builtin / pack-objects.con commit pack-objects: handle island check for "external" delta base (2fa233a)
   1#include "builtin.h"
   2#include "cache.h"
   3#include "repository.h"
   4#include "config.h"
   5#include "attr.h"
   6#include "object.h"
   7#include "blob.h"
   8#include "commit.h"
   9#include "tag.h"
  10#include "tree.h"
  11#include "delta.h"
  12#include "pack.h"
  13#include "pack-revindex.h"
  14#include "csum-file.h"
  15#include "tree-walk.h"
  16#include "diff.h"
  17#include "revision.h"
  18#include "list-objects.h"
  19#include "list-objects-filter.h"
  20#include "list-objects-filter-options.h"
  21#include "pack-objects.h"
  22#include "progress.h"
  23#include "refs.h"
  24#include "streaming.h"
  25#include "thread-utils.h"
  26#include "pack-bitmap.h"
  27#include "delta-islands.h"
  28#include "reachable.h"
  29#include "sha1-array.h"
  30#include "argv-array.h"
  31#include "list.h"
  32#include "packfile.h"
  33#include "object-store.h"
  34#include "dir.h"
  35#include "midx.h"
  36
  37#define IN_PACK(obj) oe_in_pack(&to_pack, obj)
  38#define SIZE(obj) oe_size(&to_pack, obj)
  39#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
  40#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
  41#define DELTA(obj) oe_delta(&to_pack, obj)
  42#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
  43#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
  44#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
  45#define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid)
  46#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
  47#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
  48#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
  49
  50static const char *pack_usage[] = {
  51        N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
  52        N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
  53        NULL
  54};
  55
  56/*
  57 * Objects we are going to pack are collected in the `to_pack` structure.
  58 * It contains an array (dynamically expanded) of the object data, and a map
  59 * that can resolve SHA1s to their position in the array.
  60 */
  61static struct packing_data to_pack;
  62
  63static struct pack_idx_entry **written_list;
  64static uint32_t nr_result, nr_written, nr_seen;
  65static struct bitmap_index *bitmap_git;
  66static uint32_t write_layer;
  67
  68static int non_empty;
  69static int reuse_delta = 1, reuse_object = 1;
  70static int keep_unreachable, unpack_unreachable, include_tag;
  71static timestamp_t unpack_unreachable_expiration;
  72static int pack_loose_unreachable;
  73static int local;
  74static int have_non_local_packs;
  75static int incremental;
  76static int ignore_packed_keep_on_disk;
  77static int ignore_packed_keep_in_core;
  78static int allow_ofs_delta;
  79static struct pack_idx_option pack_idx_opts;
  80static const char *base_name;
  81static int progress = 1;
  82static int window = 10;
  83static unsigned long pack_size_limit;
  84static int depth = 50;
  85static int delta_search_threads;
  86static int pack_to_stdout;
  87static int thin;
  88static int num_preferred_base;
  89static struct progress *progress_state;
  90
  91static struct packed_git *reuse_packfile;
  92static uint32_t reuse_packfile_objects;
  93static off_t reuse_packfile_offset;
  94
  95static int use_bitmap_index_default = 1;
  96static int use_bitmap_index = -1;
  97static int write_bitmap_index;
  98static uint16_t write_bitmap_options;
  99
 100static int exclude_promisor_objects;
 101
 102static int use_delta_islands;
 103
 104static unsigned long delta_cache_size = 0;
 105static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
 106static unsigned long cache_max_small_delta_size = 1000;
 107
 108static unsigned long window_memory_limit = 0;
 109
 110static struct list_objects_filter_options filter_options;
 111
 112enum missing_action {
 113        MA_ERROR = 0,      /* fail if any missing objects are encountered */
 114        MA_ALLOW_ANY,      /* silently allow ALL missing objects */
 115        MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
 116};
 117static enum missing_action arg_missing_action;
 118static show_object_fn fn_show_object;
 119
 120/*
 121 * stats
 122 */
 123static uint32_t written, written_delta;
 124static uint32_t reused, reused_delta;
 125
 126/*
 127 * Indexed commits
 128 */
 129static struct commit **indexed_commits;
 130static unsigned int indexed_commits_nr;
 131static unsigned int indexed_commits_alloc;
 132
 133static void index_commit_for_bitmap(struct commit *commit)
 134{
 135        if (indexed_commits_nr >= indexed_commits_alloc) {
 136                indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
 137                REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
 138        }
 139
 140        indexed_commits[indexed_commits_nr++] = commit;
 141}
 142
 143static void *get_delta(struct object_entry *entry)
 144{
 145        unsigned long size, base_size, delta_size;
 146        void *buf, *base_buf, *delta_buf;
 147        enum object_type type;
 148
 149        buf = read_object_file(&entry->idx.oid, &type, &size);
 150        if (!buf)
 151                die(_("unable to read %s"), oid_to_hex(&entry->idx.oid));
 152        base_buf = read_object_file(&DELTA(entry)->idx.oid, &type,
 153                                    &base_size);
 154        if (!base_buf)
 155                die("unable to read %s",
 156                    oid_to_hex(&DELTA(entry)->idx.oid));
 157        delta_buf = diff_delta(base_buf, base_size,
 158                               buf, size, &delta_size, 0);
 159        /*
 160         * We succesfully computed this delta once but dropped it for
 161         * memory reasons. Something is very wrong if this time we
 162         * recompute and create a different delta.
 163         */
 164        if (!delta_buf || delta_size != DELTA_SIZE(entry))
 165                BUG("delta size changed");
 166        free(buf);
 167        free(base_buf);
 168        return delta_buf;
 169}
 170
 171static unsigned long do_compress(void **pptr, unsigned long size)
 172{
 173        git_zstream stream;
 174        void *in, *out;
 175        unsigned long maxsize;
 176
 177        git_deflate_init(&stream, pack_compression_level);
 178        maxsize = git_deflate_bound(&stream, size);
 179
 180        in = *pptr;
 181        out = xmalloc(maxsize);
 182        *pptr = out;
 183
 184        stream.next_in = in;
 185        stream.avail_in = size;
 186        stream.next_out = out;
 187        stream.avail_out = maxsize;
 188        while (git_deflate(&stream, Z_FINISH) == Z_OK)
 189                ; /* nothing */
 190        git_deflate_end(&stream);
 191
 192        free(in);
 193        return stream.total_out;
 194}
 195
 196static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
 197                                           const struct object_id *oid)
 198{
 199        git_zstream stream;
 200        unsigned char ibuf[1024 * 16];
 201        unsigned char obuf[1024 * 16];
 202        unsigned long olen = 0;
 203
 204        git_deflate_init(&stream, pack_compression_level);
 205
 206        for (;;) {
 207                ssize_t readlen;
 208                int zret = Z_OK;
 209                readlen = read_istream(st, ibuf, sizeof(ibuf));
 210                if (readlen == -1)
 211                        die(_("unable to read %s"), oid_to_hex(oid));
 212
 213                stream.next_in = ibuf;
 214                stream.avail_in = readlen;
 215                while ((stream.avail_in || readlen == 0) &&
 216                       (zret == Z_OK || zret == Z_BUF_ERROR)) {
 217                        stream.next_out = obuf;
 218                        stream.avail_out = sizeof(obuf);
 219                        zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
 220                        hashwrite(f, obuf, stream.next_out - obuf);
 221                        olen += stream.next_out - obuf;
 222                }
 223                if (stream.avail_in)
 224                        die(_("deflate error (%d)"), zret);
 225                if (readlen == 0) {
 226                        if (zret != Z_STREAM_END)
 227                                die(_("deflate error (%d)"), zret);
 228                        break;
 229                }
 230        }
 231        git_deflate_end(&stream);
 232        return olen;
 233}
 234
 235/*
 236 * we are going to reuse the existing object data as is.  make
 237 * sure it is not corrupt.
 238 */
 239static int check_pack_inflate(struct packed_git *p,
 240                struct pack_window **w_curs,
 241                off_t offset,
 242                off_t len,
 243                unsigned long expect)
 244{
 245        git_zstream stream;
 246        unsigned char fakebuf[4096], *in;
 247        int st;
 248
 249        memset(&stream, 0, sizeof(stream));
 250        git_inflate_init(&stream);
 251        do {
 252                in = use_pack(p, w_curs, offset, &stream.avail_in);
 253                stream.next_in = in;
 254                stream.next_out = fakebuf;
 255                stream.avail_out = sizeof(fakebuf);
 256                st = git_inflate(&stream, Z_FINISH);
 257                offset += stream.next_in - in;
 258        } while (st == Z_OK || st == Z_BUF_ERROR);
 259        git_inflate_end(&stream);
 260        return (st == Z_STREAM_END &&
 261                stream.total_out == expect &&
 262                stream.total_in == len) ? 0 : -1;
 263}
 264
 265static void copy_pack_data(struct hashfile *f,
 266                struct packed_git *p,
 267                struct pack_window **w_curs,
 268                off_t offset,
 269                off_t len)
 270{
 271        unsigned char *in;
 272        unsigned long avail;
 273
 274        while (len) {
 275                in = use_pack(p, w_curs, offset, &avail);
 276                if (avail > len)
 277                        avail = (unsigned long)len;
 278                hashwrite(f, in, avail);
 279                offset += avail;
 280                len -= avail;
 281        }
 282}
 283
 284/* Return 0 if we will bust the pack-size limit */
 285static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
 286                                           unsigned long limit, int usable_delta)
 287{
 288        unsigned long size, datalen;
 289        unsigned char header[MAX_PACK_OBJECT_HEADER],
 290                      dheader[MAX_PACK_OBJECT_HEADER];
 291        unsigned hdrlen;
 292        enum object_type type;
 293        void *buf;
 294        struct git_istream *st = NULL;
 295        const unsigned hashsz = the_hash_algo->rawsz;
 296
 297        if (!usable_delta) {
 298                if (oe_type(entry) == OBJ_BLOB &&
 299                    oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
 300                    (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL)
 301                        buf = NULL;
 302                else {
 303                        buf = read_object_file(&entry->idx.oid, &type, &size);
 304                        if (!buf)
 305                                die(_("unable to read %s"),
 306                                    oid_to_hex(&entry->idx.oid));
 307                }
 308                /*
 309                 * make sure no cached delta data remains from a
 310                 * previous attempt before a pack split occurred.
 311                 */
 312                FREE_AND_NULL(entry->delta_data);
 313                entry->z_delta_size = 0;
 314        } else if (entry->delta_data) {
 315                size = DELTA_SIZE(entry);
 316                buf = entry->delta_data;
 317                entry->delta_data = NULL;
 318                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 319                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 320        } else {
 321                buf = get_delta(entry);
 322                size = DELTA_SIZE(entry);
 323                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 324                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 325        }
 326
 327        if (st) /* large blob case, just assume we don't compress well */
 328                datalen = size;
 329        else if (entry->z_delta_size)
 330                datalen = entry->z_delta_size;
 331        else
 332                datalen = do_compress(&buf, size);
 333
 334        /*
 335         * The object header is a byte of 'type' followed by zero or
 336         * more bytes of length.
 337         */
 338        hdrlen = encode_in_pack_object_header(header, sizeof(header),
 339                                              type, size);
 340
 341        if (type == OBJ_OFS_DELTA) {
 342                /*
 343                 * Deltas with relative base contain an additional
 344                 * encoding of the relative offset for the delta
 345                 * base from this object's position in the pack.
 346                 */
 347                off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
 348                unsigned pos = sizeof(dheader) - 1;
 349                dheader[pos] = ofs & 127;
 350                while (ofs >>= 7)
 351                        dheader[--pos] = 128 | (--ofs & 127);
 352                if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
 353                        if (st)
 354                                close_istream(st);
 355                        free(buf);
 356                        return 0;
 357                }
 358                hashwrite(f, header, hdrlen);
 359                hashwrite(f, dheader + pos, sizeof(dheader) - pos);
 360                hdrlen += sizeof(dheader) - pos;
 361        } else if (type == OBJ_REF_DELTA) {
 362                /*
 363                 * Deltas with a base reference contain
 364                 * additional bytes for the base object ID.
 365                 */
 366                if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
 367                        if (st)
 368                                close_istream(st);
 369                        free(buf);
 370                        return 0;
 371                }
 372                hashwrite(f, header, hdrlen);
 373                hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
 374                hdrlen += hashsz;
 375        } else {
 376                if (limit && hdrlen + datalen + hashsz >= limit) {
 377                        if (st)
 378                                close_istream(st);
 379                        free(buf);
 380                        return 0;
 381                }
 382                hashwrite(f, header, hdrlen);
 383        }
 384        if (st) {
 385                datalen = write_large_blob_data(st, f, &entry->idx.oid);
 386                close_istream(st);
 387        } else {
 388                hashwrite(f, buf, datalen);
 389                free(buf);
 390        }
 391
 392        return hdrlen + datalen;
 393}
 394
 395/* Return 0 if we will bust the pack-size limit */
 396static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
 397                                unsigned long limit, int usable_delta)
 398{
 399        struct packed_git *p = IN_PACK(entry);
 400        struct pack_window *w_curs = NULL;
 401        struct revindex_entry *revidx;
 402        off_t offset;
 403        enum object_type type = oe_type(entry);
 404        off_t datalen;
 405        unsigned char header[MAX_PACK_OBJECT_HEADER],
 406                      dheader[MAX_PACK_OBJECT_HEADER];
 407        unsigned hdrlen;
 408        const unsigned hashsz = the_hash_algo->rawsz;
 409        unsigned long entry_size = SIZE(entry);
 410
 411        if (DELTA(entry))
 412                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 413                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 414        hdrlen = encode_in_pack_object_header(header, sizeof(header),
 415                                              type, entry_size);
 416
 417        offset = entry->in_pack_offset;
 418        revidx = find_pack_revindex(p, offset);
 419        datalen = revidx[1].offset - offset;
 420        if (!pack_to_stdout && p->index_version > 1 &&
 421            check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
 422                error(_("bad packed object CRC for %s"),
 423                      oid_to_hex(&entry->idx.oid));
 424                unuse_pack(&w_curs);
 425                return write_no_reuse_object(f, entry, limit, usable_delta);
 426        }
 427
 428        offset += entry->in_pack_header_size;
 429        datalen -= entry->in_pack_header_size;
 430
 431        if (!pack_to_stdout && p->index_version == 1 &&
 432            check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
 433                error(_("corrupt packed object for %s"),
 434                      oid_to_hex(&entry->idx.oid));
 435                unuse_pack(&w_curs);
 436                return write_no_reuse_object(f, entry, limit, usable_delta);
 437        }
 438
 439        if (type == OBJ_OFS_DELTA) {
 440                off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
 441                unsigned pos = sizeof(dheader) - 1;
 442                dheader[pos] = ofs & 127;
 443                while (ofs >>= 7)
 444                        dheader[--pos] = 128 | (--ofs & 127);
 445                if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
 446                        unuse_pack(&w_curs);
 447                        return 0;
 448                }
 449                hashwrite(f, header, hdrlen);
 450                hashwrite(f, dheader + pos, sizeof(dheader) - pos);
 451                hdrlen += sizeof(dheader) - pos;
 452                reused_delta++;
 453        } else if (type == OBJ_REF_DELTA) {
 454                if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
 455                        unuse_pack(&w_curs);
 456                        return 0;
 457                }
 458                hashwrite(f, header, hdrlen);
 459                hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
 460                hdrlen += hashsz;
 461                reused_delta++;
 462        } else {
 463                if (limit && hdrlen + datalen + hashsz >= limit) {
 464                        unuse_pack(&w_curs);
 465                        return 0;
 466                }
 467                hashwrite(f, header, hdrlen);
 468        }
 469        copy_pack_data(f, p, &w_curs, offset, datalen);
 470        unuse_pack(&w_curs);
 471        reused++;
 472        return hdrlen + datalen;
 473}
 474
 475/* Return 0 if we will bust the pack-size limit */
 476static off_t write_object(struct hashfile *f,
 477                          struct object_entry *entry,
 478                          off_t write_offset)
 479{
 480        unsigned long limit;
 481        off_t len;
 482        int usable_delta, to_reuse;
 483
 484        if (!pack_to_stdout)
 485                crc32_begin(f);
 486
 487        /* apply size limit if limited packsize and not first object */
 488        if (!pack_size_limit || !nr_written)
 489                limit = 0;
 490        else if (pack_size_limit <= write_offset)
 491                /*
 492                 * the earlier object did not fit the limit; avoid
 493                 * mistaking this with unlimited (i.e. limit = 0).
