1#include"builtin.h" 2#include"cache.h" 3#include"attr.h" 4#include"object.h" 5#include"blob.h" 6#include"commit.h" 7#include"tag.h" 8#include"tree.h" 9#include"delta.h" 10#include"pack.h" 11#include"pack-revindex.h" 12#include"csum-file.h" 13#include"tree-walk.h" 14#include"diff.h" 15#include"revision.h" 16#include"list-objects.h" 17#include"pack-objects.h" 18#include"progress.h" 19#include"refs.h" 20#include"streaming.h" 21#include"thread-utils.h" 22#include"pack-bitmap.h" 23#include"reachable.h" 24#include"sha1-array.h" 25#include"argv-array.h" 26#include"mru.h" 27 28static const char*pack_usage[] = { 29N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), 30N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), 31 NULL 32}; 33 34/* 35 * Objects we are going to pack are collected in the `to_pack` structure. 36 * It contains an array (dynamically expanded) of the object data, and a map 37 * that can resolve SHA1s to their position in the array. 38 */ 39static struct packing_data to_pack; 40 41static struct pack_idx_entry **written_list; 42static uint32_t nr_result, nr_written; 43 44static int non_empty; 45static int reuse_delta =1, reuse_object =1; 46static int keep_unreachable, unpack_unreachable, include_tag; 47static unsigned long unpack_unreachable_expiration; 48static int pack_loose_unreachable; 49static int local; 50static int have_non_local_packs; 51static int incremental; 52static int ignore_packed_keep; 53static int allow_ofs_delta; 54static struct pack_idx_option pack_idx_opts; 55static const char*base_name; 56static int progress =1; 57static int window =10; 58static unsigned long pack_size_limit; 59static int depth =50; 60static int delta_search_threads; 61static int pack_to_stdout; 62static int num_preferred_base; 63static struct progress *progress_state; 64static int pack_compression_level = Z_DEFAULT_COMPRESSION; 65static int pack_compression_seen; 66 67static struct packed_git *reuse_packfile; 68static uint32_t reuse_packfile_objects; 69static off_t reuse_packfile_offset; 70 71static int use_bitmap_index =1; 72static int write_bitmap_index; 73static uint16_t write_bitmap_options; 74 75static unsigned long delta_cache_size =0; 76static unsigned long max_delta_cache_size =256*1024*1024; 77static unsigned long cache_max_small_delta_size =1000; 78 79static unsigned long window_memory_limit =0; 80 81/* 82 * stats 83 */ 84static uint32_t written, written_delta; 85static uint32_t reused, reused_delta; 86 87/* 88 * Indexed commits 89 */ 90static struct commit **indexed_commits; 91static unsigned int indexed_commits_nr; 92static unsigned int indexed_commits_alloc; 93 94static voidindex_commit_for_bitmap(struct commit *commit) 95{ 96if(indexed_commits_nr >= indexed_commits_alloc) { 97 indexed_commits_alloc = (indexed_commits_alloc +32) *2; 98REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); 99} 100 101 indexed_commits[indexed_commits_nr++] = commit; 102} 103 104static void*get_delta(struct object_entry *entry) 105{ 106unsigned long size, base_size, delta_size; 107void*buf, *base_buf, *delta_buf; 108enum object_type type; 109 110 buf =read_sha1_file(entry->idx.sha1, &type, &size); 111if(!buf) 112die("unable to read%s",sha1_to_hex(entry->idx.sha1)); 113 base_buf =read_sha1_file(entry->delta->idx.sha1, &type, &base_size); 114if(!base_buf) 115die("unable to read%s",sha1_to_hex(entry->delta->idx.sha1)); 116 delta_buf =diff_delta(base_buf, base_size, 117 buf, size, &delta_size,0); 118if(!delta_buf || delta_size != entry->delta_size) 119die("delta size changed"); 120free(buf); 121free(base_buf); 122return delta_buf; 123} 124 125static unsigned longdo_compress(void**pptr,unsigned long size) 126{ 127 git_zstream stream; 128void*in, *out; 129unsigned long maxsize; 130 131git_deflate_init(&stream, pack_compression_level); 132 maxsize =git_deflate_bound(&stream, size); 133 134 in = *pptr; 135 out =xmalloc(maxsize); 136*pptr = out; 137 138 stream.next_in = in; 139 stream.avail_in = size; 140 stream.next_out = out; 141 stream.avail_out = maxsize; 142while(git_deflate(&stream, Z_FINISH) == Z_OK) 143;/* nothing */ 144git_deflate_end(&stream); 145 146free(in); 147return stream.total_out; 148} 149 150static unsigned longwrite_large_blob_data(struct git_istream *st,struct sha1file *f, 151const unsigned char*sha1) 152{ 153 git_zstream stream; 154unsigned char ibuf[1024*16]; 155unsigned char obuf[1024*16]; 156unsigned long olen =0; 157 158git_deflate_init(&stream, pack_compression_level); 159 160for(;;) { 161 ssize_t readlen; 162int zret = Z_OK; 163 readlen =read_istream(st, ibuf,sizeof(ibuf)); 164if(readlen == -1) 165die(_("unable to read%s"),sha1_to_hex(sha1)); 166 167 stream.next_in = ibuf; 168 stream.avail_in = readlen; 169while((stream.avail_in || readlen ==0) && 170(zret == Z_OK || zret == Z_BUF_ERROR)) { 171 stream.next_out = obuf; 172 stream.avail_out =sizeof(obuf); 173 zret =git_deflate(&stream, readlen ?0: Z_FINISH); 174sha1write(f, obuf, stream.next_out - obuf); 175 olen += stream.next_out - obuf; 176} 177if(stream.avail_in) 178die(_("deflate error (%d)"), zret); 179if(readlen ==0) { 180if(zret != Z_STREAM_END) 181die(_("deflate error (%d)"), zret); 182break; 183} 184} 185git_deflate_end(&stream); 186return olen; 187} 188 189/* 190 * we are going to reuse the existing object data as is. make 191 * sure it is not corrupt. 192 */ 193static intcheck_pack_inflate(struct packed_git *p, 194struct pack_window **w_curs, 195 off_t offset, 196 off_t len, 197unsigned long expect) 198{ 199 git_zstream stream; 200unsigned char fakebuf[4096], *in; 201int st; 202 203memset(&stream,0,sizeof(stream)); 204git_inflate_init(&stream); 205do{ 206 in =use_pack(p, w_curs, offset, &stream.avail_in); 207 stream.next_in = in; 208 stream.next_out = fakebuf; 209 stream.avail_out =sizeof(fakebuf); 210 st =git_inflate(&stream, Z_FINISH); 211 offset += stream.next_in - in; 212}while(st == Z_OK || st == Z_BUF_ERROR); 213git_inflate_end(&stream); 214return(st == Z_STREAM_END && 215 stream.total_out == expect && 216 stream.total_in == len) ?0: -1; 217} 218 219static voidcopy_pack_data(struct sha1file *f, 220struct packed_git *p, 221struct pack_window **w_curs, 222 off_t offset, 223 off_t len) 224{ 225unsigned char*in; 226unsigned long avail; 227 228while(len) { 229 in =use_pack(p, w_curs, offset, &avail); 230if(avail > len) 231 avail = (unsigned long)len; 232sha1write(f, in, avail); 233 offset += avail; 234 len -= avail; 235} 236} 237 238/* Return 0 if we will bust the pack-size limit */ 239static unsigned longwrite_no_reuse_object(struct sha1file *f,struct object_entry *entry, 240unsigned long limit,int usable_delta) 241{ 242unsigned long size, datalen; 243unsigned char header[10], dheader[10]; 244unsigned hdrlen; 245enum object_type type; 246void*buf; 247struct git_istream *st = NULL; 248 249if(!usable_delta) { 250if(entry->type == OBJ_BLOB && 251 entry->size > big_file_threshold && 252(st =open_istream(entry->idx.sha1, &type, &size, NULL)) != NULL) 253 buf = NULL; 254else{ 255 buf =read_sha1_file(entry->idx.sha1, &type, &size); 256if(!buf) 257die(_("unable to read%s"),sha1_to_hex(entry->idx.sha1)); 258} 259/* 260 * make sure no cached delta data remains from a 261 * previous attempt before a pack split occurred. 262 */ 263free(entry->delta_data); 264 entry->delta_data = NULL; 265 entry->z_delta_size =0; 266}else if(entry->delta_data) { 267 size = entry->delta_size; 268 buf = entry->delta_data; 269 entry->delta_data = NULL; 270 type = (allow_ofs_delta && entry->delta->idx.offset) ? 271 OBJ_OFS_DELTA : OBJ_REF_DELTA; 272}else{ 273 buf =get_delta(entry); 274 size = entry->delta_size; 275 type = (allow_ofs_delta && entry->delta->idx.offset) ? 276 OBJ_OFS_DELTA : OBJ_REF_DELTA; 277} 278 279if(st)/* large blob case, just assume we don't compress well */ 280 datalen = size; 281else if(entry->z_delta_size) 282 datalen = entry->z_delta_size; 283else 284 datalen =do_compress(&buf, size); 285 286/* 287 * The object header is a byte of 'type' followed by zero or 288 * more bytes of length. 289 */ 290 hdrlen =encode_in_pack_object_header(type, size, header); 291 292if(type == OBJ_OFS_DELTA) { 293/* 294 * Deltas with relative base contain an additional 295 * encoding of the relative offset for the delta 296 * base from this object's position in the pack. 297 */ 298 off_t ofs = entry->idx.offset - entry->delta->idx.offset; 299unsigned pos =sizeof(dheader) -1; 300 dheader[pos] = ofs &127; 301while(ofs >>=7) 302 dheader[--pos] =128| (--ofs &127); 303if(limit && hdrlen +sizeof(dheader) - pos + datalen +20>= limit) { 304if(st) 305close_istream(st); 306free(buf); 307return0; 308} 309sha1write(f, header, hdrlen); 310sha1write(f, dheader + pos,sizeof(dheader) - pos); 311 hdrlen +=sizeof(dheader) - pos; 312}else if(type == OBJ_REF_DELTA) { 313/* 314 * Deltas with a base reference contain 315 * an additional 20 bytes for the base sha1. 316 */ 317if(limit && hdrlen +20+ datalen +20>= limit) { 318if(st) 319close_istream(st); 320free(buf); 321return0; 322} 323sha1write(f, header, hdrlen); 324sha1write(f, entry->delta->idx.sha1,20); 325 hdrlen +=20; 326}else{ 327if(limit && hdrlen + datalen +20>= limit) { 328if(st) 329close_istream(st); 330free(buf); 331return0; 332} 333sha1write(f, header, hdrlen); 334} 335if(st) { 336 datalen =write_large_blob_data(st, f, entry->idx.sha1); 337close_istream(st); 338}else{ 339sha1write(f, buf, datalen); 340free(buf); 341} 342 343return hdrlen + datalen; 344} 345 346/* Return 0 if we will bust the pack-size limit */ 347static unsigned longwrite_reuse_object(struct sha1file *f,struct object_entry *entry, 348unsigned long limit,int usable_delta) 349{ 350struct packed_git *p = entry->in_pack; 351struct pack_window *w_curs = NULL; 352struct revindex_entry *revidx; 353 off_t offset; 354enum object_type type = entry->type; 355unsigned long datalen; 356unsigned char header[10], dheader[10]; 357unsigned hdrlen; 358 359if(entry->delta) 360 type = (allow_ofs_delta && entry->delta->idx.offset) ? 