= sub_size;
2183                }
2184                target->list_size = sub_size;
2185                target->remaining = sub_size;
2186                target->working = 1;
2187                progress_unlock();
2188
2189                pthread_mutex_lock(&target->mutex);
2190                target->data_ready = 1;
2191                pthread_cond_signal(&target->cond);
2192                pthread_mutex_unlock(&target->mutex);
2193
2194                if (!sub_size) {
2195                        pthread_join(target->thread, NULL);
2196                        pthread_cond_destroy(&target->cond);
2197                        pthread_mutex_destroy(&target->mutex);
2198                        active_threads--;
2199                }
2200        }
2201        cleanup_threaded_search();
2202        free(p);
2203}
2204
2205#else
2206#define ll_find_deltas(l, s, w, d, p)   find_deltas(l, &s, w, d, p)
2207#endif
2208
2209static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2210{
2211        struct object_id peeled;
2212
2213        if (starts_with(path, "refs/tags/") && /* is a tag? */
2214            !peel_ref(path, peeled.hash)    && /* peelable? */
2215            packlist_find(&to_pack, peeled.hash, NULL))      /* object packed? */
2216                add_object_entry(oid->hash, OBJ_TAG, NULL, 0);
2217        return 0;
2218}
2219
2220static void prepare_pack(int window, int depth)
2221{
2222        struct object_entry **delta_list;
2223        uint32_t i, nr_deltas;
2224        unsigned n;
2225
2226        get_object_details();
2227
2228        /*
2229         * If we're locally repacking then we need to be doubly careful
2230         * from now on in order to make sure no stealth corruption gets
2231         * propagated to the new pack.  Clients receiving streamed packs
2232         * should validate everything they get anyway so no need to incur
2233         * the additional cost here in that case.
2234         */
2235        if (!pack_to_stdout)
2236                do_check_packed_object_crc = 1;
2237
2238        if (!to_pack.nr_objects || !window || !depth)
2239                return;
2240
2241        ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2242        nr_deltas = n = 0;
2243
2244        for (i = 0; i < to_pack.nr_objects; i++) {
2245                struct object_entry *entry = to_pack.objects + i;
2246
2247                if (entry->delta)
2248                        /* This happens if we decided to reuse existing
2249                         * delta from a pack.  "reuse_delta &&" is implied.
2250                         */
2251                        continue;
2252
2253                if (entry->size < 50)
2254                        continue;
2255
2256                if (entry->no_try_delta)
2257                        continue;
2258
2259                if (!entry->preferred_base) {
2260                        nr_deltas++;
2261                        if (entry->type < 0)
2262                                die("unable to get type of object %s",
2263                                    sha1_to_hex(entry->idx.sha1));
2264                } else {
2265                        if (entry->type < 0) {
2266                                /*
2267                                 * This object is not found, but we
2268                                 * don't have to include it anyway.
2269                                 */
2270                                continue;
2271                        }
2272                }
2273
2274                delta_list[n++] = entry;
2275        }
2276
2277        if (nr_deltas && n > 1) {
2278                unsigned nr_done = 0;
2279                if (progress)
2280                        progress_state = start_progress(_("Compressing objects"),
2281                                                        nr_deltas);
2282                qsort(delta_list, n, sizeof(*delta_list), type_size_sort);
2283                ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2284                stop_progress(&progress_state);
2285                if (nr_done != nr_deltas)
2286                        die("inconsistency with delta count");
2287        }
2288        free(delta_list);
2289}
2290
2291static int git_pack_config(const char *k, const char *v, void *cb)
2292{
2293        if (!strcmp(k, "pack.window")) {
2294                window = git_config_int(k, v);
2295                return 0;
2296        }
2297        if (!strcmp(k, "pack.windowmemory")) {
2298                window_memory_limit = git_config_ulong(k, v);
2299                return 0;
2300        }
2301        if (!strcmp(k, "pack.depth")) {
2302                depth = git_config_int(k, v);
2303                return 0;
2304        }
