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"delta-islands.h" 28#include"reachable.h" 29#include"sha1-array.h" 30#include"argv-array.h" 31#include"list.h" 32#include"packfile.h" 33#include"object-store.h" 34#include"dir.h" 35#include"midx.h" 36#include"trace2.h" 37 38#define IN_PACK(obj) oe_in_pack(&to_pack, obj) 39#define SIZE(obj) oe_size(&to_pack, obj) 40#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size) 41#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj) 42#define DELTA(obj) oe_delta(&to_pack, obj) 43#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj) 44#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj) 45#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val) 46#define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid) 47#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val) 48#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val) 49#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val) 50 51static const char*pack_usage[] = { 52N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), 53N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), 54 NULL 55}; 56 57/* 58 * Objects we are going to pack are collected in the `to_pack` structure. 59 * It contains an array (dynamically expanded) of the object data, and a map 60 * that can resolve SHA1s to their position in the array. 61 */ 62static struct packing_data to_pack; 63 64static struct pack_idx_entry **written_list; 65static uint32_t nr_result, nr_written, nr_seen; 66static struct bitmap_index *bitmap_git; 67static uint32_t write_layer; 68 69static int non_empty; 70static int reuse_delta =1, reuse_object =1; 71static int keep_unreachable, unpack_unreachable, include_tag; 72static timestamp_t unpack_unreachable_expiration; 73static int pack_loose_unreachable; 74static int local; 75static int have_non_local_packs; 76static int incremental; 77static int ignore_packed_keep_on_disk; 78static int ignore_packed_keep_in_core; 79static int allow_ofs_delta; 80static struct pack_idx_option pack_idx_opts; 81static const char*base_name; 82static int progress =1; 83static int window =10; 84static unsigned long pack_size_limit; 85static int depth =50; 86static int delta_search_threads; 87static int pack_to_stdout; 88static int sparse; 89static int thin; 90static int num_preferred_base; 91static struct progress *progress_state; 92 93static struct packed_git *reuse_packfile; 94static uint32_t reuse_packfile_objects; 95static off_t reuse_packfile_offset; 96 97static int use_bitmap_index_default =1; 98static int use_bitmap_index = -1; 99static int write_bitmap_index; 100static uint16_t write_bitmap_options = BITMAP_OPT_HASH_CACHE; 101 102static int exclude_promisor_objects; 103 104static int use_delta_islands; 105 106static unsigned long delta_cache_size =0; 107static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; 108static unsigned long cache_max_small_delta_size =1000; 109 110static unsigned long window_memory_limit =0; 111 112static struct list_objects_filter_options filter_options; 113 114enum missing_action { 115 MA_ERROR =0,/* fail if any missing objects are encountered */ 116 MA_ALLOW_ANY,/* silently allow ALL missing objects */ 117 MA_ALLOW_PROMISOR,/* silently allow all missing PROMISOR objects */ 118}; 119static enum missing_action arg_missing_action; 120static show_object_fn fn_show_object; 121 122/* 123 * stats 124 */ 125static uint32_t written, written_delta; 126static uint32_t reused, reused_delta; 127 128/* 129 * Indexed commits 130 */ 131static struct commit **indexed_commits; 132static unsigned int indexed_commits_nr; 133static unsigned int indexed_commits_alloc; 134 135static voidindex_commit_for_bitmap(struct commit *commit) 136{ 137if(indexed_commits_nr >= indexed_commits_alloc) { 138 indexed_commits_alloc = (indexed_commits_alloc +32) *2; 139REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); 140} 141 142 indexed_commits[indexed_commits_nr++] = commit; 143} 144 145static void*get_delta(struct object_entry *entry) 146{ 147unsigned long size, base_size, delta_size; 148void*buf, *base_buf, *delta_buf; 149enum object_type type; 150 151 buf =read_object_file(&entry->idx.oid, &type, &size); 152if(!buf) 153die(_("unable to read%s"),oid_to_hex(&entry->idx.oid)); 154 base_buf =read_object_file(&DELTA(entry)->idx.oid, &type, 155&base_size); 156if(!base_buf) 157die("unable to read%s", 158oid_to_hex(&DELTA(entry)->idx.oid)); 159 delta_buf =diff_delta(base_buf, base_size, 160 buf, size, &delta_size,0); 161/* 162 * We succesfully computed this delta once but dropped it for 163 * memory reasons. Something is very wrong if this time we 164 * recompute and create a different delta. 165 */ 166if(!delta_buf || delta_size !