361 OBJ_OFS_DELTA : OBJ_REF_DELTA; 362 hdrlen =encode_in_pack_object_header(type, entry->size, header); 363 364 offset = entry->in_pack_offset; 365 revidx =find_pack_revindex(p, offset); 366 datalen = revidx[1].offset - offset; 367if(!pack_to_stdout && p->index_version >1&& 368check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 369error("bad packed object CRC for%s",sha1_to_hex(entry->idx.sha1)); 370unuse_pack(&w_curs); 371returnwrite_no_reuse_object(f, entry, limit, usable_delta); 372} 373 374 offset += entry->in_pack_header_size; 375 datalen -= entry->in_pack_header_size; 376 377if(!pack_to_stdout && p->index_version ==1&& 378check_pack_inflate(p, &w_curs, offset, datalen, entry->size)) { 379error("corrupt packed object for%s",sha1_to_hex(entry->idx.sha1)); 380unuse_pack(&w_curs); 381returnwrite_no_reuse_object(f, entry, limit, usable_delta); 382} 383 384if(type == OBJ_OFS_DELTA) { 385 off_t ofs = entry->idx.offset - entry->delta->idx.offset; 386unsigned pos =sizeof(dheader) -1; 387 dheader[pos] = ofs &127; 388while(ofs >>=7) 389 dheader[--pos] =128| (--ofs &127); 390if(limit && hdrlen +sizeof(dheader) - pos + datalen +20>= limit) { 391unuse_pack(&w_curs); 392return0; 393} 394sha1write(f, header, hdrlen); 395sha1write(f, dheader + pos,sizeof(dheader) - pos); 396 hdrlen +=sizeof(dheader) - pos; 397 reused_delta++; 398}else if(type == OBJ_REF_DELTA) { 399if(limit && hdrlen +20+ datalen +20>= limit) { 400unuse_pack(&w_curs); 401return0; 402} 403sha1write(f, header, hdrlen); 404sha1write(f, entry->delta->idx.sha1,20); 405 hdrlen +=20; 406 reused_delta++; 407}else{ 408if(limit && hdrlen + datalen +20>= limit) { 409unuse_pack(&w_curs); 410return0; 411} 412sha1write(f, header, hdrlen); 413} 414copy_pack_data(f, p, &w_curs, offset, datalen); 415unuse_pack(&w_curs); 416 reused++; 417return hdrlen + datalen; 418} 419 420/* Return 0 if we will bust the pack-size limit */ 421static unsigned longwrite_object(struct sha1file *f, 422struct object_entry *entry, 423 off_t write_offset) 424{ 425unsigned long limit, len; 426int usable_delta, to_reuse; 427 428if(!pack_to_stdout) 429crc32_begin(f); 430 431/* apply size limit if limited packsize and not first object */ 432if(!pack_size_limit || !nr_written) 433 limit =0; 434else if(pack_size_limit <= write_offset) 435/* 436 * the earlier object did not fit the limit; avoid 437 * mistaking this with unlimited (i.e. limit = 0). 438 */ 439 limit =1; 440else 441 limit = pack_size_limit - write_offset; 442 443if(!entry->delta) 444 usable_delta =0;/* no delta */ 445else if(!pack_size_limit) 446 usable_delta =1;/* unlimited packfile */ 447else if(entry->delta->idx.offset == (off_t)-1) 448 usable_delta =0;/* base was written to another pack */ 449else if(entry->delta->idx.offset) 450 usable_delta =1;/* base already exists in this pack */ 451else 452 usable_delta =0;/* base could end up in another pack */ 453 454if(!reuse_object) 455 to_reuse =0;/* explicit */ 456else if(!entry->in_pack) 457 to_reuse =0;/* can't reuse what we don't have */ 458else if(entry->type == OBJ_REF_DELTA || entry->type == OBJ_OFS_DELTA) 459/* check_object() decided it for us ... */ 460 to_reuse = usable_delta; 461/* ... but pack split may override that */ 462else if(entry->type != entry->in_pack_type) 463 to_reuse =0;/* pack has delta which is unusable */ 464else if(entry->delta) 465 to_reuse =0;/* we want to pack afresh */ 466else 467 to_reuse =1;/* we have it in-pack undeltified, 468 * and we do not need to deltify it. 469 */ 470 471if(!to_reuse) 472 len =write_no_reuse_object(f, entry, limit, usable_delta); 473else 474 len =write_reuse_object(f, entry, limit, usable_delta); 475if(!len) 476return0; 477 478if(usable_delta) 479 written_delta++; 480 written++; 481if(!pack_to_stdout) 482 entry->idx.crc32 =crc32_end(f); 483return len; 484} 485 486enum write_one_status { 487 WRITE_ONE_SKIP = -1,/* already written */ 488 WRITE_ONE_BREAK =0,/* writing this will bust the limit; not written */ 489 WRITE_ONE_WRITTEN =1,/* normal */ 490 WRITE_ONE_RECURSIVE =2/* already scheduled to be written */ 491}; 492 493static enum write_one_status write_one(struct sha1file *f, 494struct object_entry *e, 495 off_t *offset) 496{ 497unsigned long size; 498int recursing; 499 500/* 501 * we set offset to 1 (which is an impossible value) to mark 502 * the fact that this object is involved in "write its base 503 * first before writing a deltified object" recursion. 504 */ 505 recursing = (e->idx.offset ==1); 506if(recursing) { 507warning("recursive delta detected for object%s", 508sha1_to_hex(e->idx.sha1)); 509return WRITE_ONE_RECURSIVE; 510}else if(e->idx.offset || e->preferred_base) { 511/* offset is non zero if object is written already. */ 512return WRITE_ONE_SKIP; 513} 514 515/* if we are deltified, write out base object first. */ 516if(e->delta) { 517 e->idx.offset =1;/* now recurse */ 518switch(write_one(f, e->delta, offset)) { 519case WRITE_ONE_RECURSIVE: 520/* we cannot depend on this one */ 521 e->delta = NULL; 522break; 523default: 524break; 525case WRITE_ONE_BREAK: 526 e->idx.offset = recursing; 527return WRITE_ONE_BREAK; 528} 529} 530 531 e->idx.offset = *offset; 532 size =write_object(f, e, *offset); 533if(!size) { 534 e->idx.offset = recursing; 535return WRITE_ONE_BREAK; 536} 537 written_list[nr_written++] = &e->idx; 538 539/* make sure off_t is sufficiently large not to wrap */ 540if(signed_add_overflows(*offset, size)) 541die("pack too large for current definition of off_t"); 542*offset += size; 543return WRITE_ONE_WRITTEN; 544} 545 546static intmark_tagged(const char*path,const struct object_id *oid,int flag, 547void*cb_data) 548{ 549unsigned char peeled[20]; 550struct object_entry *entry =packlist_find(&to_pack, oid->hash, NULL); 551 552if(entry) 553 entry->tagged =1; 554if(!peel_ref(path, peeled)) { 555 entry =packlist_find(&to_pack, peeled, NULL); 556if(entry) 557 entry->tagged =1; 558} 559return0; 560} 561 562staticinlinevoidadd_to_write_order(struct object_entry **wo, 563unsigned int*endp, 564struct object_entry *e) 565{ 566if(e->filled) 567return; 568 wo[(*endp)++] = e; 569 e->filled =1; 570} 571 572static voidadd_descendants_to_write_order(struct object_entry **wo, 573unsigned int*endp, 574struct object_entry *e) 575{ 576int add_to_order =1; 577while(e) { 578if(add_to_order) { 579struct object_entry *s; 580/* add this node... */ 581add_to_write_order(wo, endp, e); 582/* all its siblings... */ 583for(s = e->delta_sibling; s; s = s->delta_sibling) { 584add_to_write_order(wo, endp, s); 585} 586} 587/* drop down a level to add left subtree nodes if possible */ 588if(e->delta_child) { 589 add_to_order =1; 590 e = e->delta_child; 591}else{ 592 add_to_order =0; 593/* our sibling might have some children, it is next */ 594if(e->delta_sibling) { 595 e = e->delta_sibling; 596continue; 597} 598/* go back to our parent node */ 599 e = e->delta; 600while(e && !e->delta_sibling) { 601/* we're on the right side of a subtree, keep 602 * going up until we can go right again */ 603 e = e->delta; 604} 605if(!e) { 606/* done- we hit our original root node */ 607return; 608} 609/* pass it off to sibling at this level */ 610 e = e->delta_sibling; 611} 612}; 613} 614 615static voidadd_family_to_write_order(struct object_entry **wo, 616unsigned int*endp, 617struct object_entry *e) 618{ 619struct object_entry *root; 620 621for(root = e; root->delta; root = root->delta) 622;/* nothing */ 623add_descendants_to_write_order(wo, endp, root); 624} 625 626static struct object_entry **compute_write_order(void) 627{ 628unsigned int i, wo_end, last_untagged; 629 630struct object_entry **wo; 631struct object_entry *objects = to_pack.objects; 632 633for(i =0; i < to_pack.nr_objects; i++) { 634 objects[i].tagged =0; 635 objects[i].filled =0; 636 objects[i].delta_child = NULL; 637 objects[i].delta_sibling = NULL; 638} 639 640/* 641 * Fully connect delta_child/delta_sibling network. 642 * Make sure delta_sibling is sorted in the original 643 * recency order. 644 */ 645for(i = to_pack.nr_objects; i >0;) { 646struct object_entry *e = &objects[--i]; 647if(!e->delta) 648continue; 649/* Mark me as the first child */ 650 e->delta_sibling = e->delta->delta_child; 651 e->delta->delta_child = e; 652} 653 654/* 655 * Mark objects that are at the tip of tags. 656 */ 657for_each_tag_ref(mark_tagged, NULL); 658 659/* 660 * Give the objects in the original recency order until 661 * we see a tagged tip. 662 */ 663ALLOC_ARRAY(wo, to_pack.nr_objects); 664for(i = wo_end =0; i < to_pack.nr_objects; i++) { 665if(objects[i].tagged) 666break; 667add_to_write_order(wo, &wo_end, &objects[i]); 668} 669 last_untagged = i; 670 671/* 672 * Then fill all the tagged tips. 673 */ 674for(; i < to_pack.nr_objects; i++) { 675if(objects[i].tagged) 676add_to_write_order(wo, &wo_end, &objects[i]); 677} 678 679/* 680 * And then all remaining commits and tags. 681 */ 682for(i = last_untagged; i < to_pack.nr_objects; i++) { 683if(objects[i].type != OBJ_COMMIT && 684 objects[i].type != OBJ_TAG) 685continue; 686add_to_write_order(wo, &wo_end, &objects[i]); 687} 688 689/* 690 * And then all the trees. 691 */ 692for(i = last_untagged; i < to_pack.nr_objects; i++) { 693if(objects[i].type != OBJ_TREE) 694continue; 695add_to_write_order(wo, &wo_end, &objects[i]); 696} 697 698/* 699 * Finally all the rest in really tight order 700 */ 701for(i = last_untagged; i < to_pack.nr_objects; i++) { 702if(!objects[i].filled) 703add_family_to_write_order(wo, &wo_end, &objects[i]); 704} 705 706if(wo_end != to_pack.nr_objects) 707die("ordered%uobjects, expected %"PRIu32, wo_end, to_pack.nr_objects); 708 709return wo; 710} 711 712static off_t write_reused_pack(struct sha1file *f) 713{ 714unsigned char buffer[8192]; 715 off_t to_write, total; 716int fd; 717 718if(!is_pack_valid(reuse_packfile)) 719die("packfile is invalid:%s", reuse_packfile->pack_name); 720 721 fd =git_open_noatime(reuse_packfile->pack_name); 722if(fd <0) 723die_errno("unable to open packfile for reuse:%s", 724 reuse_packfile->pack_name); 725 726if(lseek(fd,sizeof(struct pack_header), SEEK_SET) == -1) 727die_errno("unable to seek in reused packfile"); 728 729if(reuse_packfile_offset <0) 730 reuse_packfile_offset = reuse_packfile->pack_size -20; 731 732 total = to_write = reuse_packfile_offset -sizeof(struct pack_header); 733 734while(to_write) { 735int read_pack =xread(fd, buffer,sizeof(buffer)); 736 737if(read_pack <=0) 738die_errno("unable to read from reused packfile"); 739 740if(read_pack > to_write) 741 read_pack = to_write; 742 743sha1write(f, buffer, read_pack); 744 to_write -= read_pack; 745 746/* 747 * We don't know the actual number of objects written, 748 * only how many bytes written, how many bytes total, and 749 * how many objects total. So we can fake it by pretending all 750 * objects we are writing are the same size. This gives us a 751 * smooth progress meter, and at the end it matches the true 752 * answer. 