builtin / pack-objects.con commit Merge branch 'sb/blame-color' (3d24129)
   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 "reachable.h"
  28#include "sha1-array.h"
  29#include "argv-array.h"
  30#include "list.h"
  31#include "packfile.h"
  32#include "object-store.h"
  33#include "dir.h"
  34
  35#define IN_PACK(obj) oe_in_pack(&to_pack, obj)
  36#define SIZE(obj) oe_size(&to_pack, obj)
  37#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size)
  38#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj)
  39#define DELTA(obj) oe_delta(&to_pack, obj)
  40#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj)
  41#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj)
  42#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val)
  43#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val)
  44#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val)
  45#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val)
  46
  47static const char *pack_usage[] = {
  48        N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"),
  49        N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"),
  50        NULL
  51};
  52
  53/*
  54 * Objects we are going to pack are collected in the `to_pack` structure.
  55 * It contains an array (dynamically expanded) of the object data, and a map
  56 * that can resolve SHA1s to their position in the array.
  57 */
  58static struct packing_data to_pack;
  59
  60static struct pack_idx_entry **written_list;
  61static uint32_t nr_result, nr_written, nr_seen;
  62
  63static int non_empty;
  64static int reuse_delta = 1, reuse_object = 1;
  65static int keep_unreachable, unpack_unreachable, include_tag;
  66static timestamp_t unpack_unreachable_expiration;
  67static int pack_loose_unreachable;
  68static int local;
  69static int have_non_local_packs;
  70static int incremental;
  71static int ignore_packed_keep_on_disk;
  72static int ignore_packed_keep_in_core;
  73static int allow_ofs_delta;
  74static struct pack_idx_option pack_idx_opts;
  75static const char *base_name;
  76static int progress = 1;
  77static int window = 10;
  78static unsigned long pack_size_limit;
  79static int depth = 50;
  80static int delta_search_threads;
  81static int pack_to_stdout;
  82static int num_preferred_base;
  83static struct progress *progress_state;
  84
  85static struct packed_git *reuse_packfile;
  86static uint32_t reuse_packfile_objects;
  87static off_t reuse_packfile_offset;
  88
  89static int use_bitmap_index_default = 1;
  90static int use_bitmap_index = -1;
  91static int write_bitmap_index;
  92static uint16_t write_bitmap_options;
  93
  94static int exclude_promisor_objects;
  95
  96static unsigned long delta_cache_size = 0;
  97static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE;
  98static unsigned long cache_max_small_delta_size = 1000;
  99
 100static unsigned long window_memory_limit = 0;
 101
 102static struct list_objects_filter_options filter_options;
 103
 104enum missing_action {
 105        MA_ERROR = 0,      /* fail if any missing objects are encountered */
 106        MA_ALLOW_ANY,      /* silently allow ALL missing objects */
 107        MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */
 108};
 109static enum missing_action arg_missing_action;
 110static show_object_fn fn_show_object;
 111
 112/*
 113 * stats
 114 */
 115static uint32_t written, written_delta;
 116static uint32_t reused, reused_delta;
 117
 118/*
 119 * Indexed commits
 120 */
 121static struct commit **indexed_commits;
 122static unsigned int indexed_commits_nr;
 123static unsigned int indexed_commits_alloc;
 124
 125static void index_commit_for_bitmap(struct commit *commit)
 126{
 127        if (indexed_commits_nr >= indexed_commits_alloc) {
 128                indexed_commits_alloc = (indexed_commits_alloc + 32) * 2;
 129                REALLOC_ARRAY(indexed_commits, indexed_commits_alloc);
 130        }
 131
 132        indexed_commits[indexed_commits_nr++] = commit;
 133}
 134
 135static void *get_delta(struct object_entry *entry)
 136{
 137        unsigned long size, base_size, delta_size;
 138        void *buf, *base_buf, *delta_buf;
 139        enum object_type type;
 140
 141        buf = read_object_file(&entry->idx.oid, &type, &size);
 142        if (!buf)
 143                die("unable to read %s", oid_to_hex(&entry->idx.oid));
 144        base_buf = read_object_file(&DELTA(entry)->idx.oid, &type,
 145                                    &base_size);
 146        if (!base_buf)
 147                die("unable to read %s",
 148                    oid_to_hex(&DELTA(entry)->idx.oid));
 149        delta_buf = diff_delta(base_buf, base_size,
 150                               buf, size, &delta_size, 0);
 151        if (!delta_buf || delta_size != DELTA_SIZE(entry))
 152                die("delta size changed");
 153        free(buf);
 154        free(base_buf);
 155        return delta_buf;
 156}
 157
 158static unsigned long do_compress(void **pptr, unsigned long size)
 159{
 160        git_zstream stream;
 161        void *in, *out;
 162        unsigned long maxsize;
 163
 164        git_deflate_init(&stream, pack_compression_level);
 165        maxsize = git_deflate_bound(&stream, size);
 166
 167        in = *pptr;
 168        out = xmalloc(maxsize);
 169        *pptr = out;
 170
 171        stream.next_in = in;
 172        stream.avail_in = size;
 173        stream.next_out = out;
 174        stream.avail_out = maxsize;
 175        while (git_deflate(&stream, Z_FINISH) == Z_OK)
 176                ; /* nothing */
 177        git_deflate_end(&stream);
 178
 179        free(in);
 180        return stream.total_out;
 181}
 182
 183static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
 184                                           const struct object_id *oid)
 185{
 186        git_zstream stream;
 187        unsigned char ibuf[1024 * 16];
 188        unsigned char obuf[1024 * 16];
 189        unsigned long olen = 0;
 190
 191        git_deflate_init(&stream, pack_compression_level);
 192
 193        for (;;) {
 194                ssize_t readlen;
 195                int zret = Z_OK;
 196                readlen = read_istream(st, ibuf, sizeof(ibuf));
 197                if (readlen == -1)
 198                        die(_("unable to read %s"), oid_to_hex(oid));
 199
 200                stream.next_in = ibuf;
 201                stream.avail_in = readlen;
 202                while ((stream.avail_in || readlen == 0) &&
 203                       (zret == Z_OK || zret == Z_BUF_ERROR)) {
 204                        stream.next_out = obuf;
 205                        stream.avail_out = sizeof(obuf);
 206                        zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
 207                        hashwrite(f, obuf, stream.next_out - obuf);
 208                        olen += stream.next_out - obuf;
 209                }
 210                if (stream.avail_in)
 211                        die(_("deflate error (%d)"), zret);
 212                if (readlen == 0) {
 213                        if (zret != Z_STREAM_END)
 214                                die(_("deflate error (%d)"), zret);
 215                        break;
 216                }
 217        }
 218        git_deflate_end(&stream);
 219        return olen;
 220}
 221
 222/*
 223 * we are going to reuse the existing object data as is.  make
 224 * sure it is not corrupt.
 225 */
 226static int check_pack_inflate(struct packed_git *p,
 227                struct pack_window **w_curs,
 228                off_t offset,
 229                off_t len,
 230                unsigned long expect)
 231{
 232        git_zstream stream;
 233        unsigned char fakebuf[4096], *in;
 234        int st;
 235
 236        memset(&stream, 0, sizeof(stream));
 237        git_inflate_init(&stream);
 238        do {
 239                in = use_pack(p, w_curs, offset, &stream.avail_in);
 240                stream.next_in = in;
 241                stream.next_out = fakebuf;
 242                stream.avail_out = sizeof(fakebuf);
 243                st = git_inflate(&stream, Z_FINISH);
 244                offset += stream.next_in - in;
 245        } while (st == Z_OK || st == Z_BUF_ERROR);
 246        git_inflate_end(&stream);
 247        return (st == Z_STREAM_END &&
 248                stream.total_out == expect &&
 249                stream.total_in == len) ? 0 : -1;
 250}
 251
 252static void copy_pack_data(struct hashfile *f,
 253                struct packed_git *p,
 254                struct pack_window **w_curs,
 255                off_t offset,
 256                off_t len)
 257{
 258        unsigned char *in;
 259        unsigned long avail;
 260
 261        while (len) {
 262                in = use_pack(p, w_curs, offset, &avail);
 263                if (avail > len)
 264                        avail = (unsigned long)len;
 265                hashwrite(f, in, avail);
 266                offset += avail;
 267                len -= avail;
 268        }
 269}
 270
 271/* Return 0 if we will bust the pack-size limit */
 272static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
 273                                           unsigned long limit, int usable_delta)
 274{
 275        unsigned long size, datalen;
 276        unsigned char header[MAX_PACK_OBJECT_HEADER],
 277                      dheader[MAX_PACK_OBJECT_HEADER];
 278        unsigned hdrlen;
 279        enum object_type type;
 280        void *buf;
 281        struct git_istream *st = NULL;
 282
 283        if (!usable_delta) {
 284                if (oe_type(entry) == OBJ_BLOB &&
 285                    oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
 286                    (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL)
 287                        buf = NULL;
 288                else {
 289                        buf = read_object_file(&entry->idx.oid, &type, &size);
 290                        if (!buf)
 291                                die(_("unable to read %s"),
 292                                    oid_to_hex(&entry->idx.oid));
 293                }
 294                /*
 295                 * make sure no cached delta data remains from a
 296                 * previous attempt before a pack split occurred.
 297                 */
 298                FREE_AND_NULL(entry->delta_data);
 299                entry->z_delta_size = 0;
 300        } else if (entry->delta_data) {
 301                size = DELTA_SIZE(entry);
 302                buf = entry->delta_data;
 303                entry->delta_data = NULL;
 304                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 305                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 306        } else {
 307                buf = get_delta(entry);
 308                size = DELTA_SIZE(entry);
 309                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 310                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 311        }
 312
 313        if (st) /* large blob case, just assume we don't compress well */
 314                datalen = size;
 315        else if (entry->z_delta_size)
 316                datalen = entry->z_delta_size;
 317        else
 318                datalen = do_compress(&buf, size);
 319
 320        /*
 321         * The object header is a byte of 'type' followed by zero or
 322         * more bytes of length.
 323         */
 324        hdrlen = encode_in_pack_object_header(header, sizeof(header),
 325                                              type, size);
 326
 327        if (type == OBJ_OFS_DELTA) {
 328                /*
 329                 * Deltas with relative base contain an additional
 330                 * encoding of the relative offset for the delta
 331                 * base from this object's position in the pack.
 332                 */
 333                off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
 334                unsigned pos = sizeof(dheader) - 1;
 335                dheader[pos] = ofs & 127;
 336                while (ofs >>= 7)
 337                        dheader[--pos] = 128 | (--ofs & 127);
 338                if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 339                        if (st)
 340                                close_istream(st);
 341                        free(buf);
 342                        return 0;
 343                }
 344                hashwrite(f, header, hdrlen);
 345                hashwrite(f, dheader + pos, sizeof(dheader) - pos);
 346                hdrlen += sizeof(dheader) - pos;
 347        } else if (type == OBJ_REF_DELTA) {
 348                /*
 349                 * Deltas with a base reference contain
 350                 * an additional 20 bytes for the base sha1.
 351                 */
 352                if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 353                        if (st)
 354                                close_istream(st);
 355                        free(buf);
 356                        return 0;
 357                }
 358                hashwrite(f, header, hdrlen);
 359                hashwrite(f, DELTA(entry)->idx.oid.hash, 20);
 360                hdrlen += 20;
 361        } else {
 362                if (limit && hdrlen + datalen + 20 >= limit) {
 363                        if (st)
 364                                close_istream(st);
 365                        free(buf);
 366                        return 0;
 367                }
 368                hashwrite(f, header, hdrlen);
 369        }
 370        if (st) {
 371                datalen = write_large_blob_data(st, f, &entry->idx.oid);
 372                close_istream(st);
 373        } else {
 374                hashwrite(f, buf, datalen);
 375                free(buf);
 376        }
 377
 378        return hdrlen + datalen;
 379}
 380
 381/* Return 0 if we will bust the pack-size limit */
 382static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
 383                                unsigned long limit, int usable_delta)
 384{
 385        struct packed_git *p = IN_PACK(entry);
 386        struct pack_window *w_curs = NULL;
 387        struct revindex_entry *revidx;
 388        off_t offset;
 389        enum object_type type = oe_type(entry);
 390        off_t datalen;
 391        unsigned char header[MAX_PACK_OBJECT_HEADER],
 392                      dheader[MAX_PACK_OBJECT_HEADER];
 393        unsigned hdrlen;
 394        unsigned long entry_size = SIZE(entry);
 395
 396        if (DELTA(entry))
 397                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 398                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 399        hdrlen = encode_in_pack_object_header(header, sizeof(header),
 400                                              type, entry_size);
 401
 402        offset = entry->in_pack_offset;
 403        revidx = find_pack_revindex(p, offset);
 404        datalen = revidx[1].offset - offset;
 405        if (!pack_to_stdout && p->index_version > 1 &&
 406            check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
 407                error("bad packed object CRC for %s",
 408                      oid_to_hex(&entry->idx.oid));
 409                unuse_pack(&w_curs);
 410                return write_no_reuse_object(f, entry, limit, usable_delta);
 411        }
 412
 413        offset += entry->in_pack_header_size;
 414        datalen -= entry->in_pack_header_size;
 415
 416        if (!pack_to_stdout && p->index_version == 1 &&
 417            check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
 418                error("corrupt packed object for %s",
 419                      oid_to_hex(&entry->idx.oid));
 420                unuse_pack(&w_curs);
 421                return write_no_reuse_object(f, entry, limit, usable_delta);
 422        }
 423
 424        if (type == OBJ_OFS_DELTA) {
 425                off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
 426                unsigned pos = sizeof(dheader) - 1;
 427                dheader[pos] = ofs & 127;
 428                while (ofs >>= 7)
 429                        dheader[--pos] = 128 | (--ofs & 127);
 430                if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 431                        unuse_pack(&w_curs);
 432                        return 0;
 433                }
 434                hashwrite(f, header, hdrlen);
 435                hashwrite(f, dheader + pos, sizeof(dheader) - pos);
 436                hdrlen += sizeof(dheader) - pos;
 437                reused_delta++;
 438        } else if (type == OBJ_REF_DELTA) {
 439                if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 440                        unuse_pack(&w_curs);
 441                        return 0;
 442                }
 443                hashwrite(f, header, hdrlen);
 444                hashwrite(f, DELTA(entry)->idx.oid.hash, 20);
 445                hdrlen += 20;
 446                reused_delta++;
 447        } else {
 448                if (limit && hdrlen + datalen + 20 >= limit) {
 449                        unuse_pack(&w_curs);
 450                        return 0;
 451                }
 452                hashwrite(f, header, hdrlen);
 453        }
 454        copy_pack_data(f, p, &w_curs, offset, datalen);
 455        unuse_pack(&w_curs);
 456        reused++;
 457        return hdrlen + datalen;
 458}
 459
 460/* Return 0 if we will bust the pack-size limit */
 461static off_t write_object(struct hashfile *f,
 462                          struct object_entry *entry,
 463                          off_t write_offset)
 464{
 465        unsigned long limit;
 466        off_t len;
 467        int usable_delta, to_reuse;
 468
 469        if (!pack_to_stdout)
 470                crc32_begin(f);
 471
 472        /* apply size limit if limited packsize and not first object */
 473        if (!pack_size_limit || !nr_written)
 474                limit = 0;
 475        else if (pack_size_limit <= write_offset)
 476                /*
 477                 * the earlier object did not fit the limit; avoid
 478                 * mistaking this with unlimited (i.e. limit = 0).