 494                 */
 495                limit = 1;
 496        else
 497                limit = pack_size_limit - write_offset;
 498
 499        if (!DELTA(entry))
 500                usable_delta = 0;       /* no delta */
 501        else if (!pack_size_limit)
 502               usable_delta = 1;        /* unlimited packfile */
 503        else if (DELTA(entry)->idx.offset == (off_t)-1)
 504                usable_delta = 0;       /* base was written to another pack */
 505        else if (DELTA(entry)->idx.offset)
 506                usable_delta = 1;       /* base already exists in this pack */
 507        else
 508                usable_delta = 0;       /* base could end up in another pack */
 509
 510        if (!reuse_object)
 511                to_reuse = 0;   /* explicit */
 512        else if (!IN_PACK(entry))
 513                to_reuse = 0;   /* can't reuse what we don't have */
 514        else if (oe_type(entry) == OBJ_REF_DELTA ||
 515                 oe_type(entry) == OBJ_OFS_DELTA)
 516                                /* check_object() decided it for us ... */
 517                to_reuse = usable_delta;
 518                                /* ... but pack split may override that */
 519        else if (oe_type(entry) != entry->in_pack_type)
 520                to_reuse = 0;   /* pack has delta which is unusable */
 521        else if (DELTA(entry))
 522                to_reuse = 0;   /* we want to pack afresh */
 523        else
 524                to_reuse = 1;   /* we have it in-pack undeltified,
 525                                 * and we do not need to deltify it.
 526                                 */
 527
 528        if (!to_reuse)
 529                len = write_no_reuse_object(f, entry, limit, usable_delta);
 530        else
 531                len = write_reuse_object(f, entry, limit, usable_delta);
 532        if (!len)
 533                return 0;
 534
 535        if (usable_delta)
 536                written_delta++;
 537        written++;
 538        if (!pack_to_stdout)
 539                entry->idx.crc32 = crc32_end(f);
 540        return len;
 541}
 542
 543enum write_one_status {
 544        WRITE_ONE_SKIP = -1, /* already written */
 545        WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
 546        WRITE_ONE_WRITTEN = 1, /* normal */
 547        WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
 548};
 549
 550static enum write_one_status write_one(struct hashfile *f,
 551                                       struct object_entry *e,
 552                                       off_t *offset)
 553{
 554        off_t size;
 555        int recursing;
 556
 557        /*
 558         * we set offset to 1 (which is an impossible value) to mark
 559         * the fact that this object is involved in "write its base
 560         * first before writing a deltified object" recursion.
 561         */
 562        recursing = (e->idx.offset == 1);
 563        if (recursing) {
 564                warning(_("recursive delta detected for object %s"),
 565                        oid_to_hex(&e->idx.oid));
 566                return WRITE_ONE_RECURSIVE;
 567        } else if (e->idx.offset || e->preferred_base) {
 568                /* offset is non zero if object is written already. */
 569                return WRITE_ONE_SKIP;
 570        }
 571
 572        /* if we are deltified, write out base object first. */
 573        if (DELTA(e)) {
 574                e->idx.offset = 1; /* now recurse */
 575                switch (write_one(f, DELTA(e), offset)) {
 576                case WRITE_ONE_RECURSIVE:
 577                        /* we cannot depend on this one */
 578                        SET_DELTA(e, NULL);
 579                        break;
 580                default:
 581                        break;
 582                case WRITE_ONE_BREAK:
 583                        e->idx.offset = recursing;
 584                        return WRITE_ONE_BREAK;
 585                }
 586        }
 587
 588        e->idx.offset = *offset;
 589        size = write_object(f, e, *offset);
 590        if (!size) {
 591                e->idx.offset = recursing;
 592                return WRITE_ONE_BREAK;
 593        }
 594        written_list[nr_written++] = &e->idx;
 595
 596        /* make sure off_t is sufficiently large not to wrap */
 597        if (signed_add_overflows(*offset, size))
 598                die(_("pack too large for current definition of off_t"));
 599        *offset += size;
 600        return WRITE_ONE_WRITTEN;
 601}
 602
 603static int mark_tagged(const char *path, const struct object_id *oid, int flag,
 604                       void *cb_data)
 605{
 606        struct object_id peeled;
 607        struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
 608
 609        if (entry)
 610                entry->tagged = 1;
 611        if (!peel_ref(path, &peeled)) {
 612                entry = packlist_find(&to_pack, peeled.hash, NULL);
 613                if (entry)
 614                        entry->tagged = 1;
 615        }
 616        return 0;
 617}
 618
 619static inline void add_to_write_order(struct object_entry **wo,
 620                               unsigned int *endp,
 621                               struct object_entry *e)
 622{
 623        if (e->filled || oe_layer(&to_pack, e) != write_layer)
 624                return;
 625        wo[(*endp)++] = e;
 626        e->filled = 1;
 627}
 628
 629static void add_descendants_to_write_order(struct object_entry **wo,
 630                                           unsigned int *endp,
 631                                           struct object_entry *e)
 632{
 633        int add_to_order = 1;
 634        while (e) {
 635                if (add_to_order) {
 636                        struct object_entry *s;
 637                        /* add this node... */
 638                        add_to_write_order(wo, endp, e);
 639                        /* all its siblings... */
 640                        for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
 641                                add_to_write_order(wo, endp, s);
 642                        }
 643                }
 644                /* drop down a level to add left subtree nodes if possible */
 645                if (DELTA_CHILD(e)) {
 646                        add_to_order = 1;
 647                        e = DELTA_CHILD(e);
 648                } else {
 649                        add_to_order = 0;
 650                        /* our sibling might have some children, it is next */
 651                        if (DELTA_SIBLING(e)) {
 652                                e = DELTA_SIBLING(e);
 653                                continue;
 654                        }
 655                        /* go back to our parent node */
 656                        e = DELTA(e);
 657                        while (e && !DELTA_SIBLING(e)) {
 658                                /* we're on the right side of a subtree, keep
 659                                 * going up until we can go right again */
 660                                e = DELTA(e);
 661                        }
 662                        if (!e) {
 663                                /* done- we hit our original root node */
 664                                return;
 665                        }
 666                        /* pass it off to sibling at this level */
 667                        e = DELTA_SIBLING(e);
 668                }
 669        };
 670}
 671
 672static void add_family_to_write_order(struct object_entry **wo,
 673                                      unsigned int *endp,
 674                                      struct object_entry *e)
 675{
 676        struct object_entry *root;
 677
 678        for (root = e; DELTA(root); root = DELTA(root))
 679                ; /* nothing */
 680        add_descendants_to_write_order(wo, endp, root);
 681}
 682
 683static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end)
 684{
 685        unsigned int i, last_untagged;
 686        struct object_entry *objects = to_pack.objects;
 687
 688        for (i = 0; i < to_pack.nr_objects; i++) {
 689                if (objects[i].tagged)
 690                        break;
 691                add_to_write_order(wo, wo_end, &objects[i]);
 692        }
 693        last_untagged = i;
 694
 695        /*
 696         * Then fill all the tagged tips.
 697         */
 698        for (; i < to_pack.nr_objects; i++) {
 699                if (objects[i].tagged)
 700                        add_to_write_order(wo, wo_end, &objects[i]);
 701        }
 702
 703        /*
 704         * And then all remaining commits and tags.
 705         */
 706        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 707                if (oe_type(&objects[i]) != OBJ_COMMIT &&
 708                    oe_type(&objects[i]) != OBJ_TAG)
 709                        continue;
 710                add_to_write_order(wo, wo_end, &objects[i]);
 711        }
 712
 713        /*
 714         * And then all the trees.
 715         */
 716        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 717                if (oe_type(&objects[i]) != OBJ_TREE)
 718                        continue;
 719                add_to_write_order(wo, wo_end, &objects[i]);
 720        }
 721
 722        /*
 723         * Finally all the rest in really tight order
 724         */
 725        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 726                if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer)
 727                        add_family_to_write_order(wo, wo_end, &objects[i]);
 728        }
 729}
 730
 731static struct object_entry **compute_write_order(void)
 732{
 733        uint32_t max_layers = 1;
 734        unsigned int i, wo_end;
 735
 736        struct object_entry **wo;
 737        struct object_entry *objects = to_pack.objects;
 738
 739        for (i = 0; i < to_pack.nr_objects; i++) {
 740                objects[i].tagged = 0;
 741                objects[i].filled = 0;
 742                SET_DELTA_CHILD(&objects[i], NULL);
 743                SET_DELTA_SIBLING(&objects[i], NULL);
 744        }
 745
 746        /*
 747         * Fully connect delta_child/delta_sibling network.
 748         * Make sure delta_sibling is sorted in the original
 749         * recency order.
 750         */
 751        for (i = to_pack.nr_objects; i > 0;) {
 752                struct object_entry *e = &objects[--i];
 753                if (!DELTA(e))
 754                        continue;
 755                /* Mark me as the first child */
 756                e->delta_sibling_idx = DELTA(e)->delta_child_idx;
 757                SET_DELTA_CHILD(DELTA(e), e);
 758        }
 759
 760        /*
 761         * Mark objects that are at the tip of tags.
 762         */
 763        for_each_tag_ref(mark_tagged, NULL);
 764
 765        if (use_delta_islands)
 766                max_layers = compute_pack_layers(&to_pack);
 767
 768        ALLOC_ARRAY(wo, to_pack.nr_objects);
 769        wo_end = 0;
 770
 771        for (; write_layer < max_layers; ++write_layer)
 772                compute_layer_order(wo, &wo_end);
 773
 774        if (wo_end != to_pack.nr_objects)
 775                die(_("ordered %u objects, expected %"PRIu32),
 776                    wo_end, to_pack.nr_objects);
 777
 778        return wo;
 779}
 780
 781static off_t write_reused_pack(struct hashfile *f)
 782{
 783        unsigned char buffer[8192];
 784        off_t to_write, total;
 785        int fd;
 786
 787        if (!is_pack_valid(reuse_packfile))
 788                die(_("packfile is invalid: %s"), reuse_packfile->pack_name);
 789
 790        fd = git_open(reuse_packfile->pack_name);
 791        if (fd < 0)
 792                die_errno(_("unable to open packfile for reuse: %s"),
 793                          reuse_packfile->pack_name);
 794
 795        if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
 796                die_errno(_("unable to seek in reused packfile"));
 797
 798        if (reuse_packfile_offset < 0)
 799                reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz;
 800
 801        total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
 802
 803        while (to_write) {
 804                int read_pack = xread(fd, buffer, sizeof(buffer));
 805
 806                if (read_pack <= 0)
 807                        die_errno(_("unable to read from reused packfile"));
 808
 809                if (read_pack > to_write)
 810                        read_pack = to_write;
 811
 812                hashwrite(f, buffer, read_pack);
 813                to_write -= read_pack;
 814
 815                /*
 816                 * We don't know the actual number of objects written,
 817                 * only how many bytes written, how many bytes total, and
 818                 * how many objects total. So we can fake it by pretending all
 819                 * objects we are writing are the same size. This gives us a
 820                 * smooth progress meter, and at the end it matches the true
 821                 * answer.
 822                 */
 823                written = reuse_packfile_objects *
 824                                (((double)(total - to_write)) / total);
 825                display_progress(progress_state, written);
 826        }
 827
 828        close(fd);
 829        written = reuse_packfile_objects;
 830        display_progress(progress_state, written);
 831        return reuse_packfile_offset - sizeof(struct pack_header);
 832}
 833
 834static const char no_split_warning[] = N_(
 835"disabling bitmap writing, packs are split due to pack.packSizeLimit"
 836);
 837
 838static void write_pack_file(void)
 839{
 840        uint32_t i = 0, j;
 841        struct hashfile *f;
 842        off_t offset;
 843        uint32_t nr_remaining = nr_result;
 844        time_t last_mtime = 0;
 845        struct object_entry **write_order;
 846
 847        if (progress > pack_to_stdout)
 848                progress_state = start_progress(_("Writing objects"), nr_result);
 849        ALLOC_ARRAY(written_list, to_pack.nr_objects);
 850        write_order = compute_write_order();
 851
 852        do {
 853                struct object_id oid;
 854                char *pack_tmp_name = NULL;
 855
 856                if (pack_to_stdout)
 857                        f = hashfd_throughput(1, "<stdout>", progress_state);
 858                else
 859                        f = create_tmp_packfile(&pack_tmp_name);
 860
 861                offset = write_pack_header(f, nr_remaining);
 862
 863                if (reuse_packfile) {
 864                        off_t packfile_size;
 865                        assert(pack_to_stdout);
 866
 867                        packfile_size = write_reused_pack(f);
 868                        offset += packfile_size;
 869                }
 870
 871                nr_written = 0;
 872                for (; i < to_pack.nr_objects; i++) {
 873                        struct object_entry *e = write_order[i];
 874                        if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
 875                                break;
 876                        display_progress(progress_state, written);
 877                }
 878
 879                /*
 880                 * Did we write the wrong # entries in the header?
