41472fcbd09b4ab5c1732d7656ebc70e4fb563dd
   1#include "builtin.h"
   2#include "cache.h"
   3#include "attr.h"
   4#include "object.h"
   5#include "blob.h"
   6#include "commit.h"
   7#include "tag.h"
   8#include "tree.h"
   9#include "delta.h"
  10#include "pack.h"
  11#include "csum-file.h"
  12#include "tree-walk.h"
  13#include "diff.h"
  14#include "revision.h"
  15#include "list-objects.h"
  16#include "progress.h"
  17
  18static const char pack_usage[] = "\
  19git-pack-objects [{ -q | --progress | --all-progress }] [--max-pack-size=N] \n\
  20        [--local] [--incremental] [--window=N] [--depth=N] \n\
  21        [--no-reuse-delta] [--no-reuse-object] [--delta-base-offset] \n\
  22        [--non-empty] [--revs [--unpacked | --all]*] [--reflog] \n\
  23        [--stdout | base-name] [<ref-list | <object-list]";
  24
  25struct object_entry {
  26        unsigned char sha1[20];
  27        uint32_t crc32;         /* crc of raw pack data for this object */
  28        off_t offset;           /* offset into the final pack file */
  29        unsigned long size;     /* uncompressed size */
  30        unsigned int hash;      /* name hint hash */
  31        unsigned int depth;     /* delta depth */
  32        struct packed_git *in_pack;     /* already in pack */
  33        off_t in_pack_offset;
  34        struct object_entry *delta;     /* delta base object */
  35        struct object_entry *delta_child; /* deltified objects who bases me */
  36        struct object_entry *delta_sibling; /* other deltified objects who
  37                                             * uses the same base as me
  38                                             */
  39        void *delta_data;       /* cached delta (uncompressed) */
  40        unsigned long delta_size;       /* delta data size (uncompressed) */
  41        enum object_type type;
  42        enum object_type in_pack_type;  /* could be delta */
  43        unsigned char in_pack_header_size;
  44        unsigned char preferred_base; /* we do not pack this, but is available
  45                                       * to be used as the base object to delta
  46                                       * objects against.
  47                                       */
  48        unsigned char no_try_delta;
  49};
  50
  51/*
  52 * Objects we are going to pack are collected in objects array (dynamically
  53 * expanded).  nr_objects & nr_alloc controls this array.  They are stored
  54 * in the order we see -- typically rev-list --objects order that gives us
  55 * nice "minimum seek" order.
  56 */
  57static struct object_entry *objects;
  58static struct object_entry **written_list;
  59static uint32_t nr_objects, nr_alloc, nr_result, nr_written;
  60
  61static int non_empty;
  62static int no_reuse_delta, no_reuse_object;
  63static int local;
  64static int incremental;
  65static int allow_ofs_delta;
  66static const char *pack_tmp_name, *idx_tmp_name;
  67static char tmpname[PATH_MAX];
  68static const char *base_name;
  69static unsigned char pack_file_sha1[20];
  70static int progress = 1;
  71static int window = 10;
  72static uint32_t pack_size_limit;
  73static int depth = 50;
  74static int pack_to_stdout;
  75static int num_preferred_base;
  76static struct progress progress_state;
  77static int pack_compression_level = Z_DEFAULT_COMPRESSION;
  78static int pack_compression_seen;
  79
  80static unsigned long delta_cache_size = 0;
  81static unsigned long max_delta_cache_size = 0;
  82static unsigned long cache_max_small_delta_size = 1000;
  83
  84/*
  85 * The object names in objects array are hashed with this hashtable,
  86 * to help looking up the entry by object name.
  87 * This hashtable is built after all the objects are seen.
  88 */
  89static int *object_ix;
  90static int object_ix_hashsz;
  91
  92/*
  93 * Pack index for existing packs give us easy access to the offsets into
  94 * corresponding pack file where each object's data starts, but the entries
  95 * do not store the size of the compressed representation (uncompressed
  96 * size is easily available by examining the pack entry header).  It is
  97 * also rather expensive to find the sha1 for an object given its offset.
  98 *
  99 * We build a hashtable of existing packs (pack_revindex), and keep reverse
 100 * index here -- pack index file is sorted by object name mapping to offset;
 101 * this pack_revindex[].revindex array is a list of offset/index_nr pairs
 102 * ordered by offset, so if you know the offset of an object, next offset
 103 * is where its packed representation ends and the index_nr can be used to
 104 * get the object sha1 from the main index.
 105 */
 106struct revindex_entry {
 107        off_t offset;
 108        unsigned int nr;
 109};
 110struct pack_revindex {
 111        struct packed_git *p;
 112        struct revindex_entry *revindex;
 113};
 114static struct  pack_revindex *pack_revindex;
 115static int pack_revindex_hashsz;
 116
 117/*
 118 * stats
 119 */
 120static uint32_t written, written_delta;
 121static uint32_t reused, reused_delta;
 122
 123static int pack_revindex_ix(struct packed_git *p)
 124{
 125        unsigned long ui = (unsigned long)p;
 126        int i;
 127
 128        ui = ui ^ (ui >> 16); /* defeat structure alignment */
 129        i = (int)(ui % pack_revindex_hashsz);
 130        while (pack_revindex[i].p) {
 131                if (pack_revindex[i].p == p)
 132                        return i;
 133                if (++i == pack_revindex_hashsz)
 134                        i = 0;
 135        }
 136        return -1 - i;
 137}
 138
 139static void prepare_pack_ix(void)
 140{
 141        int num;
 142        struct packed_git *p;
 143        for (num = 0, p = packed_git; p; p = p->next)
 144                num++;
 145        if (!num)
 146                return;
 147        pack_revindex_hashsz = num * 11;
 148        pack_revindex = xcalloc(sizeof(*pack_revindex), pack_revindex_hashsz);
 149        for (p = packed_git; p; p = p->next) {
 150                num = pack_revindex_ix(p);
 151                num = - 1 - num;
 152                pack_revindex[num].p = p;
 153        }
 154        /* revindex elements are lazily initialized */
 155}
 156
 157static int cmp_offset(const void *a_, const void *b_)
 158{
 159        const struct revindex_entry *a = a_;
 160        const struct revindex_entry *b = b_;
 161        return (a->offset < b->offset) ? -1 : (a->offset > b->offset) ? 1 : 0;
 162}
 163
 164/*
 165 * Ordered list of offsets of objects in the pack.
 166 */
 167static void prepare_pack_revindex(struct pack_revindex *rix)
 168{
 169        struct packed_git *p = rix->p;
 170        int num_ent = p->num_objects;
 171        int i;
 172        const char *index = p->index_data;
 173
 174        rix->revindex = xmalloc(sizeof(*rix->revindex) * (num_ent + 1));
 175        index += 4 * 256;
 176
 177        if (p->index_version > 1) {
 178                const uint32_t *off_32 =
 179                        (uint32_t *)(index + 8 + p->num_objects * (20 + 4));
 180                const uint32_t *off_64 = off_32 + p->num_objects;
 181                for (i = 0; i < num_ent; i++) {
 182                        uint32_t off = ntohl(*off_32++);
 183                        if (!(off & 0x80000000)) {
 184                                rix->revindex[i].offset = off;
 185                        } else {
 186                                rix->revindex[i].offset =
 187                                        ((uint64_t)ntohl(*off_64++)) << 32;
 188                                rix->revindex[i].offset |=
 189                                        ntohl(*off_64++);
 190                        }
 191                        rix->revindex[i].nr = i;
 192                }
 193        } else {
 194                for (i = 0; i < num_ent; i++) {
 195                        uint32_t hl = *((uint32_t *)(index + 24 * i));
 196                        rix->revindex[i].offset = ntohl(hl);
 197                        rix->revindex[i].nr = i;
 198                }
 199        }
 200
 201        /* This knows the pack format -- the 20-byte trailer
 202         * follows immediately after the last object data.