 479                 */
 480                limit = 1;
 481        else
 482                limit = pack_size_limit - write_offset;
 483
 484        if (!DELTA(entry))
 485                usable_delta = 0;       /* no delta */
 486        else if (!pack_size_limit)
 487               usable_delta = 1;        /* unlimited packfile */
 488        else if (DELTA(entry)->idx.offset == (off_t)-1)
 489                usable_delta = 0;       /* base was written to another pack */
 490        else if (DELTA(entry)->idx.offset)
 491                usable_delta = 1;       /* base already exists in this pack */
 492        else
 493                usable_delta = 0;       /* base could end up in another pack */
 494
 495        if (!reuse_object)
 496                to_reuse = 0;   /* explicit */
 497        else if (!IN_PACK(entry))
 498                to_reuse = 0;   /* can't reuse what we don't have */
 499        else if (oe_type(entry) == OBJ_REF_DELTA ||
 500                 oe_type(entry) == OBJ_OFS_DELTA)
 501                                /* check_object() decided it for us ... */
 502                to_reuse = usable_delta;
 503                                /* ... but pack split may override that */
 504        else if (oe_type(entry) != entry->in_pack_type)
 505                to_reuse = 0;   /* pack has delta which is unusable */
 506        else if (DELTA(entry))
 507                to_reuse = 0;   /* we want to pack afresh */
 508        else
 509                to_reuse = 1;   /* we have it in-pack undeltified,
 510                                 * and we do not need to deltify it.
 511                                 */
 512
 513        if (!to_reuse)
 514                len = write_no_reuse_object(f, entry, limit, usable_delta);
 515        else
 516                len = write_reuse_object(f, entry, limit, usable_delta);
 517        if (!len)
 518                return 0;
 519
 520        if (usable_delta)
 521                written_delta++;
 522        written++;
 523        if (!pack_to_stdout)
 524                entry->idx.crc32 = crc32_end(f);
 525        return len;
 526}
 527
 528enum write_one_status {
 529        WRITE_ONE_SKIP = -1, /* already written */
 530        WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
 531        WRITE_ONE_WRITTEN = 1, /* normal */
 532        WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
 533};
 534
 535static enum write_one_status write_one(struct hashfile *f,
 536                                       struct object_entry *e,
 537                                       off_t *offset)
 538{
 539        off_t size;
 540        int recursing;
 541
 542        /*
 543         * we set offset to 1 (which is an impossible value) to mark
 544         * the fact that this object is involved in "write its base
 545         * first before writing a deltified object" recursion.
 546         */
 547        recursing = (e->idx.offset == 1);
 548        if (recursing) {
 549                warning("recursive delta detected for object %s",
 550                        oid_to_hex(&e->idx.oid));
 551                return WRITE_ONE_RECURSIVE;
 552        } else if (e->idx.offset || e->preferred_base) {
 553                /* offset is non zero if object is written already. */
 554                return WRITE_ONE_SKIP;
 555        }
 556
 557        /* if we are deltified, write out base object first. */
 558        if (DELTA(e)) {
 559                e->idx.offset = 1; /* now recurse */
 560                switch (write_one(f, DELTA(e), offset)) {
 561                case WRITE_ONE_RECURSIVE:
 562                        /* we cannot depend on this one */
 563                        SET_DELTA(e, NULL);
 564                        break;
 565                default:
 566                        break;
 567                case WRITE_ONE_BREAK:
 568                        e->idx.offset = recursing;
 569                        return WRITE_ONE_BREAK;
 570                }
 571        }
 572
 573        e->idx.offset = *offset;
 574        size = write_object(f, e, *offset);
 575        if (!size) {
 576                e->idx.offset = recursing;
 577                return WRITE_ONE_BREAK;
 578        }
 579        written_list[nr_written++] = &e->idx;
 580
 581        /* make sure off_t is sufficiently large not to wrap */
 582        if (signed_add_overflows(*offset, size))
 583                die("pack too large for current definition of off_t");
 584        *offset += size;
 585        return WRITE_ONE_WRITTEN;
 586}
 587
 588static int mark_tagged(const char *path, const struct object_id *oid, int flag,
 589                       void *cb_data)
 590{
 591        struct object_id peeled;
 592        struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
 593
 594        if (entry)
 595                entry->tagged = 1;
 596        if (!peel_ref(path, &peeled)) {
 597                entry = packlist_find(&to_pack, peeled.hash, NULL);
 598                if (entry)
 599                        entry->tagged = 1;
 600        }
 601        return 0;
 602}
 603
 604static inline void add_to_write_order(struct object_entry **wo,
 605                               unsigned int *endp,
 606                               struct object_entry *e)
 607{
 608        if (e->filled)
 609                return;
 610        wo[(*endp)++] = e;
 611        e->filled = 1;
 612}
 613
 614static void add_descendants_to_write_order(struct object_entry **wo,
 615                                           unsigned int *endp,
 616                                           struct object_entry *e)
 617{
 618        int add_to_order = 1;
 619        while (e) {
 620                if (add_to_order) {
 621                        struct object_entry *s;
 622                        /* add this node... */
 623                        add_to_write_order(wo, endp, e);
 624                        /* all its siblings... */
 625                        for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
 626                                add_to_write_order(wo, endp, s);
 627                        }
 628                }
 629                /* drop down a level to add left subtree nodes if possible */
 630                if (DELTA_CHILD(e)) {
 631                        add_to_order = 1;
 632                        e = DELTA_CHILD(e);
 633                } else {
 634                        add_to_order = 0;
 635                        /* our sibling might have some children, it is next */
 636                        if (DELTA_SIBLING(e)) {
 637                                e = DELTA_SIBLING(e);
 638                                continue;
 639                        }
 640                        /* go back to our parent node */
 641                        e = DELTA(e);
 642                        while (e && !DELTA_SIBLING(e)) {
 643                                /* we're on the right side of a subtree, keep
 644                                 * going up until we can go right again */
 645                                e = DELTA(e);
 646                        }
 647                        if (!e) {
 648                                /* done- we hit our original root node */
 649                                return;
 650                        }
 651                        /* pass it off to sibling at this level */
 652                        e = DELTA_SIBLING(e);
 653                }
 654        };
 655}
 656
 657static void add_family_to_write_order(struct object_entry **wo,
 658                                      unsigned int *endp,
 659                                      struct object_entry *e)
 660{
 661        struct object_entry *root;
 662
 663        for (root = e; DELTA(root); root = DELTA(root))
 664                ; /* nothing */
 665        add_descendants_to_write_order(wo, endp, root);
 666}
 667
 668static struct object_entry **compute_write_order(void)
 669{
 670        unsigned int i, wo_end, last_untagged;
 671
 672        struct object_entry **wo;
 673        struct object_entry *objects = to_pack.objects;
 674
 675        for (i = 0; i < to_pack.nr_objects; i++) {
 676                objects[i].tagged = 0;
 677                objects[i].filled = 0;
 678                SET_DELTA_CHILD(&objects[i], NULL);
 679                SET_DELTA_SIBLING(&objects[i], NULL);
 680        }
 681
 682        /*
 683         * Fully connect delta_child/delta_sibling network.
 684         * Make sure delta_sibling is sorted in the original
 685         * recency order.
 686         */
 687        for (i = to_pack.nr_objects; i > 0;) {
 688                struct object_entry *e = &objects[--i];
 689                if (!DELTA(e))
 690                        continue;
 691                /* Mark me as the first child */
 692                e->delta_sibling_idx = DELTA(e)->delta_child_idx;
 693                SET_DELTA_CHILD(DELTA(e), e);
 694        }
 695
 696        /*
 697         * Mark objects that are at the tip of tags.
 698         */
 699        for_each_tag_ref(mark_tagged, NULL);
 700
 701        /*
 702         * Give the objects in the original recency order until
 703         * we see a tagged tip.
 704         */
 705        ALLOC_ARRAY(wo, to_pack.nr_objects);
 706        for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
 707                if (objects[i].tagged)
 708                        break;
 709                add_to_write_order(wo, &wo_end, &objects[i]);
 710        }
 711        last_untagged = i;
 712
 713        /*
 714         * Then fill all the tagged tips.
 715         */
 716        for (; i < to_pack.nr_objects; i++) {
 717                if (objects[i].tagged)
 718                        add_to_write_order(wo, &wo_end, &objects[i]);
 719        }
 720
 721        /*
 722         * And then all remaining commits and tags.
 723         */
 724        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 725                if (oe_type(&objects[i]) != OBJ_COMMIT &&
 726                    oe_type(&objects[i]) != OBJ_TAG)
 727                        continue;
 728                add_to_write_order(wo, &wo_end, &objects[i]);
 729        }
 730
 731        /*
 732         * And then all the trees.
 733         */
 734        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 735                if (oe_type(&objects[i]) != OBJ_TREE)
 736                        continue;
 737                add_to_write_order(wo, &wo_end, &objects[i]);
 738        }
 739
 740        /*
 741         * Finally all the rest in really tight order
 742         */
 743        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 744                if (!objects[i].filled)
 745                        add_family_to_write_order(wo, &wo_end, &objects[i]);
 746        }
 747
 748        if (wo_end != to_pack.nr_objects)
 749                die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
 750
 751        return wo;
 752}
 753
 754static off_t write_reused_pack(struct hashfile *f)
 755{
 756        unsigned char buffer[8192];
 757        off_t to_write, total;
 758        int fd;
 759
 760        if (!is_pack_valid(reuse_packfile))
 761                die("packfile is invalid: %s", reuse_packfile->pack_name);
 762
 763        fd = git_open(reuse_packfile->pack_name);
 764        if (fd < 0)
 765                die_errno("unable to open packfile for reuse: %s",
 766                          reuse_packfile->pack_name);
 767
 768        if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
 769                die_errno("unable to seek in reused packfile");
 770
 771        if (reuse_packfile_offset < 0)
 772                reuse_packfile_offset = reuse_packfile->pack_size - 20;
 773
 774        total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
 775
 776        while (to_write) {
 777                int read_pack = xread(fd, buffer, sizeof(buffer));
 778
 779                if (read_pack <= 0)
 780                        die_errno("unable to read from reused packfile");
 781
 782                if (read_pack > to_write)
 783                        read_pack = to_write;
 784
 785                hashwrite(f, buffer, read_pack);
 786                to_write -= read_pack;
 787
 788                /*
 789                 * We don't know the actual number of objects written,
 790                 * only how many bytes written, how many bytes total, and
 791                 * how many objects total. So we can fake it by pretending all
 792                 * objects we are writing are the same size. This gives us a
 793                 * smooth progress meter, and at the end it matches the true
 794                 * answer.
 795                 */
 796                written = reuse_packfile_objects *
 797                                (((double)(total - to_write)) / total);
 798                display_progress(progress_state, written);
 799        }
 800
 801        close(fd);
 802        written = reuse_packfile_objects;
 803        display_progress(progress_state, written);
 804        return reuse_packfile_offset - sizeof(struct pack_header);
 805}
 806
 807static const char no_split_warning[] = N_(
 808"disabling bitmap writing, packs are split due to pack.packSizeLimit"
 809);
 810
 811static void write_pack_file(void)
 812{
 813        uint32_t i = 0, j;
 814        struct hashfile *f;
 815        off_t offset;
 816        uint32_t nr_remaining = nr_result;
 817        time_t last_mtime = 0;
 818        struct object_entry **write_order;
 819
 820        if (progress > pack_to_stdout)
 821                progress_state = start_progress(_("Writing objects"), nr_result);
 822        ALLOC_ARRAY(written_list, to_pack.nr_objects);
 823        write_order = compute_write_order();
 824
 825        do {
 826                struct object_id oid;
 827                char *pack_tmp_name = NULL;
 828
 829                if (pack_to_stdout)
 830                        f = hashfd_throughput(1, "<stdout>", progress_state);
 831                else
 832                        f = create_tmp_packfile(&pack_tmp_name);
 833
 834                offset = write_pack_header(f, nr_remaining);
 835
 836                if (reuse_packfile) {
 837                        off_t packfile_size;
 838                        assert(pack_to_stdout);
 839
 840                        packfile_size = write_reused_pack(f);
 841                        offset += packfile_size;
 842                }
 843
 844                nr_written = 0;
 845                for (; i < to_pack.nr_objects; i++) {
 846                        struct object_entry *e = write_order[i];
 847                        if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
 848                                break;
 849                        display_progress(progress_state, written);
 850                }
 851
 852                /*
 853                 * Did we write the wrong # entries in the header?
