builtin / pack-objects.con commit builtin/config.c: mark more strings for translation (1d28ff4)
   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        /*
 152         * We succesfully computed this delta once but dropped it for
 153         * memory reasons. Something is very wrong if this time we
 154         * recompute and create a different delta.
 155         */
 156        if (!delta_buf || delta_size != DELTA_SIZE(entry))
 157                BUG("delta size changed");
 158        free(buf);
 159        free(base_buf);
 160        return delta_buf;
 161}
 162
 163static unsigned long do_compress(void **pptr, unsigned long size)
 164{
 165        git_zstream stream;
 166        void *in, *out;
 167        unsigned long maxsize;
 168
 169        git_deflate_init(&stream, pack_compression_level);
 170        maxsize = git_deflate_bound(&stream, size);
 171
 172        in = *pptr;
 173        out = xmalloc(maxsize);
 174        *pptr = out;
 175
 176        stream.next_in = in;
 177        stream.avail_in = size;
 178        stream.next_out = out;
 179        stream.avail_out = maxsize;
 180        while (git_deflate(&stream, Z_FINISH) == Z_OK)
 181                ; /* nothing */
 182        git_deflate_end(&stream);
 183
 184        free(in);
 185        return stream.total_out;
 186}
 187
 188static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f,
 189                                           const struct object_id *oid)
 190{
 191        git_zstream stream;
 192        unsigned char ibuf[1024 * 16];
 193        unsigned char obuf[1024 * 16];
 194        unsigned long olen = 0;
 195
 196        git_deflate_init(&stream, pack_compression_level);
 197
 198        for (;;) {
 199                ssize_t readlen;
 200                int zret = Z_OK;
 201                readlen = read_istream(st, ibuf, sizeof(ibuf));
 202                if (readlen == -1)
 203                        die(_("unable to read %s"), oid_to_hex(oid));
 204
 205                stream.next_in = ibuf;
 206                stream.avail_in = readlen;
 207                while ((stream.avail_in || readlen == 0) &&
 208                       (zret == Z_OK || zret == Z_BUF_ERROR)) {
 209                        stream.next_out = obuf;
 210                        stream.avail_out = sizeof(obuf);
 211                        zret = git_deflate(&stream, readlen ? 0 : Z_FINISH);
 212                        hashwrite(f, obuf, stream.next_out - obuf);
 213                        olen += stream.next_out - obuf;
 214                }
 215                if (stream.avail_in)
 216                        die(_("deflate error (%d)"), zret);
 217                if (readlen == 0) {
 218                        if (zret != Z_STREAM_END)
 219                                die(_("deflate error (%d)"), zret);
 220                        break;
 221                }
 222        }
 223        git_deflate_end(&stream);
 224        return olen;
 225}
 226
 227/*
 228 * we are going to reuse the existing object data as is.  make
 229 * sure it is not corrupt.
 230 */
 231static int check_pack_inflate(struct packed_git *p,
 232                struct pack_window **w_curs,
 233                off_t offset,
 234                off_t len,
 235                unsigned long expect)
 236{
 237        git_zstream stream;
 238        unsigned char fakebuf[4096], *in;
 239        int st;
 240
 241        memset(&stream, 0, sizeof(stream));
 242        git_inflate_init(&stream);
 243        do {
 244                in = use_pack(p, w_curs, offset, &stream.avail_in);
 245                stream.next_in = in;
 246                stream.next_out = fakebuf;
 247                stream.avail_out = sizeof(fakebuf);
 248                st = git_inflate(&stream, Z_FINISH);
 249                offset += stream.next_in - in;
 250        } while (st == Z_OK || st == Z_BUF_ERROR);
 251        git_inflate_end(&stream);
 252        return (st == Z_STREAM_END &&
 253                stream.total_out == expect &&
 254                stream.total_in == len) ? 0 : -1;
 255}
 256
 257static void copy_pack_data(struct hashfile *f,
 258                struct packed_git *p,
 259                struct pack_window **w_curs,
 260                off_t offset,
 261                off_t len)
 262{
 263        unsigned char *in;
 264        unsigned long avail;
 265
 266        while (len) {
 267                in = use_pack(p, w_curs, offset, &avail);
 268                if (avail > len)
 269                        avail = (unsigned long)len;
 270                hashwrite(f, in, avail);
 271                offset += avail;
 272                len -= avail;
 273        }
 274}
 275
 276/* Return 0 if we will bust the pack-size limit */
 277static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry,
 278                                           unsigned long limit, int usable_delta)
 279{
 280        unsigned long size, datalen;
 281        unsigned char header[MAX_PACK_OBJECT_HEADER],
 282                      dheader[MAX_PACK_OBJECT_HEADER];
 283        unsigned hdrlen;
 284        enum object_type type;
 285        void *buf;
 286        struct git_istream *st = NULL;
 287        const unsigned hashsz = the_hash_algo->rawsz;
 288
 289        if (!usable_delta) {
 290                if (oe_type(entry) == OBJ_BLOB &&
 291                    oe_size_greater_than(&to_pack, entry, big_file_threshold) &&
 292                    (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL)
 293                        buf = NULL;
 294                else {
 295                        buf = read_object_file(&entry->idx.oid, &type, &size);
 296                        if (!buf)
 297                                die(_("unable to read %s"),
 298                                    oid_to_hex(&entry->idx.oid));
 299                }
 300                /*
 301                 * make sure no cached delta data remains from a
 302                 * previous attempt before a pack split occurred.
 303                 */
 304                FREE_AND_NULL(entry->delta_data);
 305                entry->z_delta_size = 0;
 306        } else if (entry->delta_data) {
 307                size = DELTA_SIZE(entry);
 308                buf = entry->delta_data;
 309                entry->delta_data = NULL;
 310                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 311                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 312        } else {
 313                buf = get_delta(entry);
 314                size = DELTA_SIZE(entry);
 315                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 316                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 317        }
 318
 319        if (st) /* large blob case, just assume we don't compress well */
 320                datalen = size;
 321        else if (entry->z_delta_size)
 322                datalen = entry->z_delta_size;
 323        else
 324                datalen = do_compress(&buf, size);
 325
 326        /*
 327         * The object header is a byte of 'type' followed by zero or
 328         * more bytes of length.
 329         */
 330        hdrlen = encode_in_pack_object_header(header, sizeof(header),
 331                                              type, size);
 332
 333        if (type == OBJ_OFS_DELTA) {
 334                /*
 335                 * Deltas with relative base contain an additional
 336                 * encoding of the relative offset for the delta
 337                 * base from this object's position in the pack.
 338                 */
 339                off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
 340                unsigned pos = sizeof(dheader) - 1;
 341                dheader[pos] = ofs & 127;
 342                while (ofs >>= 7)
 343                        dheader[--pos] = 128 | (--ofs & 127);
 344                if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
 345                        if (st)
 346                                close_istream(st);
 347                        free(buf);
 348                        return 0;
 349                }
 350                hashwrite(f, header, hdrlen);
 351                hashwrite(f, dheader + pos, sizeof(dheader) - pos);
 352                hdrlen += sizeof(dheader) - pos;
 353        } else if (type == OBJ_REF_DELTA) {
 354                /*
 355                 * Deltas with a base reference contain
 356                 * additional bytes for the base object ID.
 357                 */
 358                if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
 359                        if (st)
 360                                close_istream(st);
 361                        free(buf);
 362                        return 0;
 363                }
 364                hashwrite(f, header, hdrlen);
 365                hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
 366                hdrlen += hashsz;
 367        } else {
 368                if (limit && hdrlen + datalen + hashsz >= limit) {
 369                        if (st)
 370                                close_istream(st);
 371                        free(buf);
 372                        return 0;
 373                }
 374                hashwrite(f, header, hdrlen);
 375        }
 376        if (st) {
 377                datalen = write_large_blob_data(st, f, &entry->idx.oid);
 378                close_istream(st);
 379        } else {
 380                hashwrite(f, buf, datalen);
 381                free(buf);
 382        }
 383
 384        return hdrlen + datalen;
 385}
 386
 387/* Return 0 if we will bust the pack-size limit */
 388static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry,
 389                                unsigned long limit, int usable_delta)
 390{
 391        struct packed_git *p = IN_PACK(entry);
 392        struct pack_window *w_curs = NULL;
 393        struct revindex_entry *revidx;
 394        off_t offset;
 395        enum object_type type = oe_type(entry);
 396        off_t datalen;
 397        unsigned char header[MAX_PACK_OBJECT_HEADER],
 398                      dheader[MAX_PACK_OBJECT_HEADER];
 399        unsigned hdrlen;
 400        const unsigned hashsz = the_hash_algo->rawsz;
 401        unsigned long entry_size = SIZE(entry);
 402
 403        if (DELTA(entry))
 404                type = (allow_ofs_delta && DELTA(entry)->idx.offset) ?
 405                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 406        hdrlen = encode_in_pack_object_header(header, sizeof(header),
 407                                              type, entry_size);
 408
 409        offset = entry->in_pack_offset;
 410        revidx = find_pack_revindex(p, offset);
 411        datalen = revidx[1].offset - offset;
 412        if (!pack_to_stdout && p->index_version > 1 &&
 413            check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) {
 414                error("bad packed object CRC for %s",
 415                      oid_to_hex(&entry->idx.oid));
 416                unuse_pack(&w_curs);
 417                return write_no_reuse_object(f, entry, limit, usable_delta);
 418        }
 419
 420        offset += entry->in_pack_header_size;
 421        datalen -= entry->in_pack_header_size;
 422
 423        if (!pack_to_stdout && p->index_version == 1 &&
 424            check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) {
 425                error("corrupt packed object for %s",
 426                      oid_to_hex(&entry->idx.oid));
 427                unuse_pack(&w_curs);
 428                return write_no_reuse_object(f, entry, limit, usable_delta);
 429        }
 430
 431        if (type == OBJ_OFS_DELTA) {
 432                off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset;
 433                unsigned pos = sizeof(dheader) - 1;
 434                dheader[pos] = ofs & 127;
 435                while (ofs >>= 7)
 436                        dheader[--pos] = 128 | (--ofs & 127);
 437                if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) {
 438                        unuse_pack(&w_curs);
 439                        return 0;
 440                }
 441                hashwrite(f, header, hdrlen);
 442                hashwrite(f, dheader + pos, sizeof(dheader) - pos);
 443                hdrlen += sizeof(dheader) - pos;
 444                reused_delta++;
 445        } else if (type == OBJ_REF_DELTA) {
 446                if (limit && hdrlen + hashsz + datalen + hashsz >= limit) {
 447                        unuse_pack(&w_curs);
 448                        return 0;
 449                }
 450                hashwrite(f, header, hdrlen);
 451                hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz);
 452                hdrlen += hashsz;
 453                reused_delta++;
 454        } else {
 455                if (limit && hdrlen + datalen + hashsz >= limit) {
 456                        unuse_pack(&w_curs);
 457                        return 0;
 458                }
 459                hashwrite(f, header, hdrlen);
 460        }
 461        copy_pack_data(f, p, &w_curs, offset, datalen);
 462        unuse_pack(&w_curs);
 463        reused++;
 464        return hdrlen + datalen;
 465}
 466
 467/* Return 0 if we will bust the pack-size limit */
 468static off_t write_object(struct hashfile *f,
 469                          struct object_entry *entry,
 470                          off_t write_offset)
 471{
 472        unsigned long limit;
 473        off_t len;
 474        int usable_delta, to_reuse;
 475
 476        if (!pack_to_stdout)
 477                crc32_begin(f);
 478
 479        /* apply size limit if limited packsize and not first object */
 480        if (!pack_size_limit || !nr_written)
 481                limit = 0;
 482        else if (pack_size_limit <= write_offset)
 483                /*
 484                 * the earlier object did not fit the limit; avoid
 485                 * mistaking this with unlimited (i.e. limit = 0).
 486                 */
 487                limit = 1;
 488        else
 489                limit = pack_size_limit - write_offset;
 490
 491        if (!DELTA(entry))
 492                usable_delta = 0;       /* no delta */
 493        else if (!pack_size_limit)
 494               usable_delta = 1;        /* unlimited packfile */
 495        else if (DELTA(entry)->idx.offset == (off_t)-1)
 496                usable_delta = 0;       /* base was written to another pack */
 497        else if (DELTA(entry)->idx.offset)
 498                usable_delta = 1;       /* base already exists in this pack */
 499        else
 500                usable_delta = 0;       /* base could end up in another pack */
 501
 502        if (!reuse_object)
 503                to_reuse = 0;   /* explicit */
 504        else if (!IN_PACK(entry))
 505                to_reuse = 0;   /* can't reuse what we don't have */
 506        else if (oe_type(entry) == OBJ_REF_DELTA ||
 507                 oe_type(entry) == OBJ_OFS_DELTA)
 508                                /* check_object() decided it for us ... */
 509                to_reuse = usable_delta;
 510                                /* ... but pack split may override that */
 511        else if (oe_type(entry) != entry->in_pack_type)
 512                to_reuse = 0;   /* pack has delta which is unusable */
 513        else if (DELTA(entry))
 514                to_reuse = 0;   /* we want to pack afresh */
 515        else
 516                to_reuse = 1;   /* we have it in-pack undeltified,
 517                                 * and we do not need to deltify it.
 518                                 */
 519
 520        if (!to_reuse)
 521                len = write_no_reuse_object(f, entry, limit, usable_delta);
 522        else
 523                len = write_reuse_object(f, entry, limit, usable_delta);
 524        if (!len)
 525                return 0;
 526
 527        if (usable_delta)
 528                written_delta++;
 529        written++;
 530        if (!pack_to_stdout)
 531                entry->idx.crc32 = crc32_end(f);
 532        return len;
 533}
 534
 535enum write_one_status {
 536        WRITE_ONE_SKIP = -1, /* already written */
 537        WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */
 538        WRITE_ONE_WRITTEN = 1, /* normal */
 539        WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */
 540};
 541
 542static enum write_one_status write_one(struct hashfile *f,
 543                                       struct object_entry *e,
 544                                       off_t *offset)
 545{
 546        off_t size;
 547        int recursing;
 548
 549        /*
 550         * we set offset to 1 (which is an impossible value) to mark
 551         * the fact that this object is involved in "write its base
 552         * first before writing a deltified object" recursion.
