1#include"builtin.h" 2#include"cache.h" 3#include"repository.h" 4#include"config.h" 5#include"attr.h" 6#include"object.h" 7#include"blob.h" 8#include"commit.h" 9#include"tag.h" 10#include"tree.h" 11#include"delta.h" 12#include"pack.h" 13#include"pack-revindex.h" 14#include"csum-file.h" 15#include"tree-walk.h" 16#include"diff.h" 17#include"revision.h" 18#include"list-objects.h" 19#include"list-objects-filter.h" 20#include"list-objects-filter-options.h" 21#include"pack-objects.h" 22#include"progress.h" 23#include"refs.h" 24#include"streaming.h" 25#include"thread-utils.h" 26#include"pack-bitmap.h" 27#include"reachable.h" 28#include"sha1-array.h" 29#include"argv-array.h" 30#include"list.h" 31#include"packfile.h" 32#include"object-store.h" 33#include"dir.h" 34 35#define IN_PACK(obj) oe_in_pack(&to_pack, obj) 36#define SIZE(obj) oe_size(&to_pack, obj) 37#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size) 38#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj) 39#define DELTA(obj) oe_delta(&to_pack, obj) 40#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj) 41#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj) 42#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val) 43#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val) 44#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val) 45#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val) 46 47static const char*pack_usage[] = { 48N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), 49N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), 50 NULL 51}; 52 53/* 54 * Objects we are going to pack are collected in the `to_pack` structure. 55 * It contains an array (dynamically expanded) of the object data, and a map 56 * that can resolve SHA1s to their position in the array. 57 */ 58static struct packing_data to_pack; 59 60static struct pack_idx_entry **written_list; 61static uint32_t nr_result, nr_written, nr_seen; 62 63static int non_empty; 64static int reuse_delta =1, reuse_object =1; 65static int keep_unreachable, unpack_unreachable, include_tag; 66static timestamp_t unpack_unreachable_expiration; 67static int pack_loose_unreachable; 68static int local; 69static int have_non_local_packs; 70static int incremental; 71static int ignore_packed_keep_on_disk; 72static int ignore_packed_keep_in_core; 73static int allow_ofs_delta; 74static struct pack_idx_option pack_idx_opts; 75static const char*base_name; 76static int progress =1; 77static int window =10; 78static unsigned long pack_size_limit; 79static int depth =50; 80static int delta_search_threads; 81static int pack_to_stdout; 82static int num_preferred_base; 83static struct progress *progress_state; 84 85static struct packed_git *reuse_packfile; 86static uint32_t reuse_packfile_objects; 87static off_t reuse_packfile_offset; 88 89static int use_bitmap_index_default =1; 90static int use_bitmap_index = -1; 91static int write_bitmap_index; 92static uint16_t write_bitmap_options; 93 94static int exclude_promisor_objects; 95 96static unsigned long delta_cache_size =0; 97static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; 98static unsigned long cache_max_small_delta_size =1000; 99 100static unsigned long window_memory_limit =0; 101 102static struct list_objects_filter_options filter_options; 103 104enum missing_action { 105 MA_ERROR =0,/* fail if any missing objects are encountered */ 106 MA_ALLOW_ANY,/* silently allow ALL missing objects */ 107 MA_ALLOW_PROMISOR,/* silently allow all missing PROMISOR objects */ 108}; 109static enum missing_action arg_missing_action; 110static show_object_fn fn_show_object; 111 112/* 113 * stats 114 */ 115static uint32_t written, written_delta; 116static uint32_t reused, reused_delta; 117 118/* 119 * Indexed commits 120 */ 121static struct commit **indexed_commits; 122static unsigned int indexed_commits_nr; 123static unsigned int indexed_commits_alloc; 124 125static voidindex_commit_for_bitmap(struct commit *commit) 126{ 127if(indexed_commits_nr >= indexed_commits_alloc) { 128 indexed_commits_alloc = (indexed_commits_alloc +32) *2; 129REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); 130} 131 132 indexed_commits[indexed_commits_nr++] = commit; 133} 134 135static void*get_delta(struct object_entry *entry) 136{ 137unsigned long size, base_size, delta_size; 138void*buf, *base_buf, *delta_buf; 139enum object_type type; 140 141 buf =read_object_file(&entry->idx.oid, &type, &size); 142if(!buf) 143die(_("unable to read%s"),oid_to_hex(&entry->idx.oid)); 144 base_buf =read_object_file(&DELTA(entry)->idx.oid, &type, 145&base_size); 146if(!base_buf) 147die("unable to read%s", 148oid_to_hex(&DELTA(entry)->idx.oid)); 149 delta_buf =diff_delta(base_buf, base_size, 150 buf, size, &delta_size,0); 151/* 152 * We succesfully computed this delta once but dropped it for 153 * memory reasons. Something is very wrong if this time we 154 * recompute and create a different delta. 155 */ 156if(!delta_buf || delta_size !=DELTA_SIZE(entry)) 157BUG("delta size changed"); 158free(buf); 159free(base_buf); 160return delta_buf; 161} 162 163static unsigned longdo_compress(void**pptr,unsigned long size) 164{ 165 git_zstream stream; 166void*in, *out; 167unsigned long maxsize; 168 169git_deflate_init(&stream, pack_compression_level); 170 maxsize =git_deflate_bound(&stream, size); 171 172 in = *pptr; 173 out =xmalloc(maxsize); 174*pptr = out; 175 176 stream.next_in = in; 177 stream.avail_in = size; 178 stream.next_out = out; 179 stream.avail_out = maxsize; 180while(git_deflate(&stream, Z_FINISH) == Z_OK) 181;/* nothing */ 182git_deflate_end(&stream); 183 184free(in); 185return stream.total_out; 186} 187 188static unsigned longwrite_large_blob_data(struct git_istream *st,struct hashfile *f, 189const struct object_id *oid) 190{ 191 git_zstream stream; 192unsigned char ibuf[1024*16]; 193unsigned char obuf[1024*16]; 194unsigned long olen =0; 195 196git_deflate_init(&stream, pack_compression_level); 197 198for(;;) { 199 ssize_t readlen; 200int zret = Z_OK; 201 readlen =read_istream(st, ibuf,sizeof(ibuf)); 202if(readlen == -1) 203die(_("unable to read%s"),oid_to_hex(oid)); 204 205 stream.next_in = ibuf; 206 stream.avail_in = readlen; 207while((stream.avail_in || readlen ==0) && 208(zret == Z_OK || zret == Z_BUF_ERROR)) { 209 stream.next_out = obuf; 210 stream.avail_out =sizeof(obuf); 211 zret =git_deflate(&stream, readlen ?0: Z_FINISH); 212hashwrite(f, obuf, stream.next_out - obuf); 213 olen += stream.next_out - obuf; 214} 215if(stream.avail_in) 216die(_("deflate error (%d)"), zret); 217if(readlen ==0) { 218if(zret != Z_STREAM_END) 219die(_("deflate error (%d)"), zret); 220break; 221} 222} 223git_deflate_end(&stream); 224return olen; 225} 226 227/* 228 * we are going to reuse the existing object data as is. make 229 * sure it is not corrupt. 230 */ 231static intcheck_pack_inflate(struct packed_git *p, 232struct pack_window **w_curs, 233 off_t offset, 234 off_t len, 235unsigned long expect) 236{ 237 git_zstream stream; 238unsigned char fakebuf[4096], *in; 239int st; 240 241memset(&stream,0,sizeof(stream)); 242git_inflate_init(&stream); 243do{ 244 in =use_pack(p, w_curs, offset, &stream.avail_in); 245 stream.next_in = in; 246 stream.next_out = fakebuf; 247 stream.avail_out =sizeof(fakebuf); 248 st =git_inflate(&stream, Z_FINISH); 249 offset += stream.next_in - in; 250}while(st == Z_OK || st == Z_BUF_ERROR); 251git_inflate_end(&stream); 252return(st == Z_STREAM_END && 253 stream.total_out == expect && 254 stream.total_in == len) ?0: -1; 255} 256 257static voidcopy_pack_data(struct hashfile *f, 258struct packed_git *p, 259struct pack_window **w_curs, 260 off_t offset, 261 off_t len) 262{ 263unsigned char*in; 264unsigned long avail; 265 266while(len) { 267 in =use_pack(p, w_curs, offset, &avail); 268if(avail > len) 269 avail = (unsigned long)len; 270hashwrite(f, in, avail); 271 offset += avail; 272 len -= avail; 273} 274} 275 276/* Return 0 if we will bust the pack-size limit */ 277static unsigned longwrite_no_reuse_object(struct hashfile *f,struct object_entry *entry, 278unsigned long limit,int usable_delta) 279{ 280unsigned long size, datalen; 281unsigned char header[MAX_PACK_OBJECT_HEADER], 282 dheader[MAX_PACK_OBJECT_HEADER]; 283unsigned hdrlen; 284enum object_type type; 285void*buf; 286struct git_istream *st = NULL; 287const unsigned hashsz = the_hash_algo->rawsz; 288 289if(!usable_delta) { 290if(oe_type(entry) == OBJ_BLOB && 291oe_size_greater_than(&to_pack, entry, big_file_threshold) && 292(st =open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL) 293 buf = NULL; 294else{ 295 buf =read_object_file(&entry->idx.oid, &type, &size); 296if(!buf) 297die(_("unable to read%s"), 298oid_to_hex(&entry->idx.oid)); 299} 300/* 301 * make sure no cached delta data remains from a 302 * previous attempt before a pack split occurred. 303 */ 304FREE_AND_NULL(entry->delta_data); 305 entry->z_delta_size =0; 306}else if(entry->delta_data) { 307 size =DELTA_SIZE(entry); 308 buf = entry->delta_data; 309 entry->delta_data = NULL; 310 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 311 OBJ_OFS_DELTA : OBJ_REF_DELTA; 312}else{ 313 buf =get_delta(entry); 314 size =DELTA_SIZE(entry); 315 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 316 OBJ_OFS_DELTA : OBJ_REF_DELTA; 317} 318 319if(st)/* large blob case, just assume we don't compress well */ 320 datalen = size; 321else if(entry->z_delta_size) 322 datalen = entry->z_delta_size; 323else 324 datalen =do_compress(&buf, size); 325 326/* 327 * The object header is a byte of 'type' followed by zero or 328 * more bytes of length. 329 */ 330 hdrlen =encode_in_pack_object_header(header,sizeof(header), 331 type, size); 332 333if(type == OBJ_OFS_DELTA) { 334/* 335 * Deltas with relative base contain an additional 336 * encoding of the relative offset for the delta 337 * base from this object's position in the pack. 338 */ 339 off_t ofs = entry->idx.offset -DELTA(entry)->idx.offset; 340unsigned pos =sizeof(dheader) -1; 341 dheader[pos] = ofs &127; 342while(ofs >>=7) 343 dheader[--pos] =128| (--ofs &127); 344if(limit && hdrlen +sizeof(dheader) - pos + datalen + hashsz >= limit) { 345if(st) 346close_istream(st); 347free(buf); 348return0; 349} 350hashwrite(f, header, hdrlen); 351hashwrite(f, dheader + pos,sizeof(dheader) - pos); 352 hdrlen +=sizeof(dheader) - pos; 353}else if(type == OBJ_REF_DELTA) { 354/* 355 * Deltas with a base reference contain 356 * additional bytes for the base object ID. 357 */ 358if(limit && hdrlen + hashsz + datalen + hashsz >= limit) { 359if(st) 360close_istream(st); 361free(buf); 362return0; 363} 364hashwrite(f, header, hdrlen); 365hashwrite(f,DELTA(entry)->idx.oid.hash, hashsz); 366 hdrlen += hashsz; 367}else{ 368if(limit && hdrlen + datalen + hashsz >= limit) { 369if(st) 370close_istream(st); 371free(buf); 372return0; 373} 374hashwrite(f, header, hdrlen); 375} 376if(st) { 377 datalen =write_large_blob_data(st, f, &entry->idx.oid); 378close_istream(st); 379}else{ 380hashwrite(f, buf, datalen); 381free(buf); 382} 383 384return hdrlen + datalen; 385} 386 387/* Return 0 if we will bust the pack-size limit */ 388static off_t write_reuse_object(struct hashfile *f,struct object_entry *entry, 389unsigned long limit,int usable_delta) 390{ 391struct packed_git *p =IN_PACK(entry); 392struct pack_window *w_curs = NULL; 393struct revindex_entry *revidx; 394 off_t offset; 395enum object_type type =oe_type(entry); 396 off_t datalen; 397unsigned char header[MAX_PACK_OBJECT_HEADER], 398 dheader[MAX_PACK_OBJECT_HEADER]; 399unsigned hdrlen; 400const unsigned hashsz = the_hash_algo->rawsz; 401unsigned long entry_size =SIZE(entry); 402 403if(DELTA(entry)) 404 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 405 OBJ_OFS_DELTA : OBJ_REF_DELTA; 406 hdrlen =encode_in_pack_object_header(header,sizeof(header), 407 type, entry_size); 408 409 offset = entry->in_pack_offset; 410 revidx =find_pack_revindex(p, offset); 411 datalen = revidx[1].offset - offset; 