2305        if (!strcmp(k, "pack.compression")) {
2306                int level = git_config_int(k, v);
2307                if (level == -1)
2308                        level = Z_DEFAULT_COMPRESSION;
2309                else if (level < 0 || level > Z_BEST_COMPRESSION)
2310                        die("bad pack compression level %d", level);
2311                pack_compression_level = level;
2312                pack_compression_seen = 1;
2313                return 0;
2314        }
2315        if (!strcmp(k, "pack.deltacachesize")) {
2316                max_delta_cache_size = git_config_int(k, v);
2317                return 0;
2318        }
2319        if (!strcmp(k, "pack.deltacachelimit")) {
2320                cache_max_small_delta_size = git_config_int(k, v);
2321                return 0;
2322        }
2323        if (!strcmp(k, "pack.writebitmaphashcache")) {
2324                if (git_config_bool(k, v))
2325                        write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2326                else
2327                        write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2328        }
2329        if (!strcmp(k, "pack.usebitmaps")) {
2330                use_bitmap_index = git_config_bool(k, v);
2331                return 0;
2332        }
2333        if (!strcmp(k, "pack.threads")) {
2334                delta_search_threads = git_config_int(k, v);
2335                if (delta_search_threads < 0)
2336                        die("invalid number of threads specified (%d)",
2337                            delta_search_threads);
2338#ifdef NO_PTHREADS
2339                if (delta_search_threads != 1)
2340                        warning("no threads support, ignoring %s", k);
2341#endif
2342                return 0;
2343        }
2344        if (!strcmp(k, "pack.indexversion")) {
2345                pack_idx_opts.version = git_config_int(k, v);
2346                if (pack_idx_opts.version > 2)
2347                        die("bad pack.indexversion=%"PRIu32,
2348                            pack_idx_opts.version);
2349                return 0;
2350        }
2351        return git_default_config(k, v, cb);
2352}
2353
2354static void read_object_list_from_stdin(void)
2355{
2356        char line[40 + 1 + PATH_MAX + 2];
2357        unsigned char sha1[20];
2358
2359        for (;;) {
2360                if (!fgets(line, sizeof(line), stdin)) {
2361                        if (feof(stdin))
2362                                break;
2363                        if (!ferror(stdin))
2364                                die("fgets returned NULL, not EOF, not error!");
2365                        if (errno != EINTR)
2366                                die_errno("fgets");
2367                        clearerr(stdin);
2368                        continue;
2369                }
2370                if (line[0] == '-') {
2371                        if (get_sha1_hex(line+1, sha1))
2372                                die("expected edge sha1, got garbage:\n %s",
2373                                    line);
2374                        add_preferred_base(sha1);
2375                        continue;
2376                }
2377                if (get_sha1_hex(line, sha1))
2378                        die("expected sha1, got garbage:\n %s", line);
2379
2380                add_preferred_base_object(line+41);
2381                add_object_entry(sha1, 0, line+41, 0);
2382        }
2383}
2384
2385#define OBJECT_ADDED (1u<<20)
2386
2387static void show_commit(struct commit *commit, void *data)
2388{
2389        add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL, 0);
2390        commit->object.flags |= OBJECT_ADDED;
2391
2392        if (write_bitmap_index)
2393                index_commit_for_bitmap(commit);
2394}
2395
2396static void show_object(struct object *obj, const char *name, void *data)
2397{
2398        add_preferred_base_object(name);
2399        add_object_entry(obj->oid.hash, obj->type, name, 0);
2400        obj->flags |= OBJECT_ADDED;
2401}
2402
2403static void show_edge(struct commit *commit)
2404{
2405        add_preferred_base(commit->object.oid.hash);
2406}
2407
2408struct in_pack_object {
2409        off_t offset;
2410        struct object *object;
2411};
2412
2413struct in_pack {
2414        int alloc;
2415        int nr;
2416        struct in_pack_object *array;
2417};
2418
2419static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2420{
2421        in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2422        in_pack->array[in_pack->nr].object = object;
2423        in_pack->nr++;
2424}
2425
2426/*
2427 * Compare the objects in the offset order, in order to emulate the
2428 * "git rev-list --objects" output that produced the pack originally.