 553         */
 554        recursing = (e->idx.offset == 1);
 555        if (recursing) {
 556                warning("recursive delta detected for object %s",
 557                        oid_to_hex(&e->idx.oid));
 558                return WRITE_ONE_RECURSIVE;
 559        } else if (e->idx.offset || e->preferred_base) {
 560                /* offset is non zero if object is written already. */
 561                return WRITE_ONE_SKIP;
 562        }
 563
 564        /* if we are deltified, write out base object first. */
 565        if (DELTA(e)) {
 566                e->idx.offset = 1; /* now recurse */
 567                switch (write_one(f, DELTA(e), offset)) {
 568                case WRITE_ONE_RECURSIVE:
 569                        /* we cannot depend on this one */
 570                        SET_DELTA(e, NULL);
 571                        break;
 572                default:
 573                        break;
 574                case WRITE_ONE_BREAK:
 575                        e->idx.offset = recursing;
 576                        return WRITE_ONE_BREAK;
 577                }
 578        }
 579
 580        e->idx.offset = *offset;
 581        size = write_object(f, e, *offset);
 582        if (!size) {
 583                e->idx.offset = recursing;
 584                return WRITE_ONE_BREAK;
 585        }
 586        written_list[nr_written++] = &e->idx;
 587
 588        /* make sure off_t is sufficiently large not to wrap */
 589        if (signed_add_overflows(*offset, size))
 590                die("pack too large for current definition of off_t");
 591        *offset += size;
 592        return WRITE_ONE_WRITTEN;
 593}
 594
 595static int mark_tagged(const char *path, const struct object_id *oid, int flag,
 596                       void *cb_data)
 597{
 598        struct object_id peeled;
 599        struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL);
 600
 601        if (entry)
 602                entry->tagged = 1;
 603        if (!peel_ref(path, &peeled)) {
 604                entry = packlist_find(&to_pack, peeled.hash, NULL);
 605                if (entry)
 606                        entry->tagged = 1;
 607        }
 608        return 0;
 609}
 610
 611static inline void add_to_write_order(struct object_entry **wo,
 612                               unsigned int *endp,
 613                               struct object_entry *e)
 614{
 615        if (e->filled)
 616                return;
 617        wo[(*endp)++] = e;
 618        e->filled = 1;
 619}
 620
 621static void add_descendants_to_write_order(struct object_entry **wo,
 622                                           unsigned int *endp,
 623                                           struct object_entry *e)
 624{
 625        int add_to_order = 1;
 626        while (e) {
 627                if (add_to_order) {
 628                        struct object_entry *s;
 629                        /* add this node... */
 630                        add_to_write_order(wo, endp, e);
 631                        /* all its siblings... */
 632                        for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) {
 633                                add_to_write_order(wo, endp, s);
 634                        }
 635                }
 636                /* drop down a level to add left subtree nodes if possible */
 637                if (DELTA_CHILD(e)) {
 638                        add_to_order = 1;
 639                        e = DELTA_CHILD(e);
 640                } else {
 641                        add_to_order = 0;
 642                        /* our sibling might have some children, it is next */
 643                        if (DELTA_SIBLING(e)) {
 644                                e = DELTA_SIBLING(e);
 645                                continue;
 646                        }
 647                        /* go back to our parent node */
 648                        e = DELTA(e);
 649                        while (e && !DELTA_SIBLING(e)) {
 650                                /* we're on the right side of a subtree, keep
 651                                 * going up until we can go right again */
 652                                e = DELTA(e);
 653                        }
 654                        if (!e) {
 655                                /* done- we hit our original root node */
 656                                return;
 657                        }
 658                        /* pass it off to sibling at this level */
 659                        e = DELTA_SIBLING(e);
 660                }
 661        };
 662}
 663
 664static void add_family_to_write_order(struct object_entry **wo,
 665                                      unsigned int *endp,
 666                                      struct object_entry *e)
 667{
 668        struct object_entry *root;
 669
 670        for (root = e; DELTA(root); root = DELTA(root))
 671                ; /* nothing */
 672        add_descendants_to_write_order(wo, endp, root);
 673}
 674
 675static struct object_entry **compute_write_order(void)
 676{
 677        unsigned int i, wo_end, last_untagged;
 678
 679        struct object_entry **wo;
 680        struct object_entry *objects = to_pack.objects;
 681
 682        for (i = 0; i < to_pack.nr_objects; i++) {
 683                objects[i].tagged = 0;
 684                objects[i].filled = 0;
 685                SET_DELTA_CHILD(&objects[i], NULL);
 686                SET_DELTA_SIBLING(&objects[i], NULL);
 687        }
 688
 689        /*
 690         * Fully connect delta_child/delta_sibling network.
 691         * Make sure delta_sibling is sorted in the original
 692         * recency order.
 693         */
 694        for (i = to_pack.nr_objects; i > 0;) {
 695                struct object_entry *e = &objects[--i];
 696                if (!DELTA(e))
 697                        continue;
 698                /* Mark me as the first child */
 699                e->delta_sibling_idx = DELTA(e)->delta_child_idx;
 700                SET_DELTA_CHILD(DELTA(e), e);
 701        }
 702
 703        /*
 704         * Mark objects that are at the tip of tags.
 705         */
 706        for_each_tag_ref(mark_tagged, NULL);
 707
 708        /*
 709         * Give the objects in the original recency order until
 710         * we see a tagged tip.
 711         */
 712        ALLOC_ARRAY(wo, to_pack.nr_objects);
 713        for (i = wo_end = 0; i < to_pack.nr_objects; i++) {
 714                if (objects[i].tagged)
 715                        break;
 716                add_to_write_order(wo, &wo_end, &objects[i]);
 717        }
 718        last_untagged = i;
 719
 720        /*
 721         * Then fill all the tagged tips.
 722         */
 723        for (; i < to_pack.nr_objects; i++) {
 724                if (objects[i].tagged)
 725                        add_to_write_order(wo, &wo_end, &objects[i]);
 726        }
 727
 728        /*
 729         * And then all remaining commits and tags.
 730         */
 731        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 732                if (oe_type(&objects[i]) != OBJ_COMMIT &&
 733                    oe_type(&objects[i]) != OBJ_TAG)
 734                        continue;
 735                add_to_write_order(wo, &wo_end, &objects[i]);
 736        }
 737
 738        /*
 739         * And then all the trees.
 740         */
 741        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 742                if (oe_type(&objects[i]) != OBJ_TREE)
 743                        continue;
 744                add_to_write_order(wo, &wo_end, &objects[i]);
 745        }
 746
 747        /*
 748         * Finally all the rest in really tight order
 749         */
 750        for (i = last_untagged; i < to_pack.nr_objects; i++) {
 751                if (!objects[i].filled)
 752                        add_family_to_write_order(wo, &wo_end, &objects[i]);
 753        }
 754
 755        if (wo_end != to_pack.nr_objects)
 756                die("ordered %u objects, expected %"PRIu32, wo_end, to_pack.nr_objects);
 757
 758        return wo;
 759}
 760
 761static off_t write_reused_pack(struct hashfile *f)
 762{
 763        unsigned char buffer[8192];
 764        off_t to_write, total;
 765        int fd;
 766
 767        if (!is_pack_valid(reuse_packfile))
 768                die("packfile is invalid: %s", reuse_packfile->pack_name);
 769
 770        fd = git_open(reuse_packfile->pack_name);
 771        if (fd < 0)
 772                die_errno("unable to open packfile for reuse: %s",
 773                          reuse_packfile->pack_name);
 774
 775        if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1)
 776                die_errno("unable to seek in reused packfile");
 777
 778        if (reuse_packfile_offset < 0)
 779                reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz;
 780
 781        total = to_write = reuse_packfile_offset - sizeof(struct pack_header);
 782
 783        while (to_write) {
 784                int read_pack = xread(fd, buffer, sizeof(buffer));
 785
 786                if (read_pack <= 0)
 787                        die_errno("unable to read from reused packfile");
 788
 789                if (read_pack > to_write)
 790                        read_pack = to_write;
 791
 792                hashwrite(f, buffer, read_pack);
 793                to_write -= read_pack;
 794
 795                /*
 796                 * We don't know the actual number of objects written,
 797                 * only how many bytes written, how many bytes total, and
 798                 * how many objects total. So we can fake it by pretending all
 799                 * objects we are writing are the same size. This gives us a
 800                 * smooth progress meter, and at the end it matches the true
 801                 * answer.
 802                 */
 803                written = reuse_packfile_objects *
 804                                (((double)(total - to_write)) / total);
 805                display_progress(progress_state, written);
 806        }
 807
 808        close(fd);
 809        written = reuse_packfile_objects;
 810        display_progress(progress_state, written);
 811        return reuse_packfile_offset - sizeof(struct pack_header);
 812}
 813
 814static const char no_split_warning[] = N_(
 815"disabling bitmap writing, packs are split due to pack.packSizeLimit"
 816);
 817
 818static void write_pack_file(void)
 819{
 820        uint32_t i = 0, j;
 821        struct hashfile *f;
 822        off_t offset;
 823        uint32_t nr_remaining = nr_result;
 824        time_t last_mtime = 0;
 825        struct object_entry **write_order;
 826
 827        if (progress > pack_to_stdout)
 828                progress_state = start_progress(_("Writing objects"), nr_result);
 829        ALLOC_ARRAY(written_list, to_pack.nr_objects);
 830        write_order = compute_write_order();
 831
 832        do {
 833                struct object_id oid;
 834                char *pack_tmp_name = NULL;
 835
 836                if (pack_to_stdout)
 837                        f = hashfd_throughput(1, "<stdout>", progress_state);
 838                else
 839                        f = create_tmp_packfile(&pack_tmp_name);
 840
 841                offset = write_pack_header(f, nr_remaining);
 842
 843                if (reuse_packfile) {
 844                        off_t packfile_size;
 845                        assert(pack_to_stdout);
 846
 847                        packfile_size = write_reused_pack(f);
 848                        offset += packfile_size;
 849                }
 850
 851                nr_written = 0;
 852                for (; i < to_pack.nr_objects; i++) {
 853                        struct object_entry *e = write_order[i];
 854                        if (write_one(f, e, &offset) == WRITE_ONE_BREAK)
 855                                break;
 856                        display_progress(progress_state, written);
 857                }
 858
 859                /*
 860                 * Did we write the wrong # entries in the header?
 861                 * If so, rewrite it like in fast-import
 862                 */
 863                if (pack_to_stdout) {
 864                        finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE);
 865                } else if (nr_written == nr_remaining) {
 866                        finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE);
 867                } else {
 868                        int fd = finalize_hashfile(f, oid.hash, 0);
 869                        fixup_pack_header_footer(fd, oid.hash, pack_tmp_name,
 870                                                 nr_written, oid.hash, offset);
 871                        close(fd);
 872                        if (write_bitmap_index) {
 873                                warning(_(no_split_warning));
 874                                write_bitmap_index = 0;
 875                        }
 876                }
 877
 878                if (!pack_to_stdout) {
 879                        struct stat st;
 880                        struct strbuf tmpname = STRBUF_INIT;
 881
 882                        /*
 883                         * Packs are runtime accessed in their mtime
 884                         * order since newer packs are more likely to contain
 885                         * younger objects.  So if we are creating multiple
 886                         * packs then we should modify the mtime of later ones
 887                         * to preserve this property.
 888                         */
 889                        if (stat(pack_tmp_name, &st) < 0) {
 890                                warning_errno("failed to stat %s", pack_tmp_name);
 891                        } else if (!last_mtime) {
 892                                last_mtime = st.st_mtime;
 893                        } else {
 894                                struct utimbuf utb;
 895                                utb.actime = st.st_atime;
 896                                utb.modtime = --last_mtime;
 897                                if (utime(pack_tmp_name, &utb) < 0)
 898                                        warning_errno("failed utime() on %s", pack_tmp_name);
 899                        }
 900
 901                        strbuf_addf(&tmpname, "%s-", base_name);
 902
 903                        if (write_bitmap_index) {
 904                                bitmap_writer_set_checksum(oid.hash);
 905                                bitmap_writer_build_type_index(
 906                                        &to_pack, written_list, nr_written);
 907                        }
 908
 909                        finish_tmp_packfile(&tmpname, pack_tmp_name,
 910                                            written_list, nr_written,
 911                                            &pack_idx_opts, oid.hash);
 912
 913                        if (write_bitmap_index) {
 914                                strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid));
 915
 916                                stop_progress(&progress_state);
 917
 918                                bitmap_writer_show_progress(progress);
 919                                bitmap_writer_reuse_bitmaps(&to_pack);
 920                                bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1);
 921                                bitmap_writer_build(&to_pack);
 922                                bitmap_writer_finish(written_list, nr_written,
 923                                                     tmpname.buf, write_bitmap_options);
 924                                write_bitmap_index = 0;
 925                        }
 926
 927                        strbuf_release(&tmpname);
 928                        free(pack_tmp_name);
 929                        puts(oid_to_hex(&oid));
 930                }
 931
 932                /* mark written objects as written to previous pack */
 933                for (j = 0; j < nr_written; j++) {
 934                        written_list[j]->offset = (off_t)-1;
 935                }
 936                nr_remaining -= nr_written;
 937        } while (nr_remaining && i < to_pack.nr_objects);
 938
 939        free(written_list);
 940        free(write_order);
 941        stop_progress(&progress_state);
 942        if (written != nr_result)
 943                die("wrote %"PRIu32" objects while expecting %"PRIu32,
 944                        written, nr_result);
 945}
 946
 947static int no_try_delta(const char *path)
 948{
 949        static struct attr_check *check;
 950
 951        if (!check)
 952                check = attr_check_initl("delta", NULL);
 953        if (git_check_attr(path, check))
 954                return 0;
 955        if (ATTR_FALSE(check->items[0].value))
 956                return 1;
 957        return 0;
 958}
 959
 960/*
 961 * When adding an object, check whether we have already added it
 962 * to our packing list. If so, we can skip. However, if we are
 963 * being asked to excludei t, but the previous mention was to include
 964 * it, make sure to adjust its flags and tweak our numbers accordingly.