 854                 * If so, rewrite it like in fast-import
 855                 */
 856                if (pack_to_stdout) {
 857                        finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
 858                } else if (nr_written == nr_remaining) {
 859                        finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
 860                } else {
 861                        int fd = finalize_hashfile(f, oid.hash, 0);
 862                        fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
 863                                                 nr_written, oid.hash, offset);
 864                        close(fd);
 865                        if (write_bitmap_index) {
 866                                warning(_(no_split_warning));
 867                                write_bitmap_index = 0;
 868                        }
 869                }
 870
 871                if (!pack_to_stdout) {
 872                        struct stat st;
 873                        struct strbuf tmpname = STRBUF_INIT;
 874
 875                        /*
 876                         * Packs are runtime accessed in their mtime
 877                         * order since newer packs are more likely to contain
 878                         * younger objects.  So if we are creating multiple
 879                         * packs then we should modify the mtime of later ones
 880                         * to preserve this property.
 881                         */
 882                        if (stat(pack_tmp_name, &st) < 0) {
 883                                warning_errno("failed to stat %s", pack_tmp_name);
 884                        } else if (!last_mtime) {
 885                                last_mtime = st.st_mtime;
 886                        } else {
 887                                struct utimbuf utb;
 888                                utb.actime = st.st_atime;
 889                                utb.modtime = --last_mtime;
 890                                if (utime(pack_tmp_name, &utb) < 0)
 891                                        warning_errno("failed utime() on %s", pack_tmp_name);
 892                        }
 893
 894                        strbuf_addf(&tmpname, "%s-", base_name);
 895
 896                        if (write_bitmap_index) {
 897                                bitmap_writer_set_checksum(oid.hash);
 898                                bitmap_writer_build_type_index(
 899                                        &to_pack, written_list, nr_written);
 900                        }
 901
 902                        finish_tmp_packfile(&tmpname, pack_tmp_name,
 903                                            written_list, nr_written,
 904                                            &pack_idx_opts, oid.hash);
 905
 906                        if (write_bitmap_index) {
 907                                strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
 908
 909                                stop_progress(&progress_state);
 910
 911                                bitmap_writer_show_progress(progress);
 912                                bitmap_writer_reuse_bitmaps(&to_pack);
 913                                bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
 914                                bitmap_writer_build(&to_pack);
 915                                bitmap_writer_finish(written_list, nr_written,
 916                                                     tmpname.buf, write_bitmap_options);
 917                                write_bitmap_index = 0;
 918                        }
 919
 920                        strbuf_release(&tmpname);
 921                        free(pack_tmp_name);
 922                        puts(oid_to_hex(&oid));
 923                }
 924
 925                /* mark written objects as written to previous pack */
 926                for (j = 0; j < nr_written; j++) {
 927                        written_list[j]->offset = (off_t)-1;
 928                }
 929                nr_remaining -= nr_written;
 930        } while (nr_remaining && i < to_pack.nr_objects);
 931
 932        free(written_list);
 933        free(write_order);
 934        stop_progress(&progress_state);
 935        if (written != nr_result)
 936                die("wrote %"PRIu32" objects while expecting %"PRIu32,
 937                        written, nr_result);
 938}
 939
 940static int no_try_delta(const char *path)
 941{
 942        static struct attr_check *check;
 943
 944        if (!check)
 945                check = attr_check_initl("delta", NULL);
 946        if (git_check_attr(path, check))
 947                return 0;
 948        if (ATTR_FALSE(check->items[0].value))
 949                return 1;
 950        return 0;
 951}
 952
 953/*
 954 * When adding an object, check whether we have already added it
 955 * to our packing list. If so, we can skip. However, if we are
 956 * being asked to excludei t, but the previous mention was to include
 957 * it, make sure to adjust its flags and tweak our numbers accordingly.
 958 *
 959 * As an optimization, we pass out the index position where we would have
 960 * found the item, since that saves us from having to look it up again a
 961 * few lines later when we want to add the new entry.
 962 */
 963static int have_duplicate_entry(const struct object_id *oid,
 964                                int exclude,
 965                                uint32_t *index_pos)
 966{
 967        struct object_entry *entry;
 968
 969        entry = packlist_find(&to_pack, oid->hash, index_pos);
 970        if (!entry)
 971                return 0;
 972
 973        if (exclude) {
 974                if (!entry->preferred_base)
 975                        nr_result--;
 976                entry->preferred_base = 1;
 977        }
 978
 979        return 1;
 980}
 981
 982static int want_found_object(int exclude, struct packed_git *p)
 983{
 984        if (exclude)
 985                return 1;
 986        if (incremental)
 987                return 0;
 988
 989        /*
 990         * When asked to do --local (do not include an object that appears in a
 991         * pack we borrow from elsewhere) or --honor-pack-keep (do not include
 992         * an object that appears in a pack marked with .keep), finding a pack
 993         * that matches the criteria is sufficient for us to decide to omit it.
 994         * However, even if this pack does not satisfy the criteria, we need to
 995         * make sure no copy of this object appears in _any_ pack that makes us
 996         * to omit the object, so we need to check all the packs.
 997         *
 998         * We can however first check whether these options can possible matter;
 999         * if they do not matter we know we want the object in generated pack.
1000         * Otherwise, we signal "-1" at the end to tell the caller that we do
1001         * not know either way, and it needs to check more packs.
1002         */
1003        if (!ignore_packed_keep_on_disk &&
1004            !ignore_packed_keep_in_core &&
1005            (!local || !have_non_local_packs))
1006                return 1;
1007
1008        if (local && !p->pack_local)
1009                return 0;
1010        if (p->pack_local &&
1011            ((ignore_packed_keep_on_disk && p->pack_keep) ||
1012             (ignore_packed_keep_in_core && p->pack_keep_in_core)))
1013                return 0;
1014
1015        /* we don't know yet; keep looking for more packs */
1016        return -1;
1017}
1018
1019/*
1020 * Check whether we want the object in the pack (e.g., we do not want
1021 * objects found in non-local stores if the "--local" option was used).
1022 *
1023 * If the caller already knows an existing pack it wants to take the object
1024 * from, that is passed in *found_pack and *found_offset; otherwise this
1025 * function finds if there is any pack that has the object and returns the pack
1026 * and its offset in these variables.
1027 */
1028static int want_object_in_pack(const struct object_id *oid,
1029                               int exclude,
1030                               struct packed_git **found_pack,
1031                               off_t *found_offset)
1032{
1033        int want;
1034        struct list_head *pos;
1035
1036        if (!exclude && local && has_loose_object_nonlocal(oid->hash))
1037                return 0;
1038
1039        /*
1040         * If we already know the pack object lives in, start checks from that
1041         * pack - in the usual case when neither --local was given nor .keep files
1042         * are present we will determine the answer right now.
1043         */
1044        if (*found_pack) {
1045                want = want_found_object(exclude, *found_pack);
1046                if (want != -1)
1047                        return want;
1048        }
1049        list_for_each(pos, get_packed_git_mru(the_repository)) {
1050                struct packed_git *p = list_entry(pos, struct packed_git, mru);
1051                off_t offset;
1052
1053                if (p == *found_pack)
1054                        offset = *found_offset;
1055                else
1056                        offset = find_pack_entry_one(oid->hash, p);
1057
1058                if (offset) {
1059                        if (!*found_pack) {
1060                                if (!is_pack_valid(p))
1061                                        continue;
1062                                *found_offset = offset;
1063                                *found_pack = p;
1064                        }
1065                        want = want_found_object(exclude, p);
1066                        if (!exclude && want > 0)
1067                                list_move(&p->mru,
1068                                          get_packed_git_mru(the_repository));
1069                        if (want != -1)
1070                                return want;
1071                }
1072        }
1073
1074        return 1;
1075}
1076
1077static void create_object_entry(const struct object_id *oid,
1078                                enum object_type type,
1079                                uint32_t hash,
1080                                int exclude,
1081                                int no_try_delta,
1082                                uint32_t index_pos,
1083                                struct packed_git *found_pack,
1084                                off_t found_offset)
1085{
1086        struct object_entry *entry;
1087
1088        entry = packlist_alloc(&to_pack, oid->hash, index_pos);
1089        entry->hash = hash;
1090        oe_set_type(entry, type);
1091        if (exclude)
1092                entry->preferred_base = 1;
1093        else
1094                nr_result++;
1095        if (found_pack) {
1096                oe_set_in_pack(&to_pack, entry, found_pack);
1097                entry->in_pack_offset = found_offset;
1098        }
1099
1100        entry->no_try_delta = no_try_delta;
1101}
1102
1103static const char no_closure_warning[] = N_(
1104"disabling bitmap writing, as some objects are not being packed"
1105);
1106
1107static int add_object_entry(const struct object_id *oid, enum object_type type,
1108                            const char *name, int exclude)
1109{
1110        struct packed_git *found_pack = NULL;
1111        off_t found_offset = 0;
1112        uint32_t index_pos;
1113
1114        display_progress(progress_state, ++nr_seen);
1115
1116        if (have_duplicate_entry(oid, exclude, &index_pos))
1117                return 0;
1118
1119        if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1120                /* The pack is missing an object, so it will not have closure */
1121                if (write_bitmap_index) {
1122                        warning(_(no_closure_warning));
1123                        write_bitmap_index = 0;
1124                }
1125                return 0;
1126        }
1127
1128        create_object_entry(oid, type, pack_name_hash(name),
1129                            exclude, name && no_try_delta(name),
1130                            index_pos, found_pack, found_offset);
1131        return 1;
1132}
1133
1134static int add_object_entry_from_bitmap(const struct object_id *oid,
1135                                        enum object_type type,
1136                                        int flags, uint32_t name_hash,
1137                                        struct packed_git *pack, off_t offset)
1138{
1139        uint32_t index_pos;
1140
1141        display_progress(progress_state, ++nr_seen);
1142
1143        if (have_duplicate_entry(oid, 0, &index_pos))
1144                return 0;
1145
1146        if (!want_object_in_pack(oid, 0, &pack, &offset))
1147                return 0;
1148
1149        create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);
1150        return 1;
1151}
1152
1153struct pbase_tree_cache {
1154        struct object_id oid;
1155        int ref;
1156        int temporary;
1157        void *tree_data;
1158        unsigned long tree_size;
1159};
1160
1161static struct pbase_tree_cache *(pbase_tree_cache[256]);
1162static int pbase_tree_cache_ix(const struct object_id *oid)
1163{
1164        return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1165}
1166static int pbase_tree_cache_ix_incr(int ix)
1167{
1168        return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1169}
1170
1171static struct pbase_tree {
1172        struct pbase_tree *next;
1173        /* This is a phony "cache" entry; we are not
1174         * going to evict it or find it through _get()
1175         * mechanism -- this is for the toplevel node that
1176         * would almost always change with any commit.
1177         */
1178        struct pbase_tree_cache pcache;
1179} *pbase_tree;
1180
1181static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1182{
1183        struct pbase_tree_cache *ent, *nent;
1184        void *data;
1185        unsigned long size;
1186        enum object_type type;
1187        int neigh;
1188        int my_ix = pbase_tree_cache_ix(oid);
1189        int available_ix = -1;
1190
1191        /* pbase-tree-cache acts as a limited hashtable.
1192         * your object will be found at your index or within a few
1193         * slots after that slot if it is cached.
1194         */
1195        for (neigh = 0; neigh < 8; neigh++) {
1196                ent = pbase_tree_cache[my_ix];
1197                if (ent && !oidcmp(&ent->oid, oid)) {
1198                        ent->ref++;
1199                        return ent;
1200                }
1201                else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1202                         ((0 <= available_ix) &&
1203                          (!ent && pbase_tree_cache[available_ix])))
1204                        available_ix = my_ix;
1205                if (!ent)
1206                        break;
1207                my_ix = pbase_tree_cache_ix_incr(my_ix);
1208        }
1209
1210        /* Did not find one.  Either we got a bogus request or
1211         * we need to read and perhaps cache.