412if(!pack_to_stdout && p->index_version >1&& 413check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 414error(_("bad packed object CRC for%s"), 415oid_to_hex(&entry->idx.oid)); 416unuse_pack(&w_curs); 417returnwrite_no_reuse_object(f, entry, limit, usable_delta); 418} 419 420 offset += entry->in_pack_header_size; 421 datalen -= entry->in_pack_header_size; 422 423if(!pack_to_stdout && p->index_version ==1&& 424check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { 425error(_("corrupt packed object for%s"), 426oid_to_hex(&entry->idx.oid)); 427unuse_pack(&w_curs); 428returnwrite_no_reuse_object(f, entry, limit, usable_delta); 429} 430 431if(type == OBJ_OFS_DELTA) { 432 off_t ofs = entry->idx.offset -DELTA(entry)->idx.offset; 433unsigned pos =sizeof(dheader) -1; 434 dheader[pos] = ofs &127; 435while(ofs >>=7) 436 dheader[--pos] =128| (--ofs &127); 437if(limit && hdrlen +sizeof(dheader) - pos + datalen + hashsz >= limit) { 438unuse_pack(&w_curs); 439return0; 440} 441hashwrite(f, header, hdrlen); 442hashwrite(f, dheader + pos,sizeof(dheader) - pos); 443 hdrlen +=sizeof(dheader) - pos; 444 reused_delta++; 445}else if(type == OBJ_REF_DELTA) { 446if(limit && hdrlen + hashsz + datalen + hashsz >= limit) { 447unuse_pack(&w_curs); 448return0; 449} 450hashwrite(f, header, hdrlen); 451hashwrite(f,DELTA(entry)->idx.oid.hash, hashsz); 452 hdrlen += hashsz; 453 reused_delta++; 454}else{ 455if(limit && hdrlen + datalen + hashsz >= limit) { 456unuse_pack(&w_curs); 457return0; 458} 459hashwrite(f, header, hdrlen); 460} 461copy_pack_data(f, p, &w_curs, offset, datalen); 462unuse_pack(&w_curs); 463 reused++; 464return hdrlen + datalen; 465} 466 467/* Return 0 if we will bust the pack-size limit */ 468static off_t write_object(struct hashfile *f, 469struct object_entry *entry, 470 off_t write_offset) 471{ 472unsigned long limit; 473 off_t len; 474int usable_delta, to_reuse; 475 476if(!pack_to_stdout) 477crc32_begin(f); 478 479/* apply size limit if limited packsize and not first object */ 480if(!pack_size_limit || !nr_written) 481 limit =0; 482else if(pack_size_limit <= write_offset) 483/* 484 * the earlier object did not fit the limit; avoid 485 * mistaking this with unlimited (i.e. limit = 0). 486 */ 487 limit =1; 488else 489 limit = pack_size_limit - write_offset; 490 491if(!DELTA(entry)) 492 usable_delta =0;/* no delta */ 493else if(!pack_size_limit) 494 usable_delta =1;/* unlimited packfile */ 495else if(DELTA(entry)->idx.offset == (off_t)-1) 496 usable_delta =0;/* base was written to another pack */ 497else if(DELTA(entry)->idx.offset) 498 usable_delta =1;/* base already exists in this pack */ 499else 500 usable_delta =0;/* base could end up in another pack */ 501 502if(!reuse_object) 503 to_reuse =0;/* explicit */ 504else if(!IN_PACK(entry)) 505 to_reuse =0;/* can't reuse what we don't have */ 506else if(oe_type(entry) == OBJ_REF_DELTA || 507oe_type(entry) == OBJ_OFS_DELTA) 508/* check_object() decided it for us ... */ 509 to_reuse = usable_delta; 510/* ... but pack split may override that */ 511else if(oe_type(entry) != entry->in_pack_type) 512 to_reuse =0;/* pack has delta which is unusable */ 513else if(DELTA(entry)) 514 to_reuse =0;/* we want to pack afresh */ 515else 516 to_reuse =1;/* we have it in-pack undeltified, 517 * and we do not need to deltify it. 518 */ 519 520if(!to_reuse) 521 len =write_no_reuse_object(f, entry, limit, usable_delta); 522else 523 len =write_reuse_object(f, entry, limit, usable_delta); 524if(!len) 525return0; 526 527if(usable_delta) 528 written_delta++; 529 written++; 530if(!pack_to_stdout) 531 entry->idx.crc32 =crc32_end(f); 532return len; 533} 534 535enum write_one_status { 536 WRITE_ONE_SKIP = -1,/* already written */ 537 WRITE_ONE_BREAK =0,/* writing this will bust the limit; not written */ 538 WRITE_ONE_WRITTEN =1,/* normal */ 539 WRITE_ONE_RECURSIVE =2/* already scheduled to be written */ 540}; 541 542static enum write_one_status write_one(struct hashfile *f, 543struct object_entry *e, 544 off_t *offset) 545{ 546 off_t size; 547int recursing; 548 549/* 550 * we set offset to 1 (which is an impossible value) to mark 551 * the fact that this object is involved in "write its base 552 * first before writing a deltified object" recursion. 553 */ 554 recursing = (e->idx.offset ==1); 555if(recursing) { 556warning(_("recursive delta detected for object%s"), 557oid_to_hex(&e->idx.oid)); 558return WRITE_ONE_RECURSIVE; 559}else if(e->idx.offset || e->preferred_base) { 560/* offset is non zero if object is written already. */ 561return WRITE_ONE_SKIP; 562} 563 564/* if we are deltified, write out base object first. */ 565if(DELTA(e)) { 566 e->idx.offset =1;/* now recurse */ 567switch(write_one(f,DELTA(e), offset)) { 568case WRITE_ONE_RECURSIVE: 569/* we cannot depend on this one */ 570SET_DELTA(e, NULL); 571break; 572default: 573break; 574case WRITE_ONE_BREAK: 575 e->idx.offset = recursing; 576return WRITE_ONE_BREAK; 577} 578} 579 580 e->idx.offset = *offset; 581 size =write_object(f, e, *offset); 582if(!size) { 583 e->idx.offset = recursing; 584return WRITE_ONE_BREAK; 585} 586 written_list[nr_written++] = &e->idx; 587 588/* make sure off_t is sufficiently large not to wrap */ 589if(signed_add_overflows(*offset, size)) 590die(_("pack too large for current definition of off_t")); 591*offset += size; 592return WRITE_ONE_WRITTEN; 593} 594 595static intmark_tagged(const char*path,const struct object_id *oid,int flag, 596void*cb_data) 597{ 598struct object_id peeled; 599struct object_entry *entry =packlist_find(&to_pack, oid->hash, NULL); 600 601if(entry) 602 entry->tagged =1; 603if(!peel_ref(path, &peeled)) { 604 entry =packlist_find(&to_pack, peeled.hash, NULL); 605if(entry) 606 entry->tagged =1; 607} 608return0; 609} 610 611staticinlinevoidadd_to_write_order(struct object_entry **wo, 612unsigned int*endp, 613struct object_entry *e) 614{ 615if(e->filled) 616return; 617 wo[(*endp)++] = e; 618 e->filled =1; 619} 620 621static voidadd_descendants_to_write_order(struct object_entry **wo, 622unsigned int*endp, 623struct object_entry *e) 624{ 625int add_to_order =1; 626while(e) { 627if(add_to_order) { 628struct object_entry *s; 629/* add this node... */ 630add_to_write_order(wo, endp, e); 631/* all its siblings... */ 632for(s =DELTA_SIBLING(e); s; s =DELTA_SIBLING(s)) { 633add_to_write_order(wo, endp, s); 634} 635} 636/* drop down a level to add left subtree nodes if possible */ 637if(DELTA_CHILD(e)) { 638 add_to_order =1; 639 e =DELTA_CHILD(e); 640}else{ 641 add_to_order =0; 642/* our sibling might have some children, it is next */ 643if(DELTA_SIBLING(e)) { 644 e =DELTA_SIBLING(e); 645continue; 646} 647/* go back to our parent node */ 648 e =DELTA(e); 649while(e && !DELTA_SIBLING(e)) { 650/* we're on the right side of a subtree, keep 651 * going up until we can go right again */ 652 e =DELTA(e); 653} 654if(!e) { 655/* done- we hit our original root node */ 656return; 657} 658/* pass it off to sibling at this level */ 659 e =DELTA_SIBLING(e); 660} 661}; 662} 663 664static voidadd_family_to_write_order(struct object_entry **wo, 665unsigned int*endp, 666struct object_entry *e) 667{ 668struct object_entry *root; 669 670for(root = e;DELTA(root); root =DELTA(root)) 671;/* nothing */ 672add_descendants_to_write_order(wo, endp, root); 673} 674 675static struct object_entry **compute_write_order(void) 676{ 677unsigned int i, wo_end, last_untagged; 678 679struct object_entry **wo; 680struct object_entry *objects = to_pack.objects; 681 682for(i =0; i < to_pack.nr_objects; i++) { 683 objects[i].tagged =0; 684 objects[i].filled =0; 685SET_DELTA_CHILD(&objects[i], NULL); 686SET_DELTA_SIBLING(&objects[i], NULL); 687} 688 689/* 690 * Fully connect delta_child/delta_sibling network. 691 * Make sure delta_sibling is sorted in the original 692 * recency order. 693 */ 694for(i = to_pack.nr_objects; i >0;) { 695struct object_entry *e = &objects[--i]; 696if(!DELTA(e)) 697continue; 698/* Mark me as the first child */ 699 e->delta_sibling_idx =DELTA(e)->delta_child_idx; 700SET_DELTA_CHILD(DELTA(e), e); 701} 702 703/* 704 * Mark objects that are at the tip of tags. 705 */ 706for_each_tag_ref(mark_tagged, NULL); 707 708/* 709 * Give the objects in the original recency order until 710 * we see a tagged tip. 711 */ 712ALLOC_ARRAY(wo, to_pack.nr_objects); 713for(i = wo_end =0; i < to_pack.nr_objects; i++) { 714if(objects[i].tagged) 715break; 716add_to_write_order(wo, &wo_end, &objects[i]); 717} 718 last_untagged = i; 719 720/* 721 * Then fill all the tagged tips. 722 */ 723for(; i < to_pack.nr_objects; i++) { 724if(objects[i].tagged) 725add_to_write_order(wo, &wo_end, &objects[i]); 726} 727 728/* 729 * And then all remaining commits and tags. 730 */ 731for(i = last_untagged; i < to_pack.nr_objects; i++) { 732if(oe_type(&objects[i]) != OBJ_COMMIT && 733oe_type(&objects[i]) != OBJ_TAG) 734continue; 735add_to_write_order(wo, &wo_end, &objects[i]); 736} 737 738/* 739 * And then all the trees. 740 */ 741for(i = last_untagged; i < to_pack.nr_objects; i++) { 742if(oe_type(&objects[i]) != OBJ_TREE) 743continue; 744add_to_write_order(wo, &wo_end, &objects[i]); 745} 746 747/* 748 * Finally all the rest in really tight order 749 */ 750for(i = last_untagged; i < to_pack.nr_objects; i++) { 751if(!objects[i].filled) 752add_family_to_write_order(wo, &wo_end, &objects[i]); 753} 754 755if(wo_end != to_pack.nr_objects) 756die(_("ordered%uobjects, expected %"PRIu32), 757 wo_end, to_pack.nr_objects); 758 759return wo; 760} 761 762static off_t write_reused_pack(struct hashfile *f) 763{ 764unsigned char buffer[8192]; 765 off_t to_write, total; 766int fd; 767 768if(!is_pack_valid(reuse_packfile)) 769die(_("packfile is invalid:%s"), reuse_packfile->pack_name); 770 771 fd =git_open(reuse_packfile->pack_name); 772if(fd <0) 773die_errno(_("unable to open packfile for reuse:%s"), 774 reuse_packfile->pack_name); 775 776if(lseek(fd,sizeof(struct pack_header), SEEK_SET) == -1) 777die_errno(_("unable to seek in reused packfile")); 778 779if(reuse_packfile_offset <0) 780 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz; 781 782 total = to_write = reuse_packfile_offset -sizeof(struct pack_header); 783 784while(to_write) { 785int read_pack =xread(fd, buffer,sizeof(buffer)); 786 787if(read_pack <=0) 788die_errno(_("unable to read from reused packfile")); 789 790if(read_pack > to_write) 791 read_pack = to_write; 792 793hashwrite(f, buffer, read_pack); 794 to_write -= read_pack; 795 796/* 797 * We don't know the actual number of objects written, 798 * only how many bytes written, how many bytes total, and 799 * how many objects total. So we can fake it by pretending all 800 * objects we are writing are the same size. This gives us a 801 * smooth progress meter, and at the end it matches the true 802 * answer. 803 */ 804 written = reuse_packfile_objects * 805(((double)(total - to_write)) / total); 806display_progress(progress_state, written); 807} 808 809close(fd); 810 written = reuse_packfile_objects; 811display_progress(progress_state, written); 812return reuse_packfile_offset -sizeof(struct pack_header); 813} 814 815static const char no_split_warning[] =N_( 816"disabling bitmap writing, packs are split due to pack.packSizeLimit" 817); 818 819static voidwrite_pack_file(void) 820{ 821uint32_t i =0, j; 822struct hashfile *f; 823 off_t offset; 824uint32_t nr_remaining = nr_result; 825time_t last_mtime =0; 826struct object_entry **write_order; 827 828if(progress > pack_to_stdout) 829 progress_state =start_progress(_("Writing objects"), nr_result); 830ALLOC_ARRAY(written_list, to_pack.nr_objects); 831 write_order =compute_write_order(); 832 833do{ 834struct object_id oid; 835char*pack_tmp_name = NULL; 836 837if(pack_to_stdout) 838 f =hashfd_throughput(1,"<stdout>", progress_state); 839else 840 f =create_tmp_packfile(&pack_tmp_name); 841 842 offset =write_pack_header(f, nr_remaining); 843 844if(reuse_packfile) { 845 off_t packfile_size; 846assert(pack_to_stdout); 847 848 packfile_size =write_reused_pack(f); 849 offset += packfile_size; 850} 851 852 nr_written =0; 853for(; i < to_pack.nr_objects; i++) { 854struct object_entry *e = write_order[i]; 855if(write_one(f, e, &offset) == WRITE_ONE_BREAK) 856break; 857display_progress(progress_state, written); 858} 859 860/* 861 * Did we write the wrong # entries in the header? 