 965 *
 966 * As an optimization, we pass out the index position where we would have
 967 * found the item, since that saves us from having to look it up again a
 968 * few lines later when we want to add the new entry.
 969 */
 970static int have_duplicate_entry(const struct object_id *oid,
 971                                int exclude,
 972                                uint32_t *index_pos)
 973{
 974        struct object_entry *entry;
 975
 976        entry = packlist_find(&to_pack, oid->hash, index_pos);
 977        if (!entry)
 978                return 0;
 979
 980        if (exclude) {
 981                if (!entry->preferred_base)
 982                        nr_result--;
 983                entry->preferred_base = 1;
 984        }
 985
 986        return 1;
 987}
 988
 989static int want_found_object(int exclude, struct packed_git *p)
 990{
 991        if (exclude)
 992                return 1;
 993        if (incremental)
 994                return 0;
 995
 996        /*
 997         * When asked to do --local (do not include an object that appears in a
 998         * pack we borrow from elsewhere) or --honor-pack-keep (do not include
 999         * an object that appears in a pack marked with .keep), finding a pack
1000         * that matches the criteria is sufficient for us to decide to omit it.
1001         * However, even if this pack does not satisfy the criteria, we need to
1002         * make sure no copy of this object appears in _any_ pack that makes us
1003         * to omit the object, so we need to check all the packs.
1004         *
1005         * We can however first check whether these options can possible matter;
1006         * if they do not matter we know we want the object in generated pack.
1007         * Otherwise, we signal "-1" at the end to tell the caller that we do
1008         * not know either way, and it needs to check more packs.
1009         */
1010        if (!ignore_packed_keep_on_disk &&
1011            !ignore_packed_keep_in_core &&
1012            (!local || !have_non_local_packs))
1013                return 1;
1014
1015        if (local && !p->pack_local)
1016                return 0;
1017        if (p->pack_local &&
1018            ((ignore_packed_keep_on_disk && p->pack_keep) ||
1019             (ignore_packed_keep_in_core && p->pack_keep_in_core)))
1020                return 0;
1021
1022        /* we don't know yet; keep looking for more packs */
1023        return -1;
1024}
1025
1026/*
1027 * Check whether we want the object in the pack (e.g., we do not want
1028 * objects found in non-local stores if the "--local" option was used).
1029 *
1030 * If the caller already knows an existing pack it wants to take the object
1031 * from, that is passed in *found_pack and *found_offset; otherwise this
1032 * function finds if there is any pack that has the object and returns the pack
1033 * and its offset in these variables.
1034 */
1035static int want_object_in_pack(const struct object_id *oid,
1036                               int exclude,
1037                               struct packed_git **found_pack,
1038                               off_t *found_offset)
1039{
1040        int want;
1041        struct list_head *pos;
1042
1043        if (!exclude && local && has_loose_object_nonlocal(oid))
1044                return 0;
1045
1046        /*
1047         * If we already know the pack object lives in, start checks from that
1048         * pack - in the usual case when neither --local was given nor .keep files
1049         * are present we will determine the answer right now.
1050         */
1051        if (*found_pack) {
1052                want = want_found_object(exclude, *found_pack);
1053                if (want != -1)
1054                        return want;
1055        }
1056        list_for_each(pos, get_packed_git_mru(the_repository)) {
1057                struct packed_git *p = list_entry(pos, struct packed_git, mru);
1058                off_t offset;
1059
1060                if (p == *found_pack)
1061                        offset = *found_offset;
1062                else
1063                        offset = find_pack_entry_one(oid->hash, p);
1064
1065                if (offset) {
1066                        if (!*found_pack) {
1067                                if (!is_pack_valid(p))
1068                                        continue;
1069                                *found_offset = offset;
1070                                *found_pack = p;
1071                        }
1072                        want = want_found_object(exclude, p);
1073                        if (!exclude && want > 0)
1074                                list_move(&p->mru,
1075                                          get_packed_git_mru(the_repository));
1076                        if (want != -1)
1077                                return want;
1078                }
1079        }
1080
1081        return 1;
1082}
1083
1084static void create_object_entry(const struct object_id *oid,
1085                                enum object_type type,
1086                                uint32_t hash,
1087                                int exclude,
1088                                int no_try_delta,
1089                                uint32_t index_pos,
1090                                struct packed_git *found_pack,
1091                                off_t found_offset)
1092{
1093        struct object_entry *entry;
1094
1095        entry = packlist_alloc(&to_pack, oid->hash, index_pos);
1096        entry->hash = hash;
1097        oe_set_type(entry, type);
1098        if (exclude)
1099                entry->preferred_base = 1;
1100        else
1101                nr_result++;
1102        if (found_pack) {
1103                oe_set_in_pack(&to_pack, entry, found_pack);
1104                entry->in_pack_offset = found_offset;
1105        }
1106
1107        entry->no_try_delta = no_try_delta;
1108}
1109
1110static const char no_closure_warning[] = N_(
1111"disabling bitmap writing, as some objects are not being packed"
1112);
1113
1114static int add_object_entry(const struct object_id *oid, enum object_type type,
1115                            const char *name, int exclude)
1116{
1117        struct packed_git *found_pack = NULL;
1118        off_t found_offset = 0;
1119        uint32_t index_pos;
1120
1121        display_progress(progress_state, ++nr_seen);
1122
1123        if (have_duplicate_entry(oid, exclude, &index_pos))
1124                return 0;
1125
1126        if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {
1127                /* The pack is missing an object, so it will not have closure */
1128                if (write_bitmap_index) {
1129                        warning(_(no_closure_warning));
1130                        write_bitmap_index = 0;
1131                }
1132                return 0;
1133        }
1134
1135        create_object_entry(oid, type, pack_name_hash(name),
1136                            exclude, name && no_try_delta(name),
1137                            index_pos, found_pack, found_offset);
1138        return 1;
1139}
1140
1141static int add_object_entry_from_bitmap(const struct object_id *oid,
1142                                        enum object_type type,
1143                                        int flags, uint32_t name_hash,
1144                                        struct packed_git *pack, off_t offset)
1145{
1146        uint32_t index_pos;
1147
1148        display_progress(progress_state, ++nr_seen);
1149
1150        if (have_duplicate_entry(oid, 0, &index_pos))
1151                return 0;
1152
1153        if (!want_object_in_pack(oid, 0, &pack, &offset))
1154                return 0;
1155
1156        create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);
1157        return 1;
1158}
1159
1160struct pbase_tree_cache {
1161        struct object_id oid;
1162        int ref;
1163        int temporary;
1164        void *tree_data;
1165        unsigned long tree_size;
1166};
1167
1168static struct pbase_tree_cache *(pbase_tree_cache[256]);
1169static int pbase_tree_cache_ix(const struct object_id *oid)
1170{
1171        return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);
1172}
1173static int pbase_tree_cache_ix_incr(int ix)
1174{
1175        return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1176}
1177
1178static struct pbase_tree {
1179        struct pbase_tree *next;
1180        /* This is a phony "cache" entry; we are not
1181         * going to evict it or find it through _get()
1182         * mechanism -- this is for the toplevel node that
1183         * would almost always change with any commit.
1184         */
1185        struct pbase_tree_cache pcache;
1186} *pbase_tree;
1187
1188static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)
1189{
1190        struct pbase_tree_cache *ent, *nent;
1191        void *data;
1192        unsigned long size;
1193        enum object_type type;
1194        int neigh;
1195        int my_ix = pbase_tree_cache_ix(oid);
1196        int available_ix = -1;
1197
1198        /* pbase-tree-cache acts as a limited hashtable.
1199         * your object will be found at your index or within a few
1200         * slots after that slot if it is cached.
1201         */
1202        for (neigh = 0; neigh < 8; neigh++) {
1203                ent = pbase_tree_cache[my_ix];
1204                if (ent && !oidcmp(&ent->oid, oid)) {
1205                        ent->ref++;
1206                        return ent;
1207                }
1208                else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1209                         ((0 <= available_ix) &&
1210                          (!ent && pbase_tree_cache[available_ix])))
1211                        available_ix = my_ix;
1212                if (!ent)
1213                        break;
1214                my_ix = pbase_tree_cache_ix_incr(my_ix);
1215        }
1216
1217        /* Did not find one.  Either we got a bogus request or
1218         * we need to read and perhaps cache.
1219         */
1220        data = read_object_file(oid, &type, &size);
1221        if (!data)
1222                return NULL;
1223        if (type != OBJ_TREE) {
1224                free(data);
1225                return NULL;
1226        }
1227
1228        /* We need to either cache or return a throwaway copy */
1229
1230        if (available_ix < 0)
1231                ent = NULL;
1232        else {
1233                ent = pbase_tree_cache[available_ix];
1234                my_ix = available_ix;
1235        }
1236
1237        if (!ent) {
1238                nent = xmalloc(sizeof(*nent));
1239                nent->temporary = (available_ix < 0);
1240        }
1241        else {
1242                /* evict and reuse */
1243                free(ent->tree_data);
1244                nent = ent;
1245        }
1246        oidcpy(&nent->oid, oid);
1247        nent->tree_data = data;
1248        nent->tree_size = size;
1249        nent->ref = 1;
1250        if (!nent->temporary)
1251                pbase_tree_cache[my_ix] = nent;
1252        return nent;
1253}
1254
1255static void pbase_tree_put(struct pbase_tree_cache *cache)
1256{
1257        if (!cache->temporary) {
1258                cache->ref--;
1259                return;
1260        }
1261        free(cache->tree_data);
1262        free(cache);
1263}
1264
1265static int name_cmp_len(const char *name)
1266{
1267        int i;
1268        for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1269                ;
1270        return i;
1271}
1272
1273static void add_pbase_object(struct tree_desc *tree,
1274                             const char *name,
1275                             int cmplen,
1276                             const char *fullname)
1277{
1278        struct name_entry entry;
1279        int cmp;
1280
1281        while (tree_entry(tree,&entry)) {
1282                if (S_ISGITLINK(entry.mode))
1283                        continue;
1284                cmp = tree_entry_len(&entry) != cmplen ? 1 :
1285                      memcmp(name, entry.path, cmplen);
1286                if (cmp > 0)
1287                        continue;
1288                if (cmp < 0)
1289                        return;
1290                if (name[cmplen] != '/') {
1291                        add_object_entry(entry.oid,
1292                                         object_type(entry.mode),
1293                                         fullname, 1);
1294                        return;
1295                }
1296                if (S_ISDIR(entry.mode)) {
1297                        struct tree_desc sub;
1298                        struct pbase_tree_cache *tree;
1299                        const char *down = name+cmplen+1;
1300                        int downlen = name_cmp_len(down);
1301
1302                        tree = pbase_tree_get(entry.oid);
1303                        if (!tree)
1304                                return;
1305                        init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1306
1307                        add_pbase_object(&sub, down, downlen, fullname);
1308                        pbase_tree_put(tree);
1309                }
1310        }
1311}
1312
1313static unsigned *done_pbase_paths;
1314static int done_pbase_paths_num;
1315static int done_pbase_paths_alloc;
1316static int done_pbase_path_pos(unsigned hash)
1317{
1318        int lo = 0;
1319        int hi = done_pbase_paths_num;
1320        while (lo < hi) {
1321                int mi = lo + (hi - lo) / 2;
1322                if (done_pbase_paths[mi] == hash)
1323                        return mi;
1324                if (done_pbase_paths[mi] < hash)
1325                        hi = mi;
1326                else
1327                        lo = mi + 1;
1328        }
1329        return -lo-1;
1330}
1331
1332static int check_pbase_path(unsigned hash)
1333{
1334        int pos = done_pbase_path_pos(hash);
1335        if (0 <= pos)
1336                return 1;
1337        pos = -pos - 1;
1338        ALLOC_GROW(done_pbase_paths,
1339                   done_pbase_paths_num + 1,
1340                   done_pbase_paths_alloc);
1341        done_pbase_paths_num++;
1342        if (pos < done_pbase_paths_num)
1343                MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,
1344                           done_pbase_paths_num - pos - 1);
1345        done_pbase_paths[pos] = hash;
1346        return 0;
1347}
1348
1349static void add_preferred_base_object(const char *name)
1350{
1351        struct pbase_tree *it;
1352        int cmplen;
1353        unsigned hash = pack_name_hash(name);
1354
1355        if (!num_preferred_base || check_pbase_path(hash))
1356                return;
1357
1358        cmplen = name_cmp_len(name);
1359        for (it = pbase_tree; it; it = it->next) {
1360                if (cmplen == 0) {
1361                        add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);
1362                }
1363                else {
1364                        struct tree_desc tree;
1365                        init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1366                        add_pbase_object(&tree, name, cmplen, name);
1367                }
1368        }
1369}
1370
1371static void add_preferred_base(struct object_id *oid)
1372{
1373        struct pbase_tree *it;
1374        void *data;
1375        unsigned long size;
1376        struct object_id tree_oid;
1377
1378        if (window <= num_preferred_base++)
1379                return;
1380
1381        data = read_object_with_reference(oid, tree_type, &size, &tree_oid);
1382        if (!data)
1383                return;
1384
1385        for (it = pbase_tree; it; it = it->next) {
1386                if (!oidcmp(&it->pcache.oid, &tree_oid)) {
1387                        free(data);
1388                        return;
1389                }
1390        }
1391
1392        it = xcalloc(1, sizeof(*it));
1393        it->next = pbase_tree;
1394        pbase_tree = it;
1395
1396        oidcpy(&it->pcache.oid, &tree_oid);
1397        it->pcache.tree_data = data;
1398        it->pcache.tree_size = size;
1399}
1400
1401static void cleanup_preferred_base(void)
1402{
1403        struct pbase_tree *it;
1404        unsigned i;
1405
1406        it = pbase_tree;
1407        pbase_tree = NULL;
1408        while (it) {
1409                struct pbase_tree *tmp = it;
1410                it = tmp->next;
1411                free(tmp->pcache.tree_data);
1412                free(tmp);
1413        }
1414
1415        for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {
1416                if (!pbase_tree_cache[i])
1417                        continue;
1418                free(pbase_tree_cache[i]->tree_data);
1419                FREE_AND_NULL(pbase_tree_cache[i]);
1420        }
1421
1422        FREE_AND_NULL(done_pbase_paths);
1423        done_pbase_paths_num = done_pbase_paths_alloc = 0;
1424}
1425
1426static void check_object(struct object_entry *entry)
1427{
1428        unsigned long canonical_size;
1429
1430        if (IN_PACK(entry)) {
1431                struct packed_git *p = IN_PACK(entry);
1432                struct pack_window *w_curs = NULL;
1433                const unsigned char *base_ref = NULL;
1434                struct object_entry *base_entry;
1435                unsigned long used, used_0;
1436                unsigned long avail;
1437                off_t ofs;
1438                unsigned char *buf, c;
1439                enum object_type type;
1440                unsigned long in_pack_size;
1441
1442                buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1443
1444                /*
1445                 * We want in_pack_type even if we do not reuse delta
1446                 * since non-delta representations could still be reused.