1212         */
1213        data = read_object_file(oid, &type, &size);
1214        if (!data)
1215                return NULL;
1216        if (type != OBJ_TREE) {
1217                free(data);
1218                return NULL;
1219        }
1220
1221        /* We need to either cache or return a throwaway copy */
1222
1223        if (available_ix < 0)
1224                ent = NULL;
1225        else {
1226                ent = pbase_tree_cache[available_ix];
1227                my_ix = available_ix;
1228        }
1229
1230        if (!ent) {
1231                nent = xmalloc(sizeof(*nent));
1232                nent->temporary = (available_ix < 0);
1233        }
1234        else {
1235                /* evict and reuse */
1236                free(ent->tree_data);
1237                nent = ent;
1238        }
1239        oidcpy(&nent->oid, oid);
1240        nent->tree_data = data;
1241        nent->tree_size = size;
1242        nent->ref = 1;
1243        if (!nent->temporary)
1244                pbase_tree_cache[my_ix] = nent;
1245        return nent;
1246}
1247
1248static void pbase_tree_put(struct pbase_tree_cache *cache)
1249{
1250        if (!cache->temporary) {
1251                cache->ref--;
1252                return;
1253        }
1254        free(cache->tree_data);
1255        free(cache);
1256}
1257
1258static int name_cmp_len(const char *name)
1259{
1260        int i;
1261        for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1262                ;
1263        return i;
1264}
1265
1266static void add_pbase_object(struct tree_desc *tree,
1267                             const char *name,
1268                             int cmplen,
1269                             const char *fullname)
1270{
1271        struct name_entry entry;
1272        int cmp;
1273
1274        while (tree_entry(tree,&entry)) {
1275                if (S_ISGITLINK(entry.mode))
1276                        continue;
1277                cmp = tree_entry_len(&entry) != cmplen ? 1 :
1278                      memcmp(name, entry.path, cmplen);
1279                if (cmp > 0)
1280                        continue;
1281                if (cmp < 0)
1282                        return;
1283                if (name[cmplen] != '/') {
1284                        add_object_entry(entry.oid,
1285                                         object_type(entry.mode),
1286                                         fullname, 1);
1287                        return;
1288                }
1289                if (S_ISDIR(entry.mode)) {
1290                        struct tree_desc sub;
1291                        struct pbase_tree_cache *tree;
1292                        const char *down = name+cmplen+1;
1293                        int downlen = name_cmp_len(down);
1294
1295                        tree = pbase_tree_get(entry.oid);
1296                        if (!tree)
1297                                return;
1298                        init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1299
1300                        add_pbase_object(&sub, down, downlen, fullname);
1301                        pbase_tree_put(tree);
1302                }
1303        }
1304}
1305
1306static unsigned *done_pbase_paths;
1307static int done_pbase_paths_num;
1308static int done_pbase_paths_alloc;
1309static int done_pbase_path_pos(unsigned hash)
1310{
1311        int lo = 0;
1312        int hi = done_pbase_paths_num;
1313        while (lo < hi) {
1314                int mi = lo + (hi - lo) / 2;
1315                if (done_pbase_paths[mi] == hash)
1316                        return mi;
1317                if (done_pbase_paths[mi] < hash)
1318                        hi = mi;
1319                else
1320                        lo = mi + 1;
1321        }
1322        return -lo-1;
1323}
1324
1325static int check_pbase_path(unsigned hash)
1326{
1327        int pos = done_pbase_path_pos(hash);
1328        if (0 <= pos)
1329                return 1;
1330        pos = -pos - 1;
1331        ALLOC_GROW(done_pbase_paths,
1332                   done_pbase_paths_num + 1,
1333                   done_pbase_paths_alloc);
1334        done_pbase_paths_num++;
1335        if (pos < done_pbase_paths_num)
1336                MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1337                           done_pbase_paths_num - pos - 1);
1338        done_pbase_paths[pos] = hash;
1339        return 0;
1340}
1341
1342static void add_preferred_base_object(const char *name)
1343{
1344        struct pbase_tree *it;
1345        int cmplen;
1346        unsigned hash = pack_name_hash(name);
1347
1348        if (!num_preferred_base || check_pbase_path(hash))
1349                return;
1350
1351        cmplen = name_cmp_len(name);
1352        for (it = pbase_tree; it; it = it->next) {
1353                if (cmplen == 0) {
1354                        add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1355                }
1356                else {
1357                        struct tree_desc tree;
1358                        init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1359                        add_pbase_object(&tree, name, cmplen, name);
1360                }
1361        }
1362}
1363
1364static void add_preferred_base(struct object_id *oid)
1365{
1366        struct pbase_tree *it;
1367        void *data;
1368        unsigned long size;
1369        struct object_id tree_oid;
1370
1371        if (window <= num_preferred_base++)
1372                return;
1373
1374        data = read_object_with_reference(oid, tree_type, &size, &tree_oid);
1375        if (!data)
1376                return;
1377
1378        for (it = pbase_tree; it; it = it->next) {
1379                if (!oidcmp(&it->pcache.oid, &tree_oid)) {
1380                        free(data);
1381                        return;
1382                }
1383        }
1384
1385        it = xcalloc(1, sizeof(*it));
1386        it->next = pbase_tree;
1387        pbase_tree = it;
1388
1389        oidcpy(&it->pcache.oid, &tree_oid);
1390        it->pcache.tree_data = data;
1391        it->pcache.tree_size = size;
1392}
1393
1394static void cleanup_preferred_base(void)
1395{
1396        struct pbase_tree *it;
1397        unsigned i;
1398
1399        it = pbase_tree;
1400        pbase_tree = NULL;
1401        while (it) {
1402                struct pbase_tree *tmp = it;
1403                it = tmp->next;
1404                free(tmp->pcache.tree_data);
1405                free(tmp);
1406        }
1407
1408        for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1409                if (!pbase_tree_cache[i])
1410                        continue;
1411                free(pbase_tree_cache[i]->tree_data);
1412                FREE_AND_NULL(pbase_tree_cache[i]);
1413        }
1414
1415        FREE_AND_NULL(done_pbase_paths);
1416        done_pbase_paths_num = done_pbase_paths_alloc = 0;
1417}
1418
1419static void check_object(struct object_entry *entry)
1420{
1421        unsigned long canonical_size;
1422
1423        if (IN_PACK(entry)) {
1424                struct packed_git *p = IN_PACK(entry);
1425                struct pack_window *w_curs = NULL;
1426                const unsigned char *base_ref = NULL;
1427                struct object_entry *base_entry;
1428                unsigned long used, used_0;
1429                unsigned long avail;
1430                off_t ofs;
1431                unsigned char *buf, c;
1432                enum object_type type;
1433                unsigned long in_pack_size;
1434
1435                buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1436
1437                /*
1438                 * We want in_pack_type even if we do not reuse delta
1439                 * since non-delta representations could still be reused.
1440                 */
1441                used = unpack_object_header_buffer(buf, avail,
1442                                                   &type,
1443                                                   &in_pack_size);
1444                if (used == 0)
1445                        goto give_up;
1446
1447                if (type < 0)
1448                        BUG("invalid type %d", type);
1449                entry->in_pack_type = type;
1450
1451                /*
1452                 * Determine if this is a delta and if so whether we can
1453                 * reuse it or not.  Otherwise let's find out as cheaply as
1454                 * possible what the actual type and size for this object is.
1455                 */
1456                switch (entry->in_pack_type) {
1457                default:
1458                        /* Not a delta hence we've already got all we need. */
1459                        oe_set_type(entry, entry->in_pack_type);
1460                        SET_SIZE(entry, in_pack_size);
1461                        entry->in_pack_header_size = used;
1462                        if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
1463                                goto give_up;
1464                        unuse_pack(&w_curs);
1465                        return;
1466                case OBJ_REF_DELTA:
1467                        if (reuse_delta && !entry->preferred_base)
1468                                base_ref = use_pack(p, &w_curs,
1469                                                entry->in_pack_offset + used, NULL);
1470                        entry->in_pack_header_size = used + 20;
1471                        break;
1472                case OBJ_OFS_DELTA:
1473                        buf = use_pack(p, &w_curs,
1474                                       entry->in_pack_offset + used, NULL);
1475                        used_0 = 0;
1476                        c = buf[used_0++];
1477                        ofs = c & 127;
1478                        while (c & 128) {
1479                                ofs += 1;
1480                                if (!ofs || MSB(ofs, 7)) {
1481                                        error("delta base offset overflow in pack for %s",
1482                                              oid_to_hex(&entry->idx.oid));
1483                                        goto give_up;
1484                                }
1485                                c = buf[used_0++];
1486                                ofs = (ofs << 7) + (c & 127);
1487                        }
1488                        ofs = entry->in_pack_offset - ofs;
1489                        if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1490                                error("delta base offset out of bound for %s",
1491                                      oid_to_hex(&entry->idx.oid));
1492                                goto give_up;
1493                        }
1494                        if (reuse_delta && !entry->preferred_base) {
1495                                struct revindex_entry *revidx;
1496                                revidx = find_pack_revindex(p, ofs);
1497                                if (!revidx)
1498                                        goto give_up;
1499                                base_ref = nth_packed_object_sha1(p, revidx->nr);
1500                        }
1501                        entry->in_pack_header_size = used + used_0;
1502                        break;
1503                }
1504
1505                if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1506                        /*
1507                         * If base_ref was set above that means we wish to
1508                         * reuse delta data, and we even found that base
1509                         * in the list of objects we want to pack. Goodie!
1510                         *
1511                         * Depth value does not matter - find_deltas() will
1512                         * never consider reused delta as the base object to
1513                         * deltify other objects against, in order to avoid
1514                         * circular deltas.
1515                         */
1516                        oe_set_type(entry, entry->in_pack_type);
1517                        SET_SIZE(entry, in_pack_size); /* delta size */
1518                        SET_DELTA(entry, base_entry);
1519                        SET_DELTA_SIZE(entry, in_pack_size);
1520                        entry->delta_sibling_idx = base_entry->delta_child_idx;
1521                        SET_DELTA_CHILD(base_entry, entry);
1522                        unuse_pack(&w_curs);
1523                        return;
1524                }
1525
1526                if (oe_type(entry)) {
1527                        off_t delta_pos;
1528
1529                        /*
1530                         * This must be a delta and we already know what the
1531                         * final object type is.  Let's extract the actual
1532                         * object size from the delta header.
1533                         */
1534                        delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
1535                        canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
1536                        if (canonical_size == 0)
1537                                goto give_up;
1538                        SET_SIZE(entry, canonical_size);
1539                        unuse_pack(&w_curs);
1540                        return;
1541                }
1542
1543                /*
1544                 * No choice but to fall back to the recursive delta walk
1545                 * with sha1_object_info() to find about the object type
1546                 * at this point...
1547                 */
1548                give_up:
1549                unuse_pack(&w_curs);
1550        }
1551
1552        oe_set_type(entry,
1553                    oid_object_info(the_repository, &entry->idx.oid, &canonical_size));
1554        if (entry->type_valid) {
1555                SET_SIZE(entry, canonical_size);
1556        } else {
1557                /*
1558                 * Bad object type is checked in prepare_pack().  This is
1559                 * to permit a missing preferred base object to be ignored
1560                 * as a preferred base.  Doing so can result in a larger
1561                 * pack file, but the transfer will still take place.
1562                 */
1563        }
1564}
1565
1566static int pack_offset_sort(const void *_a, const void *_b)
1567{
1568        const struct object_entry *a = *(struct object_entry **)_a;
1569        const struct object_entry *b = *(struct object_entry **)_b;
1570        const struct packed_git *a_in_pack = IN_PACK(a);
1571        const struct packed_git *b_in_pack = IN_PACK(b);
1572
1573        /* avoid filesystem trashing with loose objects */
1574        if (!a_in_pack && !b_in_pack)
1575                return oidcmp(&a->idx.oid, &b->idx.oid);
1576
1577        if (a_in_pack < b_in_pack)
1578                return -1;
1579        if (a_in_pack > b_in_pack)
1580                return 1;
1581        return a->in_pack_offset < b->in_pack_offset ? -1 :
1582                        (a->in_pack_offset > b->in_pack_offset);
1583}
1584
1585/*
1586 * Drop an on-disk delta we were planning to reuse. Naively, this would
1587 * just involve blanking out the "delta" field, but we have to deal
1588 * with some extra book-keeping:
1589 *
1590 *   1. Removing ourselves from the delta_sibling linked list.
1591 *
1592 *   2. Updating our size/type to the non-delta representation. These were
1593 *      either not recorded initially (size) or overwritten with the delta type
1594 *      (type) when check_object() decided to reuse the delta.
1595 *
1596 *   3. Resetting our delta depth, as we are now a base object.
1597 */
1598static void drop_reused_delta(struct object_entry *entry)
1599{
1600        unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
1601        struct object_info oi = OBJECT_INFO_INIT;
1602        enum object_type type;
1603        unsigned long size;
1604
1605        while (*idx) {
1606                struct object_entry *oe = &to_pack.objects[*idx - 1];
1607
1608                if (oe == entry)
1609                        *idx = oe->delta_sibling_idx;
1610                else
1611                        idx = &oe->delta_sibling_idx;
1612        }
1613        SET_DELTA(entry, NULL);
1614        entry->depth = 0;
1615
1616        oi.sizep = &size;
1617        oi.typep = &type;
1618        if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
1619                /*
1620                 * We failed to get the info from this pack for some reason;
1621                 * fall back to sha1_object_info, which may find another copy.
1622                 * And if that fails, the error will be recorded in oe_type(entry)
1623                 * and dealt with in prepare_pack().
1624                 */
1625                oe_set_type(entry,
1626                            oid_object_info(the_repository, &entry->idx.oid, &size));
1627        } else {
1628                oe_set_type(entry, type);
1629        }
1630        SET_SIZE(entry, size);
1631}
1632
1633/*
1634 * Follow the chain of deltas from this entry onward, throwing away any links
1635 * that cause us to hit a cycle (as determined by the DFS state flags in
1636 * the entries).
1637 *
1638 * We also detect too-long reused chains that would violate our --depth
1639 * limit.
1640 */
1641static void break_delta_chains(struct object_entry *entry)
1642{
1643        /*
1644         * The actual depth of each object we will write is stored as an int,
1645         * as it cannot exceed our int "depth" limit. But before we break
1646         * changes based no that limit, we may potentially go as deep as the
1647         * number of objects, which is elsewhere bounded to a uint32_t.
1648         */
1649        uint32_t total_depth;
1650        struct object_entry *cur, *next;
1651
1652        for (cur = entry, total_depth = 0;
1653             cur;
1654             cur = DELTA(cur), total_depth++) {
1655                if (cur->dfs_state == DFS_DONE) {
1656                        /*
1657                         * We've already seen this object and know it isn't
1658                         * part of a cycle. We do need to append its depth
1659                         * to our count.
1660                         */
1661                        total_depth += cur->depth;
1662                        break;
1663                }
1664
1665                /*
1666                 * We break cycles before looping, so an ACTIVE state (or any
1667                 * other cruft which made its way into the state variable)
1668                 * is a bug.
1669                 */
1670                if (cur->dfs_state != DFS_NONE)
1671                        BUG("confusing delta dfs state in first pass: %d",
1672                            cur->dfs_state);
1673
1674                /*
1675                 * Now we know this is the first time we've seen the object. If
1676                 * it's not a delta, we're done traversing, but we'll mark it
1677                 * done to save time on future traversals.
1678                 */
1679                if (!DELTA(cur)) {
1680                        cur->dfs_state = DFS_DONE;
1681                        break;
1682                }
1683
1684                /*
1685                 * Mark ourselves as active and see if the next step causes
1686                 * us to cycle to another active object. It's important to do
1687                 * this _before_ we loop, because it impacts where we make the
1688                 * cut, and thus how our total_depth counter works.
1689                 * E.g., We may see a partial loop like:
1690                 *
1691                 *   A -> B -> C -> D -> B
1692                 *
1693                 * Cutting B->C breaks the cycle. But now the depth of A is
1694                 * only 1, and our total_depth counter is at 3. The size of the
1695                 * error is always one less than the size of the cycle we
1696                 * broke. Commits C and D were "lost" from A's chain.