 881                 * If so, rewrite it like in fast-import
 882                 */
 883                if (pack_to_stdout) {
 884                        finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
 885                } else if (nr_written == nr_remaining) {
 886                        finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
 887                } else {
 888                        int fd = finalize_hashfile(f, oid.hash, 0);
 889                        fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
 890                                                 nr_written, oid.hash, offset);
 891                        close(fd);
 892                        if (write_bitmap_index) {
 893                                warning(_(no_split_warning));
 894                                write_bitmap_index = 0;
 895                        }
 896                }
 897
 898                if (!pack_to_stdout) {
 899                        struct stat st;
 900                        struct strbuf tmpname = STRBUF_INIT;
 901
 902                        /*
 903                         * Packs are runtime accessed in their mtime
 904                         * order since newer packs are more likely to contain
 905                         * younger objects.  So if we are creating multiple
 906                         * packs then we should modify the mtime of later ones
 907                         * to preserve this property.
 908                         */
 909                        if (stat(pack_tmp_name, &st) < 0) {
 910                                warning_errno(_("failed to stat %s"), pack_tmp_name);
 911                        } else if (!last_mtime) {
 912                                last_mtime = st.st_mtime;
 913                        } else {
 914                                struct utimbuf utb;
 915                                utb.actime = st.st_atime;
 916                                utb.modtime = --last_mtime;
 917                                if (utime(pack_tmp_name, &utb) < 0)
 918                                        warning_errno(_("failed utime() on %s"), pack_tmp_name);
 919                        }
 920
 921                        strbuf_addf(&tmpname, "%s-", base_name);
 922
 923                        if (write_bitmap_index) {
 924                                bitmap_writer_set_checksum(oid.hash);
 925                                bitmap_writer_build_type_index(
 926                                        &to_pack, written_list, nr_written);
 927                        }
 928
 929                        finish_tmp_packfile(&tmpname, pack_tmp_name,
 930                                            written_list, nr_written,
 931                                            &pack_idx_opts, oid.hash);
 932
 933                        if (write_bitmap_index) {
 934                                strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
 935
 936                                stop_progress(&progress_state);
 937
 938                                bitmap_writer_show_progress(progress);
 939                                bitmap_writer_reuse_bitmaps(&to_pack);
 940                                bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
 941                                bitmap_writer_build(&to_pack);
 942                                bitmap_writer_finish(written_list, nr_written,
 943                                                     tmpname.buf, write_bitmap_options);
 944                                write_bitmap_index = 0;
 945                        }
 946
 947                        strbuf_release(&tmpname);
 948                        free(pack_tmp_name);
 949                        puts(oid_to_hex(&oid));
 950                }
 951
 952                /* mark written objects as written to previous pack */
 953                for (j = 0; j < nr_written; j++) {
 954                        written_list[j]->offset = (off_t)-1;
 955                }
 956                nr_remaining -= nr_written;
 957        } while (nr_remaining && i < to_pack.nr_objects);
 958
 959        free(written_list);
 960        free(write_order);
 961        stop_progress(&progress_state);
 962        if (written != nr_result)
 963                die(_("wrote %"PRIu32" objects while expecting %"PRIu32),
 964                    written, nr_result);
 965}
 966
 967static int no_try_delta(const char *path)
 968{
 969        static struct attr_check *check;
 970
 971        if (!check)
 972                check = attr_check_initl("delta", NULL);
 973        if (git_check_attr(&the_index, path, check))
 974                return 0;
 975        if (ATTR_FALSE(check->items[0].value))
 976                return 1;
 977        return 0;
 978}
 979
 980/*
 981 * When adding an object, check whether we have already added it
 982 * to our packing list. If so, we can skip. However, if we are
 983 * being asked to excludei t, but the previous mention was to include
 984 * it, make sure to adjust its flags and tweak our numbers accordingly.
 985 *
 986 * As an optimization, we pass out the index position where we would have
 987 * found the item, since that saves us from having to look it up again a
 988 * few lines later when we want to add the new entry.
 989 */
 990static int have_duplicate_entry(const struct object_id *oid,
 991                                int exclude,
 992                                uint32_t *index_pos)
 993{
 994        struct object_entry *entry;
 995
 996        entry = packlist_find(&to_pack, oid->hash, index_pos);
 997        if (!entry)
 998                return 0;
 999
1000        if (exclude) {
1001                if (!entry->preferred_base)
1002                        nr_result--;
1003                entry->preferred_base = 1;
1004        }
1005
1006        return 1;
1007}
1008
1009static int want_found_object(int exclude, struct packed_git *p)
1010{
1011        if (exclude)
1012                return 1;
1013        if (incremental)
1014                return 0;
1015
1016        /*
1017         * When asked to do --local (do not include an object that appears in a
1018         * pack we borrow from elsewhere) or --honor-pack-keep (do not include
1019         * an object that appears in a pack marked with .keep), finding a pack
1020         * that matches the criteria is sufficient for us to decide to omit it.
1021         * However, even if this pack does not satisfy the criteria, we need to
1022         * make sure no copy of this object appears in _any_ pack that makes us
1023         * to omit the object, so we need to check all the packs.
1024         *
1025         * We can however first check whether these options can possible matter;
1026         * if they do not matter we know we want the object in generated pack.
1027         * Otherwise, we signal "-1" at the end to tell the caller that we do
1028         * not know either way, and it needs to check more packs.
1029         */
1030        if (!ignore_packed_keep_on_disk &&
1031            !ignore_packed_keep_in_core &&
1032            (!local || !have_non_local_packs))
1033                return 1;
1034
1035        if (local && !p->pack_local)
1036                return 0;
1037        if (p->pack_local &&
1038            ((ignore_packed_keep_on_disk && p->pack_keep) ||
1039             (ignore_packed_keep_in_core && p->pack_keep_in_core)))
1040                return 0;
1041
1042        /* we don't know yet; keep looking for more packs */
1043        return -1;
1044}
1045
1046/*
1047 * Check whether we want the object in the pack (e.g., we do not want
1048 * objects found in non-local stores if the "--local" option was used).
1049 *
1050 * If the caller already knows an existing pack it wants to take the object
1051 * from, that is passed in *found_pack and *found_offset; otherwise this
1052 * function finds if there is any pack that has the object and returns the pack
1053 * and its offset in these variables.
1054 */
1055static int want_object_in_pack(const struct object_id *oid,
1056                               int exclude,
1057                               struct packed_git **found_pack,
1058                               off_t *found_offset)
1059{
1060        int want;
1061        struct list_head *pos;
1062        struct multi_pack_index *m;
1063
1064        if (!exclude && local && has_loose_object_nonlocal(oid))
1065                return 0;
1066
1067        /*
1068         * If we already know the pack object lives in, start checks from that
1069         * pack - in the usual case when neither --local was given nor .keep files
1070         * are present we will determine the answer right now.
1071         */
1072        if (*found_pack) {
1073                want = want_found_object(exclude, *found_pack);
1074                if (want != -1)
1075                        return want;
1076        }
1077
1078        for (m = get_multi_pack_index(the_repository); m; m = m->next) {
1079                struct pack_entry e;
1080                if (fill_midx_entry(oid, &e, m)) {
1081                        struct packed_git *p = e.p;
1082                        off_t offset;
1083
1084                        if (p == *found_pack)
1085                                offset = *found_offset;
1086                        else
1087                                offset = find_pack_entry_one(oid->hash, p);
1088
1089                        if (offset) {
1090                                if (!*found_pack) {
1091                                        if (!is_pack_valid(p))
1092                                                continue;
1093                                        *found_offset = offset;
1094                                        *found_pack = p;
1095                                }
1096                                want = want_found_object(exclude, p);
1097                                if (want != -1)
1098                                        return want;
1099                        }
1100                }
1101        }
1102
1103        list_for_each(pos, get_packed_git_mru(the_repository)) {
1104                struct packed_git *p = list_entry(pos, struct packed_git, mru);
1105                off_t offset;
1106
1107                if (p == *found_pack)
1108                        offset = *found_offset;
1109                else
1110                        offset = find_pack_entry_one(oid->hash, p);
1111
1112                if (offset) {
1113                        if (!*found_pack) {
1114                                if (!is_pack_valid(p))
1115                                        continue;
1116                                *found_offset = offset;
1117                                *found_pack = p;
1118                        }
1119                        want = want_found_object(exclude, p);
1120                        if (!exclude && want > 0)
1121                                list_move(&p->mru,
1122                                          get_packed_git_mru(the_repository));
1123                        if (want != -1)
1124                                return want;
1125                }
1126        }
1127
1128        return 1;
1129}
1130
1131static void create_object_entry(const struct object_id *oid,
1132                                enum object_type type,
1133                                uint32_t hash,
1134                                int exclude,
1135                                int no_try_delta,
1136                                uint32_t index_pos,
1137                                struct packed_git *found_pack,
1138                                off_t found_offset)
1139{
1140        struct object_entry *entry;
1141
1142        entry = packlist_alloc(&to_pack, oid->hash, index_pos);
1143        entry->hash = hash;
1144        oe_set_type(entry, type);
1145        if (exclude)
1146                entry->preferred_base = 1;
1147        else
1148                nr_result++;
1149        if (found_pack) {
1150                oe_set_in_pack(&to_pack, entry, found_pack);
1151                entry->in_pack_offset = found_offset;
1152        }
1153
1154        entry->no_try_delta = no_try_delta;
1155}
1156
1157static const char no_closure_warning[] = N_(
1158"disabling bitmap writing, as some objects are not being packed"
1159);
1160
1161static int add_object_entry(const struct object_id *oid, enum object_type type,
1162                            const char *name, int exclude)
1163{
1164        struct packed_git *found_pack = NULL;
1165        off_t found_offset = 0;
1166        uint32_t index_pos;
1167
1168        display_progress(progress_state, ++nr_seen);
1169
1170        if (have_duplicate_entry(oid, exclude, &index_pos))
1171                return 0;
1172
1173        if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1174                /* The pack is missing an object, so it will not have closure */
1175                if (write_bitmap_index) {
1176                        warning(_(no_closure_warning));
1177                        write_bitmap_index = 0;
1178                }
1179                return 0;
1180        }
1181
1182        create_object_entry(oid, type, pack_name_hash(name),
1183                            exclude, name && no_try_delta(name),
1184                            index_pos, found_pack, found_offset);
1185        return 1;
1186}
1187
1188static int add_object_entry_from_bitmap(const struct object_id *oid,
1189                                        enum object_type type,
1190                                        int flags, uint32_t name_hash,
1191                                        struct packed_git *pack, off_t offset)
1192{
1193        uint32_t index_pos;
1194
1195        display_progress(progress_state, ++nr_seen);
1196
1197        if (have_duplicate_entry(oid, 0, &index_pos))
1198                return 0;
1199
1200        if (!want_object_in_pack(oid, 0, &pack, &offset))
1201                return 0;
1202
1203        create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);
1204        return 1;
1205}
1206
1207struct pbase_tree_cache {
1208        struct object_id oid;
1209        int ref;
1210        int temporary;
1211        void *tree_data;
1212        unsigned long tree_size;
1213};
1214
1215static struct pbase_tree_cache *(pbase_tree_cache[256]);
1216static int pbase_tree_cache_ix(const struct object_id *oid)
1217{
1218        return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1219}
1220static int pbase_tree_cache_ix_incr(int ix)
1221{
1222        return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1223}
1224
1225static struct pbase_tree {
1226        struct pbase_tree *next;
1227        /* This is a phony "cache" entry; we are not
1228         * going to evict it or find it through _get()
1229         * mechanism -- this is for the toplevel node that
1230         * would almost always change with any commit.
1231         */
1232        struct pbase_tree_cache pcache;
1233} *pbase_tree;
1234
1235static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1236{
1237        struct pbase_tree_cache *ent, *nent;
1238        void *data;
1239        unsigned long size;
1240        enum object_type type;
1241        int neigh;
1242        int my_ix = pbase_tree_cache_ix(oid);
1243        int available_ix = -1;
1244
1245        /* pbase-tree-cache acts as a limited hashtable.
1246         * your object will be found at your index or within a few
1247         * slots after that slot if it is cached.
1248         */
1249        for (neigh = 0; neigh < 8; neigh++) {
1250                ent = pbase_tree_cache[my_ix];
1251                if (ent && !oidcmp(&ent->oid, oid)) {
1252                        ent->ref++;
1253                        return ent;
1254                }
1255                else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1256                         ((0 <= available_ix) &&
1257                          (!ent && pbase_tree_cache[available_ix])))
1258                        available_ix = my_ix;
1259                if (!ent)
1260                        break;
1261                my_ix = pbase_tree_cache_ix_incr(my_ix);
1262        }
1263
1264        /* Did not find one.  Either we got a bogus request or
1265         * we need to read and perhaps cache.