 203         */
 204        rix->revindex[num_ent].offset = p->pack_size - 20;
 205        rix->revindex[num_ent].nr = -1;
 206        qsort(rix->revindex, num_ent, sizeof(*rix->revindex), cmp_offset);
 207}
 208
 209static struct revindex_entry * find_packed_object(struct packed_git *p,
 210                                                  off_t ofs)
 211{
 212        int num;
 213        int lo, hi;
 214        struct pack_revindex *rix;
 215        struct revindex_entry *revindex;
 216        num = pack_revindex_ix(p);
 217        if (num < 0)
 218                die("internal error: pack revindex uninitialized");
 219        rix = &pack_revindex[num];
 220        if (!rix->revindex)
 221                prepare_pack_revindex(rix);
 222        revindex = rix->revindex;
 223        lo = 0;
 224        hi = p->num_objects + 1;
 225        do {
 226                int mi = (lo + hi) / 2;
 227                if (revindex[mi].offset == ofs) {
 228                        return revindex + mi;
 229                }
 230                else if (ofs < revindex[mi].offset)
 231                        hi = mi;
 232                else
 233                        lo = mi + 1;
 234        } while (lo < hi);
 235        die("internal error: pack revindex corrupt");
 236}
 237
 238static const unsigned char *find_packed_object_name(struct packed_git *p,
 239                                                    off_t ofs)
 240{
 241        struct revindex_entry *entry = find_packed_object(p, ofs);
 242        return nth_packed_object_sha1(p, entry->nr);
 243}
 244
 245static void *delta_against(void *buf, unsigned long size, struct object_entry *entry)
 246{
 247        unsigned long othersize, delta_size;
 248        enum object_type type;
 249        void *otherbuf = read_sha1_file(entry->delta->sha1, &type, &othersize);
 250        void *delta_buf;
 251
 252        if (!otherbuf)
 253                die("unable to read %s", sha1_to_hex(entry->delta->sha1));
 254        delta_buf = diff_delta(otherbuf, othersize,
 255                               buf, size, &delta_size, 0);
 256        if (!delta_buf || delta_size != entry->delta_size)
 257                die("delta size changed");
 258        free(buf);
 259        free(otherbuf);
 260        return delta_buf;
 261}
 262
 263/*
 264 * The per-object header is a pretty dense thing, which is
 265 *  - first byte: low four bits are "size", then three bits of "type",
 266 *    and the high bit is "size continues".
 267 *  - each byte afterwards: low seven bits are size continuation,
 268 *    with the high bit being "size continues"
 269 */
 270static int encode_header(enum object_type type, unsigned long size, unsigned char *hdr)
 271{
 272        int n = 1;
 273        unsigned char c;
 274
 275        if (type < OBJ_COMMIT || type > OBJ_REF_DELTA)
 276                die("bad type %d", type);
 277
 278        c = (type << 4) | (size & 15);
 279        size >>= 4;
 280        while (size) {
 281                *hdr++ = c | 0x80;
 282                c = size & 0x7f;
 283                size >>= 7;
 284                n++;
 285        }
 286        *hdr = c;
 287        return n;
 288}
 289
 290/*
 291 * we are going to reuse the existing object data as is.  make
 292 * sure it is not corrupt.
 293 */
 294static int check_pack_inflate(struct packed_git *p,
 295                struct pack_window **w_curs,
 296                off_t offset,
 297                off_t len,
 298                unsigned long expect)
 299{
 300        z_stream stream;
 301        unsigned char fakebuf[4096], *in;
 302        int st;
 303
 304        memset(&stream, 0, sizeof(stream));
 305        inflateInit(&stream);
 306        do {
 307                in = use_pack(p, w_curs, offset, &stream.avail_in);
 308                stream.next_in = in;
 309                stream.next_out = fakebuf;
 310                stream.avail_out = sizeof(fakebuf);
 311                st = inflate(&stream, Z_FINISH);
 312                offset += stream.next_in - in;
 313        } while (st == Z_OK || st == Z_BUF_ERROR);
 314        inflateEnd(&stream);
 315        return (st == Z_STREAM_END &&
 316                stream.total_out == expect &&
 317                stream.total_in == len) ? 0 : -1;
 318}
 319
 320static int check_pack_crc(struct packed_git *p, struct pack_window **w_curs,
 321                          off_t offset, off_t len, unsigned int nr)
 322{
 323        const uint32_t *index_crc;
 324        uint32_t data_crc = crc32(0, Z_NULL, 0);
 325
 326        do {
 327                unsigned int avail;
 328                void *data = use_pack(p, w_curs, offset, &avail);
 329                if (avail > len)
 330                        avail = len;
 331                data_crc = crc32(data_crc, data, avail);
 332                offset += avail;
 333                len -= avail;
 334        } while (len);
 335
 336        index_crc = p->index_data;
 337        index_crc += 2 + 256 + p->num_objects * (20/4) + nr;
 338
 339        return data_crc != ntohl(*index_crc);
 340}
 341
 342static void copy_pack_data(struct sha1file *f,
 343                struct packed_git *p,
 344                struct pack_window **w_curs,
 345                off_t offset,
 346                off_t len)
 347{
 348        unsigned char *in;
 349        unsigned int avail;
 350
 351        while (len) {
 352                in = use_pack(p, w_curs, offset, &avail);
 353                if (avail > len)
 354                        avail = (unsigned int)len;
 355                sha1write(f, in, avail);
 356                offset += avail;
 357                len -= avail;
 358        }
 359}
 360
 361static unsigned long write_object(struct sha1file *f,
 362                                  struct object_entry *entry,
 363                                  off_t write_offset)
 364{
 365        unsigned long size;
 366        enum object_type type;
 367        void *buf;
 368        unsigned char header[10];
 369        unsigned char dheader[10];
 370        unsigned hdrlen;
 371        off_t datalen;
 372        enum object_type obj_type;
 373        int to_reuse = 0;
 374        /* write limit if limited packsize and not first object */
 375        unsigned long limit = pack_size_limit && nr_written ?
 376                                pack_size_limit - write_offset : 0;
 377                                /* no if no delta */
 378        int usable_delta =      !entry->delta ? 0 :
 379                                /* yes if unlimited packfile */
 380                                !pack_size_limit ? 1 :
 381                                /* no if base written to previous pack */
 382                                entry->delta->offset == (off_t)-1 ? 0 :
 383                                /* otherwise double-check written to this
 384                                 * pack,  like we do below
 385                                 */
 386                                entry->delta->offset ? 1 : 0;
 387
 388        if (!pack_to_stdout)
 389                crc32_begin(f);
 390
 391        obj_type = entry->type;
 392        if (no_reuse_object)
 393                to_reuse = 0;   /* explicit */
 394        else if (!entry->in_pack)
 395                to_reuse = 0;   /* can't reuse what we don't have */
 396        else if (obj_type == OBJ_REF_DELTA || obj_type == OBJ_OFS_DELTA)
 397                                /* check_object() decided it for us ... */
 398                to_reuse = usable_delta;
 399                                /* ... but pack split may override that */
 400        else if (obj_type != entry->in_pack_type)
 401                to_reuse = 0;   /* pack has delta which is unusable */
 402        else if (entry->delta)
 403                to_reuse = 0;   /* we want to pack afresh */
 404        else
 405                to_reuse = 1;   /* we have it in-pack undeltified,
 406                                 * and we do not need to deltify it.
 407                                 */
 408
 409        if (!to_reuse) {
 410                z_stream stream;
 411                unsigned long maxsize;
 412                void *out;
 413                if (entry->delta_data && usable_delta) {
 414                        buf = entry->delta_data;
 415                        size = entry->delta_size;
 416                        obj_type = (allow_ofs_delta && entry->delta->offset) ?
 417                                OBJ_OFS_DELTA : OBJ_REF_DELTA;
 418                } else {
 419                        buf = read_sha1_file(entry->sha1, &type, &size);
 420                        if (!buf)
 421                                die("unable to read %s", sha1_to_hex(entry->sha1));
 422                        if (size != entry->size)
 423                                die("object %s size inconsistency (%lu vs %lu)",
 424                                    sha1_to_hex(entry->sha1), size, entry->size);
 425                        if (usable_delta) {
 426                                buf = delta_against(buf, size, entry);
 427                                size = entry->delta_size;
 428                                obj_type = (allow_ofs_delta && entry->delta->offset) ?
 429                                        OBJ_OFS_DELTA : OBJ_REF_DELTA;
 430                        } else {
 431                                /*
 432                                 * recover real object type in case
 433                                 * check_object() wanted to re-use a delta,
 434                                 * but we couldn't since base was in previous split pack
 435                                 */
 436                                obj_type = type;
 437                        }
 438                }
 439                /* compress the data to store and put compressed length in datalen */
 440                memset(&stream, 0, sizeof(stream));
 441                deflateInit(&stream, pack_compression_level);
 442                maxsize = deflateBound(&stream, size);
 443                out = xmalloc(maxsize);
 444                /* Compress it */
 445                stream.next_in = buf;
 446                stream.avail_in = size;
 447                stream.next_out = out;
 448                stream.avail_out = maxsize;
 449                while (deflate(&stream, Z_FINISH) == Z_OK)
 450                        /* nothing */;
 451                deflateEnd(&stream);
 452                datalen = stream.total_out;
 453                deflateEnd(&stream);
 454                /*
 455                 * The object header is a byte of 'type' followed by zero or
 456                 * more bytes of length.
 457                 */
 458                hdrlen = encode_header(obj_type, size, header);
 459
 460                if (obj_type == OBJ_OFS_DELTA) {
 461                        /*
 462                         * Deltas with relative base contain an additional
 463                         * encoding of the relative offset for the delta
 464                         * base from this object's position in the pack.