2429 */
2430static int ofscmp(const void *a_, const void *b_)
2431{
2432        struct in_pack_object *a = (struct in_pack_object *)a_;
2433        struct in_pack_object *b = (struct in_pack_object *)b_;
2434
2435        if (a->offset < b->offset)
2436                return -1;
2437        else if (a->offset > b->offset)
2438                return 1;
2439        else
2440                return oidcmp(&a->object->oid, &b->object->oid);
2441}
2442
2443static void add_objects_in_unpacked_packs(struct rev_info *revs)
2444{
2445        struct packed_git *p;
2446        struct in_pack in_pack;
2447        uint32_t i;
2448
2449        memset(&in_pack, 0, sizeof(in_pack));
2450
2451        for (p = packed_git; p; p = p->next) {
2452                const unsigned char *sha1;
2453                struct object *o;
2454
2455                if (!p->pack_local || p->pack_keep)
2456                        continue;
2457                if (open_pack_index(p))
2458                        die("cannot open pack index");
2459
2460                ALLOC_GROW(in_pack.array,
2461                           in_pack.nr + p->num_objects,
2462                           in_pack.alloc);
2463
2464                for (i = 0; i < p->num_objects; i++) {
2465                        sha1 = nth_packed_object_sha1(p, i);
2466                        o = lookup_unknown_object(sha1);
2467                        if (!(o->flags & OBJECT_ADDED))
2468                                mark_in_pack_object(o, p, &in_pack);
2469                        o->flags |= OBJECT_ADDED;
2470                }
2471        }
2472
2473        if (in_pack.nr) {
2474                qsort(in_pack.array, in_pack.nr, sizeof(in_pack.array[0]),
2475                      ofscmp);
2476                for (i = 0; i < in_pack.nr; i++) {
2477                        struct object *o = in_pack.array[i].object;
2478                        add_object_entry(o->oid.hash, o->type, "", 0);
2479                }
2480        }
2481        free(in_pack.array);
2482}
2483
2484static int add_loose_object(const unsigned char *sha1, const char *path,
2485                            void *data)
2486{
2487        enum object_type type = sha1_object_info(sha1, NULL);
2488
2489        if (type < 0) {
2490                warning("loose object at %s could not be examined", path);
2491                return 0;
2492        }
2493
2494        add_object_entry(sha1, type, "", 0);
2495        return 0;
2496}
2497
2498/*
2499 * We actually don't even have to worry about reachability here.
2500 * add_object_entry will weed out duplicates, so we just add every
2501 * loose object we find.
2502 */
2503static void add_unreachable_loose_objects(void)
2504{
2505        for_each_loose_file_in_objdir(get_object_directory(),
2506                                      add_loose_object,
2507                                      NULL, NULL, NULL);
2508}
2509
2510static int has_sha1_pack_kept_or_nonlocal(const unsigned char *sha1)
2511{
2512        static struct packed_git *last_found = (void *)1;
2513        struct packed_git *p;
2514
2515        p = (last_found != (void *)1) ? last_found : packed_git;
2516
2517        while (p) {
2518                if ((!p->pack_local || p->pack_keep) &&
2519                        find_pack_entry_one(sha1, p)) {
2520                        last_found = p;
2521                        return 1;
2522                }
2523                if (p == last_found)
2524                        p = packed_git;
2525                else
2526                        p = p->next;
2527                if (p == last_found)
2528                        p = p->next;
2529        }
2530        return 0;
2531}
2532
2533/*
2534 * Store a list of sha1s that are should not be discarded
2535 * because they are either written too recently, or are
2536 * reachable from another object that was.
2537 *
2538 * This is filled by get_object_list.
2539 */
2540static struct sha1_array recent_objects;
2541
2542static int loosened_object_can_be_discarded(const unsigned char *sha1,
2543                                            unsigned long mtime)
2544{
2545        if (!unpack_unreachable_expiration)
2546                return 0;
2547        if (mtime > unpack_unreachable_expiration)
2548                return 0;
2549        if (sha1_array_lookup(&recent_objects, sha1) >= 0)
2550                return 0;
2551        return 1;
2552}
2553
2554static void loosen_unused_packed_objects(struct rev_info *revs)
2555{
2556        struct packed_git *p;
2557        uint32_t i;
2558        const unsigned char *sha1;
2559
2560        for (p = packed_git; p; p = p->next) {
2561                if (!p->pack_local || p->pack_keep)
2562                        continue;
2563
2564                if (open_pack_index(p))
2565                        die("cannot open pack index");
2566
2567                for (i = 0; i < p->num_objects; i++) {
2568                        sha1 = nth_packed_object_sha1(p, i);
2569                        if (!packlist_find(&to_pack, sha1, NULL) &&
2570                            !has_sha1_pack_kept_or_nonlocal(sha1) &&
2571                            !loosened_object_can_be_discarded(sha1, p->mtime))
2572                                if (force_object_loose(sha1, p->mtime))
2573                                        die("unable to force loose object");
2574                }
2575        }
2576}
2577
2578/*
2579 * This tracks any options which a reader of the pack might
2580 * not understand, and which would therefore prevent blind reuse
2581 * of what we have on disk.