1266         */
1267        data = read_object_file(oid, &type, &size);
1268        if (!data)
1269                return NULL;
1270        if (type != OBJ_TREE) {
1271                free(data);
1272                return NULL;
1273        }
1274
1275        /* We need to either cache or return a throwaway copy */
1276
1277        if (available_ix < 0)
1278                ent = NULL;
1279        else {
1280                ent = pbase_tree_cache[available_ix];
1281                my_ix = available_ix;
1282        }
1283
1284        if (!ent) {
1285                nent = xmalloc(sizeof(*nent));
1286                nent->temporary = (available_ix < 0);
1287        }
1288        else {
1289                /* evict and reuse */
1290                free(ent->tree_data);
1291                nent = ent;
1292        }
1293        oidcpy(&nent->oid, oid);
1294        nent->tree_data = data;
1295        nent->tree_size = size;
1296        nent->ref = 1;
1297        if (!nent->temporary)
1298                pbase_tree_cache[my_ix] = nent;
1299        return nent;
1300}
1301
1302static void pbase_tree_put(struct pbase_tree_cache *cache)
1303{
1304        if (!cache->temporary) {
1305                cache->ref--;
1306                return;
1307        }
1308        free(cache->tree_data);
1309        free(cache);
1310}
1311
1312static int name_cmp_len(const char *name)
1313{
1314        int i;
1315        for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1316                ;
1317        return i;
1318}
1319
1320static void add_pbase_object(struct tree_desc *tree,
1321                             const char *name,
1322                             int cmplen,
1323                             const char *fullname)
1324{
1325        struct name_entry entry;
1326        int cmp;
1327
1328        while (tree_entry(tree,&entry)) {
1329                if (S_ISGITLINK(entry.mode))
1330                        continue;
1331                cmp = tree_entry_len(&entry) != cmplen ? 1 :
1332                      memcmp(name, entry.path, cmplen);
1333                if (cmp > 0)
1334                        continue;
1335                if (cmp < 0)
1336                        return;
1337                if (name[cmplen] != '/') {
1338                        add_object_entry(entry.oid,
1339                                         object_type(entry.mode),
1340                                         fullname, 1);
1341                        return;
1342                }
1343                if (S_ISDIR(entry.mode)) {
1344                        struct tree_desc sub;
1345                        struct pbase_tree_cache *tree;
1346                        const char *down = name+cmplen+1;
1347                        int downlen = name_cmp_len(down);
1348
1349                        tree = pbase_tree_get(entry.oid);
1350                        if (!tree)
1351                                return;
1352                        init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1353
1354                        add_pbase_object(&sub, down, downlen, fullname);
1355                        pbase_tree_put(tree);
1356                }
1357        }
1358}
1359
1360static unsigned *done_pbase_paths;
1361static int done_pbase_paths_num;
1362static int done_pbase_paths_alloc;
1363static int done_pbase_path_pos(unsigned hash)
1364{
1365        int lo = 0;
1366        int hi = done_pbase_paths_num;
1367        while (lo < hi) {
1368                int mi = lo + (hi - lo) / 2;
1369                if (done_pbase_paths[mi] == hash)
1370                        return mi;
1371                if (done_pbase_paths[mi] < hash)
1372                        hi = mi;
1373                else
1374                        lo = mi + 1;
1375        }
1376        return -lo-1;
1377}
1378
1379static int check_pbase_path(unsigned hash)
1380{
1381        int pos = done_pbase_path_pos(hash);
1382        if (0 <= pos)
1383                return 1;
1384        pos = -pos - 1;
1385        ALLOC_GROW(done_pbase_paths,
1386                   done_pbase_paths_num + 1,
1387                   done_pbase_paths_alloc);
1388        done_pbase_paths_num++;
1389        if (pos < done_pbase_paths_num)
1390                MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1391                           done_pbase_paths_num - pos - 1);
1392        done_pbase_paths[pos] = hash;
1393        return 0;
1394}
1395
1396static void add_preferred_base_object(const char *name)
1397{
1398        struct pbase_tree *it;
1399        int cmplen;
1400        unsigned hash = pack_name_hash(name);
1401
1402        if (!num_preferred_base || check_pbase_path(hash))
1403                return;
1404
1405        cmplen = name_cmp_len(name);
1406        for (it = pbase_tree; it; it = it->next) {
1407                if (cmplen == 0) {
1408                        add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1409                }
1410                else {
1411                        struct tree_desc tree;
1412                        init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1413                        add_pbase_object(&tree, name, cmplen, name);
1414                }
1415        }
1416}
1417
1418static void add_preferred_base(struct object_id *oid)
1419{
1420        struct pbase_tree *it;
1421        void *data;
1422        unsigned long size;
1423        struct object_id tree_oid;
1424
1425        if (window <= num_preferred_base++)
1426                return;
1427
1428        data = read_object_with_reference(oid, tree_type, &size, &tree_oid);
1429        if (!data)
1430                return;
1431
1432        for (it = pbase_tree; it; it = it->next) {
1433                if (!oidcmp(&it->pcache.oid, &tree_oid)) {
1434                        free(data);
1435                        return;
1436                }
1437        }
1438
1439        it = xcalloc(1, sizeof(*it));
1440        it->next = pbase_tree;
1441        pbase_tree = it;
1442
1443        oidcpy(&it->pcache.oid, &tree_oid);
1444        it->pcache.tree_data = data;
1445        it->pcache.tree_size = size;
1446}
1447
1448static void cleanup_preferred_base(void)
1449{
1450        struct pbase_tree *it;
1451        unsigned i;
1452
1453        it = pbase_tree;
1454        pbase_tree = NULL;
1455        while (it) {
1456                struct pbase_tree *tmp = it;
1457                it = tmp->next;
1458                free(tmp->pcache.tree_data);
1459                free(tmp);
1460        }
1461
1462        for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1463                if (!pbase_tree_cache[i])
1464                        continue;
1465                free(pbase_tree_cache[i]->tree_data);
1466                FREE_AND_NULL(pbase_tree_cache[i]);
1467        }
1468
1469        FREE_AND_NULL(done_pbase_paths);
1470        done_pbase_paths_num = done_pbase_paths_alloc = 0;
1471}
1472
1473/*
1474 * Return 1 iff the object specified by "delta" can be sent
1475 * literally as a delta against the base in "base_sha1". If
1476 * so, then *base_out will point to the entry in our packing
1477 * list, or NULL if we must use the external-base list.
1478 *
1479 * Depth value does not matter - find_deltas() will
1480 * never consider reused delta as the base object to
1481 * deltify other objects against, in order to avoid
1482 * circular deltas.
1483 */
1484static int can_reuse_delta(const unsigned char *base_sha1,
1485                           struct object_entry *delta,
1486                           struct object_entry **base_out)
1487{
1488        struct object_entry *base;
1489
1490        if (!base_sha1)
1491                return 0;
1492
1493        /*
1494         * First see if we're already sending the base (or it's explicitly in
1495         * our "excluded" list).
1496         */
1497        base = packlist_find(&to_pack, base_sha1, NULL);
1498        if (base) {
1499                if (!in_same_island(&delta->idx.oid, &base->idx.oid))
1500                        return 0;
1501                *base_out = base;
1502                return 1;
1503        }
1504
1505        /*
1506         * Otherwise, reachability bitmaps may tell us if the receiver has it,
1507         * even if it was buried too deep in history to make it into the
1508         * packing list.
1509         */
1510        if (thin && bitmap_has_sha1_in_uninteresting(bitmap_git, base_sha1)) {
1511                if (use_delta_islands) {
1512                        struct object_id base_oid;
1513                        hashcpy(base_oid.hash, base_sha1);
1514                        if (!in_same_island(&delta->idx.oid, &base_oid))
1515                                return 0;
1516                }
1517                *base_out = NULL;
1518                return 1;
1519        }
1520
1521        return 0;
1522}
1523
1524static void check_object(struct object_entry *entry)
1525{
1526        unsigned long canonical_size;
1527
1528        if (IN_PACK(entry)) {
1529                struct packed_git *p = IN_PACK(entry);
1530                struct pack_window *w_curs = NULL;
1531                const unsigned char *base_ref = NULL;
1532                struct object_entry *base_entry;
1533                unsigned long used, used_0;
1534                unsigned long avail;
1535                off_t ofs;
1536                unsigned char *buf, c;
1537                enum object_type type;
1538                unsigned long in_pack_size;
1539
1540                buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1541
1542                /*
1543                 * We want in_pack_type even if we do not reuse delta
1544                 * since non-delta representations could still be reused.
1545                 */
1546                used = unpack_object_header_buffer(buf, avail,
1547                                                   &type,
1548                                                   &in_pack_size);
1549                if (used == 0)
1550                        goto give_up;
1551
1552                if (type < 0)
1553                        BUG("invalid type %d", type);
1554                entry->in_pack_type = type;
1555
1556                /*
1557                 * Determine if this is a delta and if so whether we can
1558                 * reuse it or not.  Otherwise let's find out as cheaply as
1559                 * possible what the actual type and size for this object is.
1560                 */
1561                switch (entry->in_pack_type) {
1562                default:
1563                        /* Not a delta hence we've already got all we need. */
1564                        oe_set_type(entry, entry->in_pack_type);
1565                        SET_SIZE(entry, in_pack_size);
1566                        entry->in_pack_header_size = used;
1567                        if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
1568                                goto give_up;
1569                        unuse_pack(&w_curs);
1570                        return;
1571                case OBJ_REF_DELTA:
1572                        if (reuse_delta && !entry->preferred_base)
1573                                base_ref = use_pack(p, &w_curs,
1574                                                entry->in_pack_offset + used, NULL);
1575                        entry->in_pack_header_size = used + the_hash_algo->rawsz;
1576                        break;
1577                case OBJ_OFS_DELTA:
1578                        buf = use_pack(p, &w_curs,
1579                                       entry->in_pack_offset + used, NULL);
1580                        used_0 = 0;
1581                        c = buf[used_0++];
1582                        ofs = c & 127;
1583                        while (c & 128) {
1584                                ofs += 1;
1585                                if (!ofs || MSB(ofs, 7)) {
1586                                        error(_("delta base offset overflow in pack for %s"),
1587                                              oid_to_hex(&entry->idx.oid));
1588                                        goto give_up;
1589                                }
1590                                c = buf[used_0++];
1591                                ofs = (ofs << 7) + (c & 127);
1592                        }
1593                        ofs = entry->in_pack_offset - ofs;
1594                        if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1595                                error(_("delta base offset out of bound for %s"),
1596                                      oid_to_hex(&entry->idx.oid));
1597                                goto give_up;
1598                        }
1599                        if (reuse_delta && !entry->preferred_base) {
1600                                struct revindex_entry *revidx;
1601                                revidx = find_pack_revindex(p, ofs);
1602                                if (!revidx)
1603                                        goto give_up;
1604                                base_ref = nth_packed_object_sha1(p, revidx->nr);
1605                        }
1606                        entry->in_pack_header_size = used + used_0;
1607                        break;
1608                }
1609
1610                if (can_reuse_delta(base_ref, entry, &base_entry)) {
1611                        oe_set_type(entry, entry->in_pack_type);
1612                        SET_SIZE(entry, in_pack_size); /* delta size */
1613                        SET_DELTA_SIZE(entry, in_pack_size);
1614
1615                        if (base_entry) {
1616                                SET_DELTA(entry, base_entry);
1617                                entry->delta_sibling_idx = base_entry->delta_child_idx;
1618                                SET_DELTA_CHILD(base_entry, entry);
1619                        } else {
1620                                SET_DELTA_EXT(entry, base_ref);
1621                        }
1622
1623                        unuse_pack(&w_curs);
1624                        return;
1625                }
1626
1627                if (oe_type(entry)) {
1628                        off_t delta_pos;
1629
1630                        /*
1631                         * This must be a delta and we already know what the
1632                         * final object type is.  Let's extract the actual
1633                         * object size from the delta header.
1634                         */
1635                        delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
1636                        canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
1637                        if (canonical_size == 0)
1638                                goto give_up;
1639                        SET_SIZE(entry, canonical_size);
1640                        unuse_pack(&w_curs);
1641                        return;
1642                }
1643
1644                /*
1645                 * No choice but to fall back to the recursive delta walk
1646                 * with sha1_object_info() to find about the object type
1647                 * at this point...
1648                 */
1649                give_up:
1650                unuse_pack(&w_curs);
1651        }
1652
1653        oe_set_type(entry,
1654                    oid_object_info(the_repository, &entry->idx.oid, &canonical_size));
1655        if (entry->type_valid) {
1656                SET_SIZE(entry, canonical_size);
1657        } else {
1658                /*
1659                 * Bad object type is checked in prepare_pack().  This is
1660                 * to permit a missing preferred base object to be ignored
1661                 * as a preferred base.  Doing so can result in a larger
1662                 * pack file, but the transfer will still take place.
1663                 */
1664        }
1665}
1666
1667static int pack_offset_sort(const void *_a, const void *_b)
1668{
1669        const struct object_entry *a = *(struct object_entry **)_a;
1670        const struct object_entry *b = *(struct object_entry **)_b;
1671        const struct packed_git *a_in_pack = IN_PACK(a);
1672        const struct packed_git *b_in_pack = IN_PACK(b);
1673
1674        /* avoid filesystem trashing with loose objects */
1675        if (!a_in_pack && !b_in_pack)
1676                return oidcmp(&a->idx.oid, &b->idx.oid);
1677
1678        if (a_in_pack < b_in_pack)
1679                return -1;
1680        if (a_in_pack > b_in_pack)
1681                return 1;
1682        return a->in_pack_offset < b->in_pack_offset ? -1 :
1683                        (a->in_pack_offset > b->in_pack_offset);
1684}
1685
1686/*
1687 * Drop an on-disk delta we were planning to reuse. Naively, this would
1688 * just involve blanking out the "delta" field, but we have to deal
1689 * with some extra book-keeping:
1690 *
1691 *   1. Removing ourselves from the delta_sibling linked list.
1692 *
1693 *   2. Updating our size/type to the non-delta representation. These were
1694 *      either not recorded initially (size) or overwritten with the delta type
1695 *      (type) when check_object() decided to reuse the delta.
1696 *
1697 *   3. Resetting our delta depth, as we are now a base object.
1698 */
1699static void drop_reused_delta(struct object_entry *entry)
1700{
1701        unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
1702        struct object_info oi = OBJECT_INFO_INIT;
1703        enum object_type type;
1704        unsigned long size;
1705
1706        while (*idx) {
1707                struct object_entry *oe = &to_pack.objects[*idx - 1];
1708
1709                if (oe == entry)
1710                        *idx = oe->delta_sibling_idx;
1711                else
1712                        idx = &oe->delta_sibling_idx;
1713        }
1714        SET_DELTA(entry, NULL);
1715        entry->depth = 0;
1716
1717        oi.sizep = &size;
1718        oi.typep = &type;
1719        if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
1720                /*
1721                 * We failed to get the info from this pack for some reason;
1722                 * fall back to sha1_object_info, which may find another copy.
1723                 * And if that fails, the error will be recorded in oe_type(entry)
1724                 * and dealt with in prepare_pack().
1725                 */
1726                oe_set_type(entry,
1727                            oid_object_info(the_repository, &entry->idx.oid, &size));
1728        } else {
1729                oe_set_type(entry, type);
1730        }
1731        SET_SIZE(entry, size);
1732}
1733
1734/*
1735 * Follow the chain of deltas from this entry onward, throwing away any links
1736 * that cause us to hit a cycle (as determined by the DFS state flags in
1737 * the entries).
1738 *
1739 * We also detect too-long reused chains that would violate our --depth
1740 * limit.
1741 */
1742static void break_delta_chains(struct object_entry *entry)
1743{
1744        /*
1745         * The actual depth of each object we will write is stored as an int,
1746         * as it cannot exceed our int "depth" limit. But before we break
1747         * changes based no that limit, we may potentially go as deep as the
1748         * number of objects, which is elsewhere bounded to a uint32_t.
1749         */
1750        uint32_t total_depth;
1751        struct object_entry *cur, *next;
1752
1753        for (cur = entry, total_depth = 0;
1754             cur;
1755             cur = DELTA(cur), total_depth++) {
1756                if (cur->dfs_state == DFS_DONE) {
1757                        /*
1758                         * We've already seen this object and know it isn't
1759                         * part of a cycle. We do need to append its depth
1760                         * to our count.
1761                         */
1762                        total_depth += cur->depth;
1763                        break;
1764                }
1765
1766                /*
1767                 * We break cycles before looping, so an ACTIVE state (or any
1768                 * other cruft which made its way into the state variable)
1769                 * is a bug.