1447                 */
1448                used = unpack_object_header_buffer(buf, avail,
1449                                                   &type,
1450                                                   &in_pack_size);
1451                if (used == 0)
1452                        goto give_up;
1453
1454                if (type < 0)
1455                        BUG("invalid type %d", type);
1456                entry->in_pack_type = type;
1457
1458                /*
1459                 * Determine if this is a delta and if so whether we can
1460                 * reuse it or not.  Otherwise let's find out as cheaply as
1461                 * possible what the actual type and size for this object is.
1462                 */
1463                switch (entry->in_pack_type) {
1464                default:
1465                        /* Not a delta hence we've already got all we need. */
1466                        oe_set_type(entry, entry->in_pack_type);
1467                        SET_SIZE(entry, in_pack_size);
1468                        entry->in_pack_header_size = used;
1469                        if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)
1470                                goto give_up;
1471                        unuse_pack(&w_curs);
1472                        return;
1473                case OBJ_REF_DELTA:
1474                        if (reuse_delta && !entry->preferred_base)
1475                                base_ref = use_pack(p, &w_curs,
1476                                                entry->in_pack_offset + used, NULL);
1477                        entry->in_pack_header_size = used + the_hash_algo->rawsz;
1478                        break;
1479                case OBJ_OFS_DELTA:
1480                        buf = use_pack(p, &w_curs,
1481                                       entry->in_pack_offset + used, NULL);
1482                        used_0 = 0;
1483                        c = buf[used_0++];
1484                        ofs = c & 127;
1485                        while (c & 128) {
1486                                ofs += 1;
1487                                if (!ofs || MSB(ofs, 7)) {
1488                                        error("delta base offset overflow in pack for %s",
1489                                              oid_to_hex(&entry->idx.oid));
1490                                        goto give_up;
1491                                }
1492                                c = buf[used_0++];
1493                                ofs = (ofs << 7) + (c & 127);
1494                        }
1495                        ofs = entry->in_pack_offset - ofs;
1496                        if (ofs <= 0 || ofs >= entry->in_pack_offset) {
1497                                error("delta base offset out of bound for %s",
1498                                      oid_to_hex(&entry->idx.oid));
1499                                goto give_up;
1500                        }
1501                        if (reuse_delta && !entry->preferred_base) {
1502                                struct revindex_entry *revidx;
1503                                revidx = find_pack_revindex(p, ofs);
1504                                if (!revidx)
1505                                        goto give_up;
1506                                base_ref = nth_packed_object_sha1(p, revidx->nr);
1507                        }
1508                        entry->in_pack_header_size = used + used_0;
1509                        break;
1510                }
1511
1512                if (base_ref && (base_entry = packlist_find(&to_pack, base_ref, NULL))) {
1513                        /*
1514                         * If base_ref was set above that means we wish to
1515                         * reuse delta data, and we even found that base
1516                         * in the list of objects we want to pack. Goodie!
1517                         *
1518                         * Depth value does not matter - find_deltas() will
1519                         * never consider reused delta as the base object to
1520                         * deltify other objects against, in order to avoid
1521                         * circular deltas.
1522                         */
1523                        oe_set_type(entry, entry->in_pack_type);
1524                        SET_SIZE(entry, in_pack_size); /* delta size */
1525                        SET_DELTA(entry, base_entry);
1526                        SET_DELTA_SIZE(entry, in_pack_size);
1527                        entry->delta_sibling_idx = base_entry->delta_child_idx;
1528                        SET_DELTA_CHILD(base_entry, entry);
1529                        unuse_pack(&w_curs);
1530                        return;
1531                }
1532
1533                if (oe_type(entry)) {
1534                        off_t delta_pos;
1535
1536                        /*
1537                         * This must be a delta and we already know what the
1538                         * final object type is.  Let's extract the actual
1539                         * object size from the delta header.
1540                         */
1541                        delta_pos = entry->in_pack_offset + entry->in_pack_header_size;
1542                        canonical_size = get_size_from_delta(p, &w_curs, delta_pos);
1543                        if (canonical_size == 0)
1544                                goto give_up;
1545                        SET_SIZE(entry, canonical_size);
1546                        unuse_pack(&w_curs);
1547                        return;
1548                }
1549
1550                /*
1551                 * No choice but to fall back to the recursive delta walk
1552                 * with sha1_object_info() to find about the object type
1553                 * at this point...
1554                 */
1555                give_up:
1556                unuse_pack(&w_curs);
1557        }
1558
1559        oe_set_type(entry,
1560                    oid_object_info(the_repository, &entry->idx.oid, &canonical_size));
1561        if (entry->type_valid) {
1562                SET_SIZE(entry, canonical_size);
1563        } else {
1564                /*
1565                 * Bad object type is checked in prepare_pack().  This is
1566                 * to permit a missing preferred base object to be ignored
1567                 * as a preferred base.  Doing so can result in a larger
1568                 * pack file, but the transfer will still take place.
1569                 */
1570        }
1571}
1572
1573static int pack_offset_sort(const void *_a, const void *_b)
1574{
1575        const struct object_entry *a = *(struct object_entry **)_a;
1576        const struct object_entry *b = *(struct object_entry **)_b;
1577        const struct packed_git *a_in_pack = IN_PACK(a);
1578        const struct packed_git *b_in_pack = IN_PACK(b);
1579
1580        /* avoid filesystem trashing with loose objects */
1581        if (!a_in_pack && !b_in_pack)
1582                return oidcmp(&a->idx.oid, &b->idx.oid);
1583
1584        if (a_in_pack < b_in_pack)
1585                return -1;
1586        if (a_in_pack > b_in_pack)
1587                return 1;
1588        return a->in_pack_offset < b->in_pack_offset ? -1 :
1589                        (a->in_pack_offset > b->in_pack_offset);
1590}
1591
1592/*
1593 * Drop an on-disk delta we were planning to reuse. Naively, this would
1594 * just involve blanking out the "delta" field, but we have to deal
1595 * with some extra book-keeping:
1596 *
1597 *   1. Removing ourselves from the delta_sibling linked list.
1598 *
1599 *   2. Updating our size/type to the non-delta representation. These were
1600 *      either not recorded initially (size) or overwritten with the delta type
1601 *      (type) when check_object() decided to reuse the delta.
1602 *
1603 *   3. Resetting our delta depth, as we are now a base object.
1604 */
1605static void drop_reused_delta(struct object_entry *entry)
1606{
1607        unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;
1608        struct object_info oi = OBJECT_INFO_INIT;
1609        enum object_type type;
1610        unsigned long size;
1611
1612        while (*idx) {
1613                struct object_entry *oe = &to_pack.objects[*idx - 1];
1614
1615                if (oe == entry)
1616                        *idx = oe->delta_sibling_idx;
1617                else
1618                        idx = &oe->delta_sibling_idx;
1619        }
1620        SET_DELTA(entry, NULL);
1621        entry->depth = 0;
1622
1623        oi.sizep = &size;
1624        oi.typep = &type;
1625        if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {
1626                /*
1627                 * We failed to get the info from this pack for some reason;
1628                 * fall back to sha1_object_info, which may find another copy.
1629                 * And if that fails, the error will be recorded in oe_type(entry)
1630                 * and dealt with in prepare_pack().
1631                 */
1632                oe_set_type(entry,
1633                            oid_object_info(the_repository, &entry->idx.oid, &size));
1634        } else {
1635                oe_set_type(entry, type);
1636        }
1637        SET_SIZE(entry, size);
1638}
1639
1640/*
1641 * Follow the chain of deltas from this entry onward, throwing away any links
1642 * that cause us to hit a cycle (as determined by the DFS state flags in
1643 * the entries).
1644 *
1645 * We also detect too-long reused chains that would violate our --depth
1646 * limit.
1647 */
1648static void break_delta_chains(struct object_entry *entry)
1649{
1650        /*
1651         * The actual depth of each object we will write is stored as an int,
1652         * as it cannot exceed our int "depth" limit. But before we break
1653         * changes based no that limit, we may potentially go as deep as the
1654         * number of objects, which is elsewhere bounded to a uint32_t.
1655         */
1656        uint32_t total_depth;
1657        struct object_entry *cur, *next;
1658
1659        for (cur = entry, total_depth = 0;
1660             cur;
1661             cur = DELTA(cur), total_depth++) {
1662                if (cur->dfs_state == DFS_DONE) {
1663                        /*
1664                         * We've already seen this object and know it isn't
1665                         * part of a cycle. We do need to append its depth
1666                         * to our count.
1667                         */
1668                        total_depth += cur->depth;
1669                        break;
1670                }
1671
1672                /*
1673                 * We break cycles before looping, so an ACTIVE state (or any
1674                 * other cruft which made its way into the state variable)
1675                 * is a bug.
1676                 */
1677                if (cur->dfs_state != DFS_NONE)
1678                        BUG("confusing delta dfs state in first pass: %d",
1679                            cur->dfs_state);
1680
1681                /*
1682                 * Now we know this is the first time we've seen the object. If
1683                 * it's not a delta, we're done traversing, but we'll mark it
1684                 * done to save time on future traversals.
1685                 */
1686                if (!DELTA(cur)) {
1687                        cur->dfs_state = DFS_DONE;
1688                        break;
1689                }
1690
1691                /*
1692                 * Mark ourselves as active and see if the next step causes
1693                 * us to cycle to another active object. It's important to do
1694                 * this _before_ we loop, because it impacts where we make the
1695                 * cut, and thus how our total_depth counter works.
1696                 * E.g., We may see a partial loop like:
1697                 *
1698                 *   A -> B -> C -> D -> B
1699                 *
1700                 * Cutting B->C breaks the cycle. But now the depth of A is
1701                 * only 1, and our total_depth counter is at 3. The size of the
1702                 * error is always one less than the size of the cycle we
1703                 * broke. Commits C and D were "lost" from A's chain.
1704                 *
1705                 * If we instead cut D->B, then the depth of A is correct at 3.
1706                 * We keep all commits in the chain that we examined.
1707                 */
1708                cur->dfs_state = DFS_ACTIVE;
1709                if (DELTA(cur)->dfs_state == DFS_ACTIVE) {
1710                        drop_reused_delta(cur);
1711                        cur->dfs_state = DFS_DONE;
1712                        break;
1713                }
1714        }
1715
1716        /*
1717         * And now that we've gone all the way to the bottom of the chain, we
1718         * need to clear the active flags and set the depth fields as
1719         * appropriate. Unlike the loop above, which can quit when it drops a
1720         * delta, we need to keep going to look for more depth cuts. So we need
1721         * an extra "next" pointer to keep going after we reset cur->delta.
1722         */
1723        for (cur = entry; cur; cur = next) {
1724                next = DELTA(cur);
1725
1726                /*
1727                 * We should have a chain of zero or more ACTIVE states down to
1728                 * a final DONE. We can quit after the DONE, because either it
1729                 * has no bases, or we've already handled them in a previous
1730                 * call.
1731                 */
1732                if (cur->dfs_state == DFS_DONE)
1733                        break;
1734                else if (cur->dfs_state != DFS_ACTIVE)
1735                        BUG("confusing delta dfs state in second pass: %d",
1736                            cur->dfs_state);
1737
1738                /*
1739                 * If the total_depth is more than depth, then we need to snip
1740                 * the chain into two or more smaller chains that don't exceed
1741                 * the maximum depth. Most of the resulting chains will contain
1742                 * (depth + 1) entries (i.e., depth deltas plus one base), and
1743                 * the last chain (i.e., the one containing entry) will contain
1744                 * whatever entries are left over, namely
1745                 * (total_depth % (depth + 1)) of them.
1746                 *
1747                 * Since we are iterating towards decreasing depth, we need to
1748                 * decrement total_depth as we go, and we need to write to the
1749                 * entry what its final depth will be after all of the
1750                 * snipping. Since we're snipping into chains of length (depth
1751                 * + 1) entries, the final depth of an entry will be its
1752                 * original depth modulo (depth + 1). Any time we encounter an
1753                 * entry whose final depth is supposed to be zero, we snip it
1754                 * from its delta base, thereby making it so.