2582 */
2583static int pack_options_allow_reuse(void)
2584{
2585        return allow_ofs_delta;
2586}
2587
2588static int get_object_list_from_bitmap(struct rev_info *revs)
2589{
2590        if (prepare_bitmap_walk(revs) < 0)
2591                return -1;
2592
2593        if (pack_options_allow_reuse() &&
2594            !reuse_partial_packfile_from_bitmap(
2595                        &reuse_packfile,
2596                        &reuse_packfile_objects,
2597                        &reuse_packfile_offset)) {
2598                assert(reuse_packfile_objects);
2599                nr_result += reuse_packfile_objects;
2600                display_progress(progress_state, nr_result);
2601        }
2602
2603        traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2604        return 0;
2605}
2606
2607static void record_recent_object(struct object *obj,
2608                                 const char *name,
2609                                 void *data)
2610{
2611        sha1_array_append(&recent_objects, obj->oid.hash);
2612}
2613
2614static void record_recent_commit(struct commit *commit, void *data)
2615{
2616        sha1_array_append(&recent_objects, commit->object.oid.hash);
2617}
2618
2619static void get_object_list(int ac, const char **av)
2620{
2621        struct rev_info revs;
2622        char line[1000];
2623        int flags = 0;
2624
2625        init_revisions(&revs, NULL);
2626        save_commit_buffer = 0;
2627        setup_revisions(ac, av, &revs, NULL);
2628
2629        /* make sure shallows are read */
2630        is_repository_shallow();
2631
2632        while (fgets(line, sizeof(line), stdin) != NULL) {
2633                int len = strlen(line);
2634                if (len && line[len - 1] == '\n')
2635                        line[--len] = 0;
2636                if (!len)
2637                        break;
2638                if (*line == '-') {
2639                        if (!strcmp(line, "--not")) {
2640                                flags ^= UNINTERESTING;
2641                                write_bitmap_index = 0;
2642                                continue;
2643                        }
2644                        if (starts_with(line, "--shallow ")) {
2645                                unsigned char sha1[20];
2646                                if (get_sha1_hex(line + 10, sha1))
2647                                        die("not an SHA-1 '%s'", line + 10);
2648                                register_shallow(sha1);
2649                                use_bitmap_index = 0;
2650                                continue;
2651                        }
2652                        die("not a rev '%s'", line);
2653                }
2654                if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2655                        die("bad revision '%s'", line);
2656        }
2657
2658        if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2659                return;
2660
2661        if (prepare_revision_walk(&revs))
2662                die("revision walk setup failed");
2663        mark_edges_uninteresting(&revs, show_edge);
2664        traverse_commit_list(&revs, show_commit, show_object, NULL);
2665
2666        if (unpack_unreachable_expiration) {
2667                revs.ignore_missing_links = 1;
2668                if (add_unseen_recent_objects_to_traversal(&revs,
2669                                unpack_unreachable_expiration))
2670                        die("unable to add recent objects");
2671                if (prepare_revision_walk(&revs))
2672                        die("revision walk setup failed");
2673                traverse_commit_list(&revs, record_recent_commit,
2674                                     record_recent_object, NULL);
2675        }
2676
2677        if (keep_unreachable)
2678                add_objects_in_unpacked_packs(&revs);
2679        if (pack_loose_unreachable)
2680                add_unreachable_loose_objects();
2681        if (unpack_unreachable)
2682                loosen_unused_packed_objects(&revs);
2683
2684        sha1_array_clear(&recent_objects);
2685}
2686
2687static int option_parse_index_version(const struct option *opt,
2688                                      const char *arg, int unset)
2689{
2690        char *c;
2691        const char *val = arg;
2692        pack_idx_opts.version = strtoul(val, &c, 10);
2693        if (pack_idx_opts.version > 2)
2694                die(_("unsupported index version %s"), val);
2695        if (*c == ',' && c[1])
2696                pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
2697        if (*c || pack_idx_opts.off32_limit & 0x80000000)
2698                die(_("bad index version '%s'"), val);
2699        return 0;
2700}
2701
2702static int option_parse_unpack_unreachable(const struct option *opt,
2703                                           const char *arg, int unset)
2704{
2705        if (unset) {
2706                unpack_unreachable = 0;
2707                unpack_unreachable_expiration = 0;
2708        }
2709        else {
2710                unpack_unreachable = 1;
2711                if (arg)
2712                        unpack_unreachable_expiration = approxidate(arg);
2713        }
2714        return 0;
2715}
2716
2717int cmd_pack_objects(int argc, const char **argv, const char *prefix)
2718{
2719        int use_internal_rev_list = 0;
2720        int thin = 0;
2721        int shallow = 0;