1770                 */
1771                if (cur->dfs_state != DFS_NONE)
1772                        BUG("confusing delta dfs state in first pass: %d",
1773                            cur->dfs_state);
1774
1775                /*
1776                 * Now we know this is the first time we've seen the object. If
1777                 * it's not a delta, we're done traversing, but we'll mark it
1778                 * done to save time on future traversals.
1779                 */
1780                if (!DELTA(cur)) {
1781                        cur->dfs_state = DFS_DONE;
1782                        break;
1783                }
1784
1785                /*
1786                 * Mark ourselves as active and see if the next step causes
1787                 * us to cycle to another active object. It's important to do
1788                 * this _before_ we loop, because it impacts where we make the
1789                 * cut, and thus how our total_depth counter works.
1790                 * E.g., We may see a partial loop like:
1791                 *
1792                 *   A -> B -> C -> D -> B
1793                 *
1794                 * Cutting B->C breaks the cycle. But now the depth of A is
1795                 * only 1, and our total_depth counter is at 3. The size of the
1796                 * error is always one less than the size of the cycle we
1797                 * broke. Commits C and D were "lost" from A's chain.
1798                 *
1799                 * If we instead cut D->B, then the depth of A is correct at 3.
1800                 * We keep all commits in the chain that we examined.
1801                 */
1802                cur->dfs_state = DFS_ACTIVE;
1803                if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
1804                        drop_reused_delta(cur);
1805                        cur->dfs_state = DFS_DONE;
1806                        break;
1807                }
1808        }
1809
1810        /*
1811         * And now that we've gone all the way to the bottom of the chain, we
1812         * need to clear the active flags and set the depth fields as
1813         * appropriate. Unlike the loop above, which can quit when it drops a
1814         * delta, we need to keep going to look for more depth cuts. So we need
1815         * an extra "next" pointer to keep going after we reset cur->delta.
1816         */
1817        for (cur = entry; cur; cur = next) {
1818                next = DELTA(cur);
1819
1820                /*
1821                 * We should have a chain of zero or more ACTIVE states down to
1822                 * a final DONE. We can quit after the DONE, because either it
1823                 * has no bases, or we've already handled them in a previous
1824                 * call.
1825                 */
1826                if (cur->dfs_state == DFS_DONE)
1827                        break;
1828                else if (cur->dfs_state != DFS_ACTIVE)
1829                        BUG("confusing delta dfs state in second pass: %d",
1830                            cur->dfs_state);
1831
1832                /*
1833                 * If the total_depth is more than depth, then we need to snip
1834                 * the chain into two or more smaller chains that don't exceed
1835                 * the maximum depth. Most of the resulting chains will contain
1836                 * (depth + 1) entries (i.e., depth deltas plus one base), and
1837                 * the last chain (i.e., the one containing entry) will contain
1838                 * whatever entries are left over, namely
1839                 * (total_depth % (depth + 1)) of them.
1840                 *
1841                 * Since we are iterating towards decreasing depth, we need to
1842                 * decrement total_depth as we go, and we need to write to the
1843                 * entry what its final depth will be after all of the
1844                 * snipping. Since we're snipping into chains of length (depth
1845                 * + 1) entries, the final depth of an entry will be its
1846                 * original depth modulo (depth + 1). Any time we encounter an
1847                 * entry whose final depth is supposed to be zero, we snip it
1848                 * from its delta base, thereby making it so.
1849                 */
1850                cur->depth = (total_depth--) % (depth + 1);
1851                if (!cur->depth)
1852                        drop_reused_delta(cur);
1853
1854                cur->dfs_state = DFS_DONE;
1855        }
1856}
1857
1858static void get_object_details(void)
1859{
1860        uint32_t i;
1861        struct object_entry **sorted_by_offset;
1862
1863        if (progress)
1864                progress_state = start_progress(_("Counting objects"),
1865                                                to_pack.nr_objects);
1866
1867        sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1868        for (i = 0; i < to_pack.nr_objects; i++)
1869                sorted_by_offset[i] = to_pack.objects + i;
1870        QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1871
1872        for (i = 0; i < to_pack.nr_objects; i++) {
1873                struct object_entry *entry = sorted_by_offset[i];
1874                check_object(entry);
1875                if (entry->type_valid &&
1876                    oe_size_greater_than(&to_pack, entry, big_file_threshold))
1877                        entry->no_try_delta = 1;
1878                display_progress(progress_state, i + 1);
1879        }
1880        stop_progress(&progress_state);
1881
1882        /*
1883         * This must happen in a second pass, since we rely on the delta
1884         * information for the whole list being completed.
1885         */
1886        for (i = 0; i < to_pack.nr_objects; i++)
1887                break_delta_chains(&to_pack.objects[i]);
1888
1889        free(sorted_by_offset);
1890}
1891
1892/*
1893 * We search for deltas in a list sorted by type, by filename hash, and then
1894 * by size, so that we see progressively smaller and smaller files.
1895 * That's because we prefer deltas to be from the bigger file
1896 * to the smaller -- deletes are potentially cheaper, but perhaps
1897 * more importantly, the bigger file is likely the more recent
1898 * one.  The deepest deltas are therefore the oldest objects which are
1899 * less susceptible to be accessed often.
1900 */
1901static int type_size_sort(const void *_a, const void *_b)
1902{
1903        const struct object_entry *a = *(struct object_entry **)_a;
1904        const struct object_entry *b = *(struct object_entry **)_b;
1905        enum object_type a_type = oe_type(a);
1906        enum object_type b_type = oe_type(b);
1907        unsigned long a_size = SIZE(a);
1908        unsigned long b_size = SIZE(b);
1909
1910        if (a_type > b_type)
1911                return -1;
1912        if (a_type < b_type)
1913                return 1;
1914        if (a->hash > b->hash)
1915                return -1;
1916        if (a->hash < b->hash)
1917                return 1;
1918        if (a->preferred_base > b->preferred_base)
1919                return -1;
1920        if (a->preferred_base < b->preferred_base)
1921                return 1;
1922        if (use_delta_islands) {
1923                int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);
1924                if (island_cmp)
1925                        return island_cmp;
1926        }
1927        if (a_size > b_size)
1928                return -1;
1929        if (a_size < b_size)
1930                return 1;
1931        return a < b ? -1 : (a > b);  /* newest first */
1932}
1933
1934struct unpacked {
1935        struct object_entry *entry;
1936        void *data;
1937        struct delta_index *index;
1938        unsigned depth;
1939};
1940
1941static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1942                           unsigned long delta_size)
1943{
1944        if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1945                return 0;
1946
1947        if (delta_size < cache_max_small_delta_size)
1948                return 1;
1949
1950        /* cache delta, if objects are large enough compared to delta size */
1951        if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1952                return 1;
1953
1954        return 0;
1955}
1956
1957#ifndef NO_PTHREADS
1958
1959/* Protect access to object database */
1960static pthread_mutex_t read_mutex;
1961#define read_lock()             pthread_mutex_lock(&read_mutex)
1962#define read_unlock()           pthread_mutex_unlock(&read_mutex)
1963
1964/* Protect delta_cache_size */
1965static pthread_mutex_t cache_mutex;
1966#define cache_lock()            pthread_mutex_lock(&cache_mutex)
1967#define cache_unlock()          pthread_mutex_unlock(&cache_mutex)
1968
1969/*
1970 * Protect object list partitioning (e.g. struct thread_param) and
1971 * progress_state
1972 */
1973static pthread_mutex_t progress_mutex;
1974#define progress_lock()         pthread_mutex_lock(&progress_mutex)
1975#define progress_unlock()       pthread_mutex_unlock(&progress_mutex)
1976
1977/*
1978 * Access to struct object_entry is unprotected since each thread owns
1979 * a portion of the main object list. Just don't access object entries
1980 * ahead in the list because they can be stolen and would need
1981 * progress_mutex for protection.
1982 */
1983#else
1984
1985#define read_lock()             (void)0
1986#define read_unlock()           (void)0
1987#define cache_lock()            (void)0
1988#define cache_unlock()          (void)0
1989#define progress_lock()         (void)0
1990#define progress_unlock()       (void)0
1991
1992#endif
1993
1994/*
1995 * Return the size of the object without doing any delta
1996 * reconstruction (so non-deltas are true object sizes, but deltas
1997 * return the size of the delta data).
1998 */
1999unsigned long oe_get_size_slow(struct packing_data *pack,
2000                               const struct object_entry *e)
2001{
2002        struct packed_git *p;
2003        struct pack_window *w_curs;
2004        unsigned char *buf;
2005        enum object_type type;
2006        unsigned long used, avail, size;
2007
2008        if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
2009                read_lock();
2010                if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
2011                        die(_("unable to get size of %s"),
2012                            oid_to_hex(&e->idx.oid));
2013                read_unlock();
2014                return size;
2015        }
2016
2017        p = oe_in_pack(pack, e);
2018        if (!p)
2019                BUG("when e->type is a delta, it must belong to a pack");
2020
2021        read_lock();
2022        w_curs = NULL;
2023        buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
2024        used = unpack_object_header_buffer(buf, avail, &type, &size);
2025        if (used == 0)
2026                die(_("unable to parse object header of %s"),
2027                    oid_to_hex(&e->idx.oid));
2028
2029        unuse_pack(&w_curs);
2030        read_unlock();
2031        return size;
2032}
2033
2034static int try_delta(struct unpacked *trg, struct unpacked *src,
2035                     unsigned max_depth, unsigned long *mem_usage)
2036{
2037        struct object_entry *trg_entry = trg->entry;
2038        struct object_entry *src_entry = src->entry;
2039        unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
2040        unsigned ref_depth;
2041        enum object_type type;
2042        void *delta_buf;
2043
2044        /* Don't bother doing diffs between different types */
2045        if (oe_type(trg_entry) != oe_type(src_entry))
2046                return -1;
2047
2048        /*
2049         * We do not bother to try a delta that we discarded on an
2050         * earlier try, but only when reusing delta data.  Note that
2051         * src_entry that is marked as the preferred_base should always
2052         * be considered, as even if we produce a suboptimal delta against
2053         * it, we will still save the transfer cost, as we already know
2054         * the other side has it and we won't send src_entry at all.
2055         */
2056        if (reuse_delta && IN_PACK(trg_entry) &&
2057            IN_PACK(trg_entry) == IN_PACK(src_entry) &&
2058            !src_entry->preferred_base &&
2059            trg_entry->in_pack_type != OBJ_REF_DELTA &&
2060            trg_entry->in_pack_type != OBJ_OFS_DELTA)
2061                return 0;
2062
2063        /* Let's not bust the allowed depth. */
2064        if (src->depth >= max_depth)
2065                return 0;
2066
2067        /* Now some size filtering heuristics. */
2068        trg_size = SIZE(trg_entry);
2069        if (!DELTA(trg_entry)) {
2070                max_size = trg_size/2 - the_hash_algo->rawsz;
2071                ref_depth = 1;
2072        } else {
2073                max_size = DELTA_SIZE(trg_entry);
2074                ref_depth = trg->depth;
2075        }
2076        max_size = (uint64_t)max_size * (max_depth - src->depth) /
2077                                                (max_depth - ref_depth + 1);
2078        if (max_size == 0)
2079                return 0;
2080        src_size = SIZE(src_entry);
2081        sizediff = src_size < trg_size ? trg_size - src_size : 0;
2082        if (sizediff >= max_size)
2083                return 0;
2084        if (trg_size < src_size / 32)
2085                return 0;
2086
2087        if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))
2088                return 0;
2089
2090        /* Load data if not already done */
2091        if (!trg->data) {
2092                read_lock();
2093                trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
2094                read_unlock();
2095                if (!trg->data)
2096                        die(_("object %s cannot be read"),
2097                            oid_to_hex(&trg_entry->idx.oid));
2098                if (sz != trg_size)
2099                        die(_("object %s inconsistent object length (%lu vs %lu)"),
2100                            oid_to_hex(&trg_entry->idx.oid), sz,
2101                            trg_size);
2102                *mem_usage += sz;
2103        }
2104        if (!src->data) {
2105                read_lock();
2106                src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
2107                read_unlock();
2108                if (!src->data) {
2109                        if (src_entry->preferred_base) {
2110                                static int warned = 0;
2111                                if (!warned++)
2112                                        warning(_("object %s cannot be read"),
2113                                                oid_to_hex(&src_entry->idx.oid));
2114                                /*
2115                                 * Those objects are not included in the
2116                                 * resulting pack.  Be resilient and ignore
2117                                 * them if they can't be read, in case the
2118                                 * pack could be created nevertheless.
2119                                 */
2120                                return 0;
2121                        }
2122                        die(_("object %s cannot be read"),
2123                            oid_to_hex(&src_entry->idx.oid));
2124                }
2125                if (sz != src_size)
2126                        die(_("object %s inconsistent object length (%lu vs %lu)"),
2127                            oid_to_hex(&src_entry->idx.oid), sz,
2128                            src_size);
2129                *mem_usage += sz;
2130        }
2131        if (!src->index) {
2132                src->index = create_delta_index(src->data, src_size);
2133                if (!src->index) {
2134                        static int warned = 0;
2135                        if (!warned++)
2136                                warning(_("suboptimal pack - out of memory"));
2137                        return 0;
2138                }
2139                *mem_usage += sizeof_delta_index(src->index);
2140        }
2141
2142        delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2143        if (!delta_buf)
2144                return 0;
2145
2146        if (DELTA(trg_entry)) {
2147                /* Prefer only shallower same-sized deltas. */
2148                if (delta_size == DELTA_SIZE(trg_entry) &&
2149                    src->depth + 1 >= trg->depth) {
2150                        free(delta_buf);
2151                        return 0;
2152                }
2153        }
2154
2155        /*
2156         * Handle memory allocation outside of the cache
2157         * accounting lock.  Compiler will optimize the strangeness
2158         * away when NO_PTHREADS is defined.