753 */ 754 written = reuse_packfile_objects * 755(((double)(total - to_write)) / total); 756display_progress(progress_state, written); 757} 758 759close(fd); 760 written = reuse_packfile_objects; 761display_progress(progress_state, written); 762return reuse_packfile_offset -sizeof(struct pack_header); 763} 764 765static const char no_split_warning[] =N_( 766"disabling bitmap writing, packs are split due to pack.packSizeLimit" 767); 768 769static voidwrite_pack_file(void) 770{ 771uint32_t i =0, j; 772struct sha1file *f; 773 off_t offset; 774uint32_t nr_remaining = nr_result; 775time_t last_mtime =0; 776struct object_entry **write_order; 777 778if(progress > pack_to_stdout) 779 progress_state =start_progress(_("Writing objects"), nr_result); 780ALLOC_ARRAY(written_list, to_pack.nr_objects); 781 write_order =compute_write_order(); 782 783do{ 784unsigned char sha1[20]; 785char*pack_tmp_name = NULL; 786 787if(pack_to_stdout) 788 f =sha1fd_throughput(1,"<stdout>", progress_state); 789else 790 f =create_tmp_packfile(&pack_tmp_name); 791 792 offset =write_pack_header(f, nr_remaining); 793 794if(reuse_packfile) { 795 off_t packfile_size; 796assert(pack_to_stdout); 797 798 packfile_size =write_reused_pack(f); 799 offset += packfile_size; 800} 801 802 nr_written =0; 803for(; i < to_pack.nr_objects; i++) { 804struct object_entry *e = write_order[i]; 805if(write_one(f, e, &offset) == WRITE_ONE_BREAK) 806break; 807display_progress(progress_state, written); 808} 809 810/* 811 * Did we write the wrong # entries in the header? 812 * If so, rewrite it like in fast-import 813 */ 814if(pack_to_stdout) { 815sha1close(f, sha1, CSUM_CLOSE); 816}else if(nr_written == nr_remaining) { 817sha1close(f, sha1, CSUM_FSYNC); 818}else{ 819int fd =sha1close(f, sha1,0); 820fixup_pack_header_footer(fd, sha1, pack_tmp_name, 821 nr_written, sha1, offset); 822close(fd); 823if(write_bitmap_index) { 824warning(_(no_split_warning)); 825 write_bitmap_index =0; 826} 827} 828 829if(!pack_to_stdout) { 830struct stat st; 831struct strbuf tmpname = STRBUF_INIT; 832 833/* 834 * Packs are runtime accessed in their mtime 835 * order since newer packs are more likely to contain 836 * younger objects. So if we are creating multiple 837 * packs then we should modify the mtime of later ones 838 * to preserve this property. 839 */ 840if(stat(pack_tmp_name, &st) <0) { 841warning_errno("failed to stat%s", pack_tmp_name); 842}else if(!last_mtime) { 843 last_mtime = st.st_mtime; 844}else{ 845struct utimbuf utb; 846 utb.actime = st.st_atime; 847 utb.modtime = --last_mtime; 848if(utime(pack_tmp_name, &utb) <0) 849warning_errno("failed utime() on%s", pack_tmp_name); 850} 851 852strbuf_addf(&tmpname,"%s-", base_name); 853 854if(write_bitmap_index) { 855bitmap_writer_set_checksum(sha1); 856bitmap_writer_build_type_index(written_list, nr_written); 857} 858 859finish_tmp_packfile(&tmpname, pack_tmp_name, 860 written_list, nr_written, 861&pack_idx_opts, sha1); 862 863if(write_bitmap_index) { 864strbuf_addf(&tmpname,"%s.bitmap",sha1_to_hex(sha1)); 865 866stop_progress(&progress_state); 867 868bitmap_writer_show_progress(progress); 869bitmap_writer_reuse_bitmaps(&to_pack); 870bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 871bitmap_writer_build(&to_pack); 872bitmap_writer_finish(written_list, nr_written, 873 tmpname.buf, write_bitmap_options); 874 write_bitmap_index =0; 875} 876 877strbuf_release(&tmpname); 878free(pack_tmp_name); 879puts(sha1_to_hex(sha1)); 880} 881 882/* mark written objects as written to previous pack */ 883for(j =0; j < nr_written; j++) { 884 written_list[j]->offset = (off_t)-1; 885} 886 nr_remaining -= nr_written; 887}while(nr_remaining && i < to_pack.nr_objects); 888 889free(written_list); 890free(write_order); 891stop_progress(&progress_state); 892if(written != nr_result) 893die("wrote %"PRIu32" objects while expecting %"PRIu32, 894 written, nr_result); 895} 896 897static voidsetup_delta_attr_check(struct git_attr_check *check) 898{ 899static struct git_attr *attr_delta; 900 901if(!attr_delta) 902 attr_delta =git_attr("delta"); 903 904 check[0].attr = attr_delta; 905} 906 907static intno_try_delta(const char*path) 908{ 909struct git_attr_check check[1]; 910 911setup_delta_attr_check(check); 912if(git_check_attr(path,ARRAY_SIZE(check), check)) 913return0; 914if(ATTR_FALSE(check->value)) 915return1; 916return0; 917} 918 919/* 920 * When adding an object, check whether we have already added it 921 * to our packing list. If so, we can skip. However, if we are 922 * being asked to excludei t, but the previous mention was to include 923 * it, make sure to adjust its flags and tweak our numbers accordingly. 924 * 925 * As an optimization, we pass out the index position where we would have 926 * found the item, since that saves us from having to look it up again a 927 * few lines later when we want to add the new entry. 928 */ 929static inthave_duplicate_entry(const unsigned char*sha1, 930int exclude, 931uint32_t*index_pos) 932{ 933struct object_entry *entry; 934 935 entry =packlist_find(&to_pack, sha1, index_pos); 936if(!entry) 937return0; 938 939if(exclude) { 940if(!entry->preferred_base) 941 nr_result--; 942 entry->preferred_base =1; 943} 944 945return1; 946} 947 948/* 949 * Check whether we want the object in the pack (e.g., we do not want 950 * objects found in non-local stores if the "--local" option was used). 951 * 952 * As a side effect of this check, we will find the packed version of this 953 * object, if any. We therefore pass out the pack information to avoid having 954 * to look it up again later. 955 */ 956static intwant_object_in_pack(const unsigned char*sha1, 957int exclude, 958struct packed_git **found_pack, 959 off_t *found_offset) 960{ 961struct mru_entry *entry; 962 963if(!exclude && local &&has_loose_object_nonlocal(sha1)) 964return0; 965 966*found_pack = NULL; 967*found_offset =0; 968 969for(entry = packed_git_mru->head; entry; entry = entry->next) { 970struct packed_git *p = entry->item; 971 off_t offset =find_pack_entry_one(sha1, p); 972if(offset) { 973if(!*found_pack) { 974if(!is_pack_valid(p)) 975continue; 976*found_offset = offset; 977*found_pack = p; 978} 979if(exclude) 980return1; 981if(incremental) 982return0; 983 984/* 985 * When asked to do --local (do not include an 986 * object that appears in a pack we borrow 987 * from elsewhere) or --honor-pack-keep (do not 988 * include an object that appears in a pack marked 989 * with .keep), we need to make sure no copy of this 990 * object come from in _any_ pack that causes us to 991 * omit it, and need to complete this loop. When 992 * neither option is in effect, we know the object 993 * we just found is going to be packed, so break 994 * out of the loop to return 1 now. 995 */ 996if(!ignore_packed_keep && 997(!local || !have_non_local_packs)) { 998mru_mark(packed_git_mru, entry); 999break;1000}10011002if(local && !p->pack_local)1003return0;1004if(ignore_packed_keep && p->pack_local && p->pack_keep)1005return0;1006}1007}10081009return1;1010}10111012static voidcreate_object_entry(const unsigned char*sha1,1013enum object_type type,1014uint32_t hash,1015int exclude,1016int no_try_delta,1017uint32_t index_pos,1018struct packed_git *found_pack,1019 off_t found_offset)1020{1021struct object_entry *entry;10221023 entry =packlist_alloc(&to_pack, sha1, index_pos);1024 entry->hash = hash;1025if(type)1026 entry->type = type;1027if(exclude)1028 entry->preferred_base =1;1029else1030 nr_result++;1031if(found_pack) {1032 entry->in_pack = found_pack;1033 entry->in_pack_offset = found_offset;1034}10351036 entry->no_try_delta = no_try_delta;1037}10381039static const char no_closure_warning[] =N_(1040"disabling bitmap writing, as some objects are not being packed"1041);10421043static intadd_object_entry(const unsigned char*sha1,enum object_type type,1044const char*name,int exclude)1045{1046struct packed_git *found_pack;1047 off_t found_offset;1048uint32_t index_pos;10491050if(have_duplicate_entry(sha1, exclude, &index_pos))1051return0;10521053if(!want_object_in_pack(sha1, exclude, &found_pack, &found_offset)) {1054/* The pack is missing an object, so it will not have closure */1055if(write_bitmap_index) {1056warning(_(no_closure_warning));1057 write_bitmap_index =0;1058}1059return0;1060}10611062create_object_entry(sha1, type,pack_name_hash(name),1063 exclude, name &&no_try_delta(name),1064 index_pos, found_pack, found_offset);10651066display_progress(progress_state, nr_result);1067return1;1068}10691070static intadd_object_entry_from_bitmap(const unsigned char*sha1,1071enum object_type type,1072int flags,uint32_t name_hash,1073struct packed_git *pack, off_t offset)1074{1075uint32_t index_pos;10761077if(have_duplicate_entry(sha1,0, &index_pos))1078return0;10791080create_object_entry(sha1, type, name_hash,0,0, index_pos, pack, offset);10811082display_progress(progress_state, nr_result);1083return1;1084}10851086struct pbase_tree_cache {1087unsigned char sha1[20];1088int ref;1089int temporary;1090void*tree_data;1091unsigned long tree_size;1092};10931094static struct pbase_tree_cache *(pbase_tree_cache[256]);1095static intpbase_tree_cache_ix(const unsigned char*sha1)1096{1097return sha1[0] %ARRAY_SIZE(pbase_tree_cache);1098}1099static intpbase_tree_cache_ix_incr(int ix)1100{1101return(ix+1) %ARRAY_SIZE(pbase_tree_cache);1102}11031104static struct pbase_tree {1105struct pbase_tree *next;1106/* This is a phony "cache" entry; we are not1107 * going to evict it or find it through _get()1108 * mechanism -- this is for the toplevel node that1109 * would almost