1697                 *
1698                 * If we instead cut D->B, then the depth of A is correct at 3.
1699                 * We keep all commits in the chain that we examined.
1700                 */
1701                cur->dfs_state = DFS_ACTIVE;
1702                if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
1703                        drop_reused_delta(cur);
1704                        cur->dfs_state = DFS_DONE;
1705                        break;
1706                }
1707        }
1708
1709        /*
1710         * And now that we've gone all the way to the bottom of the chain, we
1711         * need to clear the active flags and set the depth fields as
1712         * appropriate. Unlike the loop above, which can quit when it drops a
1713         * delta, we need to keep going to look for more depth cuts. So we need
1714         * an extra "next" pointer to keep going after we reset cur->delta.
1715         */
1716        for (cur = entry; cur; cur = next) {
1717                next = DELTA(cur);
1718
1719                /*
1720                 * We should have a chain of zero or more ACTIVE states down to
1721                 * a final DONE. We can quit after the DONE, because either it
1722                 * has no bases, or we've already handled them in a previous
1723                 * call.
1724                 */
1725                if (cur->dfs_state == DFS_DONE)
1726                        break;
1727                else if (cur->dfs_state != DFS_ACTIVE)
1728                        BUG("confusing delta dfs state in second pass: %d",
1729                            cur->dfs_state);
1730
1731                /*
1732                 * If the total_depth is more than depth, then we need to snip
1733                 * the chain into two or more smaller chains that don't exceed
1734                 * the maximum depth. Most of the resulting chains will contain
1735                 * (depth + 1) entries (i.e., depth deltas plus one base), and
1736                 * the last chain (i.e., the one containing entry) will contain
1737                 * whatever entries are left over, namely
1738                 * (total_depth % (depth + 1)) of them.
1739                 *
1740                 * Since we are iterating towards decreasing depth, we need to
1741                 * decrement total_depth as we go, and we need to write to the
1742                 * entry what its final depth will be after all of the
1743                 * snipping. Since we're snipping into chains of length (depth
1744                 * + 1) entries, the final depth of an entry will be its
1745                 * original depth modulo (depth + 1). Any time we encounter an
1746                 * entry whose final depth is supposed to be zero, we snip it
1747                 * from its delta base, thereby making it so.
1748                 */
1749                cur->depth = (total_depth--) % (depth + 1);
1750                if (!cur->depth)
1751                        drop_reused_delta(cur);
1752
1753                cur->dfs_state = DFS_DONE;
1754        }
1755}
1756
1757static void get_object_details(void)
1758{
1759        uint32_t i;
1760        struct object_entry **sorted_by_offset;
1761
1762        if (progress)
1763                progress_state = start_progress(_("Counting objects"),
1764                                                to_pack.nr_objects);
1765
1766        sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1767        for (i = 0; i < to_pack.nr_objects; i++)
1768                sorted_by_offset[i] = to_pack.objects + i;
1769        QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1770
1771        for (i = 0; i < to_pack.nr_objects; i++) {
1772                struct object_entry *entry = sorted_by_offset[i];
1773                check_object(entry);
1774                if (entry->type_valid &&
1775                    oe_size_greater_than(&to_pack, entry, big_file_threshold))
1776                        entry->no_try_delta = 1;
1777                display_progress(progress_state, i + 1);
1778        }
1779        stop_progress(&progress_state);
1780
1781        /*
1782         * This must happen in a second pass, since we rely on the delta
1783         * information for the whole list being completed.
1784         */
1785        for (i = 0; i < to_pack.nr_objects; i++)
1786                break_delta_chains(&to_pack.objects[i]);
1787
1788        free(sorted_by_offset);
1789}
1790
1791/*
1792 * We search for deltas in a list sorted by type, by filename hash, and then
1793 * by size, so that we see progressively smaller and smaller files.
1794 * That's because we prefer deltas to be from the bigger file
1795 * to the smaller -- deletes are potentially cheaper, but perhaps
1796 * more importantly, the bigger file is likely the more recent
1797 * one.  The deepest deltas are therefore the oldest objects which are
1798 * less susceptible to be accessed often.
1799 */
1800static int type_size_sort(const void *_a, const void *_b)
1801{
1802        const struct object_entry *a = *(struct object_entry **)_a;
1803        const struct object_entry *b = *(struct object_entry **)_b;
1804        enum object_type a_type = oe_type(a);
1805        enum object_type b_type = oe_type(b);
1806        unsigned long a_size = SIZE(a);
1807        unsigned long b_size = SIZE(b);
1808
1809        if (a_type > b_type)
1810                return -1;
1811        if (a_type < b_type)
1812                return 1;
1813        if (a->hash > b->hash)
1814                return -1;
1815        if (a->hash < b->hash)
1816                return 1;
1817        if (a->preferred_base > b->preferred_base)
1818                return -1;
1819        if (a->preferred_base < b->preferred_base)
1820                return 1;
1821        if (a_size > b_size)
1822                return -1;
1823        if (a_size < b_size)
1824                return 1;
1825        return a < b ? -1 : (a > b);  /* newest first */
1826}
1827
1828struct unpacked {
1829        struct object_entry *entry;
1830        void *data;
1831        struct delta_index *index;
1832        unsigned depth;
1833};
1834
1835static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1836                           unsigned long delta_size)
1837{
1838        if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1839                return 0;
1840
1841        if (delta_size < cache_max_small_delta_size)
1842                return 1;
1843
1844        /* cache delta, if objects are large enough compared to delta size */
1845        if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1846                return 1;
1847
1848        return 0;
1849}
1850
1851#ifndef NO_PTHREADS
1852
1853static pthread_mutex_t read_mutex;
1854#define read_lock()             pthread_mutex_lock(&read_mutex)
1855#define read_unlock()           pthread_mutex_unlock(&read_mutex)
1856
1857static pthread_mutex_t cache_mutex;
1858#define cache_lock()            pthread_mutex_lock(&cache_mutex)
1859#define cache_unlock()          pthread_mutex_unlock(&cache_mutex)
1860
1861static pthread_mutex_t progress_mutex;
1862#define progress_lock()         pthread_mutex_lock(&progress_mutex)
1863#define progress_unlock()       pthread_mutex_unlock(&progress_mutex)
1864
1865#else
1866
1867#define read_lock()             (void)0
1868#define read_unlock()           (void)0
1869#define cache_lock()            (void)0
1870#define cache_unlock()          (void)0
1871#define progress_lock()         (void)0
1872#define progress_unlock()       (void)0
1873
1874#endif
1875
1876/*
1877 * Return the size of the object without doing any delta
1878 * reconstruction (so non-deltas are true object sizes, but deltas
1879 * return the size of the delta data).
1880 */
1881unsigned long oe_get_size_slow(struct packing_data *pack,
1882                               const struct object_entry *e)
1883{
1884        struct packed_git *p;
1885        struct pack_window *w_curs;
1886        unsigned char *buf;
1887        enum object_type type;
1888        unsigned long used, avail, size;
1889
1890        if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
1891                read_lock();
1892                if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
1893                        die(_("unable to get size of %s"),
1894                            oid_to_hex(&e->idx.oid));
1895                read_unlock();
1896                return size;
1897        }
1898
1899        p = oe_in_pack(pack, e);
1900        if (!p)
1901                BUG("when e->type is a delta, it must belong to a pack");
1902
1903        read_lock();
1904        w_curs = NULL;
1905        buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
1906        used = unpack_object_header_buffer(buf, avail, &type, &size);
1907        if (used == 0)
1908                die(_("unable to parse object header of %s"),
1909                    oid_to_hex(&e->idx.oid));
1910
1911        unuse_pack(&w_curs);
1912        read_unlock();
1913        return size;
1914}
1915
1916static int try_delta(struct unpacked *trg, struct unpacked *src,
1917                     unsigned max_depth, unsigned long *mem_usage)
1918{
1919        struct object_entry *trg_entry = trg->entry;
1920        struct object_entry *src_entry = src->entry;
1921        unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1922        unsigned ref_depth;
1923        enum object_type type;
1924        void *delta_buf;
1925
1926        /* Don't bother doing diffs between different types */
1927        if (oe_type(trg_entry) != oe_type(src_entry))
1928                return -1;
1929
1930        /*
1931         * We do not bother to try a delta that we discarded on an
1932         * earlier try, but only when reusing delta data.  Note that
1933         * src_entry that is marked as the preferred_base should always
1934         * be considered, as even if we produce a suboptimal delta against
1935         * it, we will still save the transfer cost, as we already know
1936         * the other side has it and we won't send src_entry at all.
1937         */
1938        if (reuse_delta && IN_PACK(trg_entry) &&
1939            IN_PACK(trg_entry) == IN_PACK(src_entry) &&
1940            !src_entry->preferred_base &&
1941            trg_entry->in_pack_type != OBJ_REF_DELTA &&
1942            trg_entry->in_pack_type != OBJ_OFS_DELTA)
1943                return 0;
1944
1945        /* Let's not bust the allowed depth. */
1946        if (src->depth >= max_depth)
1947                return 0;
1948
1949        /* Now some size filtering heuristics. */
1950        trg_size = SIZE(trg_entry);
1951        if (!DELTA(trg_entry)) {
1952                max_size = trg_size/2 - 20;
1953                ref_depth = 1;
1954        } else {
1955                max_size = DELTA_SIZE(trg_entry);
1956                ref_depth = trg->depth;
1957        }
1958        max_size = (uint64_t)max_size * (max_depth - src->depth) /
1959                                                (max_depth - ref_depth + 1);
1960        if (max_size == 0)
1961                return 0;
1962        src_size = SIZE(src_entry);
1963        sizediff = src_size < trg_size ? trg_size - src_size : 0;
1964        if (sizediff >= max_size)
1965                return 0;
1966        if (trg_size < src_size / 32)
1967                return 0;
1968
1969        /* Load data if not already done */
1970        if (!trg->data) {
1971                read_lock();
1972                trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
1973                read_unlock();
1974                if (!trg->data)
1975                        die("object %s cannot be read",
1976                            oid_to_hex(&trg_entry->idx.oid));
1977                if (sz != trg_size)
1978                        die("object %s inconsistent object length (%lu vs %lu)",
1979                            oid_to_hex(&trg_entry->idx.oid), sz,
1980                            trg_size);
1981                *mem_usage += sz;
1982        }
1983        if (!src->data) {
1984                read_lock();
1985                src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
1986                read_unlock();
1987                if (!src->data) {
1988                        if (src_entry->preferred_base) {
1989                                static int warned = 0;
1990                                if (!warned++)
1991                                        warning("object %s cannot be read",
1992                                                oid_to_hex(&src_entry->idx.oid));
1993                                /*
1994                                 * Those objects are not included in the
1995                                 * resulting pack.  Be resilient and ignore
1996                                 * them if they can't be read, in case the
1997                                 * pack could be created nevertheless.
1998                                 */
1999                                return 0;
2000                        }
2001                        die("object %s cannot be read",
2002                            oid_to_hex(&src_entry->idx.oid));
2003                }
2004                if (sz != src_size)
2005                        die("object %s inconsistent object length (%lu vs %lu)",
2006                            oid_to_hex(&src_entry->idx.oid), sz,
2007                            src_size);
2008                *mem_usage += sz;
2009        }
2010        if (!src->index) {
2011                src->index = create_delta_index(src->data, src_size);
2012                if (!src->index) {
2013                        static int warned = 0;
2014                        if (!warned++)
2015                                warning("suboptimal pack - out of memory");
2016                        return 0;
2017                }
2018                *mem_usage += sizeof_delta_index(src->index);
2019        }
2020
2021        delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2022        if (!delta_buf)
2023                return 0;
2024        if (delta_size >= (1U << OE_DELTA_SIZE_BITS)) {
2025                free(delta_buf);
2026                return 0;
2027        }
2028
2029        if (DELTA(trg_entry)) {
2030                /* Prefer only shallower same-sized deltas. */
2031                if (delta_size == DELTA_SIZE(trg_entry) &&
2032                    src->depth + 1 >= trg->depth) {
2033                        free(delta_buf);
2034                        return 0;
2035                }
2036        }
2037
2038        /*
2039         * Handle memory allocation outside of the cache
2040         * accounting lock.  Compiler will optimize the strangeness
2041         * away when NO_PTHREADS is defined.
2042         */
2043        free(trg_entry->delta_data);
2044        cache_lock();
2045        if (trg_entry->delta_data) {
2046                delta_cache_size -= DELTA_SIZE(trg_entry);
2047                trg_entry->delta_data = NULL;
2048        }
2049        if (delta_cacheable(src_size, trg_size, delta_size)) {
2050                delta_cache_size += delta_size;
2051                cache_unlock();
2052                trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2053        } else {
2054                cache_unlock();
2055                free(delta_buf);
2056        }
2057
2058        SET_DELTA(trg_entry, src_entry);
2059        SET_DELTA_SIZE(trg_entry, delta_size);
2060        trg->depth = src->depth + 1;
2061
2062        return 1;
2063}
2064
2065static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2066{
2067        struct object_entry *child = DELTA_CHILD(me);
2068        unsigned int m = n;
2069        while (child) {
2070                unsigned int c = check_delta_limit(child, n + 1);
2071                if (m < c)
2072                        m = c;
2073                child = DELTA_SIBLING(child);
2074        }
2075        return m;
2076}
2077
2078static unsigned long free_unpacked(struct unpacked *n)
2079{
2080        unsigned long freed_mem = sizeof_delta_index(n->index);
2081        free_delta_index(n->index);
2082        n->index = NULL;
2083        if (n->data) {
2084                freed_mem += SIZE(n->entry);
2085                FREE_AND_NULL(n->data);
2086        }
2087        n->entry = NULL;
2088        n->depth = 0;
2089        return freed_mem;
2090}
2091
2092static void find_deltas(struct object_entry **list, unsigned *list_size,
2093                        int window, int depth, unsigned *processed)
2094{
2095        uint32_t i, idx = 0, count = 0;
2096        struct unpacked *array;
2097        unsigned long mem_usage = 0;
2098
2099        array = xcalloc(window, sizeof(struct unpacked));
2100
2101        for (;;) {
2102                struct object_entry *entry;
2103                struct unpacked *n = array + idx;
2104                int j, max_depth, best_base = -1;
2105
2106                progress_lock();
2107                if (!*list_size) {
2108                        progress_unlock();
2109                        break;
2110                }
2111                entry = *list++;
2112                (*list_size)--;
2113                if (!entry->preferred_base) {
2114                        (*processed)++;
2115                        display_progress(progress_state, *processed);
2116                }
2117                progress_unlock();
2118
2119                mem_usage -= free_unpacked(n);
2120                n->entry = entry;
2121
2122                while (window_memory_limit &&
2123                       mem_usage > window_memory_limit &&
2124                       count > 1) {
2125                        uint32_t tail = (idx + window - count) % window;
2126                        mem_usage -= free_unpacked(array + tail);
2127                        count--;
2128                }
2129
2130                /* We do not compute delta to *create* objects we are not
2131                 * going to pack.