1755                 */
1756                cur->depth = (total_depth--) % (depth + 1);
1757                if (!cur->depth)
1758                        drop_reused_delta(cur);
1759
1760                cur->dfs_state = DFS_DONE;
1761        }
1762}
1763
1764static void get_object_details(void)
1765{
1766        uint32_t i;
1767        struct object_entry **sorted_by_offset;
1768
1769        if (progress)
1770                progress_state = start_progress(_("Counting objects"),
1771                                                to_pack.nr_objects);
1772
1773        sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));
1774        for (i = 0; i < to_pack.nr_objects; i++)
1775                sorted_by_offset[i] = to_pack.objects + i;
1776        QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);
1777
1778        for (i = 0; i < to_pack.nr_objects; i++) {
1779                struct object_entry *entry = sorted_by_offset[i];
1780                check_object(entry);
1781                if (entry->type_valid &&
1782                    oe_size_greater_than(&to_pack, entry, big_file_threshold))
1783                        entry->no_try_delta = 1;
1784                display_progress(progress_state, i + 1);
1785        }
1786        stop_progress(&progress_state);
1787
1788        /*
1789         * This must happen in a second pass, since we rely on the delta
1790         * information for the whole list being completed.
1791         */
1792        for (i = 0; i < to_pack.nr_objects; i++)
1793                break_delta_chains(&to_pack.objects[i]);
1794
1795        free(sorted_by_offset);
1796}
1797
1798/*
1799 * We search for deltas in a list sorted by type, by filename hash, and then
1800 * by size, so that we see progressively smaller and smaller files.
1801 * That's because we prefer deltas to be from the bigger file
1802 * to the smaller -- deletes are potentially cheaper, but perhaps
1803 * more importantly, the bigger file is likely the more recent
1804 * one.  The deepest deltas are therefore the oldest objects which are
1805 * less susceptible to be accessed often.
1806 */
1807static int type_size_sort(const void *_a, const void *_b)
1808{
1809        const struct object_entry *a = *(struct object_entry **)_a;
1810        const struct object_entry *b = *(struct object_entry **)_b;
1811        enum object_type a_type = oe_type(a);
1812        enum object_type b_type = oe_type(b);
1813        unsigned long a_size = SIZE(a);
1814        unsigned long b_size = SIZE(b);
1815
1816        if (a_type > b_type)
1817                return -1;
1818        if (a_type < b_type)
1819                return 1;
1820        if (a->hash > b->hash)
1821                return -1;
1822        if (a->hash < b->hash)
1823                return 1;
1824        if (a->preferred_base > b->preferred_base)
1825                return -1;
1826        if (a->preferred_base < b->preferred_base)
1827                return 1;
1828        if (a_size > b_size)
1829                return -1;
1830        if (a_size < b_size)
1831                return 1;
1832        return a < b ? -1 : (a > b);  /* newest first */
1833}
1834
1835struct unpacked {
1836        struct object_entry *entry;
1837        void *data;
1838        struct delta_index *index;
1839        unsigned depth;
1840};
1841
1842static int delta_cacheable(unsigned long src_size, unsigned long trg_size,
1843                           unsigned long delta_size)
1844{
1845        if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1846                return 0;
1847
1848        if (delta_size < cache_max_small_delta_size)
1849                return 1;
1850
1851        /* cache delta, if objects are large enough compared to delta size */
1852        if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1853                return 1;
1854
1855        return 0;
1856}
1857
1858#ifndef NO_PTHREADS
1859
1860static pthread_mutex_t read_mutex;
1861#define read_lock()             pthread_mutex_lock(&read_mutex)
1862#define read_unlock()           pthread_mutex_unlock(&read_mutex)
1863
1864static pthread_mutex_t cache_mutex;
1865#define cache_lock()            pthread_mutex_lock(&cache_mutex)
1866#define cache_unlock()          pthread_mutex_unlock(&cache_mutex)
1867
1868static pthread_mutex_t progress_mutex;
1869#define progress_lock()         pthread_mutex_lock(&progress_mutex)
1870#define progress_unlock()       pthread_mutex_unlock(&progress_mutex)
1871
1872#else
1873
1874#define read_lock()             (void)0
1875#define read_unlock()           (void)0
1876#define cache_lock()            (void)0
1877#define cache_unlock()          (void)0
1878#define progress_lock()         (void)0
1879#define progress_unlock()       (void)0
1880
1881#endif
1882
1883/*
1884 * Return the size of the object without doing any delta
1885 * reconstruction (so non-deltas are true object sizes, but deltas
1886 * return the size of the delta data).
1887 */
1888unsigned long oe_get_size_slow(struct packing_data *pack,
1889                               const struct object_entry *e)
1890{
1891        struct packed_git *p;
1892        struct pack_window *w_curs;
1893        unsigned char *buf;
1894        enum object_type type;
1895        unsigned long used, avail, size;
1896
1897        if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {
1898                read_lock();
1899                if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)
1900                        die(_("unable to get size of %s"),
1901                            oid_to_hex(&e->idx.oid));
1902                read_unlock();
1903                return size;
1904        }
1905
1906        p = oe_in_pack(pack, e);
1907        if (!p)
1908                BUG("when e->type is a delta, it must belong to a pack");
1909
1910        read_lock();
1911        w_curs = NULL;
1912        buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);
1913        used = unpack_object_header_buffer(buf, avail, &type, &size);
1914        if (used == 0)
1915                die(_("unable to parse object header of %s"),
1916                    oid_to_hex(&e->idx.oid));
1917
1918        unuse_pack(&w_curs);
1919        read_unlock();
1920        return size;
1921}
1922
1923static int try_delta(struct unpacked *trg, struct unpacked *src,
1924                     unsigned max_depth, unsigned long *mem_usage)
1925{
1926        struct object_entry *trg_entry = trg->entry;
1927        struct object_entry *src_entry = src->entry;
1928        unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1929        unsigned ref_depth;
1930        enum object_type type;
1931        void *delta_buf;
1932
1933        /* Don't bother doing diffs between different types */
1934        if (oe_type(trg_entry) != oe_type(src_entry))
1935                return -1;
1936
1937        /*
1938         * We do not bother to try a delta that we discarded on an
1939         * earlier try, but only when reusing delta data.  Note that
1940         * src_entry that is marked as the preferred_base should always
1941         * be considered, as even if we produce a suboptimal delta against
1942         * it, we will still save the transfer cost, as we already know
1943         * the other side has it and we won't send src_entry at all.
1944         */
1945        if (reuse_delta && IN_PACK(trg_entry) &&
1946            IN_PACK(trg_entry) == IN_PACK(src_entry) &&
1947            !src_entry->preferred_base &&
1948            trg_entry->in_pack_type != OBJ_REF_DELTA &&
1949            trg_entry->in_pack_type != OBJ_OFS_DELTA)
1950                return 0;
1951
1952        /* Let's not bust the allowed depth. */
1953        if (src->depth >= max_depth)
1954                return 0;
1955
1956        /* Now some size filtering heuristics. */
1957        trg_size = SIZE(trg_entry);
1958        if (!DELTA(trg_entry)) {
1959                max_size = trg_size/2 - the_hash_algo->rawsz;
1960                ref_depth = 1;
1961        } else {
1962                max_size = DELTA_SIZE(trg_entry);
1963                ref_depth = trg->depth;
1964        }
1965        max_size = (uint64_t)max_size * (max_depth - src->depth) /
1966                                                (max_depth - ref_depth + 1);
1967        if (max_size == 0)
1968                return 0;
1969        src_size = SIZE(src_entry);
1970        sizediff = src_size < trg_size ? trg_size - src_size : 0;
1971        if (sizediff >= max_size)
1972                return 0;
1973        if (trg_size < src_size / 32)
1974                return 0;
1975
1976        /* Load data if not already done */
1977        if (!trg->data) {
1978                read_lock();
1979                trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);
1980                read_unlock();
1981                if (!trg->data)
1982                        die("object %s cannot be read",
1983                            oid_to_hex(&trg_entry->idx.oid));
1984                if (sz != trg_size)
1985                        die("object %s inconsistent object length (%lu vs %lu)",
1986                            oid_to_hex(&trg_entry->idx.oid), sz,
1987                            trg_size);
1988                *mem_usage += sz;
1989        }
1990        if (!src->data) {
1991                read_lock();
1992                src->data = read_object_file(&src_entry->idx.oid, &type, &sz);
1993                read_unlock();
1994                if (!src->data) {
1995                        if (src_entry->preferred_base) {
1996                                static int warned = 0;
1997                                if (!warned++)
1998                                        warning("object %s cannot be read",
1999                                                oid_to_hex(&src_entry->idx.oid));
2000                                /*
2001                                 * Those objects are not included in the
2002                                 * resulting pack.  Be resilient and ignore
2003                                 * them if they can't be read, in case the
2004                                 * pack could be created nevertheless.
2005                                 */
2006                                return 0;
2007                        }
2008                        die("object %s cannot be read",
2009                            oid_to_hex(&src_entry->idx.oid));
2010                }
2011                if (sz != src_size)
2012                        die("object %s inconsistent object length (%lu vs %lu)",
2013                            oid_to_hex(&src_entry->idx.oid), sz,
2014                            src_size);
2015                *mem_usage += sz;
2016        }
2017        if (!src->index) {
2018                src->index = create_delta_index(src->data, src_size);
2019                if (!src->index) {
2020                        static int warned = 0;
2021                        if (!warned++)
2022                                warning("suboptimal pack - out of memory");
2023                        return 0;
2024                }
2025                *mem_usage += sizeof_delta_index(src->index);
2026        }
2027
2028        delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
2029        if (!delta_buf)
2030                return 0;
2031        if (delta_size >= (1U << OE_DELTA_SIZE_BITS)) {
2032                free(delta_buf);
2033                return 0;
2034        }
2035
2036        if (DELTA(trg_entry)) {
2037                /* Prefer only shallower same-sized deltas. */
2038                if (delta_size == DELTA_SIZE(trg_entry) &&
2039                    src->depth + 1 >= trg->depth) {
2040                        free(delta_buf);
2041                        return 0;
2042                }
2043        }
2044
2045        /*
2046         * Handle memory allocation outside of the cache
2047         * accounting lock.  Compiler will optimize the strangeness
2048         * away when NO_PTHREADS is defined.
2049         */
2050        free(trg_entry->delta_data);
2051        cache_lock();
2052        if (trg_entry->delta_data) {
2053                delta_cache_size -= DELTA_SIZE(trg_entry);
2054                trg_entry->delta_data = NULL;
2055        }
2056        if (delta_cacheable(src_size, trg_size, delta_size)) {
2057                delta_cache_size += delta_size;
2058                cache_unlock();
2059                trg_entry->delta_data = xrealloc(delta_buf, delta_size);
2060        } else {
2061                cache_unlock();
2062                free(delta_buf);
2063        }
2064
2065        SET_DELTA(trg_entry, src_entry);
2066        SET_DELTA_SIZE(trg_entry, delta_size);
2067        trg->depth = src->depth + 1;
2068
2069        return 1;
2070}
2071
2072static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
2073{
2074        struct object_entry *child = DELTA_CHILD(me);
2075        unsigned int m = n;
2076        while (child) {
2077                unsigned int c = check_delta_limit(child, n + 1);
2078                if (m < c)
2079                        m = c;
2080                child = DELTA_SIBLING(child);
2081        }
2082        return m;
2083}
2084
2085static unsigned long free_unpacked(struct unpacked *n)
2086{
2087        unsigned long freed_mem = sizeof_delta_index(n->index);
2088        free_delta_index(n->index);
2089        n->index = NULL;
2090        if (n->data) {
2091                freed_mem += SIZE(n->entry);
2092                FREE_AND_NULL(n->data);
2093        }
2094        n->entry = NULL;
2095        n->depth = 0;
2096        return freed_mem;
2097}
2098
2099static void find_deltas(struct object_entry **list, unsigned *list_size,
2100                        int window, int depth, unsigned *processed)
2101{
2102        uint32_t i, idx = 0, count = 0;
2103        struct unpacked *array;
2104        unsigned long mem_usage = 0;
2105
2106        array = xcalloc(window, sizeof(struct unpacked));
2107
2108        for (;;) {
2109                struct object_entry *entry;
2110                struct unpacked *n = array + idx;
2111                int j, max_depth, best_base = -1;
2112
2113                progress_lock();
2114                if (!*list_size) {
2115                        progress_unlock();
2116                        break;
2117                }
2118                entry = *list++;
2119                (*list_size)--;
2120                if (!entry->preferred_base) {
2121                        (*processed)++;
2122                        display_progress(progress_state, *processed);
2123                }
2124                progress_unlock();
2125
2126                mem_usage -= free_unpacked(n);
2127                n->entry = entry;
2128
2129                while (window_memory_limit &&
2130                       mem_usage > window_memory_limit &&
2131                       count > 1) {
2132                        uint32_t tail = (idx + window - count) % window;
2133                        mem_usage -= free_unpacked(array + tail);
2134                        count--;
2135                }
2136
2137                /* We do not compute delta to *create* objects we are not
2138                 * going to pack.
2139                 */
2140                if (entry->preferred_base)
2141                        goto next;
2142
2143                /*
2144                 * If the current object is at pack edge, take the depth the
2145                 * objects that depend on the current object into account
2146                 * otherwise they would become too deep.