862 * If so, rewrite it like in fast-import 863 */ 864if(pack_to_stdout) { 865finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); 866}else if(nr_written == nr_remaining) { 867finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); 868}else{ 869int fd =finalize_hashfile(f, oid.hash,0); 870fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, 871 nr_written, oid.hash, offset); 872close(fd); 873if(write_bitmap_index) { 874warning(_(no_split_warning)); 875 write_bitmap_index =0; 876} 877} 878 879if(!pack_to_stdout) { 880struct stat st; 881struct strbuf tmpname = STRBUF_INIT; 882 883/* 884 * Packs are runtime accessed in their mtime 885 * order since newer packs are more likely to contain 886 * younger objects. So if we are creating multiple 887 * packs then we should modify the mtime of later ones 888 * to preserve this property. 889 */ 890if(stat(pack_tmp_name, &st) <0) { 891warning_errno(_("failed to stat%s"), pack_tmp_name); 892}else if(!last_mtime) { 893 last_mtime = st.st_mtime; 894}else{ 895struct utimbuf utb; 896 utb.actime = st.st_atime; 897 utb.modtime = --last_mtime; 898if(utime(pack_tmp_name, &utb) <0) 899warning_errno(_("failed utime() on%s"), pack_tmp_name); 900} 901 902strbuf_addf(&tmpname,"%s-", base_name); 903 904if(write_bitmap_index) { 905bitmap_writer_set_checksum(oid.hash); 906bitmap_writer_build_type_index( 907&to_pack, written_list, nr_written); 908} 909 910finish_tmp_packfile(&tmpname, pack_tmp_name, 911 written_list, nr_written, 912&pack_idx_opts, oid.hash); 913 914if(write_bitmap_index) { 915strbuf_addf(&tmpname,"%s.bitmap",oid_to_hex(&oid)); 916 917stop_progress(&progress_state); 918 919bitmap_writer_show_progress(progress); 920bitmap_writer_reuse_bitmaps(&to_pack); 921bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 922bitmap_writer_build(&to_pack); 923bitmap_writer_finish(written_list, nr_written, 924 tmpname.buf, write_bitmap_options); 925 write_bitmap_index =0; 926} 927 928strbuf_release(&tmpname); 929free(pack_tmp_name); 930puts(oid_to_hex(&oid)); 931} 932 933/* mark written objects as written to previous pack */ 934for(j =0; j < nr_written; j++) { 935 written_list[j]->offset = (off_t)-1; 936} 937 nr_remaining -= nr_written; 938}while(nr_remaining && i < to_pack.nr_objects); 939 940free(written_list); 941free(write_order); 942stop_progress(&progress_state); 943if(written != nr_result) 944die(_("wrote %"PRIu32" objects while expecting %"PRIu32), 945 written, nr_result); 946} 947 948static intno_try_delta(const char*path) 949{ 950static struct attr_check *check; 951 952if(!check) 953 check =attr_check_initl("delta", NULL); 954if(git_check_attr(&the_index, path, check)) 955return0; 956if(ATTR_FALSE(check->items[0].value)) 957return1; 958return0; 959} 960 961/* 962 * When adding an object, check whether we have already added it 963 * to our packing list. If so, we can skip. However, if we are 964 * being asked to excludei t, but the previous mention was to include 965 * it, make sure to adjust its flags and tweak our numbers accordingly. 966 * 967 * As an optimization, we pass out the index position where we would have 968 * found the item, since that saves us from having to look it up again a 969 * few lines later when we want to add the new entry. 970 */ 971static inthave_duplicate_entry(const struct object_id *oid, 972int exclude, 973uint32_t*index_pos) 974{ 975struct object_entry *entry; 976 977 entry =packlist_find(&to_pack, oid->hash, index_pos); 978if(!entry) 979return0; 980 981if(exclude) { 982if(!entry->preferred_base) 983 nr_result--; 984 entry->preferred_base =1; 985} 986 987return1; 988} 989 990static intwant_found_object(int exclude,struct packed_git *p) 991{ 992if(exclude) 993return1; 994if(incremental) 995return0; 996 997/* 998 * When asked to do --local (do not include an object that appears in a 999 * pack we borrow from elsewhere) or --honor-pack-keep (do not include1000 * an object that appears in a pack marked with .keep), finding a pack1001 * that matches the criteria is sufficient for us to decide to omit it.1002 * However, even if this pack does not satisfy the criteria, we need to1003 * make sure no copy of this object appears in _any_ pack that makes us1004 * to omit the object, so we need to check all the packs.1005 *1006 * We can however first check whether these options can possible matter;1007 * if they do not matter we know we want the object in generated pack.1008 * Otherwise, we signal "-1" at the end to tell the caller that we do1009 * not know either way, and it needs to check more packs.1010 */1011if(!ignore_packed_keep_on_disk &&1012!ignore_packed_keep_in_core &&1013(!local || !have_non_local_packs))1014return1;10151016if(local && !p->pack_local)1017return0;1018if(p->pack_local &&1019((ignore_packed_keep_on_disk && p->pack_keep) ||1020(ignore_packed_keep_in_core && p->pack_keep_in_core)))1021return0;10221023/* we don't know yet; keep looking for more packs */1024return-1;1025}10261027/*1028 * Check whether we want the object in the pack (e.g., we do not want1029 * objects found in non-local stores if the "--local" option was used).1030 *1031 * If the caller already knows an existing pack it wants to take the object1032 * from, that is passed in *found_pack and *found_offset; otherwise this1033 * function finds if there is any pack that has the object and returns the pack1034 * and its offset in these variables.1035 */1036static intwant_object_in_pack(const struct object_id *oid,1037int exclude,1038struct packed_git **found_pack,1039 off_t *found_offset)1040{1041int want;1042struct list_head *pos;10431044if(!exclude && local &&has_loose_object_nonlocal(oid))1045return0;10461047/*1048 * If we already know the pack object lives in, start checks from that1049 * pack - in the usual case when neither --local was given nor .keep files1050 * are present we will determine the answer right now.1051 */1052if(*found_pack) {1053 want =want_found_object(exclude, *found_pack);1054if(want != -1)1055return want;1056}1057list_for_each(pos,get_packed_git_mru(the_repository)) {1058struct packed_git *p =list_entry(pos,struct packed_git, mru);1059 off_t offset;10601061if(p == *found_pack)1062 offset = *found_offset;1063else1064 offset =find_pack_entry_one(oid->hash, p);10651066if(offset) {1067if(!*found_pack) {1068if(!is_pack_valid(p))1069continue;1070*found_offset = offset;1071*found_pack = p;1072}1073 want =want_found_object(exclude, p);1074if(!exclude && want >0)1075list_move(&p->mru,1076get_packed_git_mru(the_repository));1077if(want != -1)1078return want;1079}1080}10811082return1;1083}10841085static voidcreate_object_entry(const struct object_id *oid,1086enum object_type type,1087uint32_t hash,1088int exclude,1089int no_try_delta,1090uint32_t index_pos,1091struct packed_git *found_pack,1092 off_t found_offset)1093{1094struct object_entry *entry;10951096 entry =packlist_alloc(&to_pack, oid->hash, index_pos);1097 entry->hash = hash;1098oe_set_type(entry, type);1099if(exclude)1100 entry->preferred_base =1;1101else1102 nr_result++;1103if(found_pack) {1104oe_set_in_pack(&to_pack, entry, found_pack);1105 entry->in_pack_offset = found_offset;1106}11071108 entry->no_try_delta = no_try_delta;1109}11101111static const char no_closure_warning[] =N_(1112"disabling bitmap writing, as some objects are not being packed"1113);11141115static intadd_object_entry(const struct object_id *oid,enum object_type type,1116const char*name,int exclude)1117{1118struct packed_git *found_pack = NULL;1119 off_t found_offset =0;1120uint32_t index_pos;11211122display_progress(progress_state, ++nr_seen);11231124if(have_duplicate_entry(oid, exclude, &index_pos))1125return0;11261127if(!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1128/* The pack is missing an object, so it will not have closure */1129if(write_bitmap_index) {1130warning(_(no_closure_warning));1131 write_bitmap_index =0;1132}1133return0;1134}11351136create_object_entry(oid, type,pack_name_hash(name),1137 exclude, name &&no_try_delta(name),1138 index_pos, found_pack, found_offset);1139return1;1140}11411142static intadd_object_entry_from_bitmap(const struct object_id *oid,1143enum object_type type,1144int flags,uint32_t name_hash,1145struct packed_git *pack, off_t offset)1146{1147uint32_t index_pos;11481149display_progress(progress_state, ++nr_seen);11501151if(have_duplicate_entry(oid,0, &index_pos))1152return0;11531154if(!want_object_in_pack(oid,0, &pack, &offset))1155return0;11561157create_object_entry(oid, type, name_hash,0,0, index_pos, pack, offset);1158return1;1159}11601161struct pbase_tree_cache {1162struct object_id oid;1163int ref;1164int temporary;1165void*tree_data;1166unsigned long tree_size;1167};11681169static struct pbase_tree_cache *(pbase_tree_cache[256]);1170static intpbase_tree_cache_ix(const struct object_id *oid)1171{1172return oid->hash[0] %ARRAY_SIZE(pbase_tree_cache);1173}1174static intpbase_tree_cache_ix_incr(int ix)1175{1176return(ix+1) %ARRAY_SIZE(pbase_tree_cache);1177}11781179static struct pbase_tree {1180struct pbase_tree *next;1181/* This is a phony "cache" entry; we are not1182 * going to evict it or find it through _get()1183 * mechanism -- this is for the toplevel node that1184 * would almost always change with any commit.1185 */1186struct pbase_tree_cache pcache;1187} *pbase_tree;11881189static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1190{1191struct pbase_tree_cache *ent, *nent;1192void*data;1193unsigned long size;1194enum object_type type;1195int neigh;1196int my_ix =pbase_tree_cache_ix(oid);1197int available_ix = -1;11981199/* pbase-tree-cache acts as a limited hashtable.1200 * your object will be found at your index or within a few1201 * slots after that slot if it is cached.1202 */1203for(neigh =0; neigh <8; neigh++) {1204 ent = pbase_tree_cache[my_ix];1205if(ent &&oideq(&ent->oid, oid)) {1206 ent->ref++;1207return ent;1208}1209else if(((available_ix <0) && (!ent || !ent->ref)) ||1210((0<= available_ix) &&1211(!ent && pbase_tree_cache[available_ix])))1212 available_ix = my_ix;1213if(!ent)1214break;1215 my_ix =pbase_tree_cache_ix_incr(my_ix);1216}12171218/* Did not find one. Either we got a bogus request or1219 * we need to read and perhaps cache.1220 */1221 data =read_object_file(oid, &type, &size);1222if(!data)1223return NULL;1224if(type != OBJ_TREE) {1225free(data);1226return NULL;1227}12281229/* We need to either cache or return a throwaway copy */12301231if(available_ix <0)1232 ent = NULL;1233else{1234 ent = pbase_tree_cache[available_ix];1235 my_ix = available_ix;1236}12371238if(!ent) {1239 nent =xmalloc(sizeof(*nent));1240 nent->temporary = (available_ix <0);1241}1242else{1243/* evict and reuse */1244free(ent->tree_data);1245 nent = ent;1246}1247oidcpy(&nent->oid, oid);1248 nent->tree_data = data;1249 nent->tree_size = size;1250 nent->ref =1;1251if(!nent->temporary)1252 pbase_tree_cache[my_ix] = nent;1253return nent;1254}12551256static voidpbase_tree_put(struct pbase_tree_cache *cache)1257{1258if(!cache->temporary) {1259 cache->ref--;1260return;1261}1262free(cache->tree_data);1263free(cache);1264}12651266static intname_cmp_len(const char*name)1267{1268int i;1269for(i =0; name[i] && name[i] !='\n'&& name[i] !='/'; i++)1270;1271return i;1272}12731274static voidadd_pbase_object(struct tree_desc *tree,1275const char*name,1276int cmplen,1277const char*fullname)1278{1279struct name_entry entry;1280int cmp;12811282while(tree_entry(tree,&entry)) {1283if(S_ISGITLINK(entry.mode))1284continue;1285 cmp =tree_entry_len(&entry) != cmplen ?1:1286memcmp(name, entry.path, cmplen);1287if(cmp >0)1288continue;1289if(cmp <0)1290return;1291if(name[cmplen] !