2159         */
2160        free(trg_entry->delta_data);
2161        cache_lock();
2162        if (trg_entry->delta_data) {
2163                delta_cache_size -= DELTA_SIZE(trg_entry);
2164                trg_entry->delta_data = NULL;
2165        }
2166        if (delta_cacheable(src_size, trg_size, delta_size)) {
2167                delta_cache_size += delta_size;
2168                cache_unlock();
2169                trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2170        } else {
2171                cache_unlock();
2172                free(delta_buf);
2173        }
2174
2175        SET_DELTA(trg_entry, src_entry);
2176        SET_DELTA_SIZE(trg_entry, delta_size);
2177        trg->depth = src->depth + 1;
2178
2179        return 1;
2180}
2181
2182static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2183{
2184        struct object_entry *child = DELTA_CHILD(me);
2185        unsigned int m = n;
2186        while (child) {
2187                unsigned int c = check_delta_limit(child, n + 1);
2188                if (m < c)
2189                        m = c;
2190                child = DELTA_SIBLING(child);
2191        }
2192        return m;
2193}
2194
2195static unsigned long free_unpacked(struct unpacked *n)
2196{
2197        unsigned long freed_mem = sizeof_delta_index(n->index);
2198        free_delta_index(n->index);
2199        n->index = NULL;
2200        if (n->data) {
2201                freed_mem += SIZE(n->entry);
2202                FREE_AND_NULL(n->data);
2203        }
2204        n->entry = NULL;
2205        n->depth = 0;
2206        return freed_mem;
2207}
2208
2209static void find_deltas(struct object_entry **list, unsigned *list_size,
2210                        int window, int depth, unsigned *processed)
2211{
2212        uint32_t i, idx = 0, count = 0;
2213        struct unpacked *array;
2214        unsigned long mem_usage = 0;
2215
2216        array = xcalloc(window, sizeof(struct unpacked));
2217
2218        for (;;) {
2219                struct object_entry *entry;
2220                struct unpacked *n = array + idx;
2221                int j, max_depth, best_base = -1;
2222
2223                progress_lock();
2224                if (!*list_size) {
2225                        progress_unlock();
2226                        break;
2227                }
2228                entry = *list++;
2229                (*list_size)--;
2230                if (!entry->preferred_base) {
2231                        (*processed)++;
2232                        display_progress(progress_state, *processed);
2233                }
2234                progress_unlock();
2235
2236                mem_usage -= free_unpacked(n);
2237                n->entry = entry;
2238
2239                while (window_memory_limit &&
2240                       mem_usage > window_memory_limit &&
2241                       count > 1) {
2242                        uint32_t tail = (idx + window - count) % window;
2243                        mem_usage -= free_unpacked(array + tail);
2244                        count--;
2245                }
2246
2247                /* We do not compute delta to *create* objects we are not
2248                 * going to pack.
2249                 */
2250                if (entry->preferred_base)
2251                        goto next;
2252
2253                /*
2254                 * If the current object is at pack edge, take the depth the
2255                 * objects that depend on the current object into account
2256                 * otherwise they would become too deep.
2257                 */
2258                max_depth = depth;
2259                if (DELTA_CHILD(entry)) {
2260                        max_depth -= check_delta_limit(entry, 0);
2261                        if (max_depth <= 0)
2262                                goto next;
2263                }
2264
2265                j = window;
2266                while (--j > 0) {
2267                        int ret;
2268                        uint32_t other_idx = idx + j;
2269                        struct unpacked *m;
2270                        if (other_idx >= window)
2271                                other_idx -= window;
2272                        m = array + other_idx;
2273                        if (!m->entry)
2274                                break;
2275                        ret = try_delta(n, m, max_depth, &mem_usage);
2276                        if (ret < 0)
2277                                break;
2278                        else if (ret > 0)
2279                                best_base = other_idx;
2280                }
2281
2282                /*
2283                 * If we decided to cache the delta data, then it is best
2284                 * to compress it right away.  First because we have to do
2285                 * it anyway, and doing it here while we're threaded will
2286                 * save a lot of time in the non threaded write phase,
2287                 * as well as allow for caching more deltas within
2288                 * the same cache size limit.
2289                 * ...
2290                 * But only if not writing to stdout, since in that case
2291                 * the network is most likely throttling writes anyway,
2292                 * and therefore it is best to go to the write phase ASAP
2293                 * instead, as we can afford spending more time compressing
2294                 * between writes at that moment.
2295                 */
2296                if (entry->delta_data && !pack_to_stdout) {
2297                        unsigned long size;
2298
2299                        size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
2300                        if (size < (1U << OE_Z_DELTA_BITS)) {
2301                                entry->z_delta_size = size;
2302                                cache_lock();
2303                                delta_cache_size -= DELTA_SIZE(entry);
2304                                delta_cache_size += entry->z_delta_size;
2305                                cache_unlock();
2306                        } else {
2307                                FREE_AND_NULL(entry->delta_data);
2308                                entry->z_delta_size = 0;
2309                        }
2310                }
2311
2312                /* if we made n a delta, and if n is already at max
2313                 * depth, leaving it in the window is pointless.  we
2314                 * should evict it first.
2315                 */
2316                if (DELTA(entry) && max_depth <= n->depth)
2317                        continue;
2318
2319                /*
2320                 * Move the best delta base up in the window, after the
2321                 * currently deltified object, to keep it longer.  It will
2322                 * be the first base object to be attempted next.
2323                 */
2324                if (DELTA(entry)) {
2325                        struct unpacked swap = array[best_base];
2326                        int dist = (window + idx - best_base) % window;
2327                        int dst = best_base;
2328                        while (dist--) {
2329                                int src = (dst + 1) % window;
2330                                array[dst] = array[src];
2331                                dst = src;
2332                        }
2333                        array[dst] = swap;
2334                }
2335
2336                next:
2337                idx++;
2338                if (count + 1 < window)
2339                        count++;
2340                if (idx >= window)
2341                        idx = 0;
2342        }
2343
2344        for (i = 0; i < window; ++i) {
2345                free_delta_index(array[i].index);
2346                free(array[i].data);
2347        }
2348        free(array);
2349}
2350
2351#ifndef NO_PTHREADS
2352
2353static void try_to_free_from_threads(size_t size)
2354{
2355        read_lock();
2356        release_pack_memory(size);
2357        read_unlock();
2358}
2359
2360static try_to_free_t old_try_to_free_routine;
2361
2362/*
2363 * The main object list is split into smaller lists, each is handed to
2364 * one worker.
2365 *
2366 * The main thread waits on the condition that (at least) one of the workers
2367 * has stopped working (which is indicated in the .working member of
2368 * struct thread_params).
2369 *
2370 * When a work thread has completed its work, it sets .working to 0 and
2371 * signals the main thread and waits on the condition that .data_ready
2372 * becomes 1.
2373 *
2374 * The main thread steals half of the work from the worker that has
2375 * most work left to hand it to the idle worker.
2376 */
2377
2378struct thread_params {
2379        pthread_t thread;
2380        struct object_entry **list;
2381        unsigned list_size;
2382        unsigned remaining;
2383        int window;
2384        int depth;
2385        int working;
2386        int data_ready;
2387        pthread_mutex_t mutex;
2388        pthread_cond_t cond;
2389        unsigned *processed;
2390};
2391
2392static pthread_cond_t progress_cond;
2393
2394/*
2395 * Mutex and conditional variable can't be statically-initialized on Windows.
2396 */
2397static void init_threaded_search(void)
2398{
2399        init_recursive_mutex(&read_mutex);
2400        pthread_mutex_init(&cache_mutex, NULL);
2401        pthread_mutex_init(&progress_mutex, NULL);
2402        pthread_cond_init(&progress_cond, NULL);
2403        pthread_mutex_init(&to_pack.lock, NULL);
2404        old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2405}
2406
2407static void cleanup_threaded_search(void)
2408{
2409        set_try_to_free_routine(old_try_to_free_routine);
2410        pthread_cond_destroy(&progress_cond);
2411        pthread_mutex_destroy(&read_mutex);
2412        pthread_mutex_destroy(&cache_mutex);
2413        pthread_mutex_destroy(&progress_mutex);
2414}
2415
2416static void *threaded_find_deltas(void *arg)
2417{
2418        struct thread_params *me = arg;
2419
2420        progress_lock();
2421        while (me->remaining) {
2422                progress_unlock();
2423
2424                find_deltas(me->list, &me->remaining,
2425                            me->window, me->depth, me->processed);
2426
2427                progress_lock();
2428                me->working = 0;
2429                pthread_cond_signal(&progress_cond);
2430                progress_unlock();
2431
2432                /*
2433                 * We must not set ->data_ready before we wait on the
2434                 * condition because the main thread may have set it to 1
2435                 * before we get here. In order to be sure that new
2436                 * work is available if we see 1 in ->data_ready, it
2437                 * was initialized to 0 before this thread was spawned
2438                 * and we reset it to 0 right away.
2439                 */
2440                pthread_mutex_lock(&me->mutex);
2441                while (!me->data_ready)
2442                        pthread_cond_wait(&me->cond, &me->mutex);
2443                me->data_ready = 0;
2444                pthread_mutex_unlock(&me->mutex);
2445
2446                progress_lock();
2447        }
2448        progress_unlock();
2449        /* leave ->working 1 so that this doesn't get more work assigned */
2450        return NULL;
2451}
2452
2453static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2454                           int window, int depth, unsigned *processed)
2455{
2456        struct thread_params *p;
2457        int i, ret, active_threads = 0;
2458
2459        init_threaded_search();
2460
2461        if (delta_search_threads <= 1) {
2462                find_deltas(list, &list_size, window, depth, processed);
2463                cleanup_threaded_search();
2464                return;
2465        }
2466        if (progress > pack_to_stdout)
2467                fprintf_ln(stderr, _("Delta compression using up to %d threads"),
2468                           delta_search_threads);
2469        p = xcalloc(delta_search_threads, sizeof(*p));
2470
2471        /* Partition the work amongst work threads. */
2472        for (i = 0; i < delta_search_threads; i++) {
2473                unsigned sub_size = list_size / (delta_search_threads - i);
2474
2475                /* don't use too small segments or no deltas will be found */
2476                if (sub_size < 2*window && i+1 < delta_search_threads)
2477                        sub_size = 0;
2478
2479                p[i].window = window;
2480                p[i].depth = depth;
2481                p[i].processed = processed;
2482                p[i].working = 1;
2483                p[i].data_ready = 0;
2484
2485                /* try to split chunks on "path" boundaries */
2486                while (sub_size && sub_size < list_size &&
2487                       list[sub_size]->hash &&
2488                       list[sub_size]->hash == list[sub_size-1]->hash)
2489                        sub_size++;
2490
2491                p[i].list = list;
2492                p[i].list_size = sub_size;
2493                p[i].remaining = sub_size;
2494
2495                list += sub_size;
2496                list_size -= sub_size;
2497        }
2498
2499        /* Start work threads. */
2500        for (i = 0; i < delta_search_threads; i++) {
2501                if (!p[i].list_size)
2502                        continue;
2503                pthread_mutex_init(&p[i].mutex, NULL);
2504                pthread_cond_init(&p[i].cond, NULL);
2505                ret = pthread_create(&p[i].thread, NULL,
2506                                     threaded_find_deltas, &p[i]);
2507                if (ret)
2508                        die(_("unable to create thread: %s"), strerror(ret));
2509                active_threads++;
2510        }
2511
2512        /*
2513         * Now let's wait for work completion.  Each time a thread is done
2514         * with its work, we steal half of the remaining work from the
2515         * thread with the largest number of unprocessed objects and give
2516         * it to that newly idle thread.  This ensure good load balancing
2517         * until the remaining object list segments are simply too short
2518         * to be worth splitting anymore.
2519         */
2520        while (active_threads) {
2521                struct thread_params *target = NULL;
2522                struct thread_params *victim = NULL;
2523                unsigned sub_size = 0;
2524
2525                progress_lock();
2526                for (;;) {
2527                        for (i = 0; !target && i < delta_search_threads; i++)
2528                                if (!p[i].working)
2529                                        target = &p[i];
2530                        if (target)
2531                                break;
2532                        pthread_cond_wait(&progress_cond, &progress_mutex);
2533                }
2534
2535                for (i = 0; i < delta_search_threads; i++)
2536                        if (p[i].remaining > 2*window &&
2537                            (!victim || victim->remaining < p[i].remaining))
2538                                victim = &p[i];
2539                if (victim) {
2540                        sub_size = victim->remaining / 2;
2541                        list = victim->list + victim->list_size - sub_size;
2542                        while (sub_size && list[0]->hash &&
2543                               list[0]->hash == list[-1]->hash) {
2544                                list++;
2545                                sub_size--;
2546                        }
2547                        if (!sub_size) {
2548                                /*
2549                                 * It is possible for some "paths" to have
2550                                 * so many objects that no hash boundary
2551                                 * might be found.  Let's just steal the
2552                                 * exact half in that case.
2553                                 */
2554                                sub_size = victim->remaining / 2;
2555                                list -= sub_size;
2556                        }
2557                        target->list = list;
2558                        victim->list_size -= sub_size;
2559                        victim->remaining -= sub_size;
2560                }
2561                target->list_size = sub_size;
2562                target->remaining = sub_size;
2563                target->working = 1;
2564                progress_unlock();
2565
2566                pthread_mutex_lock(&target->mutex);
2567                target->data_ready = 1;
2568                pthread_cond_signal(&target->cond);
2569                pthread_mutex_unlock(&target->mutex);
2570
2571                if (!sub_size) {
2572                        pthread_join(target->thread, NULL);
2573                        pthread_cond_destroy(&target->cond);
2574                        pthread_mutex_destroy(&target->mutex);
2575                        active_threads--;
2576                }
2577        }
2578        cleanup_threaded_search();
2579        free(p);
2580}
2581
2582#else
2583#define ll_find_deltas(l, s, w, d, p)   find_deltas(l, &s, w, d, p)
2584#endif
2585
2586static void add_tag_chain(const struct object_id *oid)
2587{
2588        struct tag *tag;
2589
2590        /*
2591         * We catch duplicates already in add_object_entry(), but we'd
2592         * prefer to do this extra check to avoid having to parse the
2593         * tag at all if we already know that it's being packed (e.g., if
2594         * it was included via bitmaps, we would not have parsed it
2595         * previously).
2596         */
2597        if (packlist_find(&to_pack, oid->hash, NULL))
2598                return;
2599
2600        tag = lookup_tag(the_repository, oid);
2601        while (1) {
2602                if (!tag || parse_tag(tag) || !tag->tagged)
2603                        die(_("unable to pack objects reachable from tag %s"),
2604                            oid_to_hex(oid));
2605
2606                add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2607
2608                if (tag->tagged->type != OBJ_TAG)
2609                        return;
2610
2611                tag = (struct tag *)tag->tagged;
2612        }
2613}
2614
2615static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2616{
2617        struct object_id peeled;
2618
2619        if (starts_with(path, "refs/tags/") && /* is a tag? */
2620            !peel_ref(path, &peeled)    && /* peelable? */
2621            packlist_find(&to_pack, peeled.hash, NULL))      /* object packed? */
2622                add_tag_chain(oid);
2623        return 0;
2624}
2625
2626static void prepare_pack(int window, int depth)
2627{
2628        struct object_entry **delta_list;
2629        uint32_t i, nr_deltas;
2630        unsigned n;
2631
2632        if (use_delta_islands)
2633                resolve_tree_islands(progress, &to_pack);
2634
2635        get_object_details();
2636
2637        /*
2638         * If we're locally repacking then we need to be doubly careful
2639         * from now on in order to make sure no stealth corruption gets
2640         * propagated to the new pack.  Clients receiving streamed packs
2641         * should validate everything they get anyway so no need to incur
2642         * the additional cost here in that case.