 465                         */
 466                        off_t ofs = entry->offset - entry->delta->offset;
 467                        unsigned pos = sizeof(dheader) - 1;
 468                        dheader[pos] = ofs & 127;
 469                        while (ofs >>= 7)
 470                                dheader[--pos] = 128 | (--ofs & 127);
 471                        if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit) {
 472                                free(out);
 473                                free(buf);
 474                                return 0;
 475                        }
 476                        sha1write(f, header, hdrlen);
 477                        sha1write(f, dheader + pos, sizeof(dheader) - pos);
 478                        hdrlen += sizeof(dheader) - pos;
 479                } else if (obj_type == OBJ_REF_DELTA) {
 480                        /*
 481                         * Deltas with a base reference contain
 482                         * an additional 20 bytes for the base sha1.
 483                         */
 484                        if (limit && hdrlen + 20 + datalen + 20 >= limit) {
 485                                free(out);
 486                                free(buf);
 487                                return 0;
 488                        }
 489                        sha1write(f, header, hdrlen);
 490                        sha1write(f, entry->delta->sha1, 20);
 491                        hdrlen += 20;
 492                } else {
 493                        if (limit && hdrlen + datalen + 20 >= limit) {
 494                                free(out);
 495                                free(buf);
 496                                return 0;
 497                        }
 498                        sha1write(f, header, hdrlen);
 499                }
 500                sha1write(f, out, datalen);
 501                free(out);
 502                free(buf);
 503        }
 504        else {
 505                struct packed_git *p = entry->in_pack;
 506                struct pack_window *w_curs = NULL;
 507                struct revindex_entry *revidx;
 508                off_t offset;
 509
 510                if (entry->delta) {
 511                        obj_type = (allow_ofs_delta && entry->delta->offset) ?
 512                                OBJ_OFS_DELTA : OBJ_REF_DELTA;
 513                        reused_delta++;
 514                }
 515                hdrlen = encode_header(obj_type, entry->size, header);
 516                offset = entry->in_pack_offset;
 517                revidx = find_packed_object(p, offset);
 518                datalen = revidx[1].offset - offset;
 519                if (!pack_to_stdout && p->index_version > 1 &&
 520                    check_pack_crc(p, &w_curs, offset, datalen, revidx->nr))
 521                        die("bad packed object CRC for %s", sha1_to_hex(entry->sha1));
 522                offset += entry->in_pack_header_size;
 523                datalen -= entry->in_pack_header_size;
 524                if (obj_type == OBJ_OFS_DELTA) {
 525                        off_t ofs = entry->offset - entry->delta->offset;
 526                        unsigned pos = sizeof(dheader) - 1;
 527                        dheader[pos] = ofs & 127;
 528                        while (ofs >>= 7)
 529                                dheader[--pos] = 128 | (--ofs & 127);
 530                        if (limit && hdrlen + sizeof(dheader) - pos + datalen + 20 >= limit)
 531                                return 0;
 532                        sha1write(f, header, hdrlen);
 533                        sha1write(f, dheader + pos, sizeof(dheader) - pos);
 534                        hdrlen += sizeof(dheader) - pos;
 535                } else if (obj_type == OBJ_REF_DELTA) {
 536                        if (limit && hdrlen + 20 + datalen + 20 >= limit)
 537                                return 0;
 538                        sha1write(f, header, hdrlen);
 539                        sha1write(f, entry->delta->sha1, 20);
 540                        hdrlen += 20;
 541                } else {
 542                        if (limit && hdrlen + datalen + 20 >= limit)
 543                                return 0;
 544                        sha1write(f, header, hdrlen);
 545                }
 546
 547                if (!pack_to_stdout && p->index_version == 1 &&
 548                    check_pack_inflate(p, &w_curs, offset, datalen, entry->size))
 549                        die("corrupt packed object for %s", sha1_to_hex(entry->sha1));
 550                copy_pack_data(f, p, &w_curs, offset, datalen);
 551                unuse_pack(&w_curs);
 552                reused++;
 553        }
 554        if (usable_delta)
 555                written_delta++;
 556        written++;
 557        if (!pack_to_stdout)
 558                entry->crc32 = crc32_end(f);
 559        return hdrlen + datalen;
 560}
 561
 562static off_t write_one(struct sha1file *f,
 563                               struct object_entry *e,
 564                               off_t offset)
 565{
 566        unsigned long size;
 567
 568        /* offset is non zero if object is written already. */
 569        if (e->offset || e->preferred_base)
 570                return offset;
 571
 572        /* if we are deltified, write out base object first. */
 573        if (e->delta) {
 574                offset = write_one(f, e->delta, offset);
 575                if (!offset)
 576                        return 0;
 577        }
 578
 579        e->offset = offset;
 580        size = write_object(f, e, offset);
 581        if (!size) {
 582                e->offset = 0;
 583                return 0;
 584        }
 585        written_list[nr_written++] = e;
 586
 587        /* make sure off_t is sufficiently large not to wrap */
 588        if (offset > offset + size)
 589                die("pack too large for current definition of off_t");
 590        return offset + size;
 591}
 592
 593static int open_object_dir_tmp(const char *path)
 594{
 595    snprintf(tmpname, sizeof(tmpname), "%s/%s", get_object_directory(), path);
 596    return mkstemp(tmpname);
 597}
 598
 599/* forward declarations for write_pack_file */
 600static void write_index_file(off_t last_obj_offset, unsigned char *sha1);
 601static int adjust_perm(const char *path, mode_t mode);
 602
 603static void write_pack_file(void)
 604{
 605        uint32_t i = 0, j;
 606        struct sha1file *f;
 607        off_t offset, offset_one, last_obj_offset = 0;
 608        struct pack_header hdr;
 609        int do_progress = progress >> pack_to_stdout;
 610        uint32_t nr_remaining = nr_result;
 611
 612        if (do_progress)
 613                start_progress(&progress_state, "Writing %u objects...", "", nr_result);
 614        written_list = xmalloc(nr_objects * sizeof(struct object_entry *));
 615
 616        do {
 617                if (pack_to_stdout) {
 618                        f = sha1fd(1, "<stdout>");
 619                } else {
 620                        int fd = open_object_dir_tmp("tmp_pack_XXXXXX");
 621                        if (fd < 0)
 622                                die("unable to create %s: %s\n", tmpname, strerror(errno));
 623                        pack_tmp_name = xstrdup(tmpname);
 624                        f = sha1fd(fd, pack_tmp_name);
 625                }
 626
 627                hdr.hdr_signature = htonl(PACK_SIGNATURE);
 628                hdr.hdr_version = htonl(PACK_VERSION);
 629                hdr.hdr_entries = htonl(nr_remaining);
 630                sha1write(f, &hdr, sizeof(hdr));
 631                offset = sizeof(hdr);
 632                nr_written = 0;
 633                for (; i < nr_objects; i++) {
 634                        last_obj_offset = offset;
 635                        offset_one = write_one(f, objects + i, offset);
 636                        if (!offset_one)
 637                                break;
 638                        offset = offset_one;
 639                        if (do_progress)
 640                                display_progress(&progress_state, written);
 641                }
 642
 643                /*
 644                 * Did we write the wrong # entries in the header?
 645                 * If so, rewrite it like in fast-import
 646                 */
 647                if (pack_to_stdout || nr_written == nr_remaining) {
 648                        sha1close(f, pack_file_sha1, 1);
 649                } else {
 650                        sha1close(f, pack_file_sha1, 0);
 651                        fixup_pack_header_footer(f->fd, pack_file_sha1, pack_tmp_name, nr_written);
 652                        close(f->fd);
 653                }
 654
 655                if (!pack_to_stdout) {
 656                        unsigned char object_list_sha1[20];
 657                        mode_t mode = umask(0);
 658
 659                        umask(mode);
 660                        mode = 0444 & ~mode;
 661
 662                        write_index_file(last_obj_offset, object_list_sha1);
 663                        snprintf(tmpname, sizeof(tmpname), "%s-%s.pack",
 664                                 base_name, sha1_to_hex(object_list_sha1));
 665                        if (adjust_perm(pack_tmp_name, mode))
 666                                die("unable to make temporary pack file readable: %s",
 667                                    strerror(errno));
 668                        if (rename(pack_tmp_name, tmpname))
 669                                die("unable to rename temporary pack file: %s",
 670                                    strerror(errno));
 671                        snprintf(tmpname, sizeof(tmpname), "%s-%s.idx",
 672                                 base_name, sha1_to_hex(object_list_sha1));
 673                        if (adjust_perm(idx_tmp_name, mode))
 674                                die("unable to make temporary index file readable: %s",
 675                                    strerror(errno));
 676                        if (rename(idx_tmp_name, tmpname))
 677                                die("unable to rename temporary index file: %s",
 678                                    strerror(errno));
 679                        puts(sha1_to_hex(object_list_sha1));
 680                }
 681
 682                /* mark written objects as written to previous pack */
 683                for (j = 0; j < nr_written; j++) {
 684                        written_list[j]->offset = (off_t)-1;
 685                }
 686                nr_remaining -= nr_written;
 687        } while (nr_remaining && i < nr_objects);
 688
 689        free(written_list);
 690        if (do_progress)
 691                stop_progress(&progress_state);
 692        if (written != nr_result)
 693                die("wrote %u objects while expecting %u", written, nr_result);
 694        /*
 695         * We have scanned through [0 ... i).  Since we have written
 696         * the correct number of objects,  the remaining [i ... nr_objects)
 697         * items must be either already written (due to out-of-order delta base)
 698         * or a preferred base.  Count those which are neither and complain if any.