2147                 */
2148                max_depth = depth;
2149                if (DELTA_CHILD(entry)) {
2150                        max_depth -= check_delta_limit(entry, 0);
2151                        if (max_depth <= 0)
2152                                goto next;
2153                }
2154
2155                j = window;
2156                while (--j > 0) {
2157                        int ret;
2158                        uint32_t other_idx = idx + j;
2159                        struct unpacked *m;
2160                        if (other_idx >= window)
2161                                other_idx -= window;
2162                        m = array + other_idx;
2163                        if (!m->entry)
2164                                break;
2165                        ret = try_delta(n, m, max_depth, &mem_usage);
2166                        if (ret < 0)
2167                                break;
2168                        else if (ret > 0)
2169                                best_base = other_idx;
2170                }
2171
2172                /*
2173                 * If we decided to cache the delta data, then it is best
2174                 * to compress it right away.  First because we have to do
2175                 * it anyway, and doing it here while we're threaded will
2176                 * save a lot of time in the non threaded write phase,
2177                 * as well as allow for caching more deltas within
2178                 * the same cache size limit.
2179                 * ...
2180                 * But only if not writing to stdout, since in that case
2181                 * the network is most likely throttling writes anyway,
2182                 * and therefore it is best to go to the write phase ASAP
2183                 * instead, as we can afford spending more time compressing
2184                 * between writes at that moment.
2185                 */
2186                if (entry->delta_data && !pack_to_stdout) {
2187                        unsigned long size;
2188
2189                        size = do_compress(&entry->delta_data, DELTA_SIZE(entry));
2190                        if (size < (1U << OE_Z_DELTA_BITS)) {
2191                                entry->z_delta_size = size;
2192                                cache_lock();
2193                                delta_cache_size -= DELTA_SIZE(entry);
2194                                delta_cache_size += entry->z_delta_size;
2195                                cache_unlock();
2196                        } else {
2197                                FREE_AND_NULL(entry->delta_data);
2198                                entry->z_delta_size = 0;
2199                        }
2200                }
2201
2202                /* if we made n a delta, and if n is already at max
2203                 * depth, leaving it in the window is pointless.  we
2204                 * should evict it first.
2205                 */
2206                if (DELTA(entry) && max_depth <= n->depth)
2207                        continue;
2208
2209                /*
2210                 * Move the best delta base up in the window, after the
2211                 * currently deltified object, to keep it longer.  It will
2212                 * be the first base object to be attempted next.
2213                 */
2214                if (DELTA(entry)) {
2215                        struct unpacked swap = array[best_base];
2216                        int dist = (window + idx - best_base) % window;
2217                        int dst = best_base;
2218                        while (dist--) {
2219                                int src = (dst + 1) % window;
2220                                array[dst] = array[src];
2221                                dst = src;
2222                        }
2223                        array[dst] = swap;
2224                }
2225
2226                next:
2227                idx++;
2228                if (count + 1 < window)
2229                        count++;
2230                if (idx >= window)
2231                        idx = 0;
2232        }
2233
2234        for (i = 0; i < window; ++i) {
2235                free_delta_index(array[i].index);
2236                free(array[i].data);
2237        }
2238        free(array);
2239}
2240
2241#ifndef NO_PTHREADS
2242
2243static void try_to_free_from_threads(size_t size)
2244{
2245        read_lock();
2246        release_pack_memory(size);
2247        read_unlock();
2248}
2249
2250static try_to_free_t old_try_to_free_routine;
2251
2252/*
2253 * The main thread waits on the condition that (at least) one of the workers
2254 * has stopped working (which is indicated in the .working member of
2255 * struct thread_params).
2256 * When a work thread has completed its work, it sets .working to 0 and
2257 * signals the main thread and waits on the condition that .data_ready
2258 * becomes 1.
2259 */
2260
2261struct thread_params {
2262        pthread_t thread;
2263        struct object_entry **list;
2264        unsigned list_size;
2265        unsigned remaining;
2266        int window;
2267        int depth;
2268        int working;
2269        int data_ready;
2270        pthread_mutex_t mutex;
2271        pthread_cond_t cond;
2272        unsigned *processed;
2273};
2274
2275static pthread_cond_t progress_cond;
2276
2277/*
2278 * Mutex and conditional variable can't be statically-initialized on Windows.
2279 */
2280static void init_threaded_search(void)
2281{
2282        init_recursive_mutex(&read_mutex);
2283        pthread_mutex_init(&cache_mutex, NULL);
2284        pthread_mutex_init(&progress_mutex, NULL);
2285        pthread_cond_init(&progress_cond, NULL);
2286        old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);
2287}
2288
2289static void cleanup_threaded_search(void)
2290{
2291        set_try_to_free_routine(old_try_to_free_routine);
2292        pthread_cond_destroy(&progress_cond);
2293        pthread_mutex_destroy(&read_mutex);
2294        pthread_mutex_destroy(&cache_mutex);
2295        pthread_mutex_destroy(&progress_mutex);
2296}
2297
2298static void *threaded_find_deltas(void *arg)
2299{
2300        struct thread_params *me = arg;
2301
2302        progress_lock();
2303        while (me->remaining) {
2304                progress_unlock();
2305
2306                find_deltas(me->list, &me->remaining,
2307                            me->window, me->depth, me->processed);
2308
2309                progress_lock();
2310                me->working = 0;
2311                pthread_cond_signal(&progress_cond);
2312                progress_unlock();
2313
2314                /*
2315                 * We must not set ->data_ready before we wait on the
2316                 * condition because the main thread may have set it to 1
2317                 * before we get here. In order to be sure that new
2318                 * work is available if we see 1 in ->data_ready, it
2319                 * was initialized to 0 before this thread was spawned
2320                 * and we reset it to 0 right away.
2321                 */
2322                pthread_mutex_lock(&me->mutex);
2323                while (!me->data_ready)
2324                        pthread_cond_wait(&me->cond, &me->mutex);
2325                me->data_ready = 0;
2326                pthread_mutex_unlock(&me->mutex);
2327
2328                progress_lock();
2329        }
2330        progress_unlock();
2331        /* leave ->working 1 so that this doesn't get more work assigned */
2332        return NULL;
2333}
2334
2335static void ll_find_deltas(struct object_entry **list, unsigned list_size,
2336                           int window, int depth, unsigned *processed)
2337{
2338        struct thread_params *p;
2339        int i, ret, active_threads = 0;
2340
2341        init_threaded_search();
2342
2343        if (delta_search_threads <= 1) {
2344                find_deltas(list, &list_size, window, depth, processed);
2345                cleanup_threaded_search();
2346                return;
2347        }
2348        if (progress > pack_to_stdout)
2349                fprintf_ln(stderr, "Delta compression using up to %d threads",
2350                           delta_search_threads);
2351        p = xcalloc(delta_search_threads, sizeof(*p));
2352
2353        /* Partition the work amongst work threads. */
2354        for (i = 0; i < delta_search_threads; i++) {
2355                unsigned sub_size = list_size / (delta_search_threads - i);
2356
2357                /* don't use too small segments or no deltas will be found */
2358                if (sub_size < 2*window && i+1 < delta_search_threads)
2359                        sub_size = 0;
2360
2361                p[i].window = window;
2362                p[i].depth = depth;
2363                p[i].processed = processed;
2364                p[i].working = 1;
2365                p[i].data_ready = 0;
2366
2367                /* try to split chunks on "path" boundaries */
2368                while (sub_size && sub_size < list_size &&
2369                       list[sub_size]->hash &&
2370                       list[sub_size]->hash == list[sub_size-1]->hash)
2371                        sub_size++;
2372
2373                p[i].list = list;
2374                p[i].list_size = sub_size;
2375                p[i].remaining = sub_size;
2376
2377                list += sub_size;
2378                list_size -= sub_size;
2379        }
2380
2381        /* Start work threads. */
2382        for (i = 0; i < delta_search_threads; i++) {
2383                if (!p[i].list_size)
2384                        continue;
2385                pthread_mutex_init(&p[i].mutex, NULL);
2386                pthread_cond_init(&p[i].cond, NULL);
2387                ret = pthread_create(&p[i].thread, NULL,
2388                                     threaded_find_deltas, &p[i]);
2389                if (ret)
2390                        die("unable to create thread: %s", strerror(ret));
2391                active_threads++;
2392        }
2393
2394        /*
2395         * Now let's wait for work completion.  Each time a thread is done
2396         * with its work, we steal half of the remaining work from the
2397         * thread with the largest number of unprocessed objects and give
2398         * it to that newly idle thread.  This ensure good load balancing
2399         * until the remaining object list segments are simply too short
2400         * to be worth splitting anymore.
2401         */
2402        while (active_threads) {
2403                struct thread_params *target = NULL;
2404                struct thread_params *victim = NULL;
2405                unsigned sub_size = 0;
2406
2407                progress_lock();
2408                for (;;) {
2409                        for (i = 0; !target && i < delta_search_threads; i++)
2410                                if (!p[i].working)
2411                                        target = &p[i];
2412                        if (target)
2413                                break;
2414                        pthread_cond_wait(&progress_cond, &progress_mutex);
2415                }
2416
2417                for (i = 0; i < delta_search_threads; i++)
2418                        if (p[i].remaining > 2*window &&
2419                            (!victim || victim->remaining < p[i].remaining))
2420                                victim = &p[i];
2421                if (victim) {
2422                        sub_size = victim->remaining / 2;
2423                        list = victim->list + victim->list_size - sub_size;
2424                        while (sub_size && list[0]->hash &&
2425                               list[0]->hash == list[-1]->hash) {
2426                                list++;
2427                                sub_size--;
2428                        }
2429                        if (!sub_size) {
2430                                /*
2431                                 * It is possible for some "paths" to have
2432                                 * so many objects that no hash boundary
2433                                 * might be found.  Let's just steal the
2434                                 * exact half in that case.
2435                                 */
2436                                sub_size = victim->remaining / 2;
2437                                list -= sub_size;
2438                        }
2439                        target->list = list;
2440                        victim->list_size -= sub_size;
2441                        victim->remaining -= sub_size;
2442                }
2443                target->list_size = sub_size;
2444                target->remaining = sub_size;
2445                target->working = 1;
2446                progress_unlock();
2447
2448                pthread_mutex_lock(&target->mutex);
2449                target->data_ready = 1;
2450                pthread_cond_signal(&target->cond);
2451                pthread_mutex_unlock(&target->mutex);
2452
2453                if (!sub_size) {
2454                        pthread_join(target->thread, NULL);
2455                        pthread_cond_destroy(&target->cond);
2456                        pthread_mutex_destroy(&target->mutex);
2457                        active_threads--;
2458                }
2459        }
2460        cleanup_threaded_search();
2461        free(p);
2462}
2463
2464#else
2465#define ll_find_deltas(l, s, w, d, p)   find_deltas(l, &s, w, d, p)
2466#endif
2467
2468static void add_tag_chain(const struct object_id *oid)
2469{
2470        struct tag *tag;
2471
2472        /*
2473         * We catch duplicates already in add_object_entry(), but we'd
2474         * prefer to do this extra check to avoid having to parse the
2475         * tag at all if we already know that it's being packed (e.g., if
2476         * it was included via bitmaps, we would not have parsed it
2477         * previously).
2478         */
2479        if (packlist_find(&to_pack, oid->hash, NULL))
2480                return;
2481
2482        tag = lookup_tag(oid);
2483        while (1) {
2484                if (!tag || parse_tag(tag) || !tag->tagged)
2485                        die("unable to pack objects reachable from tag %s",
2486                            oid_to_hex(oid));
2487
2488                add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);
2489
2490                if (tag->tagged->type != OBJ_TAG)
2491                        return;
2492
2493                tag = (struct tag *)tag->tagged;
2494        }
2495}
2496
2497static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)
2498{
2499        struct object_id peeled;
2500
2501        if (starts_with(path, "refs/tags/") && /* is a tag? */
2502            !peel_ref(path, &peeled)    && /* peelable? */
2503            packlist_find(&to_pack, peeled.hash, NULL))      /* object packed? */
2504                add_tag_chain(oid);
2505        return 0;
2506}
2507
2508static void prepare_pack(int window, int depth)
2509{
2510        struct object_entry **delta_list;
2511        uint32_t i, nr_deltas;
2512        unsigned n;
2513
2514        get_object_details();
2515
2516        /*
2517         * If we're locally repacking then we need to be doubly careful
2518         * from now on in order to make sure no stealth corruption gets
2519         * propagated to the new pack.  Clients receiving streamed packs
2520         * should validate everything they get anyway so no need to incur
2521         * the additional cost here in that case.
2522         */
2523        if (!pack_to_stdout)
2524                do_check_packed_object_crc = 1;
2525
2526        if (!to_pack.nr_objects || !window || !depth)
2527                return;
2528
2529        ALLOC_ARRAY(delta_list, to_pack.nr_objects);
2530        nr_deltas = n = 0;
2531
2532        for (i = 0; i < to_pack.nr_objects; i++) {
2533                struct object_entry *entry = to_pack.objects + i;
2534
2535                if (DELTA(entry))
2536                        /* This happens if we decided to reuse existing
2537                         * delta from a pack.  "reuse_delta &&" is implied.
2538                         */
2539                        continue;
2540
2541                if (!entry->type_valid ||
2542                    oe_size_less_than(&to_pack, entry, 50))
2543                        continue;
2544
2545                if (entry->no_try_delta)
2546                        continue;
2547
2548                if (!entry->preferred_base) {
2549                        nr_deltas++;
2550                        if (oe_type(entry) < 0)
2551                                die("unable to get type of object %s",
2552                                    oid_to_hex(&entry->idx.oid));
2553                } else {
2554                        if (oe_type(entry) < 0) {
2555                                /*
2556                                 * This object is not found, but we
2557                                 * don't have to include it anyway.