2643         */
2644        if (!pack_to_stdout)
2645                do_check_packed_object_crc = 1;
2646
2647        if (!to_pack.nr_objects || !window || !depth)
2648                return;
2649
2650        ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2651        nr_deltas = n = 0;
2652
2653        for (i = 0; i < to_pack.nr_objects; i++) {
2654                struct object_entry *entry = to_pack.objects + i;
2655
2656                if (DELTA(entry))
2657                        /* This happens if we decided to reuse existing
2658                         * delta from a pack.  "reuse_delta &&" is implied.
2659                         */
2660                        continue;
2661
2662                if (!entry->type_valid ||
2663                    oe_size_less_than(&to_pack, entry, 50))
2664                        continue;
2665
2666                if (entry->no_try_delta)
2667                        continue;
2668
2669                if (!entry->preferred_base) {
2670                        nr_deltas++;
2671                        if (oe_type(entry) < 0)
2672                                die(_("unable to get type of object %s"),
2673                                    oid_to_hex(&entry->idx.oid));
2674                } else {
2675                        if (oe_type(entry) < 0) {
2676                                /*
2677                                 * This object is not found, but we
2678                                 * don't have to include it anyway.
2679                                 */
2680                                continue;
2681                        }
2682                }
2683
2684                delta_list[n++] = entry;
2685        }
2686
2687        if (nr_deltas && n > 1) {
2688                unsigned nr_done = 0;
2689                if (progress)
2690                        progress_state = start_progress(_("Compressing objects"),
2691                                                        nr_deltas);
2692                QSORT(delta_list, n, type_size_sort);
2693                ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2694                stop_progress(&progress_state);
2695                if (nr_done != nr_deltas)
2696                        die(_("inconsistency with delta count"));
2697        }
2698        free(delta_list);
2699}
2700
2701static int git_pack_config(const char *k, const char *v, void *cb)
2702{
2703        if (!strcmp(k, "pack.window")) {
2704                window = git_config_int(k, v);
2705                return 0;
2706        }
2707        if (!strcmp(k, "pack.windowmemory")) {
2708                window_memory_limit = git_config_ulong(k, v);
2709                return 0;
2710        }
2711        if (!strcmp(k, "pack.depth")) {
2712                depth = git_config_int(k, v);
2713                return 0;
2714        }
2715        if (!strcmp(k, "pack.deltacachesize")) {
2716                max_delta_cache_size = git_config_int(k, v);
2717                return 0;
2718        }
2719        if (!strcmp(k, "pack.deltacachelimit")) {
2720                cache_max_small_delta_size = git_config_int(k, v);
2721                return 0;
2722        }
2723        if (!strcmp(k, "pack.writebitmaphashcache")) {
2724                if (git_config_bool(k, v))
2725                        write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2726                else
2727                        write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2728        }
2729        if (!strcmp(k, "pack.usebitmaps")) {
2730                use_bitmap_index_default = git_config_bool(k, v);
2731                return 0;
2732        }
2733        if (!strcmp(k, "pack.threads")) {
2734                delta_search_threads = git_config_int(k, v);
2735                if (delta_search_threads < 0)
2736                        die(_("invalid number of threads specified (%d)"),
2737                            delta_search_threads);
2738#ifdef NO_PTHREADS
2739                if (delta_search_threads != 1) {
2740                        warning(_("no threads support, ignoring %s"), k);
2741                        delta_search_threads = 0;
2742                }
2743#endif
2744                return 0;
2745        }
2746        if (!strcmp(k, "pack.indexversion")) {
2747                pack_idx_opts.version = git_config_int(k, v);
2748                if (pack_idx_opts.version > 2)
2749                        die(_("bad pack.indexversion=%"PRIu32),
2750                            pack_idx_opts.version);
2751                return 0;
2752        }
2753        return git_default_config(k, v, cb);
2754}
2755
2756static void read_object_list_from_stdin(void)
2757{
2758        char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2759        struct object_id oid;
2760        const char *p;
2761
2762        for (;;) {
2763                if (!fgets(line, sizeof(line), stdin)) {
2764                        if (feof(stdin))
2765                                break;
2766                        if (!ferror(stdin))
2767                                die("BUG: fgets returned NULL, not EOF, not error!");
2768                        if (errno != EINTR)
2769                                die_errno("fgets");
2770                        clearerr(stdin);
2771                        continue;
2772                }
2773                if (line[0] == '-') {
2774                        if (get_oid_hex(line+1, &oid))
2775                                die(_("expected edge object ID, got garbage:\n %s"),
2776                                    line);
2777                        add_preferred_base(&oid);
2778                        continue;
2779                }
2780                if (parse_oid_hex(line, &oid, &p))
2781                        die(_("expected object ID, got garbage:\n %s"), line);
2782
2783                add_preferred_base_object(p + 1);
2784                add_object_entry(&oid, OBJ_NONE, p + 1, 0);
2785        }
2786}
2787
2788/* Remember to update object flag allocation in object.h */
2789#define OBJECT_ADDED (1u<<20)
2790
2791static void show_commit(struct commit *commit, void *data)
2792{
2793        add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2794        commit->object.flags |= OBJECT_ADDED;
2795
2796        if (write_bitmap_index)
2797                index_commit_for_bitmap(commit);
2798
2799        if (use_delta_islands)
2800                propagate_island_marks(commit);
2801}
2802
2803static void show_object(struct object *obj, const char *name, void *data)
2804{
2805        add_preferred_base_object(name);
2806        add_object_entry(&obj->oid, obj->type, name, 0);
2807        obj->flags |= OBJECT_ADDED;
2808
2809        if (use_delta_islands) {
2810                const char *p;
2811                unsigned depth = 0;
2812                struct object_entry *ent;
2813
2814                for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))
2815                        depth++;
2816
2817                ent = packlist_find(&to_pack, obj->oid.hash, NULL);
2818                if (ent && depth > oe_tree_depth(&to_pack, ent))
2819                        oe_set_tree_depth(&to_pack, ent, depth);
2820        }
2821}
2822
2823static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2824{
2825        assert(arg_missing_action == MA_ALLOW_ANY);
2826
2827        /*
2828         * Quietly ignore ALL missing objects.  This avoids problems with
2829         * staging them now and getting an odd error later.
2830         */
2831        if (!has_object_file(&obj->oid))
2832                return;
2833
2834        show_object(obj, name, data);
2835}
2836
2837static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2838{
2839        assert(arg_missing_action == MA_ALLOW_PROMISOR);
2840
2841        /*
2842         * Quietly ignore EXPECTED missing objects.  This avoids problems with
2843         * staging them now and getting an odd error later.
2844         */
2845        if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2846                return;
2847
2848        show_object(obj, name, data);
2849}
2850
2851static int option_parse_missing_action(const struct option *opt,
2852                                       const char *arg, int unset)
2853{
2854        assert(arg);
2855        assert(!unset);
2856
2857        if (!strcmp(arg, "error")) {
2858                arg_missing_action = MA_ERROR;
2859                fn_show_object = show_object;
2860                return 0;
2861        }
2862
2863        if (!strcmp(arg, "allow-any")) {
2864                arg_missing_action = MA_ALLOW_ANY;
2865                fetch_if_missing = 0;
2866                fn_show_object = show_object__ma_allow_any;
2867                return 0;
2868        }
2869
2870        if (!strcmp(arg, "allow-promisor")) {
2871                arg_missing_action = MA_ALLOW_PROMISOR;
2872                fetch_if_missing = 0;
2873                fn_show_object = show_object__ma_allow_promisor;
2874                return 0;
2875        }
2876
2877        die(_("invalid value for --missing"));
2878        return 0;
2879}
2880
2881static void show_edge(struct commit *commit)
2882{
2883        add_preferred_base(&commit->object.oid);
2884}
2885
2886struct in_pack_object {
2887        off_t offset;
2888        struct object *object;
2889};
2890
2891struct in_pack {
2892        unsigned int alloc;
2893        unsigned int nr;
2894        struct in_pack_object *array;
2895};
2896
2897static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2898{
2899        in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2900        in_pack->array[in_pack->nr].object = object;
2901        in_pack->nr++;
2902}
2903
2904/*
2905 * Compare the objects in the offset order, in order to emulate the
2906 * "git rev-list --objects" output that produced the pack originally.
2907 */
2908static int ofscmp(const void *a_, const void *b_)
2909{
2910        struct in_pack_object *a = (struct in_pack_object *)a_;
2911        struct in_pack_object *b = (struct in_pack_object *)b_;
2912
2913        if (a->offset < b->offset)
2914                return -1;
2915        else if (a->offset > b->offset)
2916                return 1;
2917        else
2918                return oidcmp(&a->object->oid, &b->object->oid);
2919}
2920
2921static void add_objects_in_unpacked_packs(struct rev_info *revs)
2922{
2923        struct packed_git *p;
2924        struct in_pack in_pack;
2925        uint32_t i;
2926
2927        memset(&in_pack, 0, sizeof(in_pack));
2928
2929        for (p = get_all_packs(the_repository); p; p = p->next) {
2930                struct object_id oid;
2931                struct object *o;
2932
2933                if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
2934                        continue;
2935                if (open_pack_index(p))
2936                        die(_("cannot open pack index"));
2937
2938                ALLOC_GROW(in_pack.array,
2939                           in_pack.nr + p->num_objects,
2940                           in_pack.alloc);
2941
2942                for (i = 0; i < p->num_objects; i++) {
2943                        nth_packed_object_oid(&oid, p, i);
2944                        o = lookup_unknown_object(oid.hash);
2945                        if (!(o->flags & OBJECT_ADDED))
2946                                mark_in_pack_object(o, p, &in_pack);
2947                        o->flags |= OBJECT_ADDED;
2948                }
2949        }
2950
2951        if (in_pack.nr) {
2952                QSORT(in_pack.array, in_pack.nr, ofscmp);
2953                for (i = 0; i < in_pack.nr; i++) {
2954                        struct object *o = in_pack.array[i].object;
2955                        add_object_entry(&o->oid, o->type, "", 0);
2956                }
2957        }
2958        free(in_pack.array);
2959}
2960
2961static int add_loose_object(const struct object_id *oid, const char *path,
2962                            void *data)
2963{
2964        enum object_type type = oid_object_info(the_repository, oid, NULL);
2965
2966        if (type < 0) {
2967                warning(_("loose object at %s could not be examined"), path);
2968                return 0;
2969        }
2970
2971        add_object_entry(oid, type, "", 0);
2972        return 0;
2973}
2974
2975/*
2976 * We actually don't even have to worry about reachability here.
2977 * add_object_entry will weed out duplicates, so we just add every
2978 * loose object we find.
2979 */
2980static void add_unreachable_loose_objects(void)
2981{
2982        for_each_loose_file_in_objdir(get_object_directory(),
2983                                      add_loose_object,
2984                                      NULL, NULL, NULL);
2985}
2986
2987static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
2988{
2989        static struct packed_git *last_found = (void *)1;
2990        struct packed_git *p;
2991
2992        p = (last_found != (void *)1) ? last_found :
2993                                        get_all_packs(the_repository);
2994
2995        while (p) {
2996                if ((!p->pack_local || p->pack_keep ||
2997                                p->pack_keep_in_core) &&
2998                        find_pack_entry_one(oid->hash, p)) {
2999                        last_found = p;
3000                        return 1;
3001                }
3002                if (p == last_found)
3003                        p = get_all_packs(the_repository);
3004                else
3005                        p = p->next;
3006                if (p == last_found)
3007                        p = p->next;
3008        }
3009        return 0;
3010}
3011
3012/*
3013 * Store a list of sha1s that are should not be discarded
3014 * because they are either written too recently, or are
3015 * reachable from another object that was.
3016 *
3017 * This is filled by get_object_list.
3018 */
3019static struct oid_array recent_objects;
3020
3021static int loosened_object_can_be_discarded(const struct object_id *oid,
3022                                            timestamp_t mtime)
3023{
3024        if (!unpack_unreachable_expiration)
3025                return 0;
3026        if (mtime > unpack_unreachable_expiration)
3027                return 0;
3028        if (oid_array_lookup(&recent_objects, oid) >= 0)
3029                return 0;
3030        return 1;
3031}
3032
3033static void loosen_unused_packed_objects(struct rev_info *revs)
3034{
3035        struct packed_git *p;
3036        uint32_t i;
3037        struct object_id oid;
3038
3039        for (p = get_all_packs(the_repository); p; p = p->next) {
3040                if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
3041                        continue;
3042
3043                if (open_pack_index(p))
3044                        die(_("cannot open pack index"));
3045
3046                for (i = 0; i < p->num_objects; i++) {
3047                        nth_packed_object_oid(&oid, p, i);
3048                        if (!packlist_find(&to_pack, oid.hash, NULL) &&
3049                            !has_sha1_pack_kept_or_nonlocal(&oid) &&
3050                            !loosened_object_can_be_discarded(&oid, p->mtime))
3051                                if (force_object_loose(&oid, p->mtime))
3052                                        die(_("unable to force loose object"));
3053                }
3054        }
3055}
3056
3057/*
3058 * This tracks any options which pack-reuse code expects to be on, or which a
3059 * reader of the pack might not understand, and which would therefore prevent
3060 * blind reuse of what we have on disk.