 699         */
 700        for (j = 0; i < nr_objects; i++) {
 701                struct object_entry *e = objects + i;
 702                j += !e->offset && !e->preferred_base;
 703        }
 704        if (j)
 705                die("wrote %u objects as expected but %u unwritten", written, j);
 706}
 707
 708static int sha1_sort(const void *_a, const void *_b)
 709{
 710        const struct object_entry *a = *(struct object_entry **)_a;
 711        const struct object_entry *b = *(struct object_entry **)_b;
 712        return hashcmp(a->sha1, b->sha1);
 713}
 714
 715static uint32_t index_default_version = 1;
 716static uint32_t index_off32_limit = 0x7fffffff;
 717
 718static void write_index_file(off_t last_obj_offset, unsigned char *sha1)
 719{
 720        struct sha1file *f;
 721        struct object_entry **sorted_by_sha, **list, **last;
 722        uint32_t array[256];
 723        uint32_t i, index_version;
 724        SHA_CTX ctx;
 725
 726        int fd = open_object_dir_tmp("tmp_idx_XXXXXX");
 727        if (fd < 0)
 728                die("unable to create %s: %s\n", tmpname, strerror(errno));
 729        idx_tmp_name = xstrdup(tmpname);
 730        f = sha1fd(fd, idx_tmp_name);
 731
 732        if (nr_written) {
 733                sorted_by_sha = written_list;
 734                qsort(sorted_by_sha, nr_written, sizeof(*sorted_by_sha), sha1_sort);
 735                list = sorted_by_sha;
 736                last = sorted_by_sha + nr_written;
 737        } else
 738                sorted_by_sha = list = last = NULL;
 739
 740        /* if last object's offset is >= 2^31 we should use index V2 */
 741        index_version = (last_obj_offset >> 31) ? 2 : index_default_version;
 742
 743        /* index versions 2 and above need a header */
 744        if (index_version >= 2) {
 745                struct pack_idx_header hdr;
 746                hdr.idx_signature = htonl(PACK_IDX_SIGNATURE);
 747                hdr.idx_version = htonl(index_version);
 748                sha1write(f, &hdr, sizeof(hdr));
 749        }
 750
 751        /*
 752         * Write the first-level table (the list is sorted,
 753         * but we use a 256-entry lookup to be able to avoid
 754         * having to do eight extra binary search iterations).
 755         */
 756        for (i = 0; i < 256; i++) {
 757                struct object_entry **next = list;
 758                while (next < last) {
 759                        struct object_entry *entry = *next;
 760                        if (entry->sha1[0] != i)
 761                                break;
 762                        next++;
 763                }
 764                array[i] = htonl(next - sorted_by_sha);
 765                list = next;
 766        }
 767        sha1write(f, array, 256 * 4);
 768
 769        /* Compute the SHA1 hash of sorted object names. */
 770        SHA1_Init(&ctx);
 771
 772        /* Write the actual SHA1 entries. */
 773        list = sorted_by_sha;
 774        for (i = 0; i < nr_written; i++) {
 775                struct object_entry *entry = *list++;
 776                if (index_version < 2) {
 777                        uint32_t offset = htonl(entry->offset);
 778                        sha1write(f, &offset, 4);
 779                }
 780                sha1write(f, entry->sha1, 20);
 781                SHA1_Update(&ctx, entry->sha1, 20);
 782        }
 783
 784        if (index_version >= 2) {
 785                unsigned int nr_large_offset = 0;
 786
 787                /* write the crc32 table */
 788                list = sorted_by_sha;
 789                for (i = 0; i < nr_written; i++) {
 790                        struct object_entry *entry = *list++;
 791                        uint32_t crc32_val = htonl(entry->crc32);
 792                        sha1write(f, &crc32_val, 4);
 793                }
 794
 795                /* write the 32-bit offset table */
 796                list = sorted_by_sha;
 797                for (i = 0; i < nr_written; i++) {
 798                        struct object_entry *entry = *list++;
 799                        uint32_t offset = (entry->offset <= index_off32_limit) ?
 800                                entry->offset : (0x80000000 | nr_large_offset++);
 801                        offset = htonl(offset);
 802                        sha1write(f, &offset, 4);
 803                }
 804
 805                /* write the large offset table */
 806                list = sorted_by_sha;
 807                while (nr_large_offset) {
 808                        struct object_entry *entry = *list++;
 809                        uint64_t offset = entry->offset;
 810                        if (offset > index_off32_limit) {
 811                                uint32_t split[2];
 812                                split[0]        = htonl(offset >> 32);
 813                                split[1] = htonl(offset & 0xffffffff);
 814                                sha1write(f, split, 8);
 815                                nr_large_offset--;
 816                        }
 817                }
 818        }
 819
 820        sha1write(f, pack_file_sha1, 20);
 821        sha1close(f, NULL, 1);
 822        SHA1_Final(sha1, &ctx);
 823}
 824
 825static int locate_object_entry_hash(const unsigned char *sha1)
 826{
 827        int i;
 828        unsigned int ui;
 829        memcpy(&ui, sha1, sizeof(unsigned int));
 830        i = ui % object_ix_hashsz;
 831        while (0 < object_ix[i]) {
 832                if (!hashcmp(sha1, objects[object_ix[i] - 1].sha1))
 833                        return i;
 834                if (++i == object_ix_hashsz)
 835                        i = 0;
 836        }
 837        return -1 - i;
 838}
 839
 840static struct object_entry *locate_object_entry(const unsigned char *sha1)
 841{
 842        int i;
 843
 844        if (!object_ix_hashsz)
 845                return NULL;
 846
 847        i = locate_object_entry_hash(sha1);
 848        if (0 <= i)
 849                return &objects[object_ix[i]-1];
 850        return NULL;
 851}
 852
 853static void rehash_objects(void)
 854{
 855        uint32_t i;
 856        struct object_entry *oe;
 857
 858        object_ix_hashsz = nr_objects * 3;
 859        if (object_ix_hashsz < 1024)
 860                object_ix_hashsz = 1024;
 861        object_ix = xrealloc(object_ix, sizeof(int) * object_ix_hashsz);
 862        memset(object_ix, 0, sizeof(int) * object_ix_hashsz);
 863        for (i = 0, oe = objects; i < nr_objects; i++, oe++) {
 864                int ix = locate_object_entry_hash(oe->sha1);
 865                if (0 <= ix)
 866                        continue;
 867                ix = -1 - ix;
 868                object_ix[ix] = i + 1;
 869        }
 870}
 871
 872static unsigned name_hash(const char *name)
 873{
 874        unsigned char c;
 875        unsigned hash = 0;
 876
 877        if (!name)
 878                return 0;
 879
 880        /*
 881         * This effectively just creates a sortable number from the
 882         * last sixteen non-whitespace characters. Last characters
 883         * count "most", so things that end in ".c" sort together.