2722        int all_progress_implied = 0;
2723        struct argv_array rp = ARGV_ARRAY_INIT;
2724        int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
2725        int rev_list_index = 0;
2726        struct option pack_objects_options[] = {
2727                OPT_SET_INT('q', "quiet", &progress,
2728                            N_("do not show progress meter"), 0),
2729                OPT_SET_INT(0, "progress", &progress,
2730                            N_("show progress meter"), 1),
2731                OPT_SET_INT(0, "all-progress", &progress,
2732                            N_("show progress meter during object writing phase"), 2),
2733                OPT_BOOL(0, "all-progress-implied",
2734                         &all_progress_implied,
2735                         N_("similar to --all-progress when progress meter is shown")),
2736                { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
2737                  N_("write the pack index file in the specified idx format version"),
2738                  0, option_parse_index_version },
2739                OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
2740                              N_("maximum size of each output pack file")),
2741                OPT_BOOL(0, "local", &local,
2742                         N_("ignore borrowed objects from alternate object store")),
2743                OPT_BOOL(0, "incremental", &incremental,
2744                         N_("ignore packed objects")),
2745                OPT_INTEGER(0, "window", &window,
2746                            N_("limit pack window by objects")),
2747                OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
2748                              N_("limit pack window by memory in addition to object limit")),
2749                OPT_INTEGER(0, "depth", &depth,
2750                            N_("maximum length of delta chain allowed in the resulting pack")),
2751                OPT_BOOL(0, "reuse-delta", &reuse_delta,
2752                         N_("reuse existing deltas")),
2753                OPT_BOOL(0, "reuse-object", &reuse_object,
2754                         N_("reuse existing objects")),
2755                OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
2756                         N_("use OFS_DELTA objects")),
2757                OPT_INTEGER(0, "threads", &delta_search_threads,
2758                            N_("use threads when searching for best delta matches")),
2759                OPT_BOOL(0, "non-empty", &non_empty,
2760                         N_("do not create an empty pack output")),
2761                OPT_BOOL(0, "revs", &use_internal_rev_list,
2762                         N_("read revision arguments from standard input")),
2763                { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
2764                  N_("limit the objects to those that are not yet packed"),
2765                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2766                { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
2767                  N_("include objects reachable from any reference"),
2768                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2769                { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
2770                  N_("include objects referred by reflog entries"),
2771                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2772                { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
2773                  N_("include objects referred to by the index"),
2774                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
2775                OPT_BOOL(0, "stdout", &pack_to_stdout,
2776                         N_("output pack to stdout")),
2777                OPT_BOOL(0, "include-tag", &include_tag,
2778                         N_("include tag objects that refer to objects to be packed")),
2779                OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
2780                         N_("keep unreachable objects")),
2781                OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
2782                         N_("pack loose unreachable objects")),
2783                { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
2784                  N_("unpack unreachable objects newer than <time>"),
2785                  PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
2786                OPT_BOOL(0, "thin", &thin,
2787                         N_("create thin packs")),
2788                OPT_BOOL(0, "shallow", &shallow,
2789                         N_("create packs suitable for shallow fetches")),
2790                OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep,
2791                         N_("ignore packs that have companion .keep file")),
2792                OPT_INTEGER(0, "compression", &pack_compression_level,
2793                            N_("pack compression level")),
2794                OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
2795                            N_("do not hide commits by grafts"), 0),
2796                OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
2797                         N_("use a bitmap index if available to speed up counting objects")),
2798                OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
2799                         N_("write a bitmap index together with the pack index")),
2800                OPT_END(),