always change with any commit.1110 */1111struct pbase_tree_cache pcache;1112} *pbase_tree;11131114static struct pbase_tree_cache *pbase_tree_get(const unsigned char*sha1)1115{1116struct pbase_tree_cache *ent, *nent;1117void*data;1118unsigned long size;1119enum object_type type;1120int neigh;1121int my_ix =pbase_tree_cache_ix(sha1);1122int available_ix = -1;11231124/* pbase-tree-cache acts as a limited hashtable.1125 * your object will be found at your index or within a few1126 * slots after that slot if it is cached.1127 */1128for(neigh =0; neigh <8; neigh++) {1129 ent = pbase_tree_cache[my_ix];1130if(ent && !hashcmp(ent->sha1, sha1)) {1131 ent->ref++;1132return ent;1133}1134else if(((available_ix <0) && (!ent || !ent->ref)) ||1135((0<= available_ix) &&1136(!ent && pbase_tree_cache[available_ix])))1137 available_ix = my_ix;1138if(!ent)1139break;1140 my_ix =pbase_tree_cache_ix_incr(my_ix);1141}11421143/* Did not find one. Either we got a bogus request or1144 * we need to read and perhaps cache.1145 */1146 data =read_sha1_file(sha1, &type, &size);1147if(!data)1148return NULL;1149if(type != OBJ_TREE) {1150free(data);1151return NULL;1152}11531154/* We need to either cache or return a throwaway copy */11551156if(available_ix <0)1157 ent = NULL;1158else{1159 ent = pbase_tree_cache[available_ix];1160 my_ix = available_ix;1161}11621163if(!ent) {1164 nent =xmalloc(sizeof(*nent));1165 nent->temporary = (available_ix <0);1166}1167else{1168/* evict and reuse */1169free(ent->tree_data);1170 nent = ent;1171}1172hashcpy(nent->sha1, sha1);1173 nent->tree_data = data;1174 nent->tree_size = size;1175 nent->ref =1;1176if(!nent->temporary)1177 pbase_tree_cache[my_ix] = nent;1178return nent;1179}11801181static voidpbase_tree_put(struct pbase_tree_cache *cache)1182{1183if(!cache->temporary) {1184 cache->ref--;1185return;1186}1187free(cache->tree_data);1188free(cache);1189}11901191static intname_cmp_len(const char*name)1192{1193int i;1194for(i =0; name[i] && name[i] !='\n'&& name[i] !='/'; i++)1195;1196return i;1197}11981199static voidadd_pbase_object(struct tree_desc *tree,1200const char*name,1201int cmplen,1202const char*fullname)1203{1204struct name_entry entry;1205int cmp;12061207while(tree_entry(tree,&entry)) {1208if(S_ISGITLINK(entry.mode))1209continue;1210 cmp =tree_entry_len(&entry) != cmplen ?1:1211memcmp(name, entry.path, cmplen);1212if(cmp >0)1213continue;1214if(cmp <0)1215return;1216if(name[cmplen] !='/') {1217add_object_entry(entry.oid->hash,1218object_type(entry.mode),1219 fullname,1);1220return;1221}1222if(S_ISDIR(entry.mode)) {1223struct tree_desc sub;1224struct pbase_tree_cache *tree;1225const char*down = name+cmplen+1;1226int downlen =name_cmp_len(down);12271228 tree =pbase_tree_get(entry.oid->hash);1229if(!tree)1230return;1231init_tree_desc(&sub, tree->tree_data, tree->tree_size);12321233add_pbase_object(&sub, down, downlen, fullname);1234pbase_tree_put(tree);1235}1236}1237}12381239static unsigned*done_pbase_paths;1240static int done_pbase_paths_num;1241static int done_pbase_paths_alloc;1242static intdone_pbase_path_pos(unsigned hash)1243{1244int lo =0;1245int hi = done_pbase_paths_num;1246while(lo < hi) {1247int mi = (hi + lo) /2;1248if(done_pbase_paths[mi] == hash)1249return mi;1250if(done_pbase_paths[mi] < hash)1251 hi = mi;1252else1253 lo = mi +1;1254}1255return-lo-1;1256}12571258static intcheck_pbase_path(unsigned hash)1259{1260int pos = (!done_pbase_paths) ? -1:done_pbase_path_pos(hash);1261if(0<= pos)1262return1;1263 pos = -pos -1;1264ALLOC_GROW(done_pbase_paths,1265 done_pbase_paths_num +1,1266 done_pbase_paths_alloc);1267 done_pbase_paths_num++;1268if(pos < done_pbase_paths_num)1269memmove(done_pbase_paths + pos +1,1270 done_pbase_paths + pos,1271(done_pbase_paths_num - pos -1) *sizeof(unsigned));1272 done_pbase_paths[pos] = hash;1273return0;1274}12751276static voidadd_preferred_base_object(const char*name)1277{1278struct pbase_tree *it;1279int cmplen;1280unsigned hash =pack_name_hash(name);12811282if(!num_preferred_base ||check_pbase_path(hash))1283return;12841285 cmplen =name_cmp_len(name);1286for(it = pbase_tree; it; it = it->next) {1287if(cmplen ==0) {1288add_object_entry(it->pcache.sha1, OBJ_TREE, NULL,1);1289}1290else{1291struct tree_desc tree;1292init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1293add_pbase_object(&tree, name, cmplen, name);1294}1295}1296}12971298static voidadd_preferred_base(unsigned char*sha1)1299{1300struct pbase_tree *it;1301void*data;1302unsigned long size;1303unsigned char tree_sha1[20];13041305if(window <= num_preferred_base++)1306return;13071308 data =read_object_with_reference(sha1, tree_type, &size, tree_sha1);1309if(!data)1310return;13111312for(it = pbase_tree; it; it = it->next) {1313if(!hashcmp(it->pcache.sha1, tree_sha1)) {1314free(data);1315return;1316}1317}13181319 it =xcalloc(1,sizeof(*it));1320 it->next = pbase_tree;1321 pbase_tree = it;13221323hashcpy(it->pcache.sha1, tree_sha1);1324 it->pcache.tree_data = data;1325 it->pcache.tree_size = size;1326}13271328static voidcleanup_preferred_base(void)1329{1330struct pbase_tree *it;1331unsigned i;13321333 it = pbase_tree;1334 pbase_tree = NULL;1335while(it) {1336struct pbase_tree *this= it;1337 it =this->next;1338free(this->pcache.tree_data);1339free(this);1340}13411342for(i =0; i <ARRAY_SIZE(pbase_tree_cache); i++) {1343if(!pbase_tree_cache[i])1344continue;1345free(pbase_tree_cache[i]->tree_data);1346free(pbase_tree_cache[i]);1347 pbase_tree_cache[i] = NULL;1348}13491350free(done_pbase_paths);1351 done_pbase_paths = NULL;1352 done_pbase_paths_num = done_pbase_paths_alloc =0;1353}13541355static voidcheck_object(struct object_entry *entry)1356{1357if(entry->in_pack) {1358struct packed_git *p = entry->in_pack;1359struct pack_window *w_curs = NULL;1360const unsigned char*base_ref = NULL;1361struct object_entry *base_entry;1362unsigned long used, used_0;1363unsigned long avail;1364 off_t ofs;1365unsigned char*buf, c;13661367 buf =use_pack(p, &w_curs, entry->in_pack_offset, &avail);13681369/*1370 * We want in_pack_type even if we do not reuse delta1371 * since non-delta representations could still be reused.1372 */1373 used =unpack_object_header_buffer(buf, avail,1374&entry->in_pack_type,1375&entry->size);1376if(used ==0)1377goto give_up;13781379/*1380 * Determine if this is a delta and if so whether we can1381 * reuse it or not. Otherwise let's find out as cheaply as1382 * possible what the actual type and size for this object is.1383 */1384switch(entry->in_pack_type) {1385default:1386/* Not a delta hence we've already got all we need. */1387 entry->type = entry->in_pack_type;1388 entry->in_pack_header_size = used;1389if(entry->type < OBJ_COMMIT || entry->type > OBJ_BLOB)1390goto give_up;1391unuse_pack(&w_curs);1392return;1393case OBJ_REF_DELTA:1394if(reuse_delta && !entry->preferred_base)1395 base_ref =use_pack(p, &w_curs,1396 entry->in_pack_offset + used, NULL);1397 entry->in_pack_header_size = used +20;1398break;1399case OBJ_OFS_DELTA:1400 buf =use_pack(p, &w_curs,1401 entry->in_pack_offset + used, NULL);1402 used_0 =0;1403 c = buf[used_0++];1404 ofs = c &127;1405while(c &128) {1406 ofs +=1;1407if(!ofs ||MSB(ofs,7)) {1408error("delta base offset overflow in pack for%s",1409sha1_to_hex(entry->idx.sha1));1410goto give_up;1411}1412 c = buf[used_0++];1413 ofs = (ofs <<7) + (c &127);1414}1415 ofs = entry->in_pack_offset - ofs;1416if(ofs <=0|| ofs >= entry->in_pack_offset) {1417error("delta base offset out of bound for%s",1418sha1_to_hex(entry->idx.sha1));1419goto give_up;1420}1421if(reuse_delta && !entry->preferred_base) {1422struct revindex_entry *revidx;1423 revidx =find_pack_revindex(p, ofs);1424if(!revidx)1425goto give_up;1426 base_ref =nth_packed_object_sha1(p, revidx->nr);1427}1428 entry->in_pack_header_size = used + used_0;1429break;1430}14311432if(base_ref && (base_entry =packlist_find(&to_pack, base_ref, NULL))) {1433/*1434 * If base_ref was set above that means we wish to1435 * reuse delta data, and we even found that base1436 * in the list of objects we want to pack. Goodie!1437 *1438 * Depth value does not matter - find_deltas() will1439 * never consider reused delta as the base object to1440 * deltify other objects against, in order to avoid1441 * circular deltas.1442 */1443 entry->type = entry->in_pack_type;1444 entry->delta = base_entry;1445 entry->delta_size = entry->size;1446 entry->delta_sibling = base_entry->delta_child;1447 base_entry->delta_child = entry;1448unuse_pack(&w_curs);1449return;1450}14511452if(entry->type) {1453/*1454 * This must be a delta and we already know what the1455 * final object type is. Let's extract the actual1456 * object size from the delta header.1457 */1458 entry->size =get_size_from_delta(p, &w_curs,1459 entry->in_pack_offset + entry->in_pack_header_size);1460if(entry->size ==0)1461goto give_up;1462unuse_pack(&w_curs);1463return;1464}14651466/*1467 * No choice but to fall back to the recursive delta walk1468 * with sha1_object_info() to find about the object type1469 * at this point...1470 */1471 give_up:1472unuse_pack(&w_curs);1473}14741475 entry->type =sha1_object_info(entry->idx.sha1, &entry->size);1476/*1477 * The error condition is checked in prepare_pack(). This is1478 * to permit a missing preferred base object to be ignored1479 * as a preferred base. Doing so can result in a larger1480 * pack file, but the transfer will still take place.1481 */1482}14831484static intpack_offset_sort(const void*_a,const void*_b)1485{1486const struct object_entry *a = *(struct object_entry **)_a;1487const struct object_entry *b = *(struct object_entry **)_b;14881489/* avoid filesystem trashing with loose objects */1490if(!a->in_pack && !b->in_pack)1491returnhashcmp(a->idx.sha1, b->idx.sha1);14921493if(a->in_pack < b->in_pack)1494return-1;1495if(a->in_pack > b->in_pack)1496return1;1497return a->in_pack_offset < b->in_pack_offset ? -1:1498(a->in_pack_offset > b->in_pack_offset);1499}15001501/*1502 * Drop an on-disk delta we were planning to reuse. Naively, this would1503 * just involve blanking out the "delta" field, but we have to deal1504 * with some extra book-keeping:1505 *1506 * 1. Removing ourselves from the