2132                 */
2133                if (entry->preferred_base)
2134                        goto next;
2135
2136                /*
2137                 * If the current object is at pack edge, take the depth the
2138                 * objects that depend on the current object into account
2139                 * otherwise they would become too deep.
2140                 */
2141                max_depth = depth;
2142                if (DELTA_CHILD(entry)) {
2143                        max_depth -= check_delta_limit(entry, 0);
2144                        if (max_depth <= 0)
2145                                goto next;
2146                }
2147
2148                j = window;
2149                while (--j > 0) {
2150                        int ret;
2151                        uint32_t other_idx = idx + j;
2152                        struct unpacked *m;
2153                        if (other_idx >= window)
2154                                other_idx -= window;
2155                        m = array + other_idx;
2156                        if (!m->entry)
2157                                break;
2158                        ret = try_delta(n, m, max_depth, &mem_usage);
2159                        if (ret < 0)
2160                                break;
2161                        else if (ret > 0)
2162                                best_base = other_idx;
2163                }
2164
2165                /*
2166                 * If we decided to cache the delta data, then it is best
2167                 * to compress it right away.  First because we have to do
2168                 * it anyway, and doing it here while we're threaded will
2169                 * save a lot of time in the non threaded write phase,
2170                 * as well as allow for caching more deltas within
2171                 * the same cache size limit.
2172                 * ...
2173                 * But only if not writing to stdout, since in that case
2174                 * the network is most likely throttling writes anyway,
2175                 * and therefore it is best to go to the write phase ASAP
2176                 * instead, as we can afford spending more time compressing
2177                 * between writes at that moment.
2178                 */
2179                if (entry->delta_data && !pack_to_stdout) {
2180                        unsigned long size;
2181
2182                        size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
2183                        if (size < (1U << OE_Z_DELTA_BITS)) {
2184                                entry->z_delta_size = size;
2185                                cache_lock();
2186                                delta_cache_size -= DELTA_SIZE(entry);
2187                                delta_cache_size += entry->z_delta_size;
2188                                cache_unlock();
2189                        } else {
2190                                FREE_AND_NULL(entry->delta_data);
2191                                entry->z_delta_size = 0;
2192                        }
2193                }
2194
2195                /* if we made n a delta, and if n is already at max
2196                 * depth, leaving it in the window is pointless.  we
2197                 * should evict it first.
2198                 */
2199                if (DELTA(entry) && max_depth <= n->depth)
2200                        continue;
2201
2202                /*
2203                 * Move the best delta base up in the window, after the
2204                 * currently deltified object, to keep it longer.  It will
2205                 * be the first base object to be attempted next.
2206                 */
2207                if (DELTA(entry)) {
2208                        struct unpacked swap = array[best_base];
2209                        int dist = (window + idx - best_base) % window;
2210                        int dst = best_base;
2211                        while (dist--) {
2212                                int src = (dst + 1) % window;
2213                                array[dst] = array[src];
2214                                dst = src;
2215                        }
2216                        array[dst] = swap;
2217                }
2218
2219                next:
2220                idx++;
2221                if (count + 1 < window)
2222                        count++;
2223                if (idx >= window)
2224                        idx = 0;
2225        }
2226
2227        for (i = 0; i < window; ++i) {
2228                free_delta_index(array[i].index);
2229                free(array[i].data);
2230        }
2231        free(array);
2232}
2233
2234#ifndef NO_PTHREADS
2235
2236static void try_to_free_from_threads(size_t size)
2237{
2238        read_lock();
2239        release_pack_memory(size);
2240        read_unlock();
2241}
2242
2243static try_to_free_t old_try_to_free_routine;
2244
2245/*
2246 * The main thread waits on the condition that (at least) one of the workers
2247 * has stopped working (which is indicated in the .working member of
2248 * struct thread_params).
2249 * When a work thread has completed its work, it sets .working to 0 and
2250 * signals the main thread and waits on the condition that .data_ready
2251 * becomes 1.
2252 */
2253
2254struct thread_params {
2255        pthread_t thread;
2256        struct object_entry **list;
2257        unsigned list_size;
2258        unsigned remaining;
2259        int window;
2260        int depth;
2261        int working;
2262        int data_ready;
2263        pthread_mutex_t mutex;
2264        pthread_cond_t cond;
2265        unsigned *processed;
2266};
2267
2268static pthread_cond_t progress_cond;
2269
2270/*
2271 * Mutex and conditional variable can't be statically-initialized on Windows.
2272 */
2273static void init_threaded_search(void)
2274{
2275        init_recursive_mutex(&read_mutex);
2276        pthread_mutex_init(&cache_mutex, NULL);
2277        pthread_mutex_init(&progress_mutex, NULL);
2278        pthread_cond_init(&progress_cond, NULL);
2279        old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2280}
2281
2282static void cleanup_threaded_search(void)
2283{
2284        set_try_to_free_routine(old_try_to_free_routine);
2285        pthread_cond_destroy(&progress_cond);
2286        pthread_mutex_destroy(&read_mutex);
2287        pthread_mutex_destroy(&cache_mutex);
2288        pthread_mutex_destroy(&progress_mutex);
2289}
2290
2291static void *threaded_find_deltas(void *arg)
2292{
2293        struct thread_params *me = arg;
2294
2295        progress_lock();
2296        while (me->remaining) {
2297                progress_unlock();
2298
2299                find_deltas(me->list, &me->remaining,
2300                            me->window, me->depth, me->processed);
2301
2302                progress_lock();
2303                me->working = 0;
2304                pthread_cond_signal(&progress_cond);
2305                progress_unlock();
2306
2307                /*
2308                 * We must not set ->data_ready before we wait on the
2309                 * condition because the main thread may have set it to 1
2310                 * before we get here. In order to be sure that new
2311                 * work is available if we see 1 in ->data_ready, it
2312                 * was initialized to 0 before this thread was spawned
2313                 * and we reset it to 0 right away.
2314                 */
2315                pthread_mutex_lock(&me->mutex);
2316                while (!me->data_ready)
2317                        pthread_cond_wait(&me->cond, &me->mutex);
2318                me->data_ready = 0;
2319                pthread_mutex_unlock(&me->mutex);
2320
2321                progress_lock();
2322        }
2323        progress_unlock();
2324        /* leave ->working 1 so that this doesn't get more work assigned */
2325        return NULL;
2326}
2327
2328static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2329                           int window, int depth, unsigned *processed)
2330{
2331        struct thread_params *p;
2332        int i, ret, active_threads = 0;
2333
2334        init_threaded_search();
2335
2336        if (delta_search_threads <= 1) {
2337                find_deltas(list, &list_size, window, depth, processed);
2338                cleanup_threaded_search();
2339                return;
2340        }
2341        if (progress > pack_to_stdout)
2342                fprintf(stderr, "Delta compression using up to %d threads.\n",
2343                                delta_search_threads);
2344        p = xcalloc(delta_search_threads, sizeof(*p));
2345
2346        /* Partition the work amongst work threads. */
2347        for (i = 0; i < delta_search_threads; i++) {
2348                unsigned sub_size = list_size / (delta_search_threads - i);
2349
2350                /* don't use too small segments or no deltas will be found */
2351                if (sub_size < 2*window && i+1 < delta_search_threads)
2352                        sub_size = 0;
2353
2354                p[i].window = window;
2355                p[i].depth = depth;
2356                p[i].processed = processed;
2357                p[i].working = 1;
2358                p[i].data_ready = 0;
2359
2360                /* try to split chunks on "path" boundaries */
2361                while (sub_size && sub_size < list_size &&
2362                       list[sub_size]->hash &&
2363                       list[sub_size]->hash == list[sub_size-1]->hash)
2364                        sub_size++;
2365
2366                p[i].list = list;
2367                p[i].list_size = sub_size;
2368                p[i].remaining = sub_size;
2369
2370                list += sub_size;
2371                list_size -= sub_size;
2372        }
2373
2374        /* Start work threads. */
2375        for (i = 0; i < delta_search_threads; i++) {
2376                if (!p[i].list_size)
2377                        continue;
2378                pthread_mutex_init(&p[i].mutex, NULL);
2379                pthread_cond_init(&p[i].cond, NULL);
2380                ret = pthread_create(&p[i].thread, NULL,
2381                                     threaded_find_deltas, &p[i]);
2382                if (ret)
2383                        die("unable to create thread: %s", strerror(ret));
2384                active_threads++;
2385        }
2386
2387        /*
2388         * Now let's wait for work completion.  Each time a thread is done
2389         * with its work, we steal half of the remaining work from the
2390         * thread with the largest number of unprocessed objects and give
2391         * it to that newly idle thread.  This ensure good load balancing
2392         * until the remaining object list segments are simply too short
2393         * to be worth splitting anymore.
2394         */
2395        while (active_threads) {
2396                struct thread_params *target = NULL;
2397                struct thread_params *victim = NULL;
2398                unsigned sub_size = 0;
2399
2400                progress_lock();
2401                for (;;) {
2402                        for (i = 0; !target && i < delta_search_threads; i++)
2403                                if (!p[i].working)
2404                                        target = &p[i];
2405                        if (target)
2406                                break;
2407                        pthread_cond_wait(&progress_cond, &progress_mutex);
2408                }
2409
2410                for (i = 0; i < delta_search_threads; i++)
2411                        if (p[i].remaining > 2*window &&
2412                            (!victim || victim->remaining < p[i].remaining))
2413                                victim = &p[i];
2414                if (victim) {
2415                        sub_size = victim->remaining / 2;
2416                        list = victim->list + victim->list_size - sub_size;
2417                        while (sub_size && list[0]->hash &&
2418                               list[0]->hash == list[-1]->hash) {
2419                                list++;
2420                                sub_size--;
2421                        }
2422                        if (!sub_size) {
2423                                /*
2424                                 * It is possible for some "paths" to have
2425                                 * so many objects that no hash boundary
2426                                 * might be found.  Let's just steal the
2427                                 * exact half in that case.
2428                                 */
2429                                sub_size = victim->remaining / 2;
2430                                list -= sub_size;
2431                        }
2432                        target->list = list;
2433                        victim->list_size -= sub_size;
2434                        victim->remaining -= sub_size;
2435                }
2436                target->list_size = sub_size;
2437                target->remaining = sub_size;
2438                target->working = 1;
2439                progress_unlock();
2440
2441                pthread_mutex_lock(&target->mutex);
2442                target->data_ready = 1;
2443                pthread_cond_signal(&target->cond);
2444                pthread_mutex_unlock(&target->mutex);
2445
2446                if (!sub_size) {
2447                        pthread_join(target->thread, NULL);
2448                        pthread_cond_destroy(&target->cond);
2449                        pthread_mutex_destroy(&target->mutex);
2450                        active_threads--;
2451                }
2452        }
2453        cleanup_threaded_search();
2454        free(p);
2455}
2456
2457#else
2458#define ll_find_deltas(l, s, w, d, p)   find_deltas(l, &s, w, d, p)
2459#endif
2460
2461static void add_tag_chain(const struct object_id *oid)
2462{
2463        struct tag *tag;
2464
2465        /*
2466         * We catch duplicates already in add_object_entry(), but we'd
2467         * prefer to do this extra check to avoid having to parse the
2468         * tag at all if we already know that it's being packed (e.g., if
2469         * it was included via bitmaps, we would not have parsed it
2470         * previously).
2471         */
2472        if (packlist_find(&to_pack, oid->hash, NULL))
2473                return;
2474
2475        tag = lookup_tag(oid);
2476        while (1) {
2477                if (!tag || parse_tag(tag) || !tag->tagged)
2478                        die("unable to pack objects reachable from tag %s",
2479                            oid_to_hex(oid));
2480
2481                add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2482
2483                if (tag->tagged->type != OBJ_TAG)
2484                        return;
2485
2486                tag = (struct tag *)tag->tagged;
2487        }
2488}
2489
2490static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2491{
2492        struct object_id peeled;
2493
2494        if (starts_with(path, "refs/tags/") && /* is a tag? */
2495            !peel_ref(path, &peeled)    && /* peelable? */
2496            packlist_find(&to_pack, peeled.hash, NULL))      /* object packed? */
2497                add_tag_chain(oid);
2498        return 0;
2499}
2500
2501static void prepare_pack(int window, int depth)
2502{
2503        struct object_entry **delta_list;
2504        uint32_t i, nr_deltas;
2505        unsigned n;
2506
2507        get_object_details();
2508
2509        /*
2510         * If we're locally repacking then we need to be doubly careful
2511         * from now on in order to make sure no stealth corruption gets
2512         * propagated to the new pack.  Clients receiving streamed packs
2513         * should validate everything they get anyway so no need to incur
2514         * the additional cost here in that case.
2515         */
2516        if (!pack_to_stdout)
2517                do_check_packed_object_crc = 1;
2518
2519        if (!to_pack.nr_objects || !window || !depth)
2520                return;
2521
2522        ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2523        nr_deltas = n = 0;
2524
2525        for (i = 0; i < to_pack.nr_objects; i++) {
2526                struct object_entry *entry = to_pack.objects + i;
2527
2528                if (DELTA(entry))
2529                        /* This happens if we decided to reuse existing
2530                         * delta from a pack.  "reuse_delta &&" is implied.