 884         */
 885        while ((c = *name++) != 0) {
 886                if (isspace(c))
 887                        continue;
 888                hash = (hash >> 2) + (c << 24);
 889        }
 890        return hash;
 891}
 892
 893static void setup_delta_attr_check(struct git_attr_check *check)
 894{
 895        static struct git_attr *attr_delta;
 896
 897        if (!attr_delta)
 898                attr_delta = git_attr("delta", 5);
 899
 900        check[0].attr = attr_delta;
 901}
 902
 903static int no_try_delta(const char *path)
 904{
 905        struct git_attr_check check[1];
 906
 907        setup_delta_attr_check(check);
 908        if (git_checkattr(path, ARRAY_SIZE(check), check))
 909                return 0;
 910        if (ATTR_FALSE(check->value))
 911                return 1;
 912        return 0;
 913}
 914
 915static int add_object_entry(const unsigned char *sha1, enum object_type type,
 916                            const char *name, int exclude)
 917{
 918        struct object_entry *entry;
 919        struct packed_git *p, *found_pack = NULL;
 920        off_t found_offset = 0;
 921        int ix;
 922        unsigned hash = name_hash(name);
 923
 924        ix = nr_objects ? locate_object_entry_hash(sha1) : -1;
 925        if (ix >= 0) {
 926                if (exclude) {
 927                        entry = objects + object_ix[ix] - 1;
 928                        if (!entry->preferred_base)
 929                                nr_result--;
 930                        entry->preferred_base = 1;
 931                }
 932                return 0;
 933        }
 934
 935        for (p = packed_git; p; p = p->next) {
 936                off_t offset = find_pack_entry_one(sha1, p);
 937                if (offset) {
 938                        if (!found_pack) {
 939                                found_offset = offset;
 940                                found_pack = p;
 941                        }
 942                        if (exclude)
 943                                break;
 944                        if (incremental)
 945                                return 0;
 946                        if (local && !p->pack_local)
 947                                return 0;
 948                }
 949        }
 950
 951        if (nr_objects >= nr_alloc) {
 952                nr_alloc = (nr_alloc  + 1024) * 3 / 2;
 953                objects = xrealloc(objects, nr_alloc * sizeof(*entry));
 954        }
 955
 956        entry = objects + nr_objects++;
 957        memset(entry, 0, sizeof(*entry));
 958        hashcpy(entry->sha1, sha1);
 959        entry->hash = hash;
 960        if (type)
 961                entry->type = type;
 962        if (exclude)
 963                entry->preferred_base = 1;
 964        else
 965                nr_result++;
 966        if (found_pack) {
 967                entry->in_pack = found_pack;
 968                entry->in_pack_offset = found_offset;
 969        }
 970
 971        if (object_ix_hashsz * 3 <= nr_objects * 4)
 972                rehash_objects();
 973        else
 974                object_ix[-1 - ix] = nr_objects;
 975
 976        if (progress)
 977                display_progress(&progress_state, nr_objects);
 978
 979        if (name && no_try_delta(name))
 980                entry->no_try_delta = 1;
 981
 982        return 1;
 983}
 984
 985struct pbase_tree_cache {
 986        unsigned char sha1[20];
 987        int ref;
 988        int temporary;
 989        void *tree_data;
 990        unsigned long tree_size;
 991};
 992
 993static struct pbase_tree_cache *(pbase_tree_cache[256]);
 994static int pbase_tree_cache_ix(const unsigned char *sha1)
 995{
 996        return sha1[0] % ARRAY_SIZE(pbase_tree_cache);
 997}
 998static int pbase_tree_cache_ix_incr(int ix)
 999{
1000        return (ix+1) % ARRAY_SIZE(pbase_tree_cache);
1001}
1002
1003static struct pbase_tree {
1004        struct pbase_tree *next;
1005        /* This is a phony "cache" entry; we are not
1006         * going to evict it nor find it through _get()
1007         * mechanism -- this is for the toplevel node that
1008         * would almost always change with any commit.
1009         */
1010        struct pbase_tree_cache pcache;
1011} *pbase_tree;
1012
1013static struct pbase_tree_cache *pbase_tree_get(const unsigned char *sha1)
1014{
1015        struct pbase_tree_cache *ent, *nent;
1016        void *data;
1017        unsigned long size;
1018        enum object_type type;
1019        int neigh;
1020        int my_ix = pbase_tree_cache_ix(sha1);
1021        int available_ix = -1;
1022
1023        /* pbase-tree-cache acts as a limited hashtable.
1024         * your object will be found at your index or within a few
1025         * slots after that slot if it is cached.
1026         */
1027        for (neigh = 0; neigh < 8; neigh++) {
1028                ent = pbase_tree_cache[my_ix];
1029                if (ent && !hashcmp(ent->sha1, sha1)) {
1030                        ent->ref++;
1031                        return ent;
1032                }
1033                else if (((available_ix < 0) && (!ent || !ent->ref)) ||
1034                         ((0 <= available_ix) &&
1035                          (!ent && pbase_tree_cache[available_ix])))
1036                        available_ix = my_ix;
1037                if (!ent)
1038                        break;
1039                my_ix = pbase_tree_cache_ix_incr(my_ix);
1040        }
1041
1042        /* Did not find one.  Either we got a bogus request or
1043         * we need to read and perhaps cache.
1044         */
1045        data = read_sha1_file(sha1, &type, &size);
1046        if (!data)
1047                return NULL;
1048        if (type != OBJ_TREE) {
1049                free(data);
1050                return NULL;
1051        }
1052
1053        /* We need to either cache or return a throwaway copy */
1054
1055        if (available_ix < 0)
1056                ent = NULL;
1057        else {
1058                ent = pbase_tree_cache[available_ix];
1059                my_ix = available_ix;
1060        }
1061
1062        if (!ent) {
1063                nent = xmalloc(sizeof(*nent));
1064                nent->temporary = (available_ix < 0);
1065        }
1066        else {
1067                /* evict and reuse */
1068                free(ent->tree_data);
1069                nent = ent;
1070        }
1071        hashcpy(nent->sha1, sha1);
1072        nent->tree_data = data;
1073        nent->tree_size = size;
1074        nent->ref = 1;
1075        if (!nent->temporary)
1076                pbase_tree_cache[my_ix] = nent;
1077        return nent;
1078}
1079
1080static void pbase_tree_put(struct pbase_tree_cache *cache)
1081{
1082        if (!cache->temporary) {
1083                cache->ref--;
1084                return;
1085        }
1086        free(cache->tree_data);
1087        free(cache);
1088}
1089
1090static int name_cmp_len(const char *name)
1091{
1092        int i;
1093        for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)
1094                ;
1095        return i;
1096}
1097
1098static void add_pbase_object(struct tree_desc *tree,
1099                             const char *name,
1100                             int cmplen,
1101                             const char *fullname)
1102{
1103        struct name_entry entry;
1104        int cmp;
1105
1106        while (tree_entry(tree,&entry)) {
1107                cmp = tree_entry_len(entry.path, entry.sha1) != cmplen ? 1 :
1108                      memcmp(name, entry.path, cmplen);
1109                if (cmp > 0)
1110                        continue;
1111                if (cmp < 0)
1112                        return;
1113                if (name[cmplen] != '/') {
1114                        add_object_entry(entry.sha1,
1115                                         S_ISDIR(entry.mode) ? OBJ_TREE : OBJ_BLOB,
1116                                         fullname, 1);
1117                        return;
1118                }
1119                if (S_ISDIR(entry.mode)) {
1120                        struct tree_desc sub;
1121                        struct pbase_tree_cache *tree;
1122                        const char *down = name+cmplen+1;
1123                        int downlen = name_cmp_len(down);
1124
1125                        tree = pbase_tree_get(entry.sha1);
1126                        if (!tree)
1127                                return;
1128                        init_tree_desc(&sub, tree->tree_data, tree->tree_size);
1129
1130                        add_pbase_object(&sub, down, downlen, fullname);
1131                        pbase_tree_put(tree);
1132                }
1133        }
1134}
1135
1136static unsigned *done_pbase_paths;
1137static int done_pbase_paths_num;
1138static int done_pbase_paths_alloc;
1139static int done_pbase_path_pos(unsigned hash)
1140{
1141        int lo = 0;
1142        int hi = done_pbase_paths_num;
1143        while (lo < hi) {
1144                int mi = (hi + lo) / 2;
1145                if (done_pbase_paths[mi] == hash)
1146                        return mi;
1147                if (done_pbase_paths[mi] < hash)
1148                        hi = mi;
1149                else
1150                        lo = mi + 1;
1151        }
1152        return -lo-1;
1153}
1154
1155static int check_pbase_path(unsigned hash)
1156{
1157        int pos = (!done_pbase_paths) ? -1 : done_pbase_path_pos(hash);
1158        if (0 <= pos)
1159                return 1;
1160        pos = -pos - 1;
1161        if (done_pbase_paths_alloc <= done_pbase_paths_num) {
1162                done_pbase_paths_alloc = alloc_nr(done_pbase_paths_alloc);
1163                done_pbase_paths = xrealloc(done_pbase_paths,
1164                                            done_pbase_paths_alloc *
1165                                            sizeof(unsigned));
1166        }
1167        done_pbase_paths_num++;
1168        if (pos < done_pbase_paths_num)
1169                memmove(done_pbase_paths + pos + 1,
1170                        done_pbase_paths + pos,
1171                        (done_pbase_paths_num - pos - 1) * sizeof(unsigned));
1172        done_pbase_paths[pos] = hash;
1173        return 0;
1174}
1175
1176static void add_preferred_base_object(const char *name)
1177{
1178        struct pbase_tree *it;
1179        int cmplen;
1180        unsigned hash = name_hash(name);
1181
1182        if (!num_preferred_base || check_pbase_path(hash))
1183                return;
1184
1185        cmplen = name_cmp_len(name);
1186        for (it = pbase_tree; it; it = it->next) {
1187                if (cmplen == 0) {
1188                        add_object_entry(it->pcache.sha1, OBJ_TREE, NULL, 1);
1189                }
1190                else {
1191                        struct tree_desc tree;
1192                        init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);
1193                        add_pbase_object(&tree, name, cmplen, name);
1194                }
1195        }
1196}
1197
1198static void add_preferred_base(unsigned char *sha1)
1199{
1200        struct pbase_tree *it;
1201        void *data;
1202        unsigned long size;
1203        unsigned char tree_sha1[20];
1204
1205        if (window <= num_preferred_base++)
1206                return;
1207
1208        data = read_object_with_reference(sha1, tree_type, &size, tree_sha1);
1209        if (!data)
1210                return;
1211
1212        for (it = pbase_tree; it; it = it->next) {
1213                if (!hashcmp(it->pcache.sha1, tree_sha1)) {
1214                        free(data);
1215                        return;
1216                }
1217        }
1218
1219        it = xcalloc(1, sizeof(*it));
1220        it->next = pbase_tree;
1221        pbase_tree = it;
1222
1223        hashcpy(it->pcache.sha1, tree_sha1);
1224        it->pcache.tree_data = data;
1225        it->pcache.tree_size = size;
1226}
1227
1228static void check_object(struct object_entry *entry)
1229{
1230        if (entry->in_pack) {
1231                struct packed_git *p = entry->in_pack;
1232                struct pack_window *w_curs = NULL;
1233                const unsigned char *base_ref = NULL;
1234                struct object_entry *base_entry;
1235                unsigned long used, used_0;
1236                unsigned int avail;
1237                off_t ofs;
1238                unsigned char *buf, c;
1239
1240                buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);
1241
1242                /*
1243                 * We want in_pack_type even if we do not reuse delta
1244                 * since non-delta representations could still be reused.