delta_sibling linked list.1507 *1508 * 2. Updating our size/type to the non-delta representation. These were1509 * either not recorded initially (size) or overwritten with the delta type1510 * (type) when check_object() decided to reuse the delta.1511 */1512static voiddrop_reused_delta(struct object_entry *entry)1513{1514struct object_entry **p = &entry->delta->delta_child;1515struct object_info oi = OBJECT_INFO_INIT;15161517while(*p) {1518if(*p == entry)1519*p = (*p)->delta_sibling;1520else1521 p = &(*p)->delta_sibling;1522}1523 entry->delta = NULL;15241525 oi.sizep = &entry->size;1526 oi.typep = &entry->type;1527if(packed_object_info(entry->in_pack, entry->in_pack_offset, &oi) <0) {1528/*1529 * We failed to get the info from this pack for some reason;1530 * fall back to sha1_object_info, which may find another copy.1531 * And if that fails, the error will be recorded in entry->type1532 * and dealt with in prepare_pack().1533 */1534 entry->type =sha1_object_info(entry->idx.sha1, &entry->size);1535}1536}15371538/*1539 * Follow the chain of deltas from this entry onward, throwing away any links1540 * that cause us to hit a cycle (as determined by the DFS state flags in1541 * the entries).1542 */1543static voidbreak_delta_chains(struct object_entry *entry)1544{1545/* If it's not a delta, it can't be part of a cycle. */1546if(!entry->delta) {1547 entry->dfs_state = DFS_DONE;1548return;1549}15501551switch(entry->dfs_state) {1552case DFS_NONE:1553/*1554 * This is the first time we've seen the object. We mark it as1555 * part of the active potential cycle and recurse.1556 */1557 entry->dfs_state = DFS_ACTIVE;1558break_delta_chains(entry->delta);1559 entry->dfs_state = DFS_DONE;1560break;15611562case DFS_DONE:1563/* object already examined, and not part of a cycle */1564break;15651566case DFS_ACTIVE:1567/*1568 * We found a cycle that needs broken. It would be correct to1569 * break any link in the chain, but it's convenient to1570 * break this one.1571 */1572drop_reused_delta(entry);1573 entry->dfs_state = DFS_DONE;1574break;1575}1576}15771578static voidget_object_details(void)1579{1580uint32_t i;1581struct object_entry **sorted_by_offset;15821583 sorted_by_offset =xcalloc(to_pack.nr_objects,sizeof(struct object_entry *));1584for(i =0; i < to_pack.nr_objects; i++)1585 sorted_by_offset[i] = to_pack.objects + i;1586qsort(sorted_by_offset, to_pack.nr_objects,sizeof(*sorted_by_offset), pack_offset_sort);15871588for(i =0; i < to_pack.nr_objects; i++) {1589struct object_entry *entry = sorted_by_offset[i];1590check_object(entry);1591if(big_file_threshold < entry->size)1592 entry->no_try_delta =1;1593}15941595/*1596 * This must happen in a second pass, since we rely on the delta1597 * information for the whole list being completed.1598 */1599for(i =0; i < to_pack.nr_objects; i++)1600break_delta_chains(&to_pack.objects[i]);16011602free(sorted_by_offset);1603}16041605/*1606 * We search for deltas in a list sorted by type, by filename hash, and then1607 * by size, so that we see progressively smaller and smaller files.1608 * That's because we prefer deltas to be from the bigger file1609 * to the smaller -- deletes are potentially cheaper, but perhaps1610 * more importantly, the bigger file is likely the more recent1611 * one. The deepest deltas are therefore the oldest objects which are1612 * less susceptible to be accessed often.1613 */1614static inttype_size_sort(const void*_a,const void*_b)1615{1616const struct object_entry *a = *(struct object_entry **)_a;1617const struct object_entry *b = *(struct object_entry **)_b;16181619if(a->type > b->type)1620return-1;1621if(a->type < b->type)1622return1;1623if(a->hash > b->hash)1624return-1;1625if(a->hash < b->hash)1626return1;1627if(a->preferred_base > b->preferred_base)1628return-1;1629if(a->preferred_base < b->preferred_base)1630return1;1631if(a->size > b->size)1632return-1;1633if(a->size < b->size)1634return1;1635return a < b ? -1: (a > b);/* newest first */1636}16371638struct unpacked {1639struct object_entry *entry;1640void*data;1641struct delta_index *index;1642unsigned depth;1643};16441645static intdelta_cacheable(unsigned long src_size,unsigned long trg_size,1646unsigned long delta_size)1647{1648if(max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1649return0;16501651if(delta_size < cache_max_small_delta_size)1652return1;16531654/* cache delta, if objects are large enough compared to delta size */1655if((src_size >>20) + (trg_size >>21) > (delta_size >>10))1656return1;16571658return0;1659}16601661#ifndef NO_PTHREADS16621663static pthread_mutex_t read_mutex;1664#define read_lock() pthread_mutex_lock(&read_mutex)1665#define read_unlock() pthread_mutex_unlock(&read_mutex)16661667static pthread_mutex_t cache_mutex;1668#define cache_lock() pthread_mutex_lock(&cache_mutex)1669#define cache_unlock() pthread_mutex_unlock(&cache_mutex)16701671static pthread_mutex_t progress_mutex;1672#define progress_lock() pthread_mutex_lock(&progress_mutex)1673#define progress_unlock() pthread_mutex_unlock(&progress_mutex)16741675#else16761677#define read_lock() (void)01678#define read_unlock() (void)01679#define cache_lock() (void)01680#define cache_unlock() (void)01681#define progress_lock() (void)01682#define progress_unlock() (void)016831684#endif16851686static inttry_delta(struct unpacked *trg,struct unpacked *src,1687unsigned max_depth,unsigned long*mem_usage)1688{1689struct object_entry *trg_entry = trg->entry;1690struct object_entry *src_entry = src->entry;1691unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;1692unsigned ref_depth;1693enum object_type type;1694void*delta_buf;16951696/* Don't bother doing diffs between different types */1697if(trg_entry->type != src_entry->type)1698return-1;16991700/*1701 * We do not bother to try a delta that we discarded on an1702 * earlier try, but only when reusing delta data. Note that1703 * src_entry that is marked as the preferred_base should always1704 * be considered, as even if we produce a suboptimal delta against1705 * it, we will still save the transfer cost, as we already know1706 * the other side has it and we won't send src_entry at all.1707 */1708if(reuse_delta && trg_entry->in_pack &&1709 trg_entry->in_pack == src_entry->in_pack &&1710!src_entry->preferred_base &&1711 trg_entry->in_pack_type != OBJ_REF_DELTA &&1712 trg_entry->in_pack_type != OBJ_OFS_DELTA)1713return0;17141715/* Let's not bust the allowed depth. */1716if(src->depth >= max_depth)1717return0;17181719/* Now some size filtering heuristics. */1720 trg_size = trg_entry->size;1721if(!trg_entry->delta) {1722 max_size = trg_size/2-20;1723 ref_depth =1;1724}else{1725 max_size = trg_entry->delta_size;1726 ref_depth = trg->depth;1727}1728 max_size = (uint64_t)max_size * (max_depth - src->depth) /1729(max_depth - ref_depth +1);1730if(max_size ==0)1731return0;1732 src_size = src_entry->size;1733 sizediff = src_size < trg_size ? trg_size - src_size :0;1734if(sizediff >= max_size)1735return0;1736if(trg_size < src_size /32)1737return0;17381739/* Load data if not already done */1740if(!trg->data) {1741read_lock();1742 trg->data =read_sha1_file(trg_entry->idx.sha1, &type, &sz);1743read_unlock();1744if(!trg->data)1745die("object%scannot be read",1746sha1_to_hex(trg_entry->idx.sha1));1747if(sz != trg_size)1748die("object%sinconsistent object length (%lu vs%lu)",1749sha1_to_hex(trg_entry->idx.sha1), sz, trg_size);1750*mem_usage += sz;1751}1752if(!src->data) {1753read_lock();1754 src->data =read_sha1_file(src_entry->idx.sha1, &type, &sz);1755read_unlock();1756if(!src->data) {1757if(src_entry->preferred_base) {1758static int warned =0;1759if(!warned++)1760warning("object%scannot be read",1761sha1_to_hex(src_entry->idx.sha1));1762/*1763 * Those objects are not included in the1764 * resulting pack. Be resilient and ignore1765 * them if they can't be read, in case the1766 * pack could be created nevertheless.1767 */1768return0;1769}1770die("object%scannot be read",1771sha1_to_hex(src_entry->idx.sha1));1772}1773if(sz != src_size)1774die("object%sinconsistent object length (%lu vs%lu)",1775sha1_to_hex(src_entry->idx.sha1), sz, src_size);1776*mem_usage += sz;1777}1778if(!src->index) {1779 src->index =create_delta_index(src->data, src_size);1780if(!src->index) {1781static int warned =0;1782if(!warned++)1783warning("suboptimal pack - out of memory");1784return0;1785}1786*mem_usage +=sizeof_delta_index(src->index);1787}17881789 delta_buf =create_delta(src->index, trg->data, trg_size, &delta_size, max_size);1790if(!delta_buf)1791return0;17921793if(trg_entry->delta) {1794/* Prefer only shallower same-sized deltas. */1795if(delta_size == trg_entry->delta_size &&1796 src->depth +1>= trg->depth) {1797free(delta_buf);1798return0;1799}1800}18011802/*1803 * Handle memory allocation outside of the cache1804 * accounting lock. Compiler will optimize the strangeness1805 * away when NO_PTHREADS is defined.1806 */1807free(trg_entry->delta_data);1808cache_lock();1809if(trg_entry->delta_data) {1810 delta_cache_size -= trg_entry->delta_size;1811 trg_entry->delta_data = NULL;1812}1813if(delta_cacheable(src_size, trg_size, delta_size)) {1814 delta_cache_size += delta_size;1815cache_unlock();1816 trg_entry->delta_data =xrealloc(delta_buf, delta_size);1817}else{1818cache_unlock();1819free(delta_buf);1820}18211822 trg_entry->delta = src_entry;1823 trg_entry->delta_size = delta_size;1824 trg->depth = src->depth +1;18251826return1;1827}18281829static unsigned intcheck_delta_limit(struct object_entry *me,unsigned int n)1830{1831struct object_entry *child = me->delta_child;1832unsigned int m = n;1833while(child) {1834unsigned int c =check_delta_limit(child, n +1);1835if(m < c)1836 m = c;1837 child = child->delta_sibling;1838}1839return m;1840}18411842static unsigned longfree_unpacked(struct unpacked *n)1843{1844unsigned long freed_mem =sizeof_delta_index(n->index);1845free_delta_index(n->index);1846 n->index = NULL;1847if(n->data) {1848 freed_mem += n->entry->size;1849free(n->data);1850 n->data = NULL;1851}1852 n->entry = NULL;1853 n->depth =0;1854return freed_mem;1855}18561857static voidfind_deltas(struct object_entry **list,unsigned*list_size,1858int window,int depth,unsigned*processed)1859{1860uint32_t i, idx =0, count =0;1861struct unpacked *array;1862unsigned long mem_usage =0;18631864 array =xcalloc(window,sizeof(struct unpacked));18651866for(;;) {1867struct object_entry *entry;1868struct unpacked *n = array + idx;1869int j, max_depth, best_base = -1;18701871progress_lock();1872if(!