='/') {1292add_object_entry(entry.oid,1293object_type(entry.mode),1294 fullname,1);1295return;1296}1297if(S_ISDIR(entry.mode)) {1298struct tree_desc sub;1299struct pbase_tree_cache *tree;1300const char*down = name+cmplen+1;1301int downlen =name_cmp_len(down);13021303 tree =pbase_tree_get(entry.oid);1304if(!tree)1305return;1306init_tree_desc(&sub, tree->tree_data, tree->tree_size);13071308add_pbase_object(&sub, down, downlen, fullname);1309pbase_tree_put(tree);1310}1311}1312}13131314static unsigned*done_pbase_paths;1315static int done_pbase_paths_num;1316static int done_pbase_paths_alloc;1317static intdone_pbase_path_pos(unsigned hash)1318{1319int lo =0;1320int hi = done_pbase_paths_num;1321while(lo < hi) {1322int mi = lo + (hi - lo) /2;1323if(done_pbase_paths[mi] == hash)1324return mi;1325if(done_pbase_paths[mi] < hash)1326 hi = mi;1327else1328 lo = mi +1;1329}1330return-lo-1;1331}13321333static intcheck_pbase_path(unsigned hash)1334{1335int pos =done_pbase_path_pos(hash);1336if(0<= pos)1337return1;1338 pos = -pos -1;1339ALLOC_GROW(done_pbase_paths,1340 done_pbase_paths_num +1,1341 done_pbase_paths_alloc);1342 done_pbase_paths_num++;1343if(pos < done_pbase_paths_num)1344MOVE_ARRAY(done_pbase_paths + pos +1, done_pbase_paths + pos,1345 done_pbase_paths_num - pos -1);1346 done_pbase_paths[pos] = hash;1347return0;1348}13491350static voidadd_preferred_base_object(const char*name)1351{1352struct pbase_tree *it;1353int cmplen;1354unsigned hash =pack_name_hash(name);13551356if(!num_preferred_base ||check_pbase_path(hash))1357return;13581359 cmplen =name_cmp_len(name);1360for(it = pbase_tree; it; it = it->next) {1361if(cmplen ==0) {1362add_object_entry(&it->pcache.oid, OBJ_TREE, NULL,1);1363}1364else{1365struct tree_desc tree;1366init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1367add_pbase_object(&tree, name, cmplen, name);1368}1369}1370}13711372static voidadd_preferred_base(struct object_id *oid)1373{1374struct pbase_tree *it;1375void*data;1376unsigned long size;1377struct object_id tree_oid;13781379if(window <= num_preferred_base++)1380return;13811382 data =read_object_with_reference(oid, tree_type, &size, &tree_oid);1383if(!data)1384return;13851386for(it = pbase_tree; it; it = it->next) {1387if(oideq(&it->pcache.oid, &tree_oid)) {1388free(data);1389return;1390}1391}13921393 it =xcalloc(1,sizeof(*it));1394 it->next = pbase_tree;1395 pbase_tree = it;13961397oidcpy(&it->pcache.oid, &tree_oid);1398 it->pcache.tree_data = data;1399 it->pcache.tree_size = size;1400}14011402static voidcleanup_preferred_base(void)1403{1404struct pbase_tree *it;1405unsigned i;14061407 it = pbase_tree;1408 pbase_tree = NULL;1409while(it) {1410struct pbase_tree *tmp = it;1411 it = tmp->next;1412free(tmp->pcache.tree_data);1413free(tmp);1414}14151416for(i =0; i <ARRAY_SIZE(pbase_tree_cache); i++) {1417if(!pbase_tree_cache[i])1418continue;1419free(pbase_tree_cache[i]->tree_data);1420FREE_AND_NULL(pbase_tree_cache[i]);1421}14221423FREE_AND_NULL(done_pbase_paths);1424 done_pbase_paths_num = done_pbase_paths_alloc =0;1425}14261427static voidcheck_object(struct object_entry *entry)1428{1429unsigned long canonical_size;14301431if(IN_PACK(entry)) {1432struct packed_git *p =IN_PACK(entry);1433struct pack_window *w_curs = NULL;1434const unsigned char*base_ref = NULL;1435struct object_entry *base_entry;1436unsigned long used, used_0;1437unsigned long avail;1438 off_t ofs;1439unsigned char*buf, c;1440enum object_type type;1441unsigned long in_pack_size;14421443 buf =use_pack(p, &w_curs, entry->in_pack_offset, &avail);14441445/*1446 * We want in_pack_type even if we do not reuse delta1447 * since non-delta representations could still be reused.1448 */1449 used =unpack_object_header_buffer(buf, avail,1450&type,1451&in_pack_size);1452if(used ==0)1453goto give_up;14541455if(type <0)1456BUG("invalid type%d", type);1457 entry->in_pack_type = type;14581459/*1460 * Determine if this is a delta and if so whether we can1461 * reuse it or not. Otherwise let's find out as cheaply as1462 * possible what the actual type and size for this object is.1463 */1464switch(entry->in_pack_type) {1465default:1466/* Not a delta hence we've already got all we need. */1467oe_set_type(entry, entry->in_pack_type);1468SET_SIZE(entry, in_pack_size);1469 entry->in_pack_header_size = used;1470if(oe_type(entry) < OBJ_COMMIT ||oe_type(entry) > OBJ_BLOB)1471goto give_up;1472unuse_pack(&w_curs);1473return;1474case OBJ_REF_DELTA:1475if(reuse_delta && !entry->preferred_base)1476 base_ref =use_pack(p, &w_curs,1477 entry->in_pack_offset + used, NULL);1478 entry->in_pack_header_size = used + the_hash_algo->rawsz;1479break;1480case OBJ_OFS_DELTA:1481 buf =use_pack(p, &w_curs,1482 entry->in_pack_offset + used, NULL);1483 used_0 =0;1484 c = buf[used_0++];1485 ofs = c &127;1486while(c &128) {1487 ofs +=1;1488if(!ofs ||MSB(ofs,7)) {1489error(_("delta base offset overflow in pack for%s"),1490oid_to_hex(&entry->idx.oid));1491goto give_up;1492}1493 c = buf[used_0++];1494 ofs = (ofs <<7) + (c &127);1495}1496 ofs = entry->in_pack_offset - ofs;1497if(ofs <=0|| ofs >= entry->in_pack_offset) {1498error(_("delta base offset out of bound for%s"),1499oid_to_hex(&entry->idx.oid));1500goto give_up;1501}1502if(reuse_delta && !entry->preferred_base) {1503struct revindex_entry *revidx;1504 revidx =find_pack_revindex(p, ofs);1505if(!revidx)1506goto give_up;1507 base_ref =nth_packed_object_sha1(p, revidx->nr);1508}1509 entry->in_pack_header_size = used + used_0;1510break;1511}15121513if(base_ref && (base_entry =packlist_find(&to_pack, base_ref, NULL))) {1514/*1515 * If base_ref was set above that means we wish to1516 * reuse delta data, and we even found that base1517 * in the list of objects we want to pack. Goodie!1518 *1519 * Depth value does not matter - find_deltas() will1520 * never consider reused delta as the base object to1521 * deltify other objects against, in order to avoid1522 * circular deltas.1523 */1524oe_set_type(entry, entry->in_pack_type);1525SET_SIZE(entry, in_pack_size);/* delta size */1526SET_DELTA(entry, base_entry);1527SET_DELTA_SIZE(entry, in_pack_size);1528 entry->delta_sibling_idx = base_entry->delta_child_idx;1529SET_DELTA_CHILD(base_entry, entry);1530unuse_pack(&w_curs);1531return;1532}15331534if(oe_type(entry)) {1535 off_t delta_pos;15361537/*1538 * This must be a delta and we already know what the1539 * final object type is. Let's extract the actual1540 * object size from the delta header.1541 */1542 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;1543 canonical_size =get_size_from_delta(p, &w_curs, delta_pos);1544if(canonical_size ==0)1545goto give_up;1546SET_SIZE(entry, canonical_size);1547unuse_pack(&w_curs);1548return;1549}15501551/*1552 * No choice but to fall back to the recursive delta walk1553 * with sha1_object_info() to find about the object type1554 * at this point...1555 */1556 give_up:1557unuse_pack(&w_curs);1558}15591560oe_set_type(entry,1561oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1562if(entry->type_valid) {1563SET_SIZE(entry, canonical_size);1564}else{1565/*1566 * Bad object type is checked in prepare_pack(). This is1567 * to permit a missing preferred base object to be ignored1568 * as a preferred base. Doing so can result in a larger1569 * pack file, but the transfer will still take place.1570 */1571}1572}15731574static intpack_offset_sort(const void*_a,const void*_b)1575{1576const struct object_entry *a = *(struct object_entry **)_a;1577const struct object_entry *b = *(struct object_entry **)_b;1578const struct packed_git *a_in_pack =IN_PACK(a);1579const struct packed_git *b_in_pack =IN_PACK(b);15801581/* avoid filesystem trashing with loose objects */1582if(!a_in_pack && !b_in_pack)1583returnoidcmp(&a->idx.oid, &b->idx.oid);15841585if(a_in_pack < b_in_pack)1586return-1;1587if(a_in_pack > b_in_pack)1588return1;1589return a->in_pack_offset < b->in_pack_offset ? -1:1590(a->in_pack_offset > b->in_pack_offset);1591}15921593/*1594 * Drop an on-disk delta we were planning to reuse. Naively, this would1595 * just involve blanking out the "delta" field, but we have to deal1596 * with some extra book-keeping:1597 *1598 * 1. Removing ourselves from the delta_sibling linked list.1599 *1600 * 2. Updating our size/type to the non-delta representation. These were1601 * either not recorded initially (size) or overwritten with the delta type1602 * (type) when check_object() decided to reuse the delta.1603 *1604 * 3. Resetting our delta depth, as we are now a base object.1605 */1606static voiddrop_reused_delta(struct object_entry *entry)1607{1608unsigned*idx = &to_pack.objects[entry->delta_idx -1].delta_child_idx;1609struct object_info oi = OBJECT_INFO_INIT;1610enum object_type type;1611unsigned long size;16121613while(*idx) {1614struct object_entry *oe = &to_pack.objects[*idx -1];16151616if(oe == entry)1617*idx = oe->delta_sibling_idx;1618else1619 idx = &oe->delta_sibling_idx;1620}1621SET_DELTA(entry, NULL);1622 entry->depth =0;16231624 oi.sizep = &size;1625 oi.typep = &type;1626if(packed_object_info(the_repository,IN_PACK(entry), entry->in_pack_offset, &oi) <0) {1627/*1628 * We failed to get the info from this pack for some reason;1629 * fall back to sha1_object_info, which may find another copy.1630 * And if that fails, the error will be recorded in oe_type(entry)1631 * and dealt with in prepare_pack().1632 */1633oe_set_type(entry,1634oid_object_info(the_repository, &entry->idx.oid, &size));1635}else{1636oe_set_type(entry, type);1637}1638SET_SIZE(entry, size);1639}16401641/*1642 * Follow the chain of deltas from this entry onward, throwing away any links1643 * that cause us to hit a cycle (as determined by the DFS state flags in1644 * the entries).1645 *1646 * We also detect too-long reused chains that would violate our --depth1647 * limit.1648 */1649static voidbreak_delta_chains(struct object_entry *entry)1650{1651/*1652 * The actual depth of each object we will write is stored as an int,1653 * as it cannot exceed our int "depth" limit. But before we break1654 * changes based no that limit, we may potentially go as deep as the1655 * number of objects, which is elsewhere bounded to a uint32_t.1656 */1657uint32_t total_depth;1658struct object_entry *cur, *next;16591660for(cur = entry, total_depth =0;1661 cur;1662 cur =DELTA(cur), total_depth++) {1663if(cur->dfs_state == DFS_DONE) {1664/*1665 * We've already seen this object and know it isn't1666 * part of a cycle. We do need to append its depth1667 * to our count.1668 */1669 total_depth += cur->depth;1670break;1671}16721673/*1674 * We break cycles before looping, so an ACTIVE state (or any1675 * other cruft which made its way into the state variable)1676 * is a bug.1677 */1678if(cur->dfs_state != DFS_NONE)1679BUG("confusing delta dfs state in first pass:%d",1680 cur->dfs_state);16811682/*1683 * Now we know this is the first time we've seen the object. If1684 * it's not a delta, we're done traversing, but we'll mark it1685 * done to save time on future traversals.1686 */1687if(!DELTA(cur)) {1688 cur->dfs_state = DFS_DONE;1689break;1690}16911692/*1693 * Mark ourselves as active and see if the next step causes1694 * us to cycle to another active object. It's important to do1695 * this _before_ we loop, because it impacts where we make the1696 * cut, and thus how our total_depth counter works.1697 * E.g., We may see a partial loop like:1698 *1699 * A -> B -> C -> D -> B1700 *1701 * Cutting B->C breaks the cycle. But now the depth of A is1702 * only 1, and our total_depth counter is at 3. The size of the1703 * error is always one less than the size of the cycle we1704 * broke. Commits C and D were "lost" from A's chain.1705 *1706 * If we instead cut D->B, then the depth of A is correct at 3.1707 * We keep all commits in the chain that we examined.1708 */1709 cur->dfs_state = DFS_ACTIVE;1710if(DELTA(cur)->dfs_state == DFS_ACTIVE) {1711drop_reused_delta(cur);1712 cur->dfs_state = DFS_DONE;1713break;1714}1715}17161717/*1718 * And now that we've