1245                 */
1246                used = unpack_object_header_gently(buf, avail,
1247                                                   &entry->in_pack_type,
1248                                                   &entry->size);
1249
1250                /*
1251                 * Determine if this is a delta and if so whether we can
1252                 * reuse it or not.  Otherwise let's find out as cheaply as
1253                 * possible what the actual type and size for this object is.
1254                 */
1255                switch (entry->in_pack_type) {
1256                default:
1257                        /* Not a delta hence we've already got all we need. */
1258                        entry->type = entry->in_pack_type;
1259                        entry->in_pack_header_size = used;
1260                        unuse_pack(&w_curs);
1261                        return;
1262                case OBJ_REF_DELTA:
1263                        if (!no_reuse_delta && !entry->preferred_base)
1264                                base_ref = use_pack(p, &w_curs,
1265                                                entry->in_pack_offset + used, NULL);
1266                        entry->in_pack_header_size = used + 20;
1267                        break;
1268                case OBJ_OFS_DELTA:
1269                        buf = use_pack(p, &w_curs,
1270                                       entry->in_pack_offset + used, NULL);
1271                        used_0 = 0;
1272                        c = buf[used_0++];
1273                        ofs = c & 127;
1274                        while (c & 128) {
1275                                ofs += 1;
1276                                if (!ofs || MSB(ofs, 7))
1277                                        die("delta base offset overflow in pack for %s",
1278                                            sha1_to_hex(entry->sha1));
1279                                c = buf[used_0++];
1280                                ofs = (ofs << 7) + (c & 127);
1281                        }
1282                        if (ofs >= entry->in_pack_offset)
1283                                die("delta base offset out of bound for %s",
1284                                    sha1_to_hex(entry->sha1));
1285                        ofs = entry->in_pack_offset - ofs;
1286                        if (!no_reuse_delta && !entry->preferred_base)
1287                                base_ref = find_packed_object_name(p, ofs);
1288                        entry->in_pack_header_size = used + used_0;
1289                        break;
1290                }
1291
1292                if (base_ref && (base_entry = locate_object_entry(base_ref))) {
1293                        /*
1294                         * If base_ref was set above that means we wish to
1295                         * reuse delta data, and we even found that base
1296                         * in the list of objects we want to pack. Goodie!
1297                         *
1298                         * Depth value does not matter - find_deltas() will
1299                         * never consider reused delta as the base object to
1300                         * deltify other objects against, in order to avoid
1301                         * circular deltas.
1302                         */
1303                        entry->type = entry->in_pack_type;
1304                        entry->delta = base_entry;
1305                        entry->delta_sibling = base_entry->delta_child;
1306                        base_entry->delta_child = entry;
1307                        unuse_pack(&w_curs);
1308                        return;
1309                }
1310
1311                if (entry->type) {
1312                        /*
1313                         * This must be a delta and we already know what the
1314                         * final object type is.  Let's extract the actual
1315                         * object size from the delta header.
1316                         */
1317                        entry->size = get_size_from_delta(p, &w_curs,
1318                                        entry->in_pack_offset + entry->in_pack_header_size);
1319                        unuse_pack(&w_curs);
1320                        return;
1321                }
1322
1323                /*
1324                 * No choice but to fall back to the recursive delta walk
1325                 * with sha1_object_info() to find about the object type
1326                 * at this point...
1327                 */
1328                unuse_pack(&w_curs);
1329        }
1330
1331        entry->type = sha1_object_info(entry->sha1, &entry->size);
1332        if (entry->type < 0)
1333                die("unable to get type of object %s",
1334                    sha1_to_hex(entry->sha1));
1335}
1336
1337static int pack_offset_sort(const void *_a, const void *_b)
1338{
1339        const struct object_entry *a = *(struct object_entry **)_a;
1340        const struct object_entry *b = *(struct object_entry **)_b;
1341
1342        /* avoid filesystem trashing with loose objects */
1343        if (!a->in_pack && !b->in_pack)
1344                return hashcmp(a->sha1, b->sha1);
1345
1346        if (a->in_pack < b->in_pack)
1347                return -1;
1348        if (a->in_pack > b->in_pack)
1349                return 1;
1350        return a->in_pack_offset < b->in_pack_offset ? -1 :
1351                        (a->in_pack_offset > b->in_pack_offset);
1352}
1353
1354static void get_object_details(void)
1355{
1356        uint32_t i;
1357        struct object_entry **sorted_by_offset;
1358
1359        sorted_by_offset = xcalloc(nr_objects, sizeof(struct object_entry *));
1360        for (i = 0; i < nr_objects; i++)
1361                sorted_by_offset[i] = objects + i;
1362        qsort(sorted_by_offset, nr_objects, sizeof(*sorted_by_offset), pack_offset_sort);
1363
1364        prepare_pack_ix();
1365        for (i = 0; i < nr_objects; i++)
1366                check_object(sorted_by_offset[i]);
1367        free(sorted_by_offset);
1368}
1369
1370static int type_size_sort(const void *_a, const void *_b)
1371{
1372        const struct object_entry *a = *(struct object_entry **)_a;
1373        const struct object_entry *b = *(struct object_entry **)_b;
1374
1375        if (a->type < b->type)
1376                return -1;
1377        if (a->type > b->type)
1378                return 1;
1379        if (a->hash < b->hash)
1380                return -1;
1381        if (a->hash > b->hash)
1382                return 1;
1383        if (a->preferred_base < b->preferred_base)
1384                return -1;
1385        if (a->preferred_base > b->preferred_base)
1386                return 1;
1387        if (a->size < b->size)
1388                return -1;
1389        if (a->size > b->size)
1390                return 1;
1391        return a > b ? -1 : (a < b);  /* newest last */
1392}
1393
1394struct unpacked {
1395        struct object_entry *entry;
1396        void *data;
1397        struct delta_index *index;
1398};
1399
1400static int delta_cacheable(struct unpacked *trg, struct unpacked *src,
1401                            unsigned long src_size, unsigned long trg_size,
1402                            unsigned long delta_size)
1403{
1404        if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)
1405                return 0;
1406
1407        if (delta_size < cache_max_small_delta_size)
1408                return 1;
1409
1410        /* cache delta, if objects are large enough compared to delta size */
1411        if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))
1412                return 1;
1413
1414        return 0;
1415}
1416
1417/*
1418 * We search for deltas _backwards_ in a list sorted by type and
1419 * by size, so that we see progressively smaller and smaller files.
1420 * That's because we prefer deltas to be from the bigger file
1421 * to the smaller - deletes are potentially cheaper, but perhaps
1422 * more importantly, the bigger file is likely the more recent
1423 * one.
1424 */
1425static int try_delta(struct unpacked *trg, struct unpacked *src,
1426                     unsigned max_depth)
1427{
1428        struct object_entry *trg_entry = trg->entry;
1429        struct object_entry *src_entry = src->entry;
1430        unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;
1431        enum object_type type;
1432        void *delta_buf;
1433
1434        /* Don't bother doing diffs between different types */
1435        if (trg_entry->type != src_entry->type)
1436                return -1;
1437
1438        /* We do not compute delta to *create* objects we are not
1439         * going to pack.