3061 */
3062static int pack_options_allow_reuse(void)
3063{
3064        return pack_to_stdout &&
3065               allow_ofs_delta &&
3066               !ignore_packed_keep_on_disk &&
3067               !ignore_packed_keep_in_core &&
3068               (!local || !have_non_local_packs) &&
3069               !incremental;
3070}
3071
3072static int get_object_list_from_bitmap(struct rev_info *revs)
3073{
3074        if (!(bitmap_git = prepare_bitmap_walk(revs)))
3075                return -1;
3076
3077        if (pack_options_allow_reuse() &&
3078            !reuse_partial_packfile_from_bitmap(
3079                        bitmap_git,
3080                        &reuse_packfile,
3081                        &reuse_packfile_objects,
3082                        &reuse_packfile_offset)) {
3083                assert(reuse_packfile_objects);
3084                nr_result += reuse_packfile_objects;
3085                display_progress(progress_state, nr_result);
3086        }
3087
3088        traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);
3089        return 0;
3090}
3091
3092static void record_recent_object(struct object *obj,
3093                                 const char *name,
3094                                 void *data)
3095{
3096        oid_array_append(&recent_objects, &obj->oid);
3097}
3098
3099static void record_recent_commit(struct commit *commit, void *data)
3100{
3101        oid_array_append(&recent_objects, &commit->object.oid);
3102}
3103
3104static void get_object_list(int ac, const char **av)
3105{
3106        struct rev_info revs;
3107        char line[1000];
3108        int flags = 0;
3109
3110        init_revisions(&revs, NULL);
3111        save_commit_buffer = 0;
3112        setup_revisions(ac, av, &revs, NULL);
3113
3114        /* make sure shallows are read */
3115        is_repository_shallow(the_repository);
3116
3117        while (fgets(line, sizeof(line), stdin) != NULL) {
3118                int len = strlen(line);
3119                if (len && line[len - 1] == '\n')
3120                        line[--len] = 0;
3121                if (!len)
3122                        break;
3123                if (*line == '-') {
3124                        if (!strcmp(line, "--not")) {
3125                                flags ^= UNINTERESTING;
3126                                write_bitmap_index = 0;
3127                                continue;
3128                        }
3129                        if (starts_with(line, "--shallow ")) {
3130                                struct object_id oid;
3131                                if (get_oid_hex(line + 10, &oid))
3132                                        die("not an SHA-1 '%s'", line + 10);
3133                                register_shallow(the_repository, &oid);
3134                                use_bitmap_index = 0;
3135                                continue;
3136                        }
3137                        die(_("not a rev '%s'"), line);
3138                }
3139                if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
3140                        die(_("bad revision '%s'"), line);
3141        }
3142
3143        if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
3144                return;
3145
3146        if (use_delta_islands)
3147                load_delta_islands();
3148
3149        if (prepare_revision_walk(&revs))
3150                die(_("revision walk setup failed"));
3151        mark_edges_uninteresting(&revs, show_edge);
3152
3153        if (!fn_show_object)
3154                fn_show_object = show_object;
3155        traverse_commit_list_filtered(&filter_options, &revs,
3156                                      show_commit, fn_show_object, NULL,
3157                                      NULL);
3158
3159        if (unpack_unreachable_expiration) {
3160                revs.ignore_missing_links = 1;
3161                if (add_unseen_recent_objects_to_traversal(&revs,
3162                                unpack_unreachable_expiration))
3163                        die(_("unable to add recent objects"));
3164                if (prepare_revision_walk(&revs))
3165                        die(_("revision walk setup failed"));
3166                traverse_commit_list(&revs, record_recent_commit,
3167                                     record_recent_object, NULL);
3168        }
3169
3170        if (keep_unreachable)
3171                add_objects_in_unpacked_packs(&revs);
3172        if (pack_loose_unreachable)
3173                add_unreachable_loose_objects();
3174        if (unpack_unreachable)
3175                loosen_unused_packed_objects(&revs);
3176
3177        oid_array_clear(&recent_objects);
3178}
3179
3180static void add_extra_kept_packs(const struct string_list *names)
3181{
3182        struct packed_git *p;
3183
3184        if (!names->nr)
3185                return;
3186
3187        for (p = get_all_packs(the_repository); p; p = p->next) {
3188                const char *name = basename(p->pack_name);
3189                int i;
3190
3191                if (!p->pack_local)
3192                        continue;
3193
3194                for (i = 0; i < names->nr; i++)
3195                        if (!fspathcmp(name, names->items[i].string))
3196                                break;
3197
3198                if (i < names->nr) {
3199                        p->pack_keep_in_core = 1;
3200                        ignore_packed_keep_in_core = 1;
3201                        continue;
3202                }
3203        }
3204}
3205
3206static int option_parse_index_version(const struct option *opt,
3207                                      const char *arg, int unset)
3208{
3209        char *c;
3210        const char *val = arg;
3211        pack_idx_opts.version = strtoul(val, &c, 10);
3212        if (pack_idx_opts.version > 2)
3213                die(_("unsupported index version %s"), val);
3214        if (*c == ',' && c[1])
3215                pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
3216        if (*c || pack_idx_opts.off32_limit & 0x80000000)
3217                die(_("bad index version '%s'"), val);
3218        return 0;
3219}
3220
3221static int option_parse_unpack_unreachable(const struct option *opt,
3222                                           const char *arg, int unset)
3223{
3224        if (unset) {
3225                unpack_unreachable = 0;
3226                unpack_unreachable_expiration = 0;
3227        }
3228        else {
3229                unpack_unreachable = 1;
3230                if (arg)
3231                        unpack_unreachable_expiration = approxidate(arg);
3232        }
3233        return 0;
3234}
3235
3236int cmd_pack_objects(int argc, const char **argv, const char *prefix)
3237{
3238        int use_internal_rev_list = 0;
3239        int shallow = 0;
3240        int all_progress_implied = 0;
3241        struct argv_array rp = ARGV_ARRAY_INIT;
3242        int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
3243        int rev_list_index = 0;
3244        struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
3245        struct option pack_objects_options[] = {
3246                OPT_SET_INT('q', "quiet", &progress,
3247                            N_("do not show progress meter"), 0),
3248                OPT_SET_INT(0, "progress", &progress,
3249                            N_("show progress meter"), 1),
3250                OPT_SET_INT(0, "all-progress", &progress,
3251                            N_("show progress meter during object writing phase"), 2),
3252                OPT_BOOL(0, "all-progress-implied",
3253                         &all_progress_implied,
3254                         N_("similar to --all-progress when progress meter is shown")),
3255                { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),
3256                  N_("write the pack index file in the specified idx format version"),
3257                  0, option_parse_index_version },
3258                OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
3259                              N_("maximum size of each output pack file")),
3260                OPT_BOOL(0, "local", &local,
3261                         N_("ignore borrowed objects from alternate object store")),
3262                OPT_BOOL(0, "incremental", &incremental,
3263                         N_("ignore packed objects")),
3264                OPT_INTEGER(0, "window", &window,
3265                            N_("limit pack window by objects")),
3266                OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
3267                              N_("limit pack window by memory in addition to object limit")),
3268                OPT_INTEGER(0, "depth", &depth,
3269                            N_("maximum length of delta chain allowed in the resulting pack")),
3270                OPT_BOOL(0, "reuse-delta", &reuse_delta,
3271                         N_("reuse existing deltas")),
3272                OPT_BOOL(0, "reuse-object", &reuse_object,
3273                         N_("reuse existing objects")),
3274                OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
3275                         N_("use OFS_DELTA objects")),
3276                OPT_INTEGER(0, "threads", &delta_search_threads,
3277                            N_("use threads when searching for best delta matches")),
3278                OPT_BOOL(0, "non-empty", &non_empty,
3279                         N_("do not create an empty pack output")),
3280                OPT_BOOL(0, "revs", &use_internal_rev_list,
3281                         N_("read revision arguments from standard input")),
3282                OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,
3283                              N_("limit the objects to those that are not yet packed"),
3284                              1, PARSE_OPT_NONEG),
3285                OPT_SET_INT_F(0, "all", &rev_list_all,
3286                              N_("include objects reachable from any reference"),
3287                              1, PARSE_OPT_NONEG),
3288                OPT_SET_INT_F(0, "reflog", &rev_list_reflog,
3289                              N_("include objects referred by reflog entries"),
3290                              1, PARSE_OPT_NONEG),
3291                OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,
3292                              N_("include objects referred to by the index"),
3293                              1, PARSE_OPT_NONEG),
3294                OPT_BOOL(0, "stdout", &pack_to_stdout,
3295                         N_("output pack to stdout")),
3296                OPT_BOOL(0, "include-tag", &include_tag,
3297                         N_("include tag objects that refer to objects to be packed")),
3298                OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3299                         N_("keep unreachable objects")),
3300                OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3301                         N_("pack loose unreachable objects")),
3302                { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3303                  N_("unpack unreachable objects newer than <time>"),
3304                  PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3305                OPT_BOOL(0, "thin", &thin,
3306                         N_("create thin packs")),
3307                OPT_BOOL(0, "shallow", &shallow,
3308                         N_("create packs suitable for shallow fetches")),
3309                OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,
3310                         N_("ignore packs that have companion .keep file")),
3311                OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
3312                                N_("ignore this pack")),
3313                OPT_INTEGER(0, "compression", &pack_compression_level,
3314                            N_("pack compression level")),
3315                OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3316                            N_("do not hide commits by grafts"), 0),
3317                OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3318                         N_("use a bitmap index if available to speed up counting objects")),
3319                OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3320                         N_("write a bitmap index together with the pack index")),
3321                OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3322                { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3323                  N_("handling for missing objects"), PARSE_OPT_NONEG,
3324                  option_parse_missing_action },
3325                OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3326                         N_("do not pack objects in promisor packfiles")),
3327                OPT_BOOL(0, "delta-islands", &use_delta_islands,
3328                         N_("respect islands during delta compression")),
3329                OPT_END(),
3330        };
3331
3332        if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))
3333                BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3334
3335        read_replace_refs = 0;
3336
3337        reset_pack_idx_option(&pack_idx_opts);
3338        git_config(git_pack_config, NULL);
3339
3340        progress = isatty(2);
3341        argc = parse_options(argc, argv, prefix, pack_objects_options,
3342                             pack_usage, 0);
3343
3344        if (argc) {
3345                base_name = argv[0];
3346                argc--;
3347        }
3348        if (pack_to_stdout != !base_name || argc)
3349                usage_with_options(pack_usage, pack_objects_options);
3350
3351        if (depth >= (1 << OE_DEPTH_BITS)) {
3352                warning(_("delta chain depth %d is too deep, forcing %d"),
3353                        depth, (1 << OE_DEPTH_BITS) - 1);
3354                depth = (1 << OE_DEPTH_BITS) - 1;
3355        }
3356        if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {
3357                warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3358                        (1U << OE_Z_DELTA_BITS) - 1);
3359                cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;
3360        }
3361
3362        argv_array_push(&rp, "pack-objects");
3363        if (thin) {
3364                use_internal_rev_list = 1;
3365                argv_array_push(&rp, shallow
3366                                ? "--objects-edge-aggressive"
3367                                : "--objects-edge");
3368        } else
3369                argv_array_push(&rp, "--objects");
3370
3371        if (rev_list_all) {
3372                use_internal_rev_list = 1;
3373                argv_array_push(&rp, "--all");
3374        }
3375        if (rev_list_reflog) {
3376                use_internal_rev_list = 1;
3377                argv_array_push(&rp, "--reflog");
3378        }
3379        if (rev_list_index) {
3380                use_internal_rev_list = 1;
3381                argv_array_push(&rp, "--indexed-objects");
3382        }
3383        if (rev_list_unpacked) {
3384                use_internal_rev_list = 1;
3385                argv_array_push(&rp, "--unpacked");
3386        }
3387
3388        if (exclude_promisor_objects) {
3389                use_internal_rev_list = 1;
3390                fetch_if_missing = 0;
3391                argv_array_push(&rp, "--exclude-promisor-objects");
3392        }
3393        if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)
3394                use_internal_rev_list = 1;
3395
3396        if (!reuse_object)
3397                reuse_delta = 0;
3398        if (pack_compression_level == -1)
3399                pack_compression_level = Z_DEFAULT_COMPRESSION;
3400        else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3401                die(_("bad pack compression level %d"), pack_compression_level);
3402
3403        if (!delta_search_threads)      /* --threads=0 means autodetect */
3404                delta_search_threads = online_cpus();
3405
3406#ifdef NO_PTHREADS
3407        if (delta_search_threads != 1)
3408                warning(_("no threads support, ignoring --threads"));
3409#endif
3410        if (!pack_to_stdout && !pack_size_limit)
3411                pack_size_limit = pack_size_limit_cfg;
3412        if (pack_to_stdout && pack_size_limit)
3413                die(_("--max-pack-size cannot be used to build a pack for transfer"));
3414        if (pack_size_limit && pack_size_limit < 1024*1024) {
3415                warning(_("minimum pack size limit is 1 MiB"));
3416                pack_size_limit = 1024*1024;
3417        }
3418
3419        if (!pack_to_stdout && thin)
3420                die(_("--thin cannot be used to build an indexable pack"));
3421
3422        if (keep_unreachable && unpack_unreachable)
3423                die(_("--keep-unreachable and --unpack-unreachable are incompatible"));
3424        if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3425                unpack_unreachable_expiration = 0;
3426
3427        if (filter_options.choice) {
3428                if (!pack_to_stdout)
3429                        die(_("cannot use --filter without --stdout"));
3430                use_bitmap_index = 0;
3431        }
3432
3433        /*
3434         * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3435         *
3436         * - to produce good pack (with bitmap index not-yet-packed objects are
3437         *   packed in suboptimal order).
3438         *
3439         * - to use more robust pack-generation codepath (avoiding possible
3440         *   bugs in bitmap code and possible bitmap index corruption).
3441         */
3442        if (!pack_to_stdout)
3443                use_bitmap_index_default = 0;
3444
3445        if (use_bitmap_index < 0)
3446                use_bitmap_index = use_bitmap_index_default;
3447
3448        /* "hard" reasons not to use bitmaps; these just won't work at all */
3449        if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))
3450                use_bitmap_index = 0;
3451
3452        if (pack_to_stdout || !rev_list_all)
3453                write_bitmap_index = 0;
3454
3455        if (use_delta_islands)
3456                argv_array_push(&rp, "--topo-order");
3457
3458        if (progress && all_progress_implied)
3459                progress = 2;
3460
3461        add_extra_kept_packs(&keep_pack_list);
3462        if (ignore_packed_keep_on_disk) {
3463                struct packed_git *p;
3464                for (p = get_all_packs(the_repository); p; p = p->next)
3465                        if (p->pack_local && p->pack_keep)
3466                                break;
3467                if (!p) /* no keep-able packs found */
3468                        ignore_packed_keep_on_disk = 0;
3469        }
3470        if (local) {
3471                /*
3472                 * unlike ignore_packed_keep_on_disk above, we do not
3473                 * want to unset "local" based on looking at packs, as
3474                 * it also covers non-local objects
3475                 */
3476                struct packed_git *p;
3477                for (p = get_all_packs(the_repository); p; p = p->next) {
3478                        if (!p->pack_local) {
3479                                have_non_local_packs = 1;
3480                                break;
3481                        }
3482                }
3483        }
3484
3485        prepare_packing_data(&to_pack);
3486
3487        if (progress)
3488                progress_state = start_progress(_("Enumerating objects"), 0);
3489        if (!use_internal_rev_list)
3490                read_object_list_from_stdin();
3491        else {
3492                get_object_list(rp.argc, rp.argv);
3493                argv_array_clear(&rp);
3494        }
3495        cleanup_preferred_base();
3496        if (include_tag && nr_result)
3497                for_each_ref(add_ref_tag, NULL);
3498        stop_progress(&progress_state);
3499
3500        if (non_empty && !nr_result)
3501                return 0;
3502        if (nr_result)
3503                prepare_pack(window, depth);
3504        write_pack_file();
3505        if (progress)
3506                fprintf_ln(stderr,
3507                           _("Total %"PRIu32" (delta %"PRIu32"),"
3508                             " reused %"PRIu32" (delta %"PRIu32")"),
3509                           written, written_delta, reused, reused_delta);
3510        return 0;
3511}