2558                                 */
2559                                continue;
2560                        }
2561                }
2562
2563                delta_list[n++] = entry;
2564        }
2565
2566        if (nr_deltas && n > 1) {
2567                unsigned nr_done = 0;
2568                if (progress)
2569                        progress_state = start_progress(_("Compressing objects"),
2570                                                        nr_deltas);
2571                QSORT(delta_list, n, type_size_sort);
2572                ll_find_deltas(delta_list, n, window+1, depth, &nr_done);
2573                stop_progress(&progress_state);
2574                if (nr_done != nr_deltas)
2575                        die("inconsistency with delta count");
2576        }
2577        free(delta_list);
2578}
2579
2580static int git_pack_config(const char *k, const char *v, void *cb)
2581{
2582        if (!strcmp(k, "pack.window")) {
2583                window = git_config_int(k, v);
2584                return 0;
2585        }
2586        if (!strcmp(k, "pack.windowmemory")) {
2587                window_memory_limit = git_config_ulong(k, v);
2588                return 0;
2589        }
2590        if (!strcmp(k, "pack.depth")) {
2591                depth = git_config_int(k, v);
2592                return 0;
2593        }
2594        if (!strcmp(k, "pack.deltacachesize")) {
2595                max_delta_cache_size = git_config_int(k, v);
2596                return 0;
2597        }
2598        if (!strcmp(k, "pack.deltacachelimit")) {
2599                cache_max_small_delta_size = git_config_int(k, v);
2600                return 0;
2601        }
2602        if (!strcmp(k, "pack.writebitmaphashcache")) {
2603                if (git_config_bool(k, v))
2604                        write_bitmap_options |= BITMAP_OPT_HASH_CACHE;
2605                else
2606                        write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;
2607        }
2608        if (!strcmp(k, "pack.usebitmaps")) {
2609                use_bitmap_index_default = git_config_bool(k, v);
2610                return 0;
2611        }
2612        if (!strcmp(k, "pack.threads")) {
2613                delta_search_threads = git_config_int(k, v);
2614                if (delta_search_threads < 0)
2615                        die("invalid number of threads specified (%d)",
2616                            delta_search_threads);
2617#ifdef NO_PTHREADS
2618                if (delta_search_threads != 1) {
2619                        warning("no threads support, ignoring %s", k);
2620                        delta_search_threads = 0;
2621                }
2622#endif
2623                return 0;
2624        }
2625        if (!strcmp(k, "pack.indexversion")) {
2626                pack_idx_opts.version = git_config_int(k, v);
2627                if (pack_idx_opts.version > 2)
2628                        die("bad pack.indexversion=%"PRIu32,
2629                            pack_idx_opts.version);
2630                return 0;
2631        }
2632        return git_default_config(k, v, cb);
2633}
2634
2635static void read_object_list_from_stdin(void)
2636{
2637        char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];
2638        struct object_id oid;
2639        const char *p;
2640
2641        for (;;) {
2642                if (!fgets(line, sizeof(line), stdin)) {
2643                        if (feof(stdin))
2644                                break;
2645                        if (!ferror(stdin))
2646                                die("BUG: fgets returned NULL, not EOF, not error!");
2647                        if (errno != EINTR)
2648                                die_errno("fgets");
2649                        clearerr(stdin);
2650                        continue;
2651                }
2652                if (line[0] == '-') {
2653                        if (get_oid_hex(line+1, &oid))
2654                                die("expected edge object ID, got garbage:\n %s",
2655                                    line);
2656                        add_preferred_base(&oid);
2657                        continue;
2658                }
2659                if (parse_oid_hex(line, &oid, &p))
2660                        die("expected object ID, got garbage:\n %s", line);
2661
2662                add_preferred_base_object(p + 1);
2663                add_object_entry(&oid, OBJ_NONE, p + 1, 0);
2664        }
2665}
2666
2667/* Remember to update object flag allocation in object.h */
2668#define OBJECT_ADDED (1u<<20)
2669
2670static void show_commit(struct commit *commit, void *data)
2671{
2672        add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);
2673        commit->object.flags |= OBJECT_ADDED;
2674
2675        if (write_bitmap_index)
2676                index_commit_for_bitmap(commit);
2677}
2678
2679static void show_object(struct object *obj, const char *name, void *data)
2680{
2681        add_preferred_base_object(name);
2682        add_object_entry(&obj->oid, obj->type, name, 0);
2683        obj->flags |= OBJECT_ADDED;
2684}
2685
2686static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)
2687{
2688        assert(arg_missing_action == MA_ALLOW_ANY);
2689
2690        /*
2691         * Quietly ignore ALL missing objects.  This avoids problems with
2692         * staging them now and getting an odd error later.
2693         */
2694        if (!has_object_file(&obj->oid))
2695                return;
2696
2697        show_object(obj, name, data);
2698}
2699
2700static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)
2701{
2702        assert(arg_missing_action == MA_ALLOW_PROMISOR);
2703
2704        /*
2705         * Quietly ignore EXPECTED missing objects.  This avoids problems with
2706         * staging them now and getting an odd error later.
2707         */
2708        if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))
2709                return;
2710
2711        show_object(obj, name, data);
2712}
2713
2714static int option_parse_missing_action(const struct option *opt,
2715                                       const char *arg, int unset)
2716{
2717        assert(arg);
2718        assert(!unset);
2719
2720        if (!strcmp(arg, "error")) {
2721                arg_missing_action = MA_ERROR;
2722                fn_show_object = show_object;
2723                return 0;
2724        }
2725
2726        if (!strcmp(arg, "allow-any")) {
2727                arg_missing_action = MA_ALLOW_ANY;
2728                fetch_if_missing = 0;
2729                fn_show_object = show_object__ma_allow_any;
2730                return 0;
2731        }
2732
2733        if (!strcmp(arg, "allow-promisor")) {
2734                arg_missing_action = MA_ALLOW_PROMISOR;
2735                fetch_if_missing = 0;
2736                fn_show_object = show_object__ma_allow_promisor;
2737                return 0;
2738        }
2739
2740        die(_("invalid value for --missing"));
2741        return 0;
2742}
2743
2744static void show_edge(struct commit *commit)
2745{
2746        add_preferred_base(&commit->object.oid);
2747}
2748
2749struct in_pack_object {
2750        off_t offset;
2751        struct object *object;
2752};
2753
2754struct in_pack {
2755        unsigned int alloc;
2756        unsigned int nr;
2757        struct in_pack_object *array;
2758};
2759
2760static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)
2761{
2762        in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);
2763        in_pack->array[in_pack->nr].object = object;
2764        in_pack->nr++;
2765}
2766
2767/*
2768 * Compare the objects in the offset order, in order to emulate the
2769 * "git rev-list --objects" output that produced the pack originally.
2770 */
2771static int ofscmp(const void *a_, const void *b_)
2772{
2773        struct in_pack_object *a = (struct in_pack_object *)a_;
2774        struct in_pack_object *b = (struct in_pack_object *)b_;
2775
2776        if (a->offset < b->offset)
2777                return -1;
2778        else if (a->offset > b->offset)
2779                return 1;
2780        else
2781                return oidcmp(&a->object->oid, &b->object->oid);
2782}
2783
2784static void add_objects_in_unpacked_packs(struct rev_info *revs)
2785{
2786        struct packed_git *p;
2787        struct in_pack in_pack;
2788        uint32_t i;
2789
2790        memset(&in_pack, 0, sizeof(in_pack));
2791
2792        for (p = get_packed_git(the_repository); p; p = p->next) {
2793                struct object_id oid;
2794                struct object *o;
2795
2796                if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
2797                        continue;
2798                if (open_pack_index(p))
2799                        die("cannot open pack index");
2800
2801                ALLOC_GROW(in_pack.array,
2802                           in_pack.nr + p->num_objects,
2803                           in_pack.alloc);
2804
2805                for (i = 0; i < p->num_objects; i++) {
2806                        nth_packed_object_oid(&oid, p, i);
2807                        o = lookup_unknown_object(oid.hash);
2808                        if (!(o->flags & OBJECT_ADDED))
2809                                mark_in_pack_object(o, p, &in_pack);
2810                        o->flags |= OBJECT_ADDED;
2811                }
2812        }
2813
2814        if (in_pack.nr) {
2815                QSORT(in_pack.array, in_pack.nr, ofscmp);
2816                for (i = 0; i < in_pack.nr; i++) {
2817                        struct object *o = in_pack.array[i].object;
2818                        add_object_entry(&o->oid, o->type, "", 0);
2819                }
2820        }
2821        free(in_pack.array);
2822}
2823
2824static int add_loose_object(const struct object_id *oid, const char *path,
2825                            void *data)
2826{
2827        enum object_type type = oid_object_info(the_repository, oid, NULL);
2828
2829        if (type < 0) {
2830                warning("loose object at %s could not be examined", path);
2831                return 0;
2832        }
2833
2834        add_object_entry(oid, type, "", 0);
2835        return 0;
2836}
2837
2838/*
2839 * We actually don't even have to worry about reachability here.
2840 * add_object_entry will weed out duplicates, so we just add every
2841 * loose object we find.
2842 */
2843static void add_unreachable_loose_objects(void)
2844{
2845        for_each_loose_file_in_objdir(get_object_directory(),
2846                                      add_loose_object,
2847                                      NULL, NULL, NULL);
2848}
2849
2850static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)
2851{
2852        static struct packed_git *last_found = (void *)1;
2853        struct packed_git *p;
2854
2855        p = (last_found != (void *)1) ? last_found :
2856                                        get_packed_git(the_repository);
2857
2858        while (p) {
2859                if ((!p->pack_local || p->pack_keep ||
2860                                p->pack_keep_in_core) &&
2861                        find_pack_entry_one(oid->hash, p)) {
2862                        last_found = p;
2863                        return 1;
2864                }
2865                if (p == last_found)
2866                        p = get_packed_git(the_repository);
2867                else
2868                        p = p->next;
2869                if (p == last_found)
2870                        p = p->next;
2871        }
2872        return 0;
2873}
2874
2875/*
2876 * Store a list of sha1s that are should not be discarded
2877 * because they are either written too recently, or are
2878 * reachable from another object that was.
2879 *
2880 * This is filled by get_object_list.
2881 */
2882static struct oid_array recent_objects;
2883
2884static int loosened_object_can_be_discarded(const struct object_id *oid,
2885                                            timestamp_t mtime)
2886{
2887        if (!unpack_unreachable_expiration)
2888                return 0;
2889        if (mtime > unpack_unreachable_expiration)
2890                return 0;
2891        if (oid_array_lookup(&recent_objects, oid) >= 0)
2892                return 0;
2893        return 1;
2894}
2895
2896static void loosen_unused_packed_objects(struct rev_info *revs)
2897{
2898        struct packed_git *p;
2899        uint32_t i;
2900        struct object_id oid;
2901
2902        for (p = get_packed_git(the_repository); p; p = p->next) {
2903                if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)
2904                        continue;
2905
2906                if (open_pack_index(p))
2907                        die("cannot open pack index");
2908
2909                for (i = 0; i < p->num_objects; i++) {
2910                        nth_packed_object_oid(&oid, p, i);
2911                        if (!packlist_find(&to_pack, oid.hash, NULL) &&
2912                            !has_sha1_pack_kept_or_nonlocal(&oid) &&
2913                            !loosened_object_can_be_discarded(&oid, p->mtime))
2914                                if (force_object_loose(&oid, p->mtime))
2915                                        die("unable to force loose object");
2916                }
2917        }
2918}
2919
2920/*
2921 * This tracks any options which pack-reuse code expects to be on, or which a
2922 * reader of the pack might not understand, and which would therefore prevent
2923 * blind reuse of what we have on disk.