1440         */
1441        if (trg_entry->preferred_base)
1442                return -1;
1443
1444        /*
1445         * We do not bother to try a delta that we discarded
1446         * on an earlier try, but only when reusing delta data.
1447         */
1448        if (!no_reuse_delta && trg_entry->in_pack &&
1449            trg_entry->in_pack == src_entry->in_pack &&
1450            trg_entry->in_pack_type != OBJ_REF_DELTA &&
1451            trg_entry->in_pack_type != OBJ_OFS_DELTA)
1452                return 0;
1453
1454        /* Let's not bust the allowed depth. */
1455        if (src_entry->depth >= max_depth)
1456                return 0;
1457
1458        /* Now some size filtering heuristics. */
1459        trg_size = trg_entry->size;
1460        max_size = trg_size/2 - 20;
1461        max_size = max_size * (max_depth - src_entry->depth) / max_depth;
1462        if (max_size == 0)
1463                return 0;
1464        if (trg_entry->delta && trg_entry->delta_size <= max_size)
1465                max_size = trg_entry->delta_size-1;
1466        src_size = src_entry->size;
1467        sizediff = src_size < trg_size ? trg_size - src_size : 0;
1468        if (sizediff >= max_size)
1469                return 0;
1470
1471        /* Load data if not already done */
1472        if (!trg->data) {
1473                trg->data = read_sha1_file(trg_entry->sha1, &type, &sz);
1474                if (sz != trg_size)
1475                        die("object %s inconsistent object length (%lu vs %lu)",
1476                            sha1_to_hex(trg_entry->sha1), sz, trg_size);
1477        }
1478        if (!src->data) {
1479                src->data = read_sha1_file(src_entry->sha1, &type, &sz);
1480                if (sz != src_size)
1481                        die("object %s inconsistent object length (%lu vs %lu)",
1482                            sha1_to_hex(src_entry->sha1), sz, src_size);
1483        }
1484        if (!src->index) {
1485                src->index = create_delta_index(src->data, src_size);
1486                if (!src->index) {
1487                        static int warned = 0;
1488                        if (!warned++)
1489                                warning("suboptimal pack - out of memory");
1490                        return 0;
1491                }
1492        }
1493
1494        delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);
1495        if (!delta_buf)
1496                return 0;
1497
1498        if (trg_entry->delta_data) {
1499                delta_cache_size -= trg_entry->delta_size;
1500                free(trg_entry->delta_data);
1501        }
1502        trg_entry->delta_data = 0;
1503        trg_entry->delta = src_entry;
1504        trg_entry->delta_size = delta_size;
1505        trg_entry->depth = src_entry->depth + 1;
1506
1507        if (delta_cacheable(src, trg, src_size, trg_size, delta_size)) {
1508                trg_entry->delta_data = xrealloc(delta_buf, delta_size);
1509                delta_cache_size += trg_entry->delta_size;
1510        } else
1511                free(delta_buf);
1512        return 1;
1513}
1514
1515static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)
1516{
1517        struct object_entry *child = me->delta_child;
1518        unsigned int m = n;
1519        while (child) {
1520                unsigned int c = check_delta_limit(child, n + 1);
1521                if (m < c)
1522                        m = c;
1523                child = child->delta_sibling;
1524        }
1525        return m;
1526}
1527
1528static void find_deltas(struct object_entry **list, int window, int depth)
1529{
1530        uint32_t i = nr_objects, idx = 0, processed = 0;
1531        unsigned int array_size = window * sizeof(struct unpacked);
1532        struct unpacked *array;
1533        int max_depth;
1534
1535        if (!nr_objects)
1536                return;
1537        array = xmalloc(array_size);
1538        memset(array, 0, array_size);
1539        if (progress)
1540                start_progress(&progress_state, "Deltifying %u objects...", "", nr_result);
1541
1542        do {
1543                struct object_entry *entry = list[--i];
1544                struct unpacked *n = array + idx;
1545                int j;
1546
1547                if (!entry->preferred_base)
1548                        processed++;
1549
1550                if (progress)
1551                        display_progress(&progress_state, processed);
1552
1553                if (entry->delta)
1554                        /* This happens if we decided to reuse existing
1555                         * delta from a pack.  "!no_reuse_delta &&" is implied.
1556                         */
1557                        continue;
1558
1559                if (entry->size < 50)
1560                        continue;
1561
1562                if (entry->no_try_delta)
1563                        continue;
1564
1565                free_delta_index(n->index);
1566                n->index = NULL;
1567                free(n->data);
1568                n->data = NULL;
1569                n->entry = entry;
1570
1571                /*
1572                 * If the current object is at pack edge, take the depth the
1573                 * objects that depend on the current object into account
1574                 * otherwise they would become too deep.
1575                 */
1576                max_depth = depth;
1577                if (entry->delta_child) {
1578                        max_depth -= check_delta_limit(entry, 0);
1579                        if (max_depth <= 0)
1580                                goto next;
1581                }
1582
1583                j = window;
1584                while (--j > 0) {
1585                        uint32_t other_idx = idx + j;
1586                        struct unpacked *m;
1587                        if (other_idx >= window)
1588                                other_idx -= window;
1589                        m = array + other_idx;
1590                        if (!m->entry)
1591                                break;
1592                        if (try_delta(n, m, max_depth) < 0)
1593                                break;
1594                }
1595
1596                /* if we made n a delta, and if n is already at max
1597                 * depth, leaving it in the window is pointless.  we
1598                 * should evict it first.
1599                 */
1600                if (entry->delta && depth <= entry->depth)
1601                        continue;
1602
1603                next:
1604                idx++;
1605                if (idx >= window)
1606                        idx = 0;
1607        } while (i > 0);
1608
1609        if (progress)
1610                stop_progress(&progress_state);
1611
1612        for (i = 0; i < window; ++i) {
1613                free_delta_index(array[i].index);
1614                free(array[i].data);
1615        }
1616        free(array);
1617}
1618
1619static void prepare_pack(int window, int depth)
1620{
1621        struct object_entry **delta_list;
1622        uint32_t i;
1623
1624        get_object_details();
1625
1626        if (!window || !depth)
1627                return;
1628
1629        delta_list = xmalloc(nr_objects * sizeof(*delta_list));
1630        for (i = 0; i < nr_objects; i++)
1631                delta_list[i] = objects + i;
1632        qsort(delta_list, nr_objects, sizeof(*delta_list), type_size_sort);
1633        find_deltas(delta_list, window+1, depth);
1634        free(delta_list);
1635}
1636
1637static int git_pack_config(const char *k, const char *v)
1638{
1639        if(!strcmp(k, "pack.window")) {
1640                window = git_config_int(k, v);
1641                return 0;
1642        }
1643        if(!strcmp(k, "pack.depth")) {
1644                depth = git_config_int(k, v);
1645                return 0;
1646        }
1647        if (!strcmp(k, "pack.compression")) {
1648                int level = git_config_int(k, v);
1649                if (level == -1)
1650                        level = Z_DEFAULT_COMPRESSION;
1651                else if (level < 0 || level > Z_BEST_COMPRESSION)
1652                        die("bad pack compression level %d", level);
1653                pack_compression_level = level;
1654                pack_compression_seen = 1;
1655                return 0;
1656        }
1657        if (!strcmp(k, "pack.deltacachesize")) {
1658                max_delta_cache_size = git_config_int(k, v);
1659                return 0;
1660        }
1661        if (!strcmp(k, "pack.deltacachelimit")) {
1662                cache_max_small_delta_size = git_config_int(k, v);
1663                return 0;
1664        }
1665        return git_default_config(k, v);
1666}
1667
1668static void read_object_list_from_stdin(void)
1669{
1670        char line[40 + 1 + PATH_MAX + 2];
1671        unsigned char sha1[20];
1672
1673        for (;;) {
1674                if (!fgets(line, sizeof(line), stdin)) {
1675                        if (feof(stdin))
1676                                break;
1677                        if (!ferror(stdin))
1678                                die("fgets returned NULL, not EOF, not error!");
1679                        if (errno != EINTR)
1680                                die("fgets: %s", strerror(errno));
1681                        clearerr(stdin);
1682                        continue;
1683                }
1684                if (line[0] == '-') {
1685                        if (get_sha1_hex(line+1, sha1))
1686                                die("expected edge sha1, got garbage:\n %s",
1687                                    line);
1688                        add_preferred_base(sha1);
1689                        continue;
1690                }
1691                if (get_sha1_hex(line, sha1))
1692                        die("expected sha1, got garbage:\n %s", line);
1693
1694                add_preferred_base_object(line+41);