*list_size) {1873progress_unlock();1874break;1875}1876 entry = *list++;1877(*list_size)--;1878if(!entry->preferred_base) {1879(*processed)++;1880display_progress(progress_state, *processed);1881}1882progress_unlock();18831884 mem_usage -=free_unpacked(n);1885 n->entry = entry;18861887while(window_memory_limit &&1888 mem_usage > window_memory_limit &&1889 count >1) {1890uint32_t tail = (idx + window - count) % window;1891 mem_usage -=free_unpacked(array + tail);1892 count--;1893}18941895/* We do not compute delta to *create* objects we are not1896 * going to pack.1897 */1898if(entry->preferred_base)1899goto next;19001901/*1902 * If the current object is at pack edge, take the depth the1903 * objects that depend on the current object into account1904 * otherwise they would become too deep.1905 */1906 max_depth = depth;1907if(entry->delta_child) {1908 max_depth -=check_delta_limit(entry,0);1909if(max_depth <=0)1910goto next;1911}19121913 j = window;1914while(--j >0) {1915int ret;1916uint32_t other_idx = idx + j;1917struct unpacked *m;1918if(other_idx >= window)1919 other_idx -= window;1920 m = array + other_idx;1921if(!m->entry)1922break;1923 ret =try_delta(n, m, max_depth, &mem_usage);1924if(ret <0)1925break;1926else if(ret >0)1927 best_base = other_idx;1928}19291930/*1931 * If we decided to cache the delta data, then it is best1932 * to compress it right away. First because we have to do1933 * it anyway, and doing it here while we're threaded will1934 * save a lot of time in the non threaded write phase,1935 * as well as allow for caching more deltas within1936 * the same cache size limit.1937 * ...1938 * But only if not writing to stdout, since in that case1939 * the network is most likely throttling writes anyway,1940 * and therefore it is best to go to the write phase ASAP1941 * instead, as we can afford spending more time compressing1942 * between writes at that moment.1943 */1944if(entry->delta_data && !pack_to_stdout) {1945 entry->z_delta_size =do_compress(&entry->delta_data,1946 entry->delta_size);1947cache_lock();1948 delta_cache_size -= entry->delta_size;1949 delta_cache_size += entry->z_delta_size;1950cache_unlock();1951}19521953/* if we made n a delta, and if n is already at max1954 * depth, leaving it in the window is pointless. we1955 * should evict it first.1956 */1957if(entry->delta && max_depth <= n->depth)1958continue;19591960/*1961 * Move the best delta base up in the window, after the1962 * currently deltified object, to keep it longer. It will1963 * be the first base object to be attempted next.1964 */1965if(entry->delta) {1966struct unpacked swap = array[best_base];1967int dist = (window + idx - best_base) % window;1968int dst = best_base;1969while(dist--) {1970int src = (dst +1) % window;1971 array[dst] = array[src];1972 dst = src;1973}1974 array[dst] = swap;1975}19761977 next:1978 idx++;1979if(count +1< window)1980 count++;1981if(idx >= window)1982 idx =0;1983}19841985for(i =0; i < window; ++i) {1986free_delta_index(array[i].index);1987free(array[i].data);1988}1989free(array);1990}19911992#ifndef NO_PTHREADS19931994static voidtry_to_free_from_threads(size_t size)1995{1996read_lock();1997release_pack_memory(size);1998read_unlock();1999}20002001static try_to_free_t old_try_to_free_routine;20022003/*2004 * The main thread waits on the condition that (at least) one of the workers2005 * has stopped working (which is indicated in the .working member of2006 * struct thread_params).2007 * When a work thread has completed its work, it sets .working to 0 and2008 * signals the main thread and waits on the condition that .data_ready2009 * becomes 1.2010 */20112012struct thread_params {2013 pthread_t thread;2014struct object_entry **list;2015unsigned list_size;2016unsigned remaining;2017int window;2018int depth;2019int working;2020int data_ready;2021 pthread_mutex_t mutex;2022 pthread_cond_t cond;2023unsigned*processed;2024};20252026static pthread_cond_t progress_cond;20272028/*2029 * Mutex and conditional variable can't be statically-initialized on Windows.2030 */2031static voidinit_threaded_search(void)2032{2033init_recursive_mutex(&read_mutex);2034pthread_mutex_init(&cache_mutex, NULL);2035pthread_mutex_init(&progress_mutex, NULL);2036pthread_cond_init(&progress_cond, NULL);2037 old_try_to_free_routine =set_try_to_free_routine(try_to_free_from_threads);2038}20392040static voidcleanup_threaded_search(void)2041{2042set_try_to_free_routine(old_try_to_free_routine);2043pthread_cond_destroy(&progress_cond);2044pthread_mutex_destroy(&read_mutex);2045pthread_mutex_destroy(&cache_mutex);2046pthread_mutex_destroy(&progress_mutex);2047}20482049static void*threaded_find_deltas(void*arg)2050{2051struct thread_params *me = arg;20522053while(me->remaining) {2054find_deltas(me->list, &me->remaining,2055 me->window, me->depth, me->processed);20562057progress_lock();2058 me->working =0;2059pthread_cond_signal(&progress_cond);2060progress_unlock();20612062/*2063 * We must not set ->data_ready before we wait on the2064 * condition because the main thread may have set it to 12065 * before we get here. In order to be sure that new2066 * work is available if we see 1 in ->data_ready, it2067 * was initialized to 0 before this thread was spawned2068 * and we reset it to 0 right away.2069 */2070pthread_mutex_lock(&me->mutex);2071while(!me->data_ready)2072pthread_cond_wait(&me->cond, &me->mutex);2073 me->data_ready =0;2074pthread_mutex_unlock(&me->mutex);2075}2076/* leave ->working 1 so that this doesn't get more work assigned */2077return NULL;2078}20792080static voidll_find_deltas(struct object_entry **list,unsigned list_size,2081int window,int depth,unsigned*processed)2082{2083struct thread_params *p;2084int i, ret, active_threads =0;20852086init_threaded_search();20872088if(delta_search_threads <=1) {2089find_deltas(list, &list_size, window, depth, processed);2090cleanup_threaded_search();2091return;2092}2093if(progress > pack_to_stdout)2094fprintf(stderr,"Delta compression using up to%dthreads.\n",2095 delta_search_threads);2096 p =xcalloc(delta_search_threads,sizeof(*p));20972098/* Partition the work amongst work threads. */2099for(i =0; i < delta_search_threads; i++) {2100unsigned sub_size = list_size / (delta_search_threads - i);21012102/* don't use too small segments or no deltas will be found */2103if(sub_size <2*window && i+1< delta_search_threads)2104 sub_size =0;21052106 p[i].window = window;2107 p[i].depth = depth;2108 p[i].processed = processed;2109 p[i].working =1;2110 p[i].data_ready =0;21112112/* try to split chunks on "path" boundaries */2113while(sub_size && sub_size < list_size &&2114 list[sub_size]->hash &&2115 list[sub_size]->hash == list[sub_size-1]->hash)2116 sub_size++;21172118 p[i].list = list;2119 p[i].list_size = sub_size;2120 p[i].remaining = sub_size;21212122 list += sub_size;2123 list_size -= sub_size;2124}21252126/* Start work threads. */2127for(i =0; i < delta_search_threads; i++) {2128if(!p[i].list_size)2129continue;2130pthread_mutex_init(&p[i].mutex, NULL);2131pthread_cond_init(&p[i].cond, NULL);2132 ret =pthread_create(&p[i].thread, NULL,2133 threaded_find_deltas, &p[i]);2134if(ret)2135die("unable to create thread:%s",strerror(ret));2136 active_threads++;2137}21382139/*2140 * Now let's wait for work completion. Each time a thread is done2141 * with its work, we steal half of the remaining work from the2142 * thread with the largest number of unprocessed objects and give2143 * it to that newly idle thread. This ensure good load balancing2144 * until the remaining object list segments are simply too short2145 * to be worth splitting anymore.2146 */2147while(active_threads) {2148struct thread_params *target = NULL;2149struct thread_params *victim = NULL;2150unsigned sub_size =0;21512152progress_lock();2153for(;;) {2154for(i =0; !target && i < delta_search_threads; i++)2155if(!p[i].working)2156 target = &p[i];2157if(target)2158break;2159pthread_cond_wait(&progress_cond, &progress_mutex);2160}21612162for(i =0; i < delta_search_threads; i++)2163if(p[i].remaining >2*window &&2164(!victim || victim->remaining < p[i].remaining))2165 victim = &p[i];2166if(victim) {2167 sub_size = victim->remaining /2;2168 list = victim->list + victim->list_size - sub_size;2169while(sub_size && list[0]->hash &&2170 list[0]->hash == list[-1]->hash) {2171 list++;2172 sub_size--;2173}2174if(!sub_size) {2175/*2176 * It is possible for some "paths" to have2177 * so many objects that no hash boundary2178 * might be found. Let's just steal the2179 * exact half in that case.2180 */2181 sub_size = victim->remaining /2;2182 list -= sub_size;2183}2184 target->list = list;2185 victim->list_size -= sub_size;2186 victim->remaining -= sub_size;2187}2188 target->list_size = sub_size;2189 target->remaining = sub_size;2190 target->working =1;2191progress_unlock();21922193pthread_mutex_lock(&target->mutex);2194 target->data_ready =1;2195pthread_cond_signal(&target->cond);2196pthread_mutex_unlock(&target->mutex);21972198if(!sub_size) {2199pthread_join(target->thread, NULL);2200pthread_cond_destroy(&target->cond);2201pthread_mutex_destroy(&target->mutex);2202 active_threads--;2203}2204}2205cleanup_threaded_search();2206free(p);2207}22082209#else2210#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)2211#endif22122213static intadd_ref_tag(const char*path,const struct object_id *oid,int flag,void*cb_data)2214{2215struct object_id peeled;22162217if(starts_with(path,"refs/tags/") &&/* is a tag? */2218!peel_ref(path, peeled.hash) &&/* peelable? */2219packlist_find(&to_pack, peeled.hash, NULL))/* object packed? */2220add_object_entry(oid->hash, OBJ_TAG, NULL,0);2221return0;2222}22232224static voidprepare_pack(int window,int depth)2225{2226struct object_entry **delta_list;2227uint32_t i, nr_deltas;2228unsigned