2531                         */
2532                        continue;
2533
2534                if (!entry->type_valid ||
2535                    oe_size_less_than(&to_pack, entry, 50))
2536                        continue;
2537
2538                if (entry->no_try_delta)
2539                        continue;
2540
2541                if (!entry->preferred_base) {
2542                        nr_deltas++;
2543                        if (oe_type(entry) < 0)
2544                                die("unable to get type of object %s",
2545                                    oid_to_hex(&entry->idx.oid));
2546                } else {
2547                        if (oe_type(entry) < 0) {
2548                                /*
2549                                 * This object is not found, but we
2550                                 * don't have to include it anyway.
2551                                 */
2552                                continue;
2553                        }
2554                }
2555
2556                delta_list[n++] = entry;
2557        }
2558
2559        if (nr_deltas && n > 1) {
2560                unsigned nr_done = 0;
2561                if (progress)
2562                        progress_state = start_progress(_("Compressing objects"),
2563                                                        nr_deltas);
2564                QSORT(delta_list, n, type_size_sort);
2565                ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2566                stop_progress(&progress_state);
2567                if (nr_done != nr_deltas)
2568                        die("inconsistency with delta count");
2569        }
2570        free(delta_list);
2571}
2572
2573static int git_pack_config(const char *k, const char *v, void *cb)
2574{
2575        if (!strcmp(k, "pack.window")) {
2576                window = git_config_int(k, v);
2577                return 0;
2578        }
2579        if (!strcmp(k, "pack.windowmemory")) {
2580                window_memory_limit = git_config_ulong(k, v);
2581                return 0;
2582        }
2583        if (!strcmp(k, "pack.depth")) {
2584                depth = git_config_int(k, v);
2585                return 0;
2586        }
2587        if (!strcmp(k, "pack.deltacachesize")) {
2588                max_delta_cache_size = git_config_int(k, v);
2589                return 0;
2590        }
2591        if (!strcmp(k, "pack.deltacachelimit")) {
2592                cache_max_small_delta_size = git_config_int(k, v);
2593                return 0;
2594        }
2595        if (!strcmp(k, "pack.writebitmaphashcache")) {
2596                if (git_config_bool(k, v))
2597                        write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2598                else
2599                        write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2600        }
2601        if (!strcmp(k, "pack.usebitmaps")) {
2602                use_bitmap_index_default = git_config_bool(k, v);
2603                return 0;
2604        }
2605        if (!strcmp(k, "pack.threads")) {
2606                delta_search_threads = git_config_int(k, v);
2607                if (delta_search_threads < 0)
2608                        die("invalid number of threads specified (%d)",
2609                            delta_search_threads);
2610#ifdef NO_PTHREADS
2611                if (delta_search_threads != 1) {
2612                        warning("no threads support, ignoring %s", k);
2613                        delta_search_threads = 0;
2614                }
2615#endif
2616                return 0;
2617        }
2618        if (!strcmp(k, "pack.indexversion")) {
2619                pack_idx_opts.version = git_config_int(k, v);
2620                if (pack_idx_opts.version > 2)
2621                        die("bad pack.indexversion=%"PRIu32,
2622                            pack_idx_opts.version);
2623                return 0;
2624        }
2625        return git_default_config(k, v, cb);
2626}
2627
2628static void read_object_list_from_stdin(void)
2629{
2630        char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2631        struct object_id oid;
2632        const char *p;
2633
2634        for (;;) {
2635                if (!fgets(line, sizeof(line), stdin)) {
2636                        if (feof(stdin))
2637                                break;
2638                        if (!ferror(stdin))
2639                                die("fgets returned NULL, not EOF, not error!");
2640                        if (errno != EINTR)
2641                                die_errno("fgets");
2642                        clearerr(stdin);
2643                        continue;
2644                }
2645                if (line[0] == '-') {
2646                        if (get_oid_hex(line+1, &oid))
2647                                die("expected edge object ID, got garbage:\n %s",
2648                                    line);
2649                        add_preferred_base(&oid);
2650                        continue;
2651                }
2652                if (parse_oid_hex(line, &oid, &p))
2653                        die("expected object ID, got garbage:\n %s", line);
2654
2655                add_preferred_base_object(p + 1);
2656                add_object_entry(&oid, OBJ_NONE, p + 1, 0);
2657        }
2658}
2659
2660/* Remember to update object flag allocation in object.h */
2661#define OBJECT_ADDED (1u<<20)
2662
2663static void show_commit(struct commit *commit, void *data)
2664{
2665        add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2666        commit->object.flags |= OBJECT_ADDED;
2667
2668        if (write_bitmap_index)
2669                index_commit_for_bitmap(commit);
2670}
2671
2672static void show_object(struct object *obj, const char *name, void *data)
2673{
2674        add_preferred_base_object(name);
2675        add_object_entry(&obj->oid, obj->type, name, 0);
2676        obj->flags |= OBJECT_ADDED;
2677}
2678
2679static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2680{
2681        assert(arg_missing_action == MA_ALLOW_ANY);
2682
2683        /*
2684         * Quietly ignore ALL missing objects.  This avoids problems with
2685         * staging them now and getting an odd error later.
2686         */
2687        if (!has_object_file(&obj->oid))
2688                return;
2689
2690        show_object(obj, name, data);
2691}
2692
2693static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2694{
2695        assert(arg_missing_action == MA_ALLOW_PROMISOR);
2696
2697        /*
2698         * Quietly ignore EXPECTED missing objects.  This avoids problems with
2699         * staging them now and getting an odd error later.
2700         */
2701        if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2702                return;
2703
2704        show_object(obj, name, data);
2705}
2706
2707static int option_parse_missing_action(const struct option *opt,
2708                                       const char *arg, int unset)
2709{
2710        assert(arg);
2711        assert(!unset);
2712
2713        if (!strcmp(arg, "error")) {
2714                arg_missing_action = MA_ERROR;
2715                fn_show_object = show_object;
2716                return 0;
2717        }
2718
2719        if (!strcmp(arg, "allow-any")) {
2720                arg_missing_action = MA_ALLOW_ANY;
2721                fetch_if_missing = 0;
2722                fn_show_object = show_object__ma_allow_any;
2723                return 0;
2724        }
2725
2726        if (!strcmp(arg, "allow-promisor")) {
2727                arg_missing_action = MA_ALLOW_PROMISOR;
2728                fetch_if_missing = 0;
2729                fn_show_object = show_object__ma_allow_promisor;
2730                return 0;
2731        }
2732
2733        die(_("invalid value for --missing"));
2734        return 0;
2735}
2736
2737static void show_edge(struct commit *commit)
2738{
2739        add_preferred_base(&commit->object.oid);
2740}
2741
2742struct in_pack_object {
2743        off_t offset;
2744        struct object *object;
2745};
2746
2747struct in_pack {
2748        unsigned int alloc;
2749        unsigned int nr;
2750        struct in_pack_object *array;
2751};
2752
2753static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2754{
2755        in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2756        in_pack->array[in_pack->nr].object = object;
2757        in_pack->nr++;
2758}
2759
2760/*
2761 * Compare the objects in the offset order, in order to emulate the
2762 * "git rev-list --objects" output that produced the pack originally.
2763 */
2764static int ofscmp(const void *a_, const void *b_)
2765{
2766        struct in_pack_object *a = (struct in_pack_object *)a_;
2767        struct in_pack_object *b = (struct in_pack_object *)b_;
2768
2769        if (a->offset < b->offset)
2770                return -1;
2771        else if (a->offset > b->offset)
2772                return 1;
2773        else
2774                return oidcmp(&a->object->oid, &b->object->oid);
2775}
2776
2777static void add_objects_in_unpacked_packs(struct rev_info *revs)
2778{
2779        struct packed_git *p;
2780        struct in_pack in_pack;
2781        uint32_t i;
2782
2783        memset(&in_pack, 0, sizeof(in_pack));
2784
2785        for (p = get_packed_git(the_repository); p; p = p->next) {
2786                struct object_id oid;
2787                struct object *o;
2788
2789                if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
2790                        continue;
2791                if (open_pack_index(p))
2792                        die("cannot open pack index");
2793
2794                ALLOC_GROW(in_pack.array,
2795                           in_pack.nr + p->num_objects,
2796                           in_pack.alloc);
2797
2798                for (i = 0; i < p->num_objects; i++) {
2799                        nth_packed_object_oid(&oid, p, i);
2800                        o = lookup_unknown_object(oid.hash);
2801                        if (!(o->flags & OBJECT_ADDED))
2802                                mark_in_pack_object(o, p, &in_pack);
2803                        o->flags |= OBJECT_ADDED;
2804                }
2805        }
2806
2807        if (in_pack.nr) {
2808                QSORT(in_pack.array, in_pack.nr, ofscmp);
2809                for (i = 0; i < in_pack.nr; i++) {
2810                        struct object *o = in_pack.array[i].object;
2811                        add_object_entry(&o->oid, o->type, "", 0);
2812                }
2813        }
2814        free(in_pack.array);
2815}
2816
2817static int add_loose_object(const struct object_id *oid, const char *path,
2818                            void *data)
2819{
2820        enum object_type type = oid_object_info(the_repository, oid, NULL);
2821
2822        if (type < 0) {
2823                warning("loose object at %s could not be examined", path);
2824                return 0;
2825        }
2826
2827        add_object_entry(oid, type, "", 0);
2828        return 0;
2829}
2830
2831/*
2832 * We actually don't even have to worry about reachability here.
2833 * add_object_entry will weed out duplicates, so we just add every
2834 * loose object we find.
2835 */
2836static void add_unreachable_loose_objects(void)
2837{
2838        for_each_loose_file_in_objdir(get_object_directory(),
2839                                      add_loose_object,
2840                                      NULL, NULL, NULL);
2841}
2842
2843static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
2844{
2845        static struct packed_git *last_found = (void *)1;
2846        struct packed_git *p;
2847
2848        p = (last_found != (void *)1) ? last_found :
2849                                        get_packed_git(the_repository);
2850
2851        while (p) {
2852                if ((!p->pack_local || p->pack_keep ||
2853                                p->pack_keep_in_core) &&
2854                        find_pack_entry_one(oid->hash, p)) {
2855                        last_found = p;
2856                        return 1;
2857                }
2858                if (p == last_found)
2859                        p = get_packed_git(the_repository);
2860                else
2861                        p = p->next;
2862                if (p == last_found)
2863                        p = p->next;
2864        }
2865        return 0;
2866}
2867
2868/*
2869 * Store a list of sha1s that are should not be discarded
2870 * because they are either written too recently, or are
2871 * reachable from another object that was.
2872 *
2873 * This is filled by get_object_list.
2874 */
2875static struct oid_array recent_objects;
2876
2877static int loosened_object_can_be_discarded(const struct object_id *oid,
2878                                            timestamp_t mtime)
2879{
2880        if (!unpack_unreachable_expiration)
2881                return 0;
2882        if (mtime > unpack_unreachable_expiration)
2883                return 0;
2884        if (oid_array_lookup(&recent_objects, oid) >= 0)
2885                return 0;
2886        return 1;
2887}
2888
2889static void loosen_unused_packed_objects(struct rev_info *revs)
2890{
2891        struct packed_git *p;
2892        uint32_t i;
2893        struct object_id oid;
2894
2895        for (p = get_packed_git(the_repository); p; p = p->next) {
2896                if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
2897                        continue;
2898
2899                if (open_pack_index(p))
2900                        die("cannot open pack index");
2901
2902                for (i = 0; i < p->num_objects; i++) {
2903                        nth_packed_object_oid(&oid, p, i);
2904                        if (!packlist_find(&to_pack, oid.hash, NULL) &&
2905                            !has_sha1_pack_kept_or_nonlocal(&oid) &&
2906                            !loosened_object_can_be_discarded(&oid, p->mtime))
2907                                if (force_object_loose(&oid, p->mtime))
2908                                        die("unable to force loose object");
2909                }
2910        }
2911}
2912
2913/*
2914 * This tracks any options which pack-reuse code expects to be on, or which a
2915 * reader of the pack might not understand, and which would therefore prevent
2916 * blind reuse of what we have on disk.