gone all the way to the bottom of the chain, we1719 * need to clear the active flags and set the depth fields as1720 * appropriate. Unlike the loop above, which can quit when it drops a1721 * delta, we need to keep going to look for more depth cuts. So we need1722 * an extra "next" pointer to keep going after we reset cur->delta.1723 */1724for(cur = entry; cur; cur = next) {1725 next =DELTA(cur);17261727/*1728 * We should have a chain of zero or more ACTIVE states down to1729 * a final DONE. We can quit after the DONE, because either it1730 * has no bases, or we've already handled them in a previous1731 * call.1732 */1733if(cur->dfs_state == DFS_DONE)1734break;1735else if(cur->dfs_state != DFS_ACTIVE)1736BUG("confusing delta dfs state in second pass:%d",1737 cur->dfs_state);17381739/*1740 * If the total_depth is more than depth, then we need to snip1741 * the chain into two or more smaller chains that don't exceed1742 * the maximum depth. Most of the resulting chains will contain1743 * (depth + 1) entries (i.e., depth deltas plus one base), and1744 * the last chain (i.e., the one containing entry) will contain1745 * whatever entries are left over, namely1746 * (total_depth % (depth + 1)) of them.1747 *1748 * Since we are iterating towards decreasing depth, we need to1749 * decrement total_depth as we go, and we need to write to the1750 * entry what its final depth will be after all of the1751 * snipping. Since we're snipping into chains of length (depth1752 * + 1) entries, the final depth of an entry will be its1753 * original depth modulo (depth + 1). Any time we encounter an1754 * entry whose final depth is supposed to be zero, we snip it1755 * from its delta base, thereby making it so.1756 */1757 cur->depth = (total_depth--) % (depth +1);1758if(!cur->depth)1759drop_reused_delta(cur);17601761 cur->dfs_state = DFS_DONE;1762}1763}17641765static voidget_object_details(void)1766{1767uint32_t i;1768struct object_entry **sorted_by_offset;17691770if(progress)1771 progress_state =start_progress(_("Counting objects"),1772 to_pack.nr_objects);17731774 sorted_by_offset =xcalloc(to_pack.nr_objects,sizeof(struct object_entry *));1775for(i =0; i < to_pack.nr_objects; i++)1776 sorted_by_offset[i] = to_pack.objects + i;1777QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);17781779for(i =0; i < to_pack.nr_objects; i++) {1780struct object_entry *entry = sorted_by_offset[i];1781check_object(entry);1782if(entry->type_valid &&1783oe_size_greater_than(&to_pack, entry, big_file_threshold))1784 entry->no_try_delta =1;1785display_progress(progress_state, i +1);1786}1787stop_progress(&progress_state);17881789/*1790 * This must happen in a second pass, since we rely on the delta1791 * information for the whole list being completed.1792 */1793for(i =0; i < to_pack.nr_objects; i++)1794break_delta_chains(&to_pack.objects[i]);17951796free(sorted_by_offset);1797}17981799/*1800 * We search for deltas in a list sorted by type, by filename hash, and then1801 * by size, so that we see progressively smaller and smaller files.1802 * That's because we prefer deltas to be from the bigger file1803 * to the smaller -- deletes are potentially cheaper, but perhaps1804 * more importantly, the bigger file is likely the more recent1805 * one. The deepest deltas are therefore the oldest objects which are1806 * less susceptible to be accessed often.1807 */1808static inttype_size_sort(const void*_a,const void*_b)1809{1810const struct object_entry *a = *(struct object_entry **)_a;1811const struct object_entry *b = *(struct object_entry **)_b;1812enum object_type a_type =oe_type(a);1813enum object_type b_type =oe_type(b);1814unsigned long a_size =SIZE(a);1815unsigned long b_size =SIZE(b);18161817if(a_type > b_type)1818return-1;1819if(a_type < b_type)1820return1;1821if(a->hash > b->hash)1822return-1;1823if(a->hash < b->hash)1824return1;1825if(a->preferred_base > b->preferred_base)1826return-1;1827if(a->preferred_base < b->preferred_base)1828return1;1829if(a_size > b_size)1830return-1;1831if(a_size < b_size)1832return1;1833return a < b ? -1: (a > b);/* newest first */1834}18351836struct unpacked {1837struct object_entry *entry;1838void*data;1839struct delta_index *index;1840unsigned depth;1841};18421843static intdelta_cacheable(unsigned long src_size,unsigned long trg_size,1844unsigned long delta_size)1845{1846if(max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1847return0;18481849if(delta_size < cache_max_small_delta_size)1850return1;18511852/* cache delta, if objects are large enough compared to delta size */1853if((src_size >>20) + (trg_size >>21) > (delta_size >>10))1854return1;18551856return0;1857}18581859#ifndef NO_PTHREADS18601861/* Protect access to object database */1862static pthread_mutex_t read_mutex;1863#define read_lock() pthread_mutex_lock(&read_mutex)1864#define read_unlock() pthread_mutex_unlock(&read_mutex)18651866/* Protect delta_cache_size */1867static pthread_mutex_t cache_mutex;1868#define cache_lock() pthread_mutex_lock(&cache_mutex)1869#define cache_unlock() pthread_mutex_unlock(&cache_mutex)18701871/*1872 * Protect object list partitioning (e.g. struct thread_param) and1873 * progress_state1874 */1875static pthread_mutex_t progress_mutex;1876#define progress_lock() pthread_mutex_lock(&progress_mutex)1877#define progress_unlock() pthread_mutex_unlock(&progress_mutex)18781879/*1880 * Access to struct object_entry is unprotected since each thread owns1881 * a portion of the main object list. Just don't access object entries1882 * ahead in the list because they can be stolen and would need1883 * progress_mutex for protection.1884 */1885#else18861887#define read_lock() (void)01888#define read_unlock() (void)01889#define cache_lock() (void)01890#define cache_unlock() (void)01891#define progress_lock() (void)01892#define progress_unlock() (void)018931894#endif18951896/*1897 * Return the size of the object without doing any delta1898 * reconstruction (so non-deltas are true object sizes, but deltas1899 * return the size of the delta data).1900 */1901unsigned longoe_get_size_slow(struct packing_data *pack,1902const struct object_entry *e)1903{1904struct packed_git *p;1905struct pack_window *w_curs;1906unsigned char*buf;1907enum object_type type;1908unsigned long used, avail, size;19091910if(e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {1911read_lock();1912if(oid_object_info(the_repository, &e->idx.oid, &size) <0)1913die(_("unable to get size of%s"),1914oid_to_hex(&e->idx.oid));1915read_unlock();1916return size;1917}19181919 p =oe_in_pack(pack, e);1920if(!p)1921BUG("when e->type is a delta, it must belong to a pack");19221923read_lock();1924 w_curs = NULL;1925 buf =use_pack(p, &w_curs, e->in_pack_offset, &avail);1926 used =unpack_object_header_buffer(buf, avail, &type, &size);1927if(used ==0)1928die(_("unable to parse object header of%s"),1929oid_to_hex(&e->idx.oid));19301931unuse_pack(&w_curs);1932read_unlock();1933return size;1934}19351936static inttry_delta(struct unpacked *trg,struct unpacked *src,1937unsigned max_depth,unsigned long*mem_usage)1938{1939struct object_entry *trg_entry = trg->entry;1940struct object_entry *src_entry = src->entry;1941unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;1942unsigned ref_depth;1943enum object_type type;1944void*delta_buf;19451946/* Don't bother doing diffs between different types */1947if(oe_type(trg_entry) !=oe_type(src_entry))1948return-1;19491950/*1951 * We do not bother to try a delta that we discarded on an1952 * earlier try, but only when reusing delta data. Note that1953 * src_entry that is marked as the preferred_base should always1954 * be considered, as even if we produce a suboptimal delta against1955 * it, we will still save the transfer cost, as we already know1956 * the other side has it and we won't send src_entry at all.1957 */1958if(reuse_delta &&IN_PACK(trg_entry) &&1959IN_PACK(trg_entry) ==IN_PACK(src_entry) &&1960!src_entry->preferred_base &&1961 trg_entry->in_pack_type != OBJ_REF_DELTA &&1962 trg_entry->in_pack_type != OBJ_OFS_DELTA)1963return0;19641965/* Let's not bust the allowed depth. */1966if(src->depth >= max_depth)1967return0;19681969/* Now some size filtering heuristics. */1970 trg_size =SIZE(trg_entry);1971if(!DELTA(trg_entry)) {1972 max_size = trg_size/2- the_hash_algo->rawsz;1973 ref_depth =1;1974}else{1975 max_size =DELTA_SIZE(trg_entry);1976 ref_depth = trg->depth;1977}1978 max_size = (uint64_t)max_size * (max_depth - src->depth) /1979(max_depth - ref_depth +1);1980if(max_size ==0)1981return0;1982 src_size =SIZE(src_entry);1983 sizediff = src_size < trg_size ? trg_size - src_size :0;1984if(sizediff >= max_size)1985return0;1986if(trg_size < src_size /32)1987return0;19881989/* Load data if not already done */1990if(!trg->data) {1991read_lock();1992 trg->data =read_object_file(&trg_entry->idx.oid, &type, &sz);1993read_unlock();1994if(!trg->data)1995die(_("object%scannot be read"),1996oid_to_hex(&trg_entry->idx.oid));1997if(sz != trg_size)1998die(_("object%sinconsistent object length (%lu vs%lu)"),1999oid_to_hex(&trg_entry->idx.oid), sz,2000 trg_size);2001*mem_usage += sz;2002}2003if(!src->data) {2004read_lock();2005 src->data =read_object_file(&src_entry->idx.oid, &type, &sz);2006read_unlock();2007if(!src->data) {2008if(src_entry->preferred_base) {2009static int warned =0;2010if(!warned++)2011warning(_("object%scannot be read"),2012oid_to_hex(&src_entry->idx.oid));2013/*2014 * Those objects are not included in the2015 * resulting pack. Be resilient and ignore2016 * them if they can't be read, in case the2017 * pack could be created nevertheless.2018 */2019return0;2020}2021die(_("object%scannot be read"),2022oid_to_hex(&src_entry->idx.oid));2023}2024if(sz != src_size)2025die(_("object%sinconsistent object length (%lu vs%lu)"),2026oid_to_hex(&src_entry->idx.oid), sz,2027 src_size);2028*mem_usage += sz;2029}2030if(!src->index) {2031 src->index =create_delta_index(src->data, src_size);2032if(!src->index) {2033static int warned =0;2034if(!warned++)2035warning(_("suboptimal pack - out of memory"));2036return0;2037}2038*mem_usage +=sizeof_delta_index(src->index);2039}20402041 delta_buf =create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2042if(!delta_buf)2043return0;20442045if(DELTA(trg_entry)) {2046/* Prefer only shallower same-sized deltas. */2047if(delta_size ==DELTA_SIZE(trg_entry) &&2048 src->depth +1>= trg->depth) {2049free(delta_buf);2050return0;2051}2052}20532054/*2055 * Handle memory allocation outside of the cache2056 * accounting lock. Compiler will optimize the strangeness2057 * away when NO_PTHREADS is defined.2058 */2059free(trg_entry->delta_data);2060cache_lock();2061if(trg_entry->delta_data) {2062 delta_cache_size -=DELTA_SIZE(trg_entry);2063 trg_entry->delta_data = NULL;2064}2065if(delta_cacheable(src_size, trg_size, delta_size)) {2066 delta_cache_size += delta_size;2067cache_unlock();2068 trg_entry->delta_data =xrealloc(delta_buf, delta_size);2069}else{2070cache_unlock();2071free(delta_buf);2072}20732074SET_DELTA(trg_entry, src_entry);2075SET_DELTA_SIZE(trg_entry, delta_size);2076 trg->depth = src->depth +1;20772078return1;2079}20802081static unsigned intcheck_delta_limit(struct object_entry *me,unsigned int n)2082{2083struct object_entry *child =DELTA_CHILD(me);2084unsigned int m = n;2085while(child) {2086unsigned int c =check_delta_limit(child, n +1);2087if(m < c)2088 m = c;2089 child =DELTA_SIBLING(child);2090}2091return m;2092}20932094static unsigned longfree_unpacked(struct unpacked *n)2095{2096unsigned long freed_mem =sizeof_delta_index(n->index);2097free_delta_index(n->index);2098 n->index = NULL;2099if(n->data) {2100 freed_mem +=SIZE(n->entry);2101FREE_AND_NULL(n->data);2102}2103 n->entry = NULL;2104 n->depth =0;2105return freed_mem;2106}21072108static voidfind_deltas(struct object_entry **list,unsigned*list_size,2109int window,int depth,unsigned*processed)2110{2111uint32_t i, idx =0, count =0;2112struct unpacked *array;2113unsigned long mem_usage =0;21142115 array =xcalloc(window,sizeof(struct unpacked));21162117for(;;) {2118struct object_entry *entry;2119struct unpacked *n = array + idx;2120int j, max_depth, best_base = -1;21212122progress_lock();2123if(!