n;22292230get_object_details();22312232/*2233 * If we're locally repacking then we need to be doubly careful2234 * from now on in order to make sure no stealth corruption gets2235 * propagated to the new pack. Clients receiving streamed packs2236 * should validate everything they get anyway so no need to incur2237 * the additional cost here in that case.2238 */2239if(!pack_to_stdout)2240 do_check_packed_object_crc =1;22412242if(!to_pack.nr_objects || !window || !depth)2243return;22442245ALLOC_ARRAY(delta_list, to_pack.nr_objects);2246 nr_deltas = n =0;22472248for(i =0; i < to_pack.nr_objects; i++) {2249struct object_entry *entry = to_pack.objects + i;22502251if(entry->delta)2252/* This happens if we decided to reuse existing2253 * delta from a pack. "reuse_delta &&" is implied.2254 */2255continue;22562257if(entry->size <50)2258continue;22592260if(entry->no_try_delta)2261continue;22622263if(!entry->preferred_base) {2264 nr_deltas++;2265if(entry->type <0)2266die("unable to get type of object%s",2267sha1_to_hex(entry->idx.sha1));2268}else{2269if(entry->type <0) {2270/*2271 * This object is not found, but we2272 * don't have to include it anyway.2273 */2274continue;2275}2276}22772278 delta_list[n++] = entry;2279}22802281if(nr_deltas && n >1) {2282unsigned nr_done =0;2283if(progress)2284 progress_state =start_progress(_("Compressing objects"),2285 nr_deltas);2286qsort(delta_list, n,sizeof(*delta_list), type_size_sort);2287ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2288stop_progress(&progress_state);2289if(nr_done != nr_deltas)2290die("inconsistency with delta count");2291}2292free(delta_list);2293}22942295static intgit_pack_config(const char*k,const char*v,void*cb)2296{2297if(!strcmp(k,"pack.window")) {2298 window =git_config_int(k, v);2299return0;2300}2301if(!strcmp(k,"pack.windowmemory")) {2302 window_memory_limit =git_config_ulong(k, v);2303return0;2304}2305if(!strcmp(k,"pack.depth")) {2306 depth =git_config_int(k, v);2307return0;2308}2309if(!strcmp(k,"pack.compression")) {2310int level =git_config_int(k, v);2311if(level == -1)2312 level = Z_DEFAULT_COMPRESSION;2313else if(level <0|| level > Z_BEST_COMPRESSION)2314die("bad pack compression level%d", level);2315 pack_compression_level = level;2316 pack_compression_seen =1;2317return0;2318}2319if(!strcmp(k,"pack.deltacachesize")) {2320 max_delta_cache_size =git_config_int(k, v);2321return0;2322}2323if(!strcmp(k,"pack.deltacachelimit")) {2324 cache_max_small_delta_size =git_config_int(k, v);2325return0;2326}2327if(!strcmp(k,"pack.writebitmaphashcache")) {2328if(git_config_bool(k, v))2329 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2330else2331 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2332}2333if(!strcmp(k,"pack.usebitmaps")) {2334 use_bitmap_index =git_config_bool(k, v);2335return0;2336}2337if(!strcmp(k,"pack.threads")) {2338 delta_search_threads =git_config_int(k, v);2339if(delta_search_threads <0)2340die("invalid number of threads specified (%d)",2341 delta_search_threads);2342#ifdef NO_PTHREADS2343if(delta_search_threads !=1)2344warning("no threads support, ignoring%s", k);2345#endif2346return0;2347}2348if(!strcmp(k,"pack.indexversion")) {2349 pack_idx_opts.version =git_config_int(k, v);2350if(pack_idx_opts.version >2)2351die("bad pack.indexversion=%"PRIu32,2352 pack_idx_opts.version);2353return0;2354}2355returngit_default_config(k, v, cb);2356}23572358static voidread_object_list_from_stdin(void)2359{2360char line[40+1+ PATH_MAX +2];2361unsigned char sha1[20];23622363for(;;) {2364if(!fgets(line,sizeof(line), stdin)) {2365if(feof(stdin))2366break;2367if(!ferror(stdin))2368die("fgets returned NULL, not EOF, not error!");2369if(errno != EINTR)2370die_errno("fgets");2371clearerr(stdin);2372continue;2373}2374if(line[0] =='-') {2375if(get_sha1_hex(line+1, sha1))2376die("expected edge sha1, got garbage:\n%s",2377 line);2378add_preferred_base(sha1);2379continue;2380}2381if(get_sha1_hex(line, sha1))2382die("expected sha1, got garbage:\n%s", line);23832384add_preferred_base_object(line+41);2385add_object_entry(sha1,0, line+41,0);2386}2387}23882389#define OBJECT_ADDED (1u<<20)23902391static voidshow_commit(struct commit *commit,void*data)2392{2393add_object_entry(commit->object.oid.hash, OBJ_COMMIT, NULL,0);2394 commit->object.flags |= OBJECT_ADDED;23952396if(write_bitmap_index)2397index_commit_for_bitmap(commit);2398}23992400static voidshow_object(struct object *obj,const char*name,void*data)2401{2402add_preferred_base_object(name);2403add_object_entry(obj->oid.hash, obj->type, name,0);2404 obj->flags |= OBJECT_ADDED;2405}24062407static voidshow_edge(struct commit *commit)2408{2409add_preferred_base(commit->object.oid.hash);2410}24112412struct in_pack_object {2413 off_t offset;2414struct object *object;2415};24162417struct in_pack {2418int alloc;2419int nr;2420struct in_pack_object *array;2421};24222423static voidmark_in_pack_object(struct object *object,struct packed_git *p,struct in_pack *in_pack)2424{2425 in_pack->array[in_pack->nr].offset =find_pack_entry_one(object->oid.hash, p);2426 in_pack->array[in_pack->nr].object = object;2427 in_pack->nr++;2428}24292430/*2431 * Compare the objects in the offset order, in order to emulate the2432 * "git rev-list --objects" output that produced the pack originally.2433 */2434static intofscmp(const void*a_,const void*b_)2435{2436struct in_pack_object *a = (struct in_pack_object *)a_;2437struct in_pack_object *b = (struct in_pack_object *)b_;24382439if(a->offset < b->offset)2440return-1;2441else if(a->offset > b->offset)2442return1;2443else2444returnoidcmp(&a->object->oid, &b->object->oid);2445}24462447static voidadd_objects_in_unpacked_packs(struct rev_info *revs)2448{2449struct packed_git *p;2450struct in_pack in_pack;2451uint32_t i;24522453memset(&in_pack,0,sizeof(in_pack));24542455for(p = packed_git; p; p = p->next) {2456const unsigned char*sha1;2457struct object *o;24582459if(!p->pack_local || p->pack_keep)2460continue;2461if(open_pack_index(p))2462die("cannot open pack index");24632464ALLOC_GROW(in_pack.array,2465 in_pack.nr + p->num_objects,2466 in_pack.alloc);24672468for(i =0; i < p->num_objects; i++) {2469 sha1 =nth_packed_object_sha1(p, i);2470 o =lookup_unknown_object(sha1);2471if(!(o->flags & OBJECT_ADDED))2472mark_in_pack_object(o, p, &in_pack);2473 o->flags |= OBJECT_ADDED;2474}2475}24762477if(in_pack.nr) {2478qsort(in_pack.array, in_pack.nr,sizeof(in_pack.array[0]),2479 ofscmp);2480for(i =0; i < in_pack.nr; i++) {2481struct object *o = in_pack.array[i].object;2482add_object_entry(o->oid.hash, o->type,"",0);2483}2484}2485free(in_pack.array);2486}24872488static intadd_loose_object(const unsigned char*sha1,const char*path,2489void*data)2490{2491enum object_type type =sha1_object_info(sha1, NULL);24922493if(type <0) {2494warning("loose object at%scould not be examined", path);2495return0;2496}24972498add_object_entry(sha1, type,"",0);2499return0;2500}25012502/*2503 * We actually don't even have to worry about reachability here.2504 * add_object_entry will weed out duplicates, so we just add every2505 * loose object we find.2506 */2507static voidadd_unreachable_loose_objects(void)2508{2509for_each_loose_file_in_objdir(get_object_directory(),2510 add_loose_object,2511 NULL, NULL, NULL);2512}25132514static inthas_sha1_pack_kept_or_nonlocal(const unsigned char*sha1)2515{2516static struct packed_git *last_found = (void*)1;2517struct packed_git *p;25182519 p = (last_found != (void*)1) ? last_found : packed_git;25202521while(p) {2522if((!p->pack_local || p->pack_keep) &&2523find_pack_entry_one(sha1, p)) {2524 last_found = p;2525return1;2526}2527if(p == last_found)2528 p = packed_git;2529else2530 p = p->next;2531if(p == last_found)2532 p = p->next;2533}2534return0;2535}25362537/*2538 * Store a list of sha1s that are should not be discarded2539 * because they are either written too recently, or are2540 * reachable from another object that was.2541 *2542 * This is filled by get_object_list.2543 */2544static struct sha1_array recent_objects;25452546static intloosened_object_can_be_discarded(const unsigned char*sha1,2547unsigned long mtime)2548{2549if(!unpack_unreachable_expiration)2550return0;2551if(mtime > unpack_unreachable_expiration)2552return0;2553if(sha1_array_lookup(&recent_objects, sha1) >=0)2554return0;2555return1;2556}25572558static voidloosen_unused_packed_objects(struct rev_info *revs)2559{2560struct packed_git *p;2561uint32_t i;2562const unsigned char*sha1;25632564for(p = packed_git; p; p = p->next) {2565if(!p->pack_local || p->pack_keep)2566continue;25672568if(open_pack_index(p))2569die("cannot open pack index");25702571for(i =0; i < p->num_objects; i++) {2572 sha1 =nth_packed_object_sha1(p, i);2573if(!packlist_find(&to_pack, sha1, NULL) &&2574!has_sha1_pack_kept_or_nonlocal(sha1) &&2575!loosened_object_can_be_discarded(sha1, p->mtime))2576if(force_object_loose(sha1, p->mtime))2577die("unable to force loose object");2578}2579}2580}25812582/*2583 * This tracks any options which a reader of the pack might2584 * not understand, and which would therefore prevent blind reuse2585 * of what we have on disk.2586 */2587static intpack_options_allow_reuse(void)2588{2589return allow_ofs_delta;2590}25912592static intget_object_list_from_bitmap(struct rev_info *revs)2593{2594if(prepare_bitmap_walk(revs) <0)2595return-1;25962597if(pack_options_allow_reuse() &&2598!reuse_partial_packfile_from_bitmap(2599&reuse_packfile,2600&reuse_packfile_objects,2601&reuse_packfile_offset)) {2602assert(reuse_packfile_objects);2603 nr_result += reuse_packfile_objects;2604display_progress(progress_state, nr_result);2605}26062607traverse_bitmap_commit_list(&add_object_entry_from_bitmap);2608return0;2609}26102611static voidrecord_recent_object(struct object *obj,2612const char*name,2613void*data)2614{2615sha1_array_append(&recent_objects, obj->oid.hash);2616}26172618static voidrecord_recent_commit(struct commit *commit,void*data)2619{2620sha1_array_append(&recent_objects, commit->object.oid.hash);2621}26222623static voidget_object_list(int ac,const char**av)2624{2625struct rev_info revs;2626char line[1000];2627int flags =0;26282629init_revisions(&revs, NULL);2630 save_commit_buffer =0;2631setup_revisions(ac, av, &revs, NULL);26322633/* make sure shallows are read */2634is_repository_shallow();26352636while(fgets(line,sizeof(line), stdin) != NULL) {2637int len =strlen(line);2638if(len && line[len -1] =='\n')2639 line[--len] =0;2640if(!len)2641break;2642if(*line =='-') {2643if(!strcmp(line,"--not")) {2644 flags ^= UNINTERESTING;2645 write_bitmap_index =0;2646continue;2647}2648if(starts_with(line,"--shallow ")) {2649unsigned char sha1[20];2650if(get_sha1_hex(line +10, sha1))2651die("not an SHA-1 '%s'", line +10);2652register_shallow(sha1);2653 use_bitmap_index =0;2654continue;2655}2656die("not a rev '%s'", line);2657}2658if(handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))2659die("bad revision '%s'", line);2660}26612662if(use_bitmap_index && !get_object_list_from_bitmap(&revs))2663return;26642665if(prepare_revision_walk(&revs))2666die("revision walk setup failed");2667mark_edges_uninteresting(&revs, show_edge);2668traverse_commit_list(&revs, show_commit, show_object, NULL);26692670if(unpack_unreachable_expiration) {2671 revs.ignore_missing_links =1;2672if(add_unseen_recent_objects_to_traversal(&revs,2673 unpack_unreachable_expiration))2674die("unable to add recent objects");2675if(prepare_revision_walk(&revs))2676die("revision walk setup failed");2677traverse_commit_list(&revs, record_recent_commit,2678 record_recent_object, NULL);2679}26802681if(keep_unreachable)2682add_objects_in_unpacked_packs(&revs);2683if(pack_loose_unreachable)2684add_unreachable_loose_objects();2685if(unpack_unreachable)2686loosen_unused_packed_objects(&revs);26872688sha1_array_clear(&recent_objects);2689}26902691static intoption_parse_index_version(const struct option *opt,2692const char*arg,int unset)2693{2694char*c;2695const char*val = arg;2696 pack_idx_opts.version =strtoul(val, &c,10);2697if(pack_idx_opts.version >2)2698die(_("unsupported index version%s"), val);2699if(*c ==','&& c[1])2700 pack_idx_opts.off32_limit =strtoul(c+1, &c,0);2701if(*c || pack_idx_opts.off32_limit &0x80000000)2702die(_("bad index version '%s'"), val);2703return0;2704}27052706static intoption_parse_unpack_unreachable(const struct option *opt,2707const char*arg,int unset)2708{2709if(unset) {2710 unpack_unreachable =0;2711 unpack_unreachable_expiration =0;2712}2713else{2714 unpack_unreachable =1;2715if(arg)2716 unpack_unreachable_expiration =approxidate(arg);2717}2718return0;2719}27202721intcmd_pack_objects(int argc,const char**argv,const char*prefix)2722{2723int use_internal_rev_list =0;2724int thin =0;2725int shallow =0;2726int all_progress_implied =0;2727struct argv_array rp = ARGV_ARRAY_INIT;2728int rev_list_unpacked =0, rev_list_all =0, rev_list_reflog =0;2729int rev_list_index =0;2730struct option pack_objects_options[] = {2731OPT_SET_INT('q',"quiet", &progress,2732N_("do not show progress meter"),0),2733OPT_SET_INT(0,"progress", &progress,2734N_("show progress meter"),1),2735OPT_SET_INT(0,"all-progress", &progress,2736N_("show progress meter during object writing phase"),2),2737OPT_BOOL(0,"all-progress-implied",2738&all_progress_implied,2739N_("similar to --all-progress when progress meter is shown")),2740{ OPTION_CALLBACK,0,"index-version", NULL,N_("version[,offset]"),2741N_("write the pack index file in the specified idx format version"),27420, option_parse_index_version },2743OPT_MAGNITUDE(0,"max-pack-size", &pack_size_limit,2744N_("maximum size of each output pack file")),2745OPT_BOOL(0,"local", &local,2746N_("ignore borrowed objects from alternate object store")),2747OPT_BOOL(0,"incremental", &incremental,2748N_("ignore packed objects")),2749OPT_INTEGER(0,"window", &window,2750N_("limit pack window by objects")),2751OPT_MAGNITUDE(0,"window-memory", &window_memory_limit,2752N_("limit pack window by memory in addition to object limit")),2753OPT_INTEGER(0,"depth", &depth,2754N_("maximum length of delta chain allowed in the resulting pack")),2755OPT_BOOL(0,"reuse-delta", &reuse_delta,2756N_("reuse existing deltas")),2757OPT_BOOL(0,"reuse-object", &reuse_object,2758N_("reuse existing objects")),2759OPT_BOOL(0,"delta-base-offset", &allow_ofs_delta,2760N_("use OFS_DELTA objects")),2761OPT_INTEGER(0,"threads", &delta_search_threads,2762N_("use threads when searching for best delta matches")),2763OPT_BOOL(0,"non-empty", &non_empty,2764N_("do not create an empty pack output")),2765OPT_BOOL(0,"revs", &use_internal_rev_list,2766N_("read revision arguments from standard input")),2767{ OPTION_SET_INT,0,"unpacked", &rev_list_unpacked, NULL,2768N_("limit the objects to those that are not yet packed"),2769 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL,1},2770{ OPTION_SET_INT,0,"all", &rev_list_all, NULL,2771N_("include objects reachable from any reference"),2772 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL,1},2773{ OPTION_SET_INT,0,"reflog", &rev_list_reflog, NULL,2774N_("include objects referred by reflog entries"),2775 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL,1},2776{ OPTION_SET_INT,0,"indexed-objects", &rev_list_index, NULL,2777N_("include objects referred to by the index"),2778 PARSE_OPT_NOARG | PARSE_OPT_NONEG, NULL,1},2779OPT_BOOL(0,"stdout", &pack_to_stdout,2780N_("output pack to stdout")),2781OPT_BOOL(0,"include-tag", &include_tag,2782N_("include tag objects that refer to objects to be packed")),2783OPT_BOOL(0,"keep-unreachable", &keep_unreachable,2784N_("keep unreachable objects")),2785OPT_BOOL(0,"pack-loose-unreachable", &pack_loose_unreachable,2786N_("pack loose unreachable objects")),2787{ OPTION_CALLBACK,0,"unpack-unreachable", NULL,N_("time"),2788N_("unpack unreachable objects newer than <time>"),2789 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },2790OPT_BOOL(0,"thin", &thin,2791N_("create thin packs")),2792OPT_BOOL(0,"shallow", &shallow,2793N_("create packs suitable for shallow fetches")),2794OPT_BOOL(0,"honor-pack-keep", &ignore_packed_keep,2795N_("ignore packs that have companion .keep file")),2796OPT_INTEGER(0,"compression", &pack_compression_level,2797N_("pack compression level")),2798OPT_SET_INT(0,"keep-true-parents", &grafts_replace_parents,2799N_("do not hide commits by grafts"),0),2800OPT_BOOL(0,"use-bitmap-index", &use_bitmap_index,2801N_("use a bitmap index if available to speed up counting objects")),2802OPT_BOOL(0,"write-bitmap-index", &write_bitmap_index,2803N_("write a bitmap index together with the pack index")),2804OPT_END(),2805};28062807 check_replace_refs =0;28082809reset_pack_idx_option(&pack_idx_opts);2810git_config(git_pack_config, NULL);2811if(!pack_compression_seen && core_compression_seen)2812 pack_compression_level = core_compression_level;28132814 progress =isatty(2);2815 argc =parse_options(argc, argv, prefix, pack_objects_options,2816 pack_usage,0);28172818if(argc) {2819 base_name = argv[0];2820 argc--;2821}2822if(pack_to_stdout != !base_name || argc)2823usage_with_options(pack_usage, pack_objects_options);28242825argv_array_push(&rp,"pack-objects");2826if(thin) {2827 use_internal_rev_list =1;2828argv_array_push(&rp, shallow2829?"--objects-edge-aggressive"2830:"--objects-edge");2831}else2832argv_array_push(&rp,"--objects");28332834if(rev_list_all) {2835 use_internal_rev_list =1;2836argv_array_push(&rp,"--all");2837}2838if(rev_list_reflog) {2839 use_internal_rev_list =1;2840argv_array_push(&rp,"--reflog");2841}2842if(rev_list_index) {2843 use_internal_rev_list =1;2844argv_array_push(&rp,"--indexed-objects");2845}2846if(rev_list_unpacked) {2847 use_internal_rev_list =1;2848argv_array_push(&rp,"--unpacked");2849}28502851if(!reuse_object)2852 reuse_delta =0;2853if(pack_compression_level == -1)2854 pack_compression_level = Z_DEFAULT_COMPRESSION;2855else if(pack_compression_level <0|| pack_compression_level > Z_BEST_COMPRESSION)2856die("bad pack compression level%d", pack_compression_level);28572858if(!delta_search_threads)/* --threads=0 means autodetect */2859 delta_search_threads =online_cpus();28602861#ifdef NO_PTHREADS2862if(delta_search_threads !=1)2863warning("no threads support, ignoring --threads");2864#endif2865if(!pack_to_stdout && !pack_size_limit)2866 pack_size_limit = pack_size_limit_cfg;2867if(pack_to_stdout && pack_size_limit)2868die("--max-pack-size cannot be used to build a pack for transfer.");2869if(pack_size_limit && pack_size_limit <1024*1024) {2870warning("minimum pack size limit is 1 MiB");2871 pack_size_limit =1024*1024;2872}28732874if(!pack_to_stdout && thin)2875die("--thin cannot be used to build an indexable pack.");28762877if(keep_unreachable && unpack_unreachable)2878die("--keep-unreachable and --unpack-unreachable are incompatible.");2879if(!rev_list_all || !rev_list_reflog || !rev_list_index)2880 unpack_unreachable_expiration =0;28812882if(!use_internal_rev_list || !pack_to_stdout ||is_repository_shallow())2883 use_bitmap_index =0;28842885if(pack_to_stdout || !rev_list_all)2886 write_bitmap_index =0;28872888if(progress && all_progress_implied)2889 progress =2;28902891prepare_packed_git();2892if(ignore_packed_keep) {2893struct packed_git *p;2894for(p = packed_git; p; p = p->next)2895if(p->pack_local && p->pack_keep)2896break;2897if(!p)/* no keep-able packs found */2898 ignore_packed_keep =0;2899}2900if(local) {2901/*2902 * unlike ignore_packed_keep above, we do not want to2903 * unset "local" based on looking at packs, as it2904 * also covers non-local objects2905 */2906struct packed_git *p;2907for(p = packed_git; p; p = p->next) {2908if(!p->pack_local) {2909 have_non_local_packs =1;2910break;2911}2912}2913}29142915if(progress)2916 progress_state =start_progress(_("Counting objects"),0);2917if(!use_internal_rev_list)2918read_object_list_from_stdin();2919else{2920get_object_list(rp.argc, rp.argv);2921argv_array_clear(&rp);2922}2923cleanup_preferred_base();2924if(include_tag && nr_result)2925for_each_ref(add_ref_tag, NULL);2926stop_progress(&progress_state);29272928if(non_empty && !nr_result)2929return0;2930if(nr_result)2931prepare_pack(window, depth);2932write_pack_file();2933if(progress)2934fprintf(stderr,"Total %"PRIu32" (delta %"PRIu32"),"2935" reused %"PRIu32" (delta %"PRIu32")\n",2936 written, written_delta, reused, reused_delta);2937return0;2938}