2917 */
2918static int pack_options_allow_reuse(void)
2919{
2920        return pack_to_stdout &&
2921               allow_ofs_delta &&
2922               !ignore_packed_keep_on_disk &&
2923               !ignore_packed_keep_in_core &&
2924               (!local || !have_non_local_packs) &&
2925               !incremental;
2926}
2927
2928static int get_object_list_from_bitmap(struct rev_info *revs)
2929{
2930        if (prepare_bitmap_walk(revs) < 0)
2931                return -1;
2932
2933        if (pack_options_allow_reuse() &&
2934            !reuse_partial_packfile_from_bitmap(
2935                        &reuse_packfile,
2936                        &reuse_packfile_objects,
2937                        &reuse_packfile_offset)) {
2938                assert(reuse_packfile_objects);
2939                nr_result += reuse_packfile_objects;
2940                display_progress(progress_state, nr_result);
2941        }
2942
2943        traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2944        return 0;
2945}
2946
2947static void record_recent_object(struct object *obj,
2948                                 const char *name,
2949                                 void *data)
2950{
2951        oid_array_append(&recent_objects, &obj->oid);
2952}
2953
2954static void record_recent_commit(struct commit *commit, void *data)
2955{
2956        oid_array_append(&recent_objects, &commit->object.oid);
2957}
2958
2959static void get_object_list(int ac, const char **av)
2960{
2961        struct rev_info revs;
2962        char line[1000];
2963        int flags = 0;
2964
2965        init_revisions(&revs, NULL);
2966        save_commit_buffer = 0;
2967        setup_revisions(ac, av, &revs, NULL);
2968
2969        /* make sure shallows are read */
2970        is_repository_shallow();
2971
2972        while (fgets(line, sizeof(line), stdin) != NULL) {
2973                int len = strlen(line);
2974                if (len && line[len - 1] == '\n')
2975                        line[--len] = 0;
2976                if (!len)
2977                        break;
2978                if (*line == '-') {
2979                        if (!strcmp(line, "--not")) {
2980                                flags ^= UNINTERESTING;
2981                                write_bitmap_index = 0;
2982                                continue;
2983                        }
2984                        if (starts_with(line, "--shallow ")) {
2985                                struct object_id oid;
2986                                if (get_oid_hex(line + 10, &oid))
2987                                        die("not an SHA-1 '%s'", line + 10);
2988                                register_shallow(&oid);
2989                                use_bitmap_index = 0;
2990                                continue;
2991                        }
2992                        die("not a rev '%s'", line);
2993                }
2994                if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
2995                        die("bad revision '%s'", line);
2996        }
2997
2998        if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
2999                return;
3000
3001        if (prepare_revision_walk(&revs))
3002                die("revision walk setup failed");
3003        mark_edges_uninteresting(&revs, show_edge);
3004
3005        if (!fn_show_object)
3006                fn_show_object = show_object;
3007        traverse_commit_list_filtered(&filter_options, &revs,
3008                                      show_commit, fn_show_object, NULL,
3009                                      NULL);
3010
3011        if (unpack_unreachable_expiration) {
3012                revs.ignore_missing_links = 1;
3013                if (add_unseen_recent_objects_to_traversal(&revs,
3014                                unpack_unreachable_expiration))
3015                        die("unable to add recent objects");
3016                if (prepare_revision_walk(&revs))
3017                        die("revision walk setup failed");
3018                traverse_commit_list(&revs, record_recent_commit,
3019                                     record_recent_object, NULL);
3020        }
3021
3022        if (keep_unreachable)
3023                add_objects_in_unpacked_packs(&revs);
3024        if (pack_loose_unreachable)
3025                add_unreachable_loose_objects();
3026        if (unpack_unreachable)
3027                loosen_unused_packed_objects(&revs);
3028
3029        oid_array_clear(&recent_objects);
3030}
3031
3032static void add_extra_kept_packs(const struct string_list *names)
3033{
3034        struct packed_git *p;
3035
3036        if (!names->nr)
3037                return;
3038
3039        for (p = get_packed_git(the_repository); p; p = p->next) {
3040                const char *name = basename(p->pack_name);
3041                int i;
3042
3043                if (!p->pack_local)
3044                        continue;
3045
3046                for (i = 0; i < names->nr; i++)
3047                        if (!fspathcmp(name, names->items[i].string))
3048                                break;
3049
3050                if (i < names->nr) {
3051                        p->pack_keep_in_core = 1;
3052                        ignore_packed_keep_in_core = 1;
3053                        continue;
3054                }
3055        }
3056}
3057
3058static int option_parse_index_version(const struct option *opt,
3059                                      const char *arg, int unset)
3060{
3061        char *c;
3062        const char *val = arg;
3063        pack_idx_opts.version = strtoul(val, &c, 10);
3064        if (pack_idx_opts.version > 2)
3065                die(_("unsupported index version %s"), val);
3066        if (*c == ',' && c[1])
3067                pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
3068        if (*c || pack_idx_opts.off32_limit & 0x80000000)
3069                die(_("bad index version '%s'"), val);
3070        return 0;
3071}
3072
3073static int option_parse_unpack_unreachable(const struct option *opt,
3074                                           const char *arg, int unset)
3075{
3076        if (unset) {
3077                unpack_unreachable = 0;
3078                unpack_unreachable_expiration = 0;
3079        }
3080        else {
3081                unpack_unreachable = 1;
3082                if (arg)
3083                        unpack_unreachable_expiration = approxidate(arg);
3084        }
3085        return 0;
3086}
3087
3088int cmd_pack_objects(int argc, const char **argv, const char *prefix)
3089{
3090        int use_internal_rev_list = 0;
3091        int thin = 0;
3092        int shallow = 0;
3093        int all_progress_implied = 0;
3094        struct argv_array rp = ARGV_ARRAY_INIT;
3095        int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
3096        int rev_list_index = 0;
3097        struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
3098        struct option pack_objects_options[] = {
3099                OPT_SET_INT('q', "quiet", &progress,
3100                            N_("do not show progress meter"), 0),
3101                OPT_SET_INT(0, "progress", &progress,
3102                            N_("show progress meter"), 1),
3103                OPT_SET_INT(0, "all-progress", &progress,
3104                            N_("show progress meter during object writing phase"), 2),
3105                OPT_BOOL(0, "all-progress-implied",
3106                         &all_progress_implied,
3107                         N_("similar to --all-progress when progress meter is shown")),
3108                { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
3109                  N_("write the pack index file in the specified idx format version"),
3110                  0, option_parse_index_version },
3111                OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
3112                              N_("maximum size of each output pack file")),
3113                OPT_BOOL(0, "local", &local,
3114                         N_("ignore borrowed objects from alternate object store")),
3115                OPT_BOOL(0, "incremental", &incremental,
3116                         N_("ignore packed objects")),
3117                OPT_INTEGER(0, "window", &window,
3118                            N_("limit pack window by objects")),
3119                OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
3120                              N_("limit pack window by memory in addition to object limit")),
3121                OPT_INTEGER(0, "depth", &depth,
3122                            N_("maximum length of delta chain allowed in the resulting pack")),
3123                OPT_BOOL(0, "reuse-delta", &reuse_delta,
3124                         N_("reuse existing deltas")),
3125                OPT_BOOL(0, "reuse-object", &reuse_object,
3126                         N_("reuse existing objects")),
3127                OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
3128                         N_("use OFS_DELTA objects")),
3129                OPT_INTEGER(0, "threads", &delta_search_threads,
3130                            N_("use threads when searching for best delta matches")),
3131                OPT_BOOL(0, "non-empty", &non_empty,
3132                         N_("do not create an empty pack output")),
3133                OPT_BOOL(0, "revs", &use_internal_rev_list,
3134                         N_("read revision arguments from standard input")),
3135                { OPTION_SET_INT, 0, "unpacked", &rev_list_unpacked, NULL,
3136                  N_("limit the objects to those that are not yet packed"),
3137                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3138                { OPTION_SET_INT, 0, "all", &rev_list_all, NULL,
3139                  N_("include objects reachable from any reference"),
3140                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3141                { OPTION_SET_INT, 0, "reflog", &rev_list_reflog, NULL,
3142                  N_("include objects referred by reflog entries"),
3143                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3144                { OPTION_SET_INT, 0, "indexed-objects", &rev_list_index, NULL,
3145                  N_("include objects referred to by the index"),
3146                  PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL, 1 },
3147                OPT_BOOL(0, "stdout", &pack_to_stdout,
3148                         N_("output pack to stdout")),
3149                OPT_BOOL(0, "include-tag", &include_tag,
3150                         N_("include tag objects that refer to objects to be packed")),
3151                OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3152                         N_("keep unreachable objects")),
3153                OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3154                         N_("pack loose unreachable objects")),
3155                { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3156                  N_("unpack unreachable objects newer than <time>"),
3157                  PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3158                OPT_BOOL(0, "thin", &thin,
3159                         N_("create thin packs")),
3160                OPT_BOOL(0, "shallow", &shallow,
3161                         N_("create packs suitable for shallow fetches")),
3162                OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,
3163                         N_("ignore packs that have companion .keep file")),
3164                OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
3165                                N_("ignore this pack")),
3166                OPT_INTEGER(0, "compression", &pack_compression_level,
3167                            N_("pack compression level")),
3168                OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3169                            N_("do not hide commits by grafts"), 0),
3170                OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3171                         N_("use a bitmap index if available to speed up counting objects")),
3172                OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3173                         N_("write a bitmap index together with the pack index")),
3174                OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3175                { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3176                  N_("handling for missing objects"), PARSE_OPT_NONEG,
3177                  option_parse_missing_action },
3178                OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3179                         N_("do not pack objects in promisor packfiles")),
3180                OPT_END(),
3181        };
3182
3183        if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))
3184                BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3185
3186        check_replace_refs = 0;
3187
3188        reset_pack_idx_option(&pack_idx_opts);
3189        git_config(git_pack_config, NULL);
3190
3191        progress = isatty(2);
3192        argc = parse_options(argc, argv, prefix, pack_objects_options,
3193                             pack_usage, 0);
3194
3195        if (argc) {
3196                base_name = argv[0];
3197                argc--;
3198        }
3199        if (pack_to_stdout != !base_name || argc)
3200                usage_with_options(pack_usage, pack_objects_options);
3201
3202        if (depth >= (1 << OE_DEPTH_BITS)) {
3203                warning(_("delta chain depth %d is too deep, forcing %d"),
3204                        depth, (1 << OE_DEPTH_BITS) - 1);
3205                depth = (1 << OE_DEPTH_BITS) - 1;
3206        }
3207        if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {
3208                warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3209                        (1U << OE_Z_DELTA_BITS) - 1);
3210                cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;
3211        }
3212
3213        argv_array_push(&rp, "pack-objects");
3214        if (thin) {
3215                use_internal_rev_list = 1;
3216                argv_array_push(&rp, shallow
3217                                ? "--objects-edge-aggressive"
3218                                : "--objects-edge");
3219        } else
3220                argv_array_push(&rp, "--objects");
3221
3222        if (rev_list_all) {
3223                use_internal_rev_list = 1;
3224                argv_array_push(&rp, "--all");
3225        }
3226        if (rev_list_reflog) {
3227                use_internal_rev_list = 1;
3228                argv_array_push(&rp, "--reflog");
3229        }
3230        if (rev_list_index) {
3231                use_internal_rev_list = 1;
3232                argv_array_push(&rp, "--indexed-objects");
3233        }
3234        if (rev_list_unpacked) {
3235                use_internal_rev_list = 1;
3236                argv_array_push(&rp, "--unpacked");
3237        }
3238
3239        if (exclude_promisor_objects) {
3240                use_internal_rev_list = 1;
3241                fetch_if_missing = 0;
3242                argv_array_push(&rp, "--exclude-promisor-objects");
3243        }
3244        if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)
3245                use_internal_rev_list = 1;
3246
3247        if (!reuse_object)
3248                reuse_delta = 0;
3249        if (pack_compression_level == -1)
3250                pack_compression_level = Z_DEFAULT_COMPRESSION;
3251        else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3252                die("bad pack compression level %d", pack_compression_level);
3253
3254        if (!delta_search_threads)      /* --threads=0 means autodetect */
3255                delta_search_threads = online_cpus();
3256
3257#ifdef NO_PTHREADS
3258        if (delta_search_threads != 1)
3259                warning("no threads support, ignoring --threads");
3260#endif
3261        if (!pack_to_stdout && !pack_size_limit)
3262                pack_size_limit = pack_size_limit_cfg;
3263        if (pack_to_stdout && pack_size_limit)
3264                die("--max-pack-size cannot be used to build a pack for transfer.");
3265        if (pack_size_limit && pack_size_limit < 1024*1024) {
3266                warning("minimum pack size limit is 1 MiB");
3267                pack_size_limit = 1024*1024;
3268        }
3269
3270        if (!pack_to_stdout && thin)
3271                die("--thin cannot be used to build an indexable pack.");
3272
3273        if (keep_unreachable && unpack_unreachable)
3274                die("--keep-unreachable and --unpack-unreachable are incompatible.");
3275        if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3276                unpack_unreachable_expiration = 0;
3277
3278        if (filter_options.choice) {
3279                if (!pack_to_stdout)
3280                        die("cannot use --filter without --stdout.");
3281                use_bitmap_index = 0;
3282        }
3283
3284        /*
3285         * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3286         *
3287         * - to produce good pack (with bitmap index not-yet-packed objects are
3288         *   packed in suboptimal order).
3289         *
3290         * - to use more robust pack-generation codepath (avoiding possible
3291         *   bugs in bitmap code and possible bitmap index corruption).
3292         */
3293        if (!pack_to_stdout)
3294                use_bitmap_index_default = 0;
3295
3296        if (use_bitmap_index < 0)
3297                use_bitmap_index = use_bitmap_index_default;
3298
3299        /* "hard" reasons not to use bitmaps; these just won't work at all */
3300        if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3301                use_bitmap_index = 0;
3302
3303        if (pack_to_stdout || !rev_list_all)
3304                write_bitmap_index = 0;
3305
3306        if (progress && all_progress_implied)
3307                progress = 2;
3308
3309        add_extra_kept_packs(&keep_pack_list);
3310        if (ignore_packed_keep_on_disk) {
3311                struct packed_git *p;
3312                for (p = get_packed_git(the_repository); p; p = p->next)
3313                        if (p->pack_local && p->pack_keep)
3314                                break;
3315                if (!p) /* no keep-able packs found */
3316                        ignore_packed_keep_on_disk = 0;
3317        }
3318        if (local) {
3319                /*
3320                 * unlike ignore_packed_keep_on_disk above, we do not
3321                 * want to unset "local" based on looking at packs, as
3322                 * it also covers non-local objects
3323                 */
3324                struct packed_git *p;
3325                for (p = get_packed_git(the_repository); p; p = p->next) {
3326                        if (!p->pack_local) {
3327                                have_non_local_packs = 1;
3328                                break;
3329                        }
3330                }
3331        }
3332
3333        prepare_packing_data(&to_pack);
3334
3335        if (progress)
3336                progress_state = start_progress(_("Enumerating objects"), 0);
3337        if (!use_internal_rev_list)
3338                read_object_list_from_stdin();
3339        else {
3340                get_object_list(rp.argc, rp.argv);
3341                argv_array_clear(&rp);
3342        }
3343        cleanup_preferred_base();
3344        if (include_tag && nr_result)
3345                for_each_ref(add_ref_tag, NULL);
3346        stop_progress(&progress_state);
3347
3348        if (non_empty && !nr_result)
3349                return 0;
3350        if (nr_result)
3351                prepare_pack(window, depth);
3352        write_pack_file();
3353        if (progress)
3354                fprintf(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3355                        " reused %"PRIu32" (delta %"PRIu32")\n",
3356                        written, written_delta, reused, reused_delta);
3357        return 0;
3358}