*list_size) {2124progress_unlock();2125break;2126}2127 entry = *list++;2128(*list_size)--;2129if(!entry->preferred_base) {2130(*processed)++;2131display_progress(progress_state, *processed);2132}2133progress_unlock();21342135 mem_usage -=free_unpacked(n);2136 n->entry = entry;21372138while(window_memory_limit &&2139 mem_usage > window_memory_limit &&2140 count >1) {2141uint32_t tail = (idx + window - count) % window;2142 mem_usage -=free_unpacked(array + tail);2143 count--;2144}21452146/* We do not compute delta to *create* objects we are not2147 * going to pack.2148 */2149if(entry->preferred_base)2150goto next;21512152/*2153 * If the current object is at pack edge, take the depth the2154 * objects that depend on the current object into account2155 * otherwise they would become too deep.2156 */2157 max_depth = depth;2158if(DELTA_CHILD(entry)) {2159 max_depth -=check_delta_limit(entry,0);2160if(max_depth <=0)2161goto next;2162}21632164 j = window;2165while(--j >0) {2166int ret;2167uint32_t other_idx = idx + j;2168struct unpacked *m;2169if(other_idx >= window)2170 other_idx -= window;2171 m = array + other_idx;2172if(!m->entry)2173break;2174 ret =try_delta(n, m, max_depth, &mem_usage);2175if(ret <0)2176break;2177else if(ret >0)2178 best_base = other_idx;2179}21802181/*2182 * If we decided to cache the delta data, then it is best2183 * to compress it right away. First because we have to do2184 * it anyway, and doing it here while we're threaded will2185 * save a lot of time in the non threaded write phase,2186 * as well as allow for caching more deltas within2187 * the same cache size limit.2188 * ...2189 * But only if not writing to stdout, since in that case2190 * the network is most likely throttling writes anyway,2191 * and therefore it is best to go to the write phase ASAP2192 * instead, as we can afford spending more time compressing2193 * between writes at that moment.2194 */2195if(entry->delta_data && !pack_to_stdout) {2196unsigned long size;21972198 size =do_compress(&entry->delta_data,DELTA_SIZE(entry));2199if(size < (1U<< OE_Z_DELTA_BITS)) {2200 entry->z_delta_size = size;2201cache_lock();2202 delta_cache_size -=DELTA_SIZE(entry);2203 delta_cache_size += entry->z_delta_size;2204cache_unlock();2205}else{2206FREE_AND_NULL(entry->delta_data);2207 entry->z_delta_size =0;2208}2209}22102211/* if we made n a delta, and if n is already at max2212 * depth, leaving it in the window is pointless. we2213 * should evict it first.2214 */2215if(DELTA(entry) && max_depth <= n->depth)2216continue;22172218/*2219 * Move the best delta base up in the window, after the2220 * currently deltified object, to keep it longer. It will2221 * be the first base object to be attempted next.2222 */2223if(DELTA(entry)) {2224struct unpacked swap = array[best_base];2225int dist = (window + idx - best_base) % window;2226int dst = best_base;2227while(dist--) {2228int src = (dst +1) % window;2229 array[dst] = array[src];2230 dst = src;2231}2232 array[dst] = swap;2233}22342235 next:2236 idx++;2237if(count +1< window)2238 count++;2239if(idx >= window)2240 idx =0;2241}22422243for(i =0; i < window; ++i) {2244free_delta_index(array[i].index);2245free(array[i].data);2246}2247free(array);2248}22492250#ifndef NO_PTHREADS22512252static voidtry_to_free_from_threads(size_t size)2253{2254read_lock();2255release_pack_memory(size);2256read_unlock();2257}22582259static try_to_free_t old_try_to_free_routine;22602261/*2262 * The main object list is split into smaller lists, each is handed to2263 * one worker.2264 *2265 * The main thread waits on the condition that (at least) one of the workers2266 * has stopped working (which is indicated in the .working member of2267 * struct thread_params).2268 *2269 * When a work thread has completed its work, it sets .working to 0 and2270 * signals the main thread and waits on the condition that .data_ready2271 * becomes 1.2272 *2273 * The main thread steals half of the work from the worker that has2274 * most work left to hand it to the idle worker.2275 */22762277struct thread_params {2278 pthread_t thread;2279struct object_entry **list;2280unsigned list_size;2281unsigned remaining;2282int window;2283int depth;2284int working;2285int data_ready;2286 pthread_mutex_t mutex;2287 pthread_cond_t cond;2288unsigned*processed;2289};22902291static pthread_cond_t progress_cond;22922293/*2294 * Mutex and conditional variable can't be statically-initialized on Windows.2295 */2296static voidinit_threaded_search(void)2297{2298init_recursive_mutex(&read_mutex);2299pthread_mutex_init(&cache_mutex, NULL);2300pthread_mutex_init(&progress_mutex, NULL);2301pthread_cond_init(&progress_cond, NULL);2302pthread_mutex_init(&to_pack.lock, NULL);2303 old_try_to_free_routine =set_try_to_free_routine(try_to_free_from_threads);2304}23052306static voidcleanup_threaded_search(void)2307{2308set_try_to_free_routine(old_try_to_free_routine);2309pthread_cond_destroy(&progress_cond);2310pthread_mutex_destroy(&read_mutex);2311pthread_mutex_destroy(&cache_mutex);2312pthread_mutex_destroy(&progress_mutex);2313}23142315static void*threaded_find_deltas(void*arg)2316{2317struct thread_params *me = arg;23182319progress_lock();2320while(me->remaining) {2321progress_unlock();23222323find_deltas(me->list, &me->remaining,2324 me->window, me->depth, me->processed);23252326progress_lock();2327 me->working =0;2328pthread_cond_signal(&progress_cond);2329progress_unlock();23302331/*2332 * We must not set ->data_ready before we wait on the2333 * condition because the main thread may have set it to 12334 * before we get here. In order to be sure that new2335 * work is available if we see 1 in ->data_ready, it2336 * was initialized to 0 before this thread was spawned2337 * and we reset it to 0 right away.2338 */2339pthread_mutex_lock(&me->mutex);2340while(!me->data_ready)2341pthread_cond_wait(&me->cond, &me->mutex);2342 me->data_ready =0;2343pthread_mutex_unlock(&me->mutex);23442345progress_lock();2346}2347progress_unlock();2348/* leave ->working 1 so that this doesn't get more work assigned */2349return NULL;2350}23512352static voidll_find_deltas(struct object_entry **list,unsigned list_size,2353int window,int depth,unsigned*processed)2354{2355struct thread_params *p;2356int i, ret, active_threads =0;23572358init_threaded_search();23592360if(delta_search_threads <=1) {2361find_deltas(list, &list_size, window, depth, processed);2362cleanup_threaded_search();2363return;2364}2365if(progress > pack_to_stdout)2366fprintf_ln(stderr,_("Delta compression using up to%dthreads"),2367 delta_search_threads);2368 p =xcalloc(delta_search_threads,sizeof(*p));23692370/* Partition the work amongst work threads. */2371for(i =0; i < delta_search_threads; i++) {2372unsigned sub_size = list_size / (delta_search_threads - i);23732374/* don't use too small segments or no deltas will be found */2375if(sub_size <2*window && i+1< delta_search_threads)2376 sub_size =0;23772378 p[i].window = window;2379 p[i].depth = depth;2380 p[i].processed = processed;2381 p[i].working =1;2382 p[i].data_ready =0;23832384/* try to split chunks on "path" boundaries */2385while(sub_size && sub_size < list_size &&2386 list[sub_size]->hash &&2387 list[sub_size]->hash == list[sub_size-1]->hash)2388 sub_size++;23892390 p[i].list = list;2391 p[i].list_size = sub_size;2392 p[i].remaining = sub_size;23932394 list += sub_size;2395 list_size -= sub_size;2396}23972398/* Start work threads. */2399for(i =0; i < delta_search_threads; i++) {2400if(!p[i].list_size)2401continue;2402pthread_mutex_init(&p[i].mutex, NULL);2403pthread_cond_init(&p[i].cond, NULL);2404 ret =pthread_create(&p[i].thread, NULL,2405 threaded_find_deltas, &p[i]);2406if(ret)2407die(_("unable to create thread:%s"),strerror(ret));2408 active_threads++;2409}24102411/*2412 * Now let's wait for work completion. Each time a thread is done2413 * with its work, we steal half of the remaining work from the2414 * thread with the largest number of unprocessed objects and give2415 * it to that newly idle thread. This ensure good load balancing2416 * until the remaining object list segments are simply too short2417 * to be worth splitting anymore.2418 */2419while(active_threads) {2420struct thread_params *target = NULL;2421struct thread_params *victim = NULL;2422unsigned sub_size =0;24232424progress_lock();2425for(;;) {2426for(i =0; !target && i < delta_search_threads; i++)2427if(!p[i].working)2428 target = &p[i];2429if(target)2430break;2431pthread_cond_wait(&progress_cond, &progress_mutex);2432}24332434for(i =0; i < delta_search_threads; i++)2435if(p[i].remaining >2*window &&2436(!victim || victim->remaining < p[i].remaining))2437 victim = &p[i];2438if(victim) {2439 sub_size = victim->remaining /2;2440 list = victim->list + victim->list_size - sub_size;2441while(sub_size && list[0]->hash &&2442 list[0]->hash == list[-1]->hash) {2443 list++;2444 sub_size--;2445}2446if(!sub_size) {2447/*2448 * It is possible for some "paths" to have2449 * so many objects that no hash boundary2450 * might be found. Let's just steal the2451 * exact half in that case.2452 */2453 sub_size = victim->remaining /2;2454 list -= sub_size;2455}2456 target->list = list;2457 victim->list_size -= sub_size;2458 victim->remaining -= sub_size;2459}2460 target->list_size = sub_size;2461 target->remaining = sub_size;2462 target->working =1;2463progress_unlock();24642465pthread_mutex_lock(&target->mutex);2466 target->data_ready =1;2467pthread_cond_signal(&target->cond);2468pthread_mutex_unlock(&target->mutex);24692470if(!sub_size) {2471pthread_join(target->thread, NULL);2472pthread_cond_destroy(&target->cond);2473pthread_mutex_destroy(&target->mutex);2474 active_threads--;2475}2476}2477cleanup_threaded_search();2478free(p);2479}24802481#else2482#define ll_find_deltas(l, s, w, d, p) find_deltas(l, &s, w, d, p)2483#endif24842485static voidadd_tag_chain(const struct object_id *oid)2486{2487struct tag *tag;24882489/*2490 * We catch duplicates already in add_object_entry(), but we'd2491 * prefer to do this extra check to avoid having to parse the2492 * tag at all if we already know that it's being packed (e.g., if2493 * it was included via bitmaps, we would not have parsed it2494 * previously).2495 */2496if(packlist_find(&to_pack, oid->hash, NULL))2497return;24982499 tag =lookup_tag(the_repository, oid);2500while(1) {2501if(!tag ||parse_tag(tag) || !tag->tagged)2502die(_("unable to pack objects reachable from tag%s"),2503oid_to_hex(oid));25042505add_object_entry(&tag->object.oid, OBJ_TAG, NULL,0);25062507if(tag->tagged->type != OBJ_TAG)2508return;25092510 tag = (struct tag *)tag->tagged;2511}2512}25132514static intadd_ref_tag(const char*path,const struct object_id *oid,int flag,void*cb_data)2515{2516struct object_id peeled;25172518if(starts_with(path,"refs/tags/") &&/* is a tag? */2519!peel_ref(path, &peeled) &&/* peelable? */2520packlist_find(&to_pack, peeled.hash, NULL))/* object packed? */2521add_tag_chain(oid);2522return0;2523}25242525static voidprepare_pack(int window,int depth)2526{2527struct object_entry **delta_list;2528uint32_t i, nr_deltas;2529unsigned n;25302531get_object_details();25322533/*2534 * If we're locally repacking then we need to be doubly careful2535 * from now on in order to make sure no stealth corruption gets2536 * propagated to the new pack. Clients receiving streamed packs2537 * should validate everything they get anyway so no need to incur2538 * the additional cost here in that case.2539 */2540if(!pack_to_stdout)2541 do_check_packed_object_crc =1;25422543if(!to_pack.nr_objects || !window || !depth)2544return;25452546ALLOC_ARRAY(delta_list, to_pack.nr_objects);2547 nr_deltas = n =0;25482549for(i =0; i < to_pack.nr_objects; i++) {2550struct object_entry *entry = to_pack.objects + i;25512552if(DELTA(entry))2553/* This happens if we decided to reuse existing2554 * delta from a pack. "reuse_delta &&" is implied.2555 */2556continue;25572558if(!entry->type_valid ||2559oe_size_less_than(&to_pack, entry,50))2560continue;25612562if(entry->no_try_delta)2563continue;25642565if(!entry->preferred_base) {2566 nr_deltas++;2567if(oe_type(entry) <0)2568die(_("unable to get type of object%s"),2569oid_to_hex(&entry->idx.oid));2570}else{2571if(oe_type(entry) <0) {2572/*2573 * This object is not found, but we2574 * don't have to include it anyway.2575 */2576continue;2577}2578}25792580 delta_list[n++] = entry;2581}25822583if(nr_deltas && n >1) {2584unsigned nr_done =0;2585if(progress)2586 progress_state =start_progress(_("Compressing objects"),2587 nr_deltas);2588QSORT(delta_list, n, type_size_sort);2589ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2590stop_progress(&progress_state);2591if(nr_done != nr_deltas)2592die(_("inconsistency with delta count"));2593}2594free(delta_list);2595}25962597static intgit_pack_config(const char*k,const char*v,void*cb)2598{2599if(!strcmp(k,"pack.window")) {2600 window =git_config_int(k, v);2601return0;2602}2603if(!strcmp(k,"pack.windowmemory")) {2604 window_memory_limit =git_config_ulong(k, v);2605return0;2606}2607if(!strcmp(k,"pack.depth")) {2608 depth =git_config_int(k, v);2609return0;2610}2611if(!strcmp(k,"pack.deltacachesize")) {2612 max_delta_cache_size =git_config_int(k, v);2613return0;2614}2615if(!strcmp(k,"pack.deltacachelimit")) {2616 cache_max_small_delta_size =git_config_int(k, v);2617return0;2618}2619if(!strcmp(k,"pack.writebitmaphashcache")) {2620if(git_config_bool(k, v))2621 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2622else2623 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2624}2625if(!strcmp(k,"pack.usebitmaps")) {2626 use_bitmap_index_default =git_config_bool(k, v);2627return0;2628}2629if(!strcmp(k,"pack.threads")) {2630 delta_search_threads =git_config_int(k, v);2631if(delta_search_threads <0)2632die(_("invalid number of threads specified (%d)"),2633 delta_search_threads);2634#ifdef NO_PTHREADS2635if(delta_search_threads !=1) {2636warning(_("no threads support, ignoring%s"), k);2637 delta_search_threads =0;2638}2639#endif2640return0;2641}2642if(!strcmp(k,"pack.indexversion")) {2643 pack_idx_opts.version =git_config_int(k, v);2644if(pack_idx_opts.version >2)2645die(_("bad pack.indexversion=%"PRIu32),2646 pack_idx_opts.version);2647return0;2648}2649returngit_default_config(k, v, cb);2650}26512652static voidread_object_list_from_stdin(void)2653{2654char line[GIT_MAX_HEXSZ +1+ PATH_MAX +2];2655struct object_id oid;2656const char*p;26572658for(;;) {2659if(!fgets(line,sizeof(line), stdin)) {2660if(feof(stdin))2661break;2662if(!ferror(stdin))2663die("BUG: fgets returned NULL, not EOF, not error!");2664if(errno != EINTR)2665die_errno("fgets");2666clearerr(stdin);2667continue;2668}2669if(line[0] =='-') {2670if(get_oid_hex(line+1, &oid))2671die(_("expected edge object ID, got garbage:\n%s"),2672 line);2673add_preferred_base(&oid);2674continue;2675}2676if(parse_oid_hex(line, &oid, &p))2677die(_("expected object ID, got garbage:\n%s"), line);26782679add_preferred_base_object(p +1);2680add_object_entry(&oid, OBJ_NONE, p +1,0);2681}2682}26832684/* Remember to update object flag allocation in object.h */2685#define OBJECT_ADDED (1u<<20)26862687static voidshow_commit(struct commit *commit,void*data)2688{2689add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL,0);2690 commit->object.flags |= OBJECT_ADDED;26912692if(write_bitmap_index)2693index_commit_for_bitmap(commit);2694}26952696static voidshow_object(struct object *obj,const char*name,void*data)2697{2698add_preferred_base_object(name);2699add_object_entry(&obj->oid, obj->type, name,0);2700 obj->flags |= OBJECT_ADDED;2701}27022703static voidshow_object__ma_allow_any(struct object *obj,const char*name,void*data)2704{2705assert(arg_missing_action == MA_ALLOW_ANY);27062707/*2708 * Quietly ignore ALL missing objects. This avoids problems with2709 * staging them now and getting an odd error later.2710 */2711if(!has_object_file(&obj->oid))2712return;27132714show_object(obj, name, data);2715}27162717static voidshow_object__ma_allow_promisor(struct object *obj,const char*name,void*data)2718{2719assert(arg_missing_action == MA_ALLOW_PROMISOR);27202721/*2722 * Quietly ignore EXPECTED missing objects. This avoids problems with2723 * staging them now and getting an odd error later.2724 */2725if(!has_object_file(&obj->oid) &&is_promisor_object(&obj->oid))2726return;27272728show_object(obj, name, data);2729}27302731static intoption_parse_missing_action(const struct option *opt,2732const char*arg,int unset)2733{2734assert(arg);2735assert(!unset);27362737if(!strcmp(arg,"error")) {2738 arg_missing_action = MA_ERROR;2739 fn_show_object = show_object;2740return0;2741}27422743if(!strcmp(arg,"allow-any")) {2744 arg_missing_action = MA_ALLOW_ANY;2745 fetch_if_missing =0;2746 fn_show_object = show_object__ma_allow_any;2747return0;2748}27492750if(!strcmp(arg,"allow-promisor")) {2751 arg_missing_action = MA_ALLOW_PROMISOR;2752 fetch_if_missing =0;2753 fn_show_object = show_object__ma_allow_promisor;2754return0;2755}27562757die(_("invalid value for --missing"));2758return0;2759}27602761static voidshow_edge(struct commit *commit)2762{2763add_preferred_base(&commit->object.oid);2764}27652766struct in_pack_object {2767 off_t offset;2768struct object *object;2769};27702771struct in_pack {2772unsigned int alloc;2773unsigned int nr;2774struct in_pack_object *array;2775};27762777static voidmark_in_pack_object(struct object *object,struct packed_git *p,struct in_pack *in_pack)2778{2779 in_pack->array[in_pack->nr].offset =find_pack_entry_one(object->oid.hash, p);2780 in_pack->array[in_pack->nr].object = object;2781 in_pack->nr++;2782}27832784/*2785 * Compare the objects in the offset order, in order to emulate the2786 * "git rev-list --objects" output that produced the pack originally.2787 */2788static intofscmp(const void*a_,const void*b_)2789{2790struct in_pack_object *a = (struct in_pack_object *)a_;2791struct in_pack_object *b = (struct in_pack_object *)b_;27922793if(a->offset < b->offset)2794return-1;2795else if(a->offset > b->offset)2796return1;2797else2798returnoidcmp(&a->object->oid, &b->object->oid);2799}28002801static voidadd_objects_in_unpacked_packs(struct rev_info *revs)2802{2803struct packed_git *p;2804struct in_pack in_pack;2805uint32_t i;28062807memset(&in_pack,0,sizeof(in_pack));28082809for(p =get_packed_git(the_repository); p; p = p->next) {2810struct object_id oid;2811struct object *o;28122813if(!p->pack_local || p->pack_keep || p->pack_keep_in_core)2814continue;2815if(open_pack_index(p))2816die(_("cannot open pack index"));28172818ALLOC_GROW(in_pack.array,2819 in_pack.nr + p->num_objects,2820 in_pack.alloc);28212822for(i =0; i < p->num_objects; i++) {2823nth_packed_object_oid(&oid, p, i);2824 o =lookup_unknown_object(oid.hash);2825if(!(o->flags & OBJECT_ADDED))2826mark_in_pack_object(o, p, &in_pack);2827 o->flags |= OBJECT_ADDED;2828}2829}28302831if(in_pack.nr) {2832QSORT(in_pack.array, in_pack.nr, ofscmp);2833for(i =0; i < in_pack.nr; i++) {2834struct object *o = in_pack.array[i].object;2835add_object_entry(&o->oid, o->type,"",0);2836}2837}2838free(in_pack.array);2839}28402841static intadd_loose_object(const struct object_id *oid,const char*path,2842void*data)2843{2844enum object_type type =oid_object_info(the_repository, oid, NULL);28452846if(type <0) {2847warning(_("loose object at%scould not be examined"), path);2848return0;2849}28502851add_object_entry(oid, type,"",0);2852return0;2853}28542855/*2856 * We actually don't even have to worry about reachability here.2857 * add_object_entry will weed out duplicates, so we just add every2858 * loose object we find.2859 */2860static voidadd_unreachable_loose_objects(void)2861{2862for_each_loose_file_in_objdir(get_object_directory(),2863 add_loose_object,2864 NULL, NULL, NULL);2865}28662867static inthas_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2868{2869static struct packed_git *last_found = (void*)1;2870struct packed_git *p;28712872 p = (last_found != (void*)1) ? last_found :2873get_packed_git(the_repository);28742875while(p) {2876if((!p->pack_local || p->pack_keep ||2877 p->pack_keep_in_core) &&2878find_pack_entry_one(oid->hash, p)) {2879 last_found = p;2880return1;2881}2882if(p == last_found)2883 p =get_packed_git(the_repository);2884else2885 p = p->next;2886if(p == last_found)2887 p = p->next;2888}2889return0;2890}28912892/*2893 * Store a list of sha1s that are should not be discarded2894 * because they are either written too recently, or are2895 * reachable from another object that was.2896 *2897 * This is filled by get_object_list.2898 */2899static struct oid_array recent_objects;29002901static intloosened_object_can_be_discarded(const struct object_id *oid,2902 timestamp_t mtime)2903{2904if(!unpack_unreachable_expiration)2905return0;2906if(mtime > unpack_unreachable_expiration)2907return0;2908if(oid_array_lookup(&recent_objects, oid) >=0)2909return0;2910return1;2911}29122913static voidloosen_unused_packed_objects(struct rev_info *revs)2914{2915struct packed_git *p;2916uint32_t i;2917struct object_id oid;29182919for(p =get_packed_git(the_repository); p; p = p->next) {2920if(!p->pack_local || p->pack_keep || p->pack_keep_in_core)2921continue;29222923if(open_pack_index(p))2924die(_("cannot open pack index"));29252926for(i =0; i < p->num_objects; i++) {2927nth_packed_object_oid(&oid, p, i);2928if(!packlist_find(&to_pack, oid.hash, NULL) &&2929!has_sha1_pack_kept_or_nonlocal(&oid) &&2930!loosened_object_can_be_discarded(&oid, p->mtime))2931if(force_object_loose(&oid, p->mtime))2932die(_("unable to force loose object"));2933}2934}2935}29362937/*2938 * This tracks any options which pack-reuse code expects to be on, or which a2939 * reader of the pack might not understand, and which would therefore prevent2940 * blind reuse of what we have on disk.2941 */2942static intpack_options_allow_reuse(void)2943{2944return pack_to_stdout &&2945 allow_ofs_delta &&2946!ignore_packed_keep_on_disk &&2947!ignore_packed_keep_in_core &&2948(!local || !have_non_local_packs) &&2949!incremental;2950}29512952static intget_object_list_from_bitmap(struct rev_info *revs)2953{2954struct bitmap_index *bitmap_git;2955if(!