2801        };
2802
2803        check_replace_refs = 0;
2804
2805        reset_pack_idx_option(&pack_idx_opts);
2806        git_config(git_pack_config, NULL);
2807        if (!pack_compression_seen && core_compression_seen)
2808                pack_compression_level = core_compression_level;
2809
2810        progress = isatty(2);
2811        argc = parse_options(argc, argv, prefix, pack_objects_options,
2812                             pack_usage, 0);
2813
2814        if (argc) {
2815                base_name = argv[0];
2816                argc--;
2817        }
2818        if (pack_to_stdout != !base_name || argc)
2819                usage_with_options(pack_usage, pack_objects_options);
2820
2821        argv_array_push(&rp, "pack-objects");
2822        if (thin) {
2823                use_internal_rev_list = 1;
2824                argv_array_push(&rp, shallow
2825                                ? "--objects-edge-aggressive"
2826                                : "--objects-edge");
2827        } else
2828                argv_array_push(&rp, "--objects");
2829
2830        if (rev_list_all) {
2831                use_internal_rev_list = 1;
2832                argv_array_push(&rp, "--all");
2833        }
2834        if (rev_list_reflog) {
2835                use_internal_rev_list = 1;
2836                argv_array_push(&rp, "--reflog");
2837        }
2838        if (rev_list_index) {
2839                use_internal_rev_list = 1;
2840                argv_array_push(&rp, "--indexed-objects");
2841        }
2842        if (rev_list_unpacked) {
2843                use_internal_rev_list = 1;
2844                argv_array_push(&rp, "--unpacked");
2845        }
2846
2847        if (!reuse_object)
2848                reuse_delta = 0;
2849        if (pack_compression_level == -1)
2850                pack_compression_level = Z_DEFAULT_COMPRESSION;
2851        else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
2852                die("bad pack compression level %d", pack_compression_level);
2853
2854        if (!delta_search_threads)      /* --threads=0 means autodetect */
2855                delta_search_threads = online_cpus();
2856
2857#ifdef NO_PTHREADS
2858        if (delta_search_threads != 1)
2859                warning("no threads support, ignoring --threads");
2860#endif
2861        if (!pack_to_stdout && !pack_size_limit)
2862                pack_size_limit = pack_size_limit_cfg;
2863        if (pack_to_stdout && pack_size_limit)
2864                die("--max-pack-size cannot be used to build a pack for transfer.");
2865        if (pack_size_limit && pack_size_limit < 1024*1024) {
2866                warning("minimum pack size limit is 1 MiB");
2867                pack_size_limit = 1024*1024;
2868        }
2869
2870        if (!pack_to_stdout && thin)
2871                die("--thin cannot be used to build an indexable pack.");
2872
2873        if (keep_unreachable && unpack_unreachable)
2874                die("--keep-unreachable and --unpack-unreachable are incompatible.");
2875        if (!rev_list_all || !rev_list_reflog || !rev_list_index)
2876                unpack_unreachable_expiration = 0;
2877
2878        if (!use_internal_rev_list || !pack_to_stdout || is_repository_shallow())
2879                use_bitmap_index = 0;
2880
2881        if (pack_to_stdout || !rev_list_all)
2882                write_bitmap_index = 0;
2883
2884        if (progress && all_progress_implied)
2885                progress = 2;
2886
2887        prepare_packed_git();
2888        if (ignore_packed_keep) {
2889                struct packed_git *p;
2890                for (p = packed_git; p; p = p->next)
2891                        if (p->pack_local && p->pack_keep)
2892                                break;
2893                if (!p) /* no keep-able packs found */
2894                        ignore_packed_keep = 0;
2895        }
2896        if (local) {
2897                /*
2898                 * unlike ignore_packed_keep above, we do not want to
2899                 * unset "local" based on looking at packs, as it
2900                 * also covers non-local objects
2901                 */
2902                struct packed_git *p;
2903                for (p = packed_git; p; p = p->next) {
2904                        if (!p->pack_local) {
2905                                have_non_local_packs = 1;
2906                                break;
2907                        }
2908                }
2909        }
2910
2911        if (progress)
2912                progress_state = start_progress(_("Counting objects"), 0);
2913        if (!use_internal_rev_list)
2914                read_object_list_from_stdin();
2915        else {
2916                get_object_list(rp.argc, rp.argv);
2917                argv_array_clear(&rp);
2918        }
2919        cleanup_preferred_base();
2920        if (include_tag && nr_result)
2921                for_each_ref(add_ref_tag, NULL);
2922        stop_progress(&progress_state);
2923
2924        if (non_empty && !nr_result)
2925                return 0;
2926        if (nr_result)
2927                prepare_pack(window, depth);
2928        write_pack_file();
2929        if (progress)
2930                fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
2931                        " reused %"PRIu32" (delta %"PRIu32")\n",
2932                        written, written_delta, reused, reused_delta);
2933        return 0;
2934}