1695                add_object_entry(sha1, 0, line+41, 0);
1696        }
1697}
1698
1699static void show_commit(struct commit *commit)
1700{
1701        add_object_entry(commit->object.sha1, OBJ_COMMIT, NULL, 0);
1702}
1703
1704static void show_object(struct object_array_entry *p)
1705{
1706        add_preferred_base_object(p->name);
1707        add_object_entry(p->item->sha1, p->item->type, p->name, 0);
1708}
1709
1710static void show_edge(struct commit *commit)
1711{
1712        add_preferred_base(commit->object.sha1);
1713}
1714
1715static void get_object_list(int ac, const char **av)
1716{
1717        struct rev_info revs;
1718        char line[1000];
1719        int flags = 0;
1720
1721        init_revisions(&revs, NULL);
1722        save_commit_buffer = 0;
1723        track_object_refs = 0;
1724        setup_revisions(ac, av, &revs, NULL);
1725
1726        while (fgets(line, sizeof(line), stdin) != NULL) {
1727                int len = strlen(line);
1728                if (line[len - 1] == '\n')
1729                        line[--len] = 0;
1730                if (!len)
1731                        break;
1732                if (*line == '-') {
1733                        if (!strcmp(line, "--not")) {
1734                                flags ^= UNINTERESTING;
1735                                continue;
1736                        }
1737                        die("not a rev '%s'", line);
1738                }
1739                if (handle_revision_arg(line, &revs, flags, 1))
1740                        die("bad revision '%s'", line);
1741        }
1742
1743        prepare_revision_walk(&revs);
1744        mark_edges_uninteresting(revs.commits, &revs, show_edge);
1745        traverse_commit_list(&revs, show_commit, show_object);
1746}
1747
1748static int adjust_perm(const char *path, mode_t mode)
1749{
1750        if (chmod(path, mode))
1751                return -1;
1752        return adjust_shared_perm(path);
1753}
1754
1755int cmd_pack_objects(int argc, const char **argv, const char *prefix)
1756{
1757        int use_internal_rev_list = 0;
1758        int thin = 0;
1759        uint32_t i;
1760        const char **rp_av;
1761        int rp_ac_alloc = 64;
1762        int rp_ac;
1763
1764        rp_av = xcalloc(rp_ac_alloc, sizeof(*rp_av));
1765
1766        rp_av[0] = "pack-objects";
1767        rp_av[1] = "--objects"; /* --thin will make it --objects-edge */
1768        rp_ac = 2;
1769
1770        git_config(git_pack_config);
1771        if (!pack_compression_seen && core_compression_seen)
1772                pack_compression_level = core_compression_level;
1773
1774        progress = isatty(2);
1775        for (i = 1; i < argc; i++) {
1776                const char *arg = argv[i];
1777
1778                if (*arg != '-')
1779                        break;
1780
1781                if (!strcmp("--non-empty", arg)) {
1782                        non_empty = 1;
1783                        continue;
1784                }
1785                if (!strcmp("--local", arg)) {
1786                        local = 1;
1787                        continue;
1788                }
1789                if (!strcmp("--incremental", arg)) {
1790                        incremental = 1;
1791                        continue;
1792                }
1793                if (!prefixcmp(arg, "--compression=")) {
1794                        char *end;
1795                        int level = strtoul(arg+14, &end, 0);
1796                        if (!arg[14] || *end)
1797                                usage(pack_usage);
1798                        if (level == -1)
1799                                level = Z_DEFAULT_COMPRESSION;
1800                        else if (level < 0 || level > Z_BEST_COMPRESSION)
1801                                die("bad pack compression level %d", level);
1802                        pack_compression_level = level;
1803                        continue;
1804                }
1805                if (!prefixcmp(arg, "--max-pack-size=")) {
1806                        char *end;
1807                        pack_size_limit = strtoul(arg+16, &end, 0) * 1024 * 1024;
1808                        if (!arg[16] || *end)
1809                                usage(pack_usage);
1810                        continue;
1811                }
1812                if (!prefixcmp(arg, "--window=")) {
1813                        char *end;
1814                        window = strtoul(arg+9, &end, 0);
1815                        if (!arg[9] || *end)
1816                                usage(pack_usage);
1817                        continue;
1818                }
1819                if (!prefixcmp(arg, "--depth=")) {
1820                        char *end;
1821                        depth = strtoul(arg+8, &end, 0);
1822                        if (!arg[8] || *end)
1823                                usage(pack_usage);
1824                        continue;
1825                }
1826                if (!strcmp("--progress", arg)) {
1827                        progress = 1;
1828                        continue;
1829                }
1830                if (!strcmp("--all-progress", arg)) {
1831                        progress = 2;
1832                        continue;
1833                }
1834                if (!strcmp("-q", arg)) {
1835                        progress = 0;
1836                        continue;
1837                }
1838                if (!strcmp("--no-reuse-delta", arg)) {
1839                        no_reuse_delta = 1;
1840                        continue;
1841                }
1842                if (!strcmp("--no-reuse-object", arg)) {
1843                        no_reuse_object = no_reuse_delta = 1;
1844                        continue;
1845                }
1846                if (!strcmp("--delta-base-offset", arg)) {
1847                        allow_ofs_delta = 1;
1848                        continue;
1849                }
1850                if (!strcmp("--stdout", arg)) {
1851                        pack_to_stdout = 1;
1852                        continue;
1853                }
1854                if (!strcmp("--revs", arg)) {
1855                        use_internal_rev_list = 1;
1856                        continue;
1857                }
1858                if (!strcmp("--unpacked", arg) ||
1859                    !prefixcmp(arg, "--unpacked=") ||
1860                    !strcmp("--reflog", arg) ||
1861                    !strcmp("--all", arg)) {
1862                        use_internal_rev_list = 1;
1863                        if (rp_ac >= rp_ac_alloc - 1) {
1864                                rp_ac_alloc = alloc_nr(rp_ac_alloc);
1865                                rp_av = xrealloc(rp_av,
1866                                                 rp_ac_alloc * sizeof(*rp_av));
1867                        }
1868                        rp_av[rp_ac++] = arg;
1869                        continue;
1870                }
1871                if (!strcmp("--thin", arg)) {
1872                        use_internal_rev_list = 1;
1873                        thin = 1;
1874                        rp_av[1] = "--objects-edge";
1875                        continue;
1876                }
1877                if (!prefixcmp(arg, "--index-version=")) {
1878                        char *c;
1879                        index_default_version = strtoul(arg + 16, &c, 10);
1880                        if (index_default_version > 2)
1881                                die("bad %s", arg);
1882                        if (*c == ',')
1883                                index_off32_limit = strtoul(c+1, &c, 0);
1884                        if (*c || index_off32_limit & 0x80000000)
1885                                die("bad %s", arg);
1886                        continue;
1887                }
1888                usage(pack_usage);
1889        }
1890
1891        /* Traditionally "pack-objects [options] base extra" failed;
1892         * we would however want to take refs parameter that would
1893         * have been given to upstream rev-list ourselves, which means
1894         * we somehow want to say what the base name is.  So the
1895         * syntax would be:
1896         *
1897         * pack-objects [options] base <refs...>
1898         *
1899         * in other words, we would treat the first non-option as the
1900         * base_name and send everything else to the internal revision
1901         * walker.
1902         */
1903
1904        if (!pack_to_stdout)
1905                base_name = argv[i++];
1906
1907        if (pack_to_stdout != !base_name)
1908                usage(pack_usage);
1909
1910        if (pack_to_stdout && pack_size_limit)
1911                die("--max-pack-size cannot be used to build a pack for transfer.");
1912
1913        if (!pack_to_stdout && thin)
1914                die("--thin cannot be used to build an indexable pack.");
1915
1916        prepare_packed_git();
1917
1918        if (progress)
1919                start_progress(&progress_state, "Generating pack...",
1920                               "Counting objects: ", 0);
1921        if (!use_internal_rev_list)
1922                read_object_list_from_stdin();
1923        else {
1924                rp_av[rp_ac] = NULL;
1925                get_object_list(rp_ac, rp_av);
1926        }
1927        if (progress) {
1928                stop_progress(&progress_state);
1929                fprintf(stderr, "Done counting %u objects.\n", nr_objects);
1930        }
1931
1932        if (non_empty && !nr_result)
1933                return 0;
1934        if (progress && (nr_objects != nr_result))
1935                fprintf(stderr, "Result has %u objects.\n", nr_result);
1936        if (nr_result)
1937                prepare_pack(window, depth);
1938        write_pack_file();
1939        if (progress)
1940                fprintf(stderr, "Total %u (delta %u), reused %u (delta %u)\n",
1941                        written, written_delta, reused, reused_delta);
1942        return 0;
1943}