(bitmap_git =prepare_bitmap_walk(revs)))2956return-1;29572958if(pack_options_allow_reuse() &&2959!reuse_partial_packfile_from_bitmap(2960 bitmap_git,2961&reuse_packfile,2962&reuse_packfile_objects,2963&reuse_packfile_offset)) {2964assert(reuse_packfile_objects);2965 nr_result += reuse_packfile_objects;2966display_progress(progress_state, nr_result);2967}29682969traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);2970free_bitmap_index(bitmap_git);2971return0;2972}29732974static voidrecord_recent_object(struct object *obj,2975const char*name,2976void*data)2977{2978oid_array_append(&recent_objects, &obj->oid);2979}29802981static voidrecord_recent_commit(struct commit *commit,void*data)2982{2983oid_array_append(&recent_objects, &commit->object.oid);2984}29852986static voidget_object_list(int ac,const char**av)2987{2988struct rev_info revs;2989char line[1000];2990int flags =0;29912992init_revisions(&revs, NULL);2993 save_commit_buffer =0;2994setup_revisions(ac, av, &revs, NULL);29952996/* make sure shallows are read */2997is_repository_shallow(the_repository);29982999while(fgets(line,sizeof(line), stdin) != NULL) {3000int len =strlen(line);3001if(len && line[len -1] =='\n')3002 line[--len] =0;3003if(!len)3004break;3005if(*line =='-') {3006if(!strcmp(line,"--not")) {3007 flags ^= UNINTERESTING;3008 write_bitmap_index =0;3009continue;3010}3011if(starts_with(line,"--shallow ")) {3012struct object_id oid;3013if(get_oid_hex(line +10, &oid))3014die("not an SHA-1 '%s'", line +10);3015register_shallow(the_repository, &oid);3016 use_bitmap_index =0;3017continue;3018}3019die(_("not a rev '%s'"), line);3020}3021if(handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3022die(_("bad revision '%s'"), line);3023}30243025if(use_bitmap_index && !get_object_list_from_bitmap(&revs))3026return;30273028if(prepare_revision_walk(&revs))3029die(_("revision walk setup failed"));3030mark_edges_uninteresting(&revs, show_edge);30313032if(!fn_show_object)3033 fn_show_object = show_object;3034traverse_commit_list_filtered(&filter_options, &revs,3035 show_commit, fn_show_object, NULL,3036 NULL);30373038if(unpack_unreachable_expiration) {3039 revs.ignore_missing_links =1;3040if(add_unseen_recent_objects_to_traversal(&revs,3041 unpack_unreachable_expiration))3042die(_("unable to add recent objects"));3043if(prepare_revision_walk(&revs))3044die(_("revision walk setup failed"));3045traverse_commit_list(&revs, record_recent_commit,3046 record_recent_object, NULL);3047}30483049if(keep_unreachable)3050add_objects_in_unpacked_packs(&revs);3051if(pack_loose_unreachable)3052add_unreachable_loose_objects();3053if(unpack_unreachable)3054loosen_unused_packed_objects(&revs);30553056oid_array_clear(&recent_objects);3057}30583059static voidadd_extra_kept_packs(const struct string_list *names)3060{3061struct packed_git *p;30623063if(!names->nr)3064return;30653066for(p =get_packed_git(the_repository); p; p = p->next) {3067const char*name =basename(p->pack_name);3068int i;30693070if(!p->pack_local)3071continue;30723073for(i =0; i < names->nr; i++)3074if(!fspathcmp(name, names->items[i].string))3075break;30763077if(i < names->nr) {3078 p->pack_keep_in_core =1;3079 ignore_packed_keep_in_core =1;3080continue;3081}3082}3083}30843085static intoption_parse_index_version(const struct option *opt,3086const char*arg,int unset)3087{3088char*c;3089const char*val = arg;3090 pack_idx_opts.version =strtoul(val, &c,10);3091if(pack_idx_opts.version >2)3092die(_("unsupported index version%s"), val);3093if(*c ==','&& c[1])3094 pack_idx_opts.off32_limit =strtoul(c+1, &c,0);3095if(*c || pack_idx_opts.off32_limit &0x80000000)3096die(_("bad index version '%s'"), val);3097return0;3098}30993100static intoption_parse_unpack_unreachable(const struct option *opt,3101const char*arg,int unset)3102{3103if(unset) {3104 unpack_unreachable =0;3105 unpack_unreachable_expiration =0;3106}3107else{3108 unpack_unreachable =1;3109if(arg)3110 unpack_unreachable_expiration =approxidate(arg);3111}3112return0;3113}31143115intcmd_pack_objects(int argc,const char**argv,const char*prefix)3116{3117int use_internal_rev_list =0;3118int thin =0;3119int shallow =0;3120int all_progress_implied =0;3121struct argv_array rp = ARGV_ARRAY_INIT;3122int rev_list_unpacked =0, rev_list_all =0, rev_list_reflog =0;3123int rev_list_index =0;3124struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3125struct option pack_objects_options[] = {3126OPT_SET_INT('q',"quiet", &progress,3127N_("do not show progress meter"),0),3128OPT_SET_INT(0,"progress", &progress,3129N_("show progress meter"),1),3130OPT_SET_INT(0,"all-progress", &progress,3131N_("show progress meter during object writing phase"),2),3132OPT_BOOL(0,"all-progress-implied",3133&all_progress_implied,3134N_("similar to --all-progress when progress meter is shown")),3135{ OPTION_CALLBACK,0,"index-version", NULL,N_("<version>[,<offset>]"),3136N_("write the pack index file in the specified idx format version"),31370, option_parse_index_version },3138OPT_MAGNITUDE(0,"max-pack-size", &pack_size_limit,3139N_("maximum size of each output pack file")),3140OPT_BOOL(0,"local", &local,3141N_("ignore borrowed objects from alternate object store")),3142OPT_BOOL(0,"incremental", &incremental,3143N_("ignore packed objects")),3144OPT_INTEGER(0,"window", &window,3145N_("limit pack window by objects")),3146OPT_MAGNITUDE(0,"window-memory", &window_memory_limit,3147N_("limit pack window by memory in addition to object limit")),3148OPT_INTEGER(0,"depth", &depth,3149N_("maximum length of delta chain allowed in the resulting pack")),3150OPT_BOOL(0,"reuse-delta", &reuse_delta,3151N_("reuse existing deltas")),3152OPT_BOOL(0,"reuse-object", &reuse_object,3153N_("reuse existing objects")),3154OPT_BOOL(0,"delta-base-offset", &allow_ofs_delta,3155N_("use OFS_DELTA objects")),3156OPT_INTEGER(0,"threads", &delta_search_threads,3157N_("use threads when searching for best delta matches")),3158OPT_BOOL(0,"non-empty", &non_empty,3159N_("do not create an empty pack output")),3160OPT_BOOL(0,"revs", &use_internal_rev_list,3161N_("read revision arguments from standard input")),3162OPT_SET_INT_F(0,"unpacked", &rev_list_unpacked,3163N_("limit the objects to those that are not yet packed"),31641, PARSE_OPT_NONEG),3165OPT_SET_INT_F(0,"all", &rev_list_all,3166N_("include objects reachable from any reference"),31671, PARSE_OPT_NONEG),3168OPT_SET_INT_F(0,"reflog", &rev_list_reflog,3169N_("include objects referred by reflog entries"),31701, PARSE_OPT_NONEG),3171OPT_SET_INT_F(0,"indexed-objects", &rev_list_index,3172N_("include objects referred to by the index"),31731, PARSE_OPT_NONEG),3174OPT_BOOL(0,"stdout", &pack_to_stdout,3175N_("output pack to stdout")),3176OPT_BOOL(0,"include-tag", &include_tag,3177N_("include tag objects that refer to objects to be packed")),3178OPT_BOOL(0,"keep-unreachable", &keep_unreachable,3179N_("keep unreachable objects")),3180OPT_BOOL(0,"pack-loose-unreachable", &pack_loose_unreachable,3181N_("pack loose unreachable objects")),3182{ OPTION_CALLBACK,0,"unpack-unreachable", NULL,N_("time"),3183N_("unpack unreachable objects newer than <time>"),3184 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3185OPT_BOOL(0,"thin", &thin,3186N_("create thin packs")),3187OPT_BOOL(0,"shallow", &shallow,3188N_("create packs suitable for shallow fetches")),3189OPT_BOOL(0,"honor-pack-keep", &ignore_packed_keep_on_disk,3190N_("ignore packs that have companion .keep file")),3191OPT_STRING_LIST(0,"keep-pack", &keep_pack_list,N_("name"),3192N_("ignore this pack")),3193OPT_INTEGER(0,"compression", &pack_compression_level,3194N_("pack compression level")),3195OPT_SET_INT(0,"keep-true-parents", &grafts_replace_parents,3196N_("do not hide commits by grafts"),0),3197OPT_BOOL(0,"use-bitmap-index", &use_bitmap_index,3198N_("use a bitmap index if available to speed up counting objects")),3199OPT_BOOL(0,"write-bitmap-index", &write_bitmap_index,3200N_("write a bitmap index together with the pack index")),3201OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3202{ OPTION_CALLBACK,0,"missing", NULL,N_("action"),3203N_("handling for missing objects"), PARSE_OPT_NONEG,3204 option_parse_missing_action },3205OPT_BOOL(0,"exclude-promisor-objects", &exclude_promisor_objects,3206N_("do not pack objects in promisor packfiles")),3207OPT_END(),3208};32093210if(DFS_NUM_STATES > (1<< OE_DFS_STATE_BITS))3211BUG("too many dfs states, increase OE_DFS_STATE_BITS");32123213 read_replace_refs =0;32143215reset_pack_idx_option(&pack_idx_opts);3216git_config(git_pack_config, NULL);32173218 progress =isatty(2);3219 argc =parse_options(argc, argv, prefix, pack_objects_options,3220 pack_usage,0);32213222if(argc) {3223 base_name = argv[0];3224 argc--;3225}3226if(pack_to_stdout != !base_name || argc)3227usage_with_options(pack_usage, pack_objects_options);32283229if(depth >= (1<< OE_DEPTH_BITS)) {3230warning(_("delta chain depth%dis too deep, forcing%d"),3231 depth, (1<< OE_DEPTH_BITS) -1);3232 depth = (1<< OE_DEPTH_BITS) -1;3233}3234if(cache_max_small_delta_size >= (1U<< OE_Z_DELTA_BITS)) {3235warning(_("pack.deltaCacheLimit is too high, forcing%d"),3236(1U<< OE_Z_DELTA_BITS) -1);3237 cache_max_small_delta_size = (1U<< OE_Z_DELTA_BITS) -1;3238}32393240argv_array_push(&rp,"pack-objects");3241if(thin) {3242 use_internal_rev_list =1;3243argv_array_push(&rp, shallow3244?"--objects-edge-aggressive"3245:"--objects-edge");3246}else3247argv_array_push(&rp,"--objects");32483249if(rev_list_all) {3250 use_internal_rev_list =1;3251argv_array_push(&rp,"--all");3252}3253if(rev_list_reflog) {3254 use_internal_rev_list =1;3255argv_array_push(&rp,"--reflog");3256}3257if(rev_list_index) {3258 use_internal_rev_list =1;3259argv_array_push(&rp,"--indexed-objects");3260}3261if(rev_list_unpacked) {3262 use_internal_rev_list =1;3263argv_array_push(&rp,"--unpacked");3264}32653266if(exclude_promisor_objects) {3267 use_internal_rev_list =1;3268 fetch_if_missing =0;3269argv_array_push(&rp,"--exclude-promisor-objects");3270}3271if(unpack_unreachable || keep_unreachable || pack_loose_unreachable)3272 use_internal_rev_list =1;32733274if(!reuse_object)3275 reuse_delta =0;3276if(pack_compression_level == -1)3277 pack_compression_level = Z_DEFAULT_COMPRESSION;3278else if(pack_compression_level <0|| pack_compression_level > Z_BEST_COMPRESSION)3279die(_("bad pack compression level%d"), pack_compression_level);32803281if(!delta_search_threads)/* --threads=0 means autodetect */3282 delta_search_threads =online_cpus();32833284#ifdef NO_PTHREADS3285if(delta_search_threads !=1)3286warning(_("no threads support, ignoring --threads"));3287#endif3288if(!pack_to_stdout && !pack_size_limit)3289 pack_size_limit = pack_size_limit_cfg;3290if(pack_to_stdout && pack_size_limit)3291die(_("--max-pack-size cannot be used to build a pack for transfer"));3292if(pack_size_limit && pack_size_limit <1024*1024) {3293warning(_("minimum pack size limit is 1 MiB"));3294 pack_size_limit =1024*1024;3295}32963297if(!pack_to_stdout && thin)3298die(_("--thin cannot be used to build an indexable pack"));32993300if(keep_unreachable && unpack_unreachable)3301die(_("--keep-unreachable and --unpack-unreachable are incompatible"));3302if(!rev_list_all || !rev_list_reflog || !rev_list_index)3303 unpack_unreachable_expiration =0;33043305if(filter_options.choice) {3306if(!pack_to_stdout)3307die(_("cannot use --filter without --stdout"));3308 use_bitmap_index =0;3309}33103311/*3312 * "soft" reasons not to use bitmaps - for on-disk repack by default we want3313 *3314 * - to produce good pack (with bitmap index not-yet-packed objects are3315 * packed in suboptimal order).3316 *3317 * - to use more robust pack-generation codepath (avoiding possible3318 * bugs in bitmap code and possible bitmap index corruption).3319 */3320if(!pack_to_stdout)3321 use_bitmap_index_default =0;33223323if(use_bitmap_index <0)3324 use_bitmap_index = use_bitmap_index_default;33253326/* "hard" reasons not to use bitmaps; these just won't work at all */3327if(!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) ||is_repository_shallow(the_repository))3328 use_bitmap_index =0;33293330if(pack_to_stdout || !rev_list_all)3331 write_bitmap_index =0;33323333if(progress && all_progress_implied)3334 progress =2;33353336add_extra_kept_packs(&keep_pack_list);3337if(ignore_packed_keep_on_disk) {3338struct packed_git *p;3339for(p =get_packed_git(the_repository); p; p = p->next)3340if(p->pack_local && p->pack_keep)3341break;3342if(!p)/* no keep-able packs found */3343 ignore_packed_keep_on_disk =0;3344}3345if(local) {3346/*3347 * unlike ignore_packed_keep_on_disk above, we do not3348 * want to unset "local" based on looking at packs, as3349 * it also covers non-local objects3350 */3351struct packed_git *p;3352for(p =get_packed_git(the_repository); p; p = p->next) {3353if(!p->pack_local) {3354 have_non_local_packs =1;3355break;3356}3357}3358}33593360prepare_packing_data(&to_pack);33613362if(progress)3363 progress_state =start_progress(_("Enumerating objects"),0);3364if(!use_internal_rev_list)3365read_object_list_from_stdin();3366else{3367get_object_list(rp.argc, rp.argv);3368argv_array_clear(&rp);3369}3370cleanup_preferred_base();3371if(include_tag && nr_result)3372for_each_ref(add_ref_tag, NULL);3373stop_progress(&progress_state);33743375if(non_empty && !nr_result)3376return0;3377if(nr_result)3378prepare_pack(window, depth);3379write_pack_file();3380if(progress)3381fprintf_ln(stderr,3382_("Total %"PRIu32" (delta %"PRIu32"),"3383" reused %"PRIu32" (delta %"PRIu32")"),3384 written, written_delta, reused, reused_delta);3385return0;3386}