2924 */
2925static int pack_options_allow_reuse(void)
2926{
2927        return pack_to_stdout &&
2928               allow_ofs_delta &&
2929               !ignore_packed_keep_on_disk &&
2930               !ignore_packed_keep_in_core &&
2931               (!local || !have_non_local_packs) &&
2932               !incremental;
2933}
2934
2935static int get_object_list_from_bitmap(struct rev_info *revs)
2936{
2937        if (prepare_bitmap_walk(revs) < 0)
2938                return -1;
2939
2940        if (pack_options_allow_reuse() &&
2941            !reuse_partial_packfile_from_bitmap(
2942                        &reuse_packfile,
2943                        &reuse_packfile_objects,
2944                        &reuse_packfile_offset)) {
2945                assert(reuse_packfile_objects);
2946                nr_result += reuse_packfile_objects;
2947                display_progress(progress_state, nr_result);
2948        }
2949
2950        traverse_bitmap_commit_list(&add_object_entry_from_bitmap);
2951        return 0;
2952}
2953
2954static void record_recent_object(struct object *obj,
2955                                 const char *name,
2956                                 void *data)
2957{
2958        oid_array_append(&recent_objects, &obj->oid);
2959}
2960
2961static void record_recent_commit(struct commit *commit, void *data)
2962{
2963        oid_array_append(&recent_objects, &commit->object.oid);
2964}
2965
2966static void get_object_list(int ac, const char **av)
2967{
2968        struct rev_info revs;
2969        char line[1000];
2970        int flags = 0;
2971
2972        init_revisions(&revs, NULL);
2973        save_commit_buffer = 0;
2974        setup_revisions(ac, av, &revs, NULL);
2975
2976        /* make sure shallows are read */
2977        is_repository_shallow();
2978
2979        while (fgets(line, sizeof(line), stdin) != NULL) {
2980                int len = strlen(line);
2981                if (len && line[len - 1] == '\n')
2982                        line[--len] = 0;
2983                if (!len)
2984                        break;
2985                if (*line == '-') {
2986                        if (!strcmp(line, "--not")) {
2987                                flags ^= UNINTERESTING;
2988                                write_bitmap_index = 0;
2989                                continue;
2990                        }
2991                        if (starts_with(line, "--shallow ")) {
2992                                struct object_id oid;
2993                                if (get_oid_hex(line + 10, &oid))
2994                                        die("not an SHA-1 '%s'", line + 10);
2995                                register_shallow(&oid);
2996                                use_bitmap_index = 0;
2997                                continue;
2998                        }
2999                        die("not a rev '%s'", line);
3000                }
3001                if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))
3002                        die("bad revision '%s'", line);
3003        }
3004
3005        if (use_bitmap_index && !get_object_list_from_bitmap(&revs))
3006                return;
3007
3008        if (prepare_revision_walk(&revs))
3009                die("revision walk setup failed");
3010        mark_edges_uninteresting(&revs, show_edge);
3011
3012        if (!fn_show_object)
3013                fn_show_object = show_object;
3014        traverse_commit_list_filtered(&filter_options, &revs,
3015                                      show_commit, fn_show_object, NULL,
3016                                      NULL);
3017
3018        if (unpack_unreachable_expiration) {
3019                revs.ignore_missing_links = 1;
3020                if (add_unseen_recent_objects_to_traversal(&revs,
3021                                unpack_unreachable_expiration))
3022                        die("unable to add recent objects");
3023                if (prepare_revision_walk(&revs))
3024                        die("revision walk setup failed");
3025                traverse_commit_list(&revs, record_recent_commit,
3026                                     record_recent_object, NULL);
3027        }
3028
3029        if (keep_unreachable)
3030                add_objects_in_unpacked_packs(&revs);
3031        if (pack_loose_unreachable)
3032                add_unreachable_loose_objects();
3033        if (unpack_unreachable)
3034                loosen_unused_packed_objects(&revs);
3035
3036        oid_array_clear(&recent_objects);
3037}
3038
3039static void add_extra_kept_packs(const struct string_list *names)
3040{
3041        struct packed_git *p;
3042
3043        if (!names->nr)
3044                return;
3045
3046        for (p = get_packed_git(the_repository); p; p = p->next) {
3047                const char *name = basename(p->pack_name);
3048                int i;
3049
3050                if (!p->pack_local)
3051                        continue;
3052
3053                for (i = 0; i < names->nr; i++)
3054                        if (!fspathcmp(name, names->items[i].string))
3055                                break;
3056
3057                if (i < names->nr) {
3058                        p->pack_keep_in_core = 1;
3059                        ignore_packed_keep_in_core = 1;
3060                        continue;
3061                }
3062        }
3063}
3064
3065static int option_parse_index_version(const struct option *opt,
3066                                      const char *arg, int unset)
3067{
3068        char *c;
3069        const char *val = arg;
3070        pack_idx_opts.version = strtoul(val, &c, 10);
3071        if (pack_idx_opts.version > 2)
3072                die(_("unsupported index version %s"), val);
3073        if (*c == ',' && c[1])
3074                pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);
3075        if (*c || pack_idx_opts.off32_limit & 0x80000000)
3076                die(_("bad index version '%s'"), val);
3077        return 0;
3078}
3079
3080static int option_parse_unpack_unreachable(const struct option *opt,
3081                                           const char *arg, int unset)
3082{
3083        if (unset) {
3084                unpack_unreachable = 0;
3085                unpack_unreachable_expiration = 0;
3086        }
3087        else {
3088                unpack_unreachable = 1;
3089                if (arg)
3090                        unpack_unreachable_expiration = approxidate(arg);
3091        }
3092        return 0;
3093}
3094
3095int cmd_pack_objects(int argc, const char **argv, const char *prefix)
3096{
3097        int use_internal_rev_list = 0;
3098        int thin = 0;
3099        int shallow = 0;
3100        int all_progress_implied = 0;
3101        struct argv_array rp = ARGV_ARRAY_INIT;
3102        int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;
3103        int rev_list_index = 0;
3104        struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;
3105        struct option pack_objects_options[] = {
3106                OPT_SET_INT('q', "quiet", &progress,
3107                            N_("do not show progress meter"), 0),
3108                OPT_SET_INT(0, "progress", &progress,
3109                            N_("show progress meter"), 1),
3110                OPT_SET_INT(0, "all-progress", &progress,
3111                            N_("show progress meter during object writing phase"), 2),
3112                OPT_BOOL(0, "all-progress-implied",
3113                         &all_progress_implied,
3114                         N_("similar to --all-progress when progress meter is shown")),
3115                { OPTION_CALLBACK, 0, "index-version", NULL, N_("version[,offset]"),
3116                  N_("write the pack index file in the specified idx format version"),
3117                  0, option_parse_index_version },
3118                OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,
3119                              N_("maximum size of each output pack file")),
3120                OPT_BOOL(0, "local", &local,
3121                         N_("ignore borrowed objects from alternate object store")),
3122                OPT_BOOL(0, "incremental", &incremental,
3123                         N_("ignore packed objects")),
3124                OPT_INTEGER(0, "window", &window,
3125                            N_("limit pack window by objects")),
3126                OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,
3127                              N_("limit pack window by memory in addition to object limit")),
3128                OPT_INTEGER(0, "depth", &depth,
3129                            N_("maximum length of delta chain allowed in the resulting pack")),
3130                OPT_BOOL(0, "reuse-delta", &reuse_delta,
3131                         N_("reuse existing deltas")),
3132                OPT_BOOL(0, "reuse-object", &reuse_object,
3133                         N_("reuse existing objects")),
3134                OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,
3135                         N_("use OFS_DELTA objects")),
3136                OPT_INTEGER(0, "threads", &delta_search_threads,
3137                            N_("use threads when searching for best delta matches")),
3138                OPT_BOOL(0, "non-empty", &non_empty,
3139                         N_("do not create an empty pack output")),
3140                OPT_BOOL(0, "revs", &use_internal_rev_list,
3141                         N_("read revision arguments from standard input")),
3142                OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,
3143                              N_("limit the objects to those that are not yet packed"),
3144                              1, PARSE_OPT_NONEG),
3145                OPT_SET_INT_F(0, "all", &rev_list_all,
3146                              N_("include objects reachable from any reference"),
3147                              1, PARSE_OPT_NONEG),
3148                OPT_SET_INT_F(0, "reflog", &rev_list_reflog,
3149                              N_("include objects referred by reflog entries"),
3150                              1, PARSE_OPT_NONEG),
3151                OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,
3152                              N_("include objects referred to by the index"),
3153                              1, PARSE_OPT_NONEG),
3154                OPT_BOOL(0, "stdout", &pack_to_stdout,
3155                         N_("output pack to stdout")),
3156                OPT_BOOL(0, "include-tag", &include_tag,
3157                         N_("include tag objects that refer to objects to be packed")),
3158                OPT_BOOL(0, "keep-unreachable", &keep_unreachable,
3159                         N_("keep unreachable objects")),
3160                OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,
3161                         N_("pack loose unreachable objects")),
3162                { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),
3163                  N_("unpack unreachable objects newer than <time>"),
3164                  PARSE_OPT_OPTARG, option_parse_unpack_unreachable },
3165                OPT_BOOL(0, "thin", &thin,
3166                         N_("create thin packs")),
3167                OPT_BOOL(0, "shallow", &shallow,
3168                         N_("create packs suitable for shallow fetches")),
3169                OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,
3170                         N_("ignore packs that have companion .keep file")),
3171                OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),
3172                                N_("ignore this pack")),
3173                OPT_INTEGER(0, "compression", &pack_compression_level,
3174                            N_("pack compression level")),
3175                OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,
3176                            N_("do not hide commits by grafts"), 0),
3177                OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,
3178                         N_("use a bitmap index if available to speed up counting objects")),
3179                OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,
3180                         N_("write a bitmap index together with the pack index")),
3181                OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),
3182                { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),
3183                  N_("handling for missing objects"), PARSE_OPT_NONEG,
3184                  option_parse_missing_action },
3185                OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,
3186                         N_("do not pack objects in promisor packfiles")),
3187                OPT_END(),
3188        };
3189
3190        if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))
3191                BUG("too many dfs states, increase OE_DFS_STATE_BITS");
3192
3193        check_replace_refs = 0;
3194
3195        reset_pack_idx_option(&pack_idx_opts);
3196        git_config(git_pack_config, NULL);
3197
3198        progress = isatty(2);
3199        argc = parse_options(argc, argv, prefix, pack_objects_options,
3200                             pack_usage, 0);
3201
3202        if (argc) {
3203                base_name = argv[0];
3204                argc--;
3205        }
3206        if (pack_to_stdout != !base_name || argc)
3207                usage_with_options(pack_usage, pack_objects_options);
3208
3209        if (depth >= (1 << OE_DEPTH_BITS)) {
3210                warning(_("delta chain depth %d is too deep, forcing %d"),
3211                        depth, (1 << OE_DEPTH_BITS) - 1);
3212                depth = (1 << OE_DEPTH_BITS) - 1;
3213        }
3214        if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {
3215                warning(_("pack.deltaCacheLimit is too high, forcing %d"),
3216                        (1U << OE_Z_DELTA_BITS) - 1);
3217                cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;
3218        }
3219
3220        argv_array_push(&rp, "pack-objects");
3221        if (thin) {
3222                use_internal_rev_list = 1;
3223                argv_array_push(&rp, shallow
3224                                ? "--objects-edge-aggressive"
3225                                : "--objects-edge");
3226        } else
3227                argv_array_push(&rp, "--objects");
3228
3229        if (rev_list_all) {
3230                use_internal_rev_list = 1;
3231                argv_array_push(&rp, "--all");
3232        }
3233        if (rev_list_reflog) {
3234                use_internal_rev_list = 1;
3235                argv_array_push(&rp, "--reflog");
3236        }
3237        if (rev_list_index) {
3238                use_internal_rev_list = 1;
3239                argv_array_push(&rp, "--indexed-objects");
3240        }
3241        if (rev_list_unpacked) {
3242                use_internal_rev_list = 1;
3243                argv_array_push(&rp, "--unpacked");
3244        }
3245
3246        if (exclude_promisor_objects) {
3247                use_internal_rev_list = 1;
3248                fetch_if_missing = 0;
3249                argv_array_push(&rp, "--exclude-promisor-objects");
3250        }
3251        if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)
3252                use_internal_rev_list = 1;
3253
3254        if (!reuse_object)
3255                reuse_delta = 0;
3256        if (pack_compression_level == -1)
3257                pack_compression_level = Z_DEFAULT_COMPRESSION;
3258        else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)
3259                die("bad pack compression level %d", pack_compression_level);
3260
3261        if (!delta_search_threads)      /* --threads=0 means autodetect */
3262                delta_search_threads = online_cpus();
3263
3264#ifdef NO_PTHREADS
3265        if (delta_search_threads != 1)
3266                warning("no threads support, ignoring --threads");
3267#endif
3268        if (!pack_to_stdout && !pack_size_limit)
3269                pack_size_limit = pack_size_limit_cfg;
3270        if (pack_to_stdout && pack_size_limit)
3271                die("--max-pack-size cannot be used to build a pack for transfer");
3272        if (pack_size_limit && pack_size_limit < 1024*1024) {
3273                warning("minimum pack size limit is 1 MiB");
3274                pack_size_limit = 1024*1024;
3275        }
3276
3277        if (!pack_to_stdout && thin)
3278                die("--thin cannot be used to build an indexable pack.");
3279
3280        if (keep_unreachable && unpack_unreachable)
3281                die("--keep-unreachable and --unpack-unreachable are incompatible");
3282        if (!rev_list_all || !rev_list_reflog || !rev_list_index)
3283                unpack_unreachable_expiration = 0;
3284
3285        if (filter_options.choice) {
3286                if (!pack_to_stdout)
3287                        die("cannot use --filter without --stdout");
3288                use_bitmap_index = 0;
3289        }
3290
3291        /*
3292         * "soft" reasons not to use bitmaps - for on-disk repack by default we want
3293         *
3294         * - to produce good pack (with bitmap index not-yet-packed objects are
3295         *   packed in suboptimal order).
3296         *
3297         * - to use more robust pack-generation codepath (avoiding possible
3298         *   bugs in bitmap code and possible bitmap index corruption).
3299         */
3300        if (!pack_to_stdout)
3301                use_bitmap_index_default = 0;
3302
3303        if (use_bitmap_index < 0)
3304                use_bitmap_index = use_bitmap_index_default;
3305
3306        /* "hard" reasons not to use bitmaps; these just won't work at all */
3307        if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow())
3308                use_bitmap_index = 0;
3309
3310        if (pack_to_stdout || !rev_list_all)
3311                write_bitmap_index = 0;
3312
3313        if (progress && all_progress_implied)
3314                progress = 2;
3315
3316        add_extra_kept_packs(&keep_pack_list);
3317        if (ignore_packed_keep_on_disk) {
3318                struct packed_git *p;
3319                for (p = get_packed_git(the_repository); p; p = p->next)
3320                        if (p->pack_local && p->pack_keep)
3321                                break;
3322                if (!p) /* no keep-able packs found */
3323                        ignore_packed_keep_on_disk = 0;
3324        }
3325        if (local) {
3326                /*
3327                 * unlike ignore_packed_keep_on_disk above, we do not
3328                 * want to unset "local" based on looking at packs, as
3329                 * it also covers non-local objects
3330                 */
3331                struct packed_git *p;
3332                for (p = get_packed_git(the_repository); p; p = p->next) {
3333                        if (!p->pack_local) {
3334                                have_non_local_packs = 1;
3335                                break;
3336                        }
3337                }
3338        }
3339
3340        prepare_packing_data(&to_pack);
3341
3342        if (progress)
3343                progress_state = start_progress(_("Enumerating objects"), 0);
3344        if (!use_internal_rev_list)
3345                read_object_list_from_stdin();
3346        else {
3347                get_object_list(rp.argc, rp.argv);
3348                argv_array_clear(&rp);
3349        }
3350        cleanup_preferred_base();
3351        if (include_tag && nr_result)
3352                for_each_ref(add_ref_tag, NULL);
3353        stop_progress(&progress_state);
3354
3355        if (non_empty && !nr_result)
3356                return 0;
3357        if (nr_result)
3358                prepare_pack(window, depth);
3359        write_pack_file();
3360        if (progress)
3361                fprintf_ln(stderr, "Total %"PRIu32" (delta %"PRIu32"),"
3362                           " reused %"PRIu32" (delta %"PRIu32")",
3363                           written, written_delta, reused, reused_delta);
3364        return 0;
3365}