=DELTA_SIZE(entry)) 167BUG("delta size changed"); 168free(buf); 169free(base_buf); 170return delta_buf; 171} 172 173static unsigned longdo_compress(void**pptr,unsigned long size) 174{ 175 git_zstream stream; 176void*in, *out; 177unsigned long maxsize; 178 179git_deflate_init(&stream, pack_compression_level); 180 maxsize =git_deflate_bound(&stream, size); 181 182 in = *pptr; 183 out =xmalloc(maxsize); 184*pptr = out; 185 186 stream.next_in = in; 187 stream.avail_in = size; 188 stream.next_out = out; 189 stream.avail_out = maxsize; 190while(git_deflate(&stream, Z_FINISH) == Z_OK) 191;/* nothing */ 192git_deflate_end(&stream); 193 194free(in); 195return stream.total_out; 196} 197 198static unsigned longwrite_large_blob_data(struct git_istream *st,struct hashfile *f, 199const struct object_id *oid) 200{ 201 git_zstream stream; 202unsigned char ibuf[1024*16]; 203unsigned char obuf[1024*16]; 204unsigned long olen =0; 205 206git_deflate_init(&stream, pack_compression_level); 207 208for(;;) { 209 ssize_t readlen; 210int zret = Z_OK; 211 readlen =read_istream(st, ibuf,sizeof(ibuf)); 212if(readlen == -1) 213die(_("unable to read%s"),oid_to_hex(oid)); 214 215 stream.next_in = ibuf; 216 stream.avail_in = readlen; 217while((stream.avail_in || readlen ==0) && 218(zret == Z_OK || zret == Z_BUF_ERROR)) { 219 stream.next_out = obuf; 220 stream.avail_out =sizeof(obuf); 221 zret =git_deflate(&stream, readlen ?0: Z_FINISH); 222hashwrite(f, obuf, stream.next_out - obuf); 223 olen += stream.next_out - obuf; 224} 225if(stream.avail_in) 226die(_("deflate error (%d)"), zret); 227if(readlen ==0) { 228if(zret != Z_STREAM_END) 229die(_("deflate error (%d)"), zret); 230break; 231} 232} 233git_deflate_end(&stream); 234return olen; 235} 236 237/* 238 * we are going to reuse the existing object data as is. make 239 * sure it is not corrupt. 240 */ 241static intcheck_pack_inflate(struct packed_git *p, 242struct pack_window **w_curs, 243 off_t offset, 244 off_t len, 245unsigned long expect) 246{ 247 git_zstream stream; 248unsigned char fakebuf[4096], *in; 249int st; 250 251memset(&stream,0,sizeof(stream)); 252git_inflate_init(&stream); 253do{ 254 in =use_pack(p, w_curs, offset, &stream.avail_in); 255 stream.next_in = in; 256 stream.next_out = fakebuf; 257 stream.avail_out =sizeof(fakebuf); 258 st =git_inflate(&stream, Z_FINISH); 259 offset += stream.next_in - in; 260}while(st == Z_OK || st == Z_BUF_ERROR); 261git_inflate_end(&stream); 262return(st == Z_STREAM_END && 263 stream.total_out == expect && 264 stream.total_in == len) ?0: -1; 265} 266 267static voidcopy_pack_data(struct hashfile *f, 268struct packed_git *p, 269struct pack_window **w_curs, 270 off_t offset, 271 off_t len) 272{ 273unsigned char*in; 274unsigned long avail; 275 276while(len) { 277 in =use_pack(p, w_curs, offset, &avail); 278if(avail > len) 279 avail = (unsigned long)len; 280hashwrite(f, in, avail); 281 offset += avail; 282 len -= avail; 283} 284} 285 286/* Return 0 if we will bust the pack-size limit */ 287static unsigned longwrite_no_reuse_object(struct hashfile *f,struct object_entry *entry, 288unsigned long limit,int usable_delta) 289{ 290unsigned long size, datalen; 291unsigned char header[MAX_PACK_OBJECT_HEADER], 292 dheader[MAX_PACK_OBJECT_HEADER]; 293unsigned hdrlen; 294enum object_type type; 295void*buf; 296struct git_istream *st = NULL; 297const unsigned hashsz = the_hash_algo->rawsz; 298 299if(!usable_delta) { 300if(oe_type(entry) == OBJ_BLOB && 301oe_size_greater_than(&to_pack, entry, big_file_threshold) && 302(st =open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL) 303 buf = NULL; 304else{ 305 buf =read_object_file(&entry->idx.oid, &type, &size); 306if(!buf) 307die(_("unable to read%s"), 308oid_to_hex(&entry->idx.oid)); 309} 310/* 311 * make sure no cached delta data remains from a 312 * previous attempt before a pack split occurred. 313 */ 314FREE_AND_NULL(entry->delta_data); 315 entry->z_delta_size =0; 316}else if(entry->delta_data) { 317 size =DELTA_SIZE(entry); 318 buf = entry->delta_data; 319 entry->delta_data = NULL; 320 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 321 OBJ_OFS_DELTA : OBJ_REF_DELTA; 322}else{ 323 buf =get_delta(entry); 324 size =DELTA_SIZE(entry); 325 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 326 OBJ_OFS_DELTA : OBJ_REF_DELTA; 327} 328 329if(st)/* large blob case, just assume we don't compress well */ 330 datalen = size; 331else if(entry->z_delta_size) 332 datalen = entry->z_delta_size; 333else 334 datalen =do_compress(&buf, size); 335 336/* 337 * The object header is a byte of 'type' followed by zero or 338 * more bytes of length. 339 */ 340 hdrlen =encode_in_pack_object_header(header,sizeof(header), 341 type, size); 342 343if(type == OBJ_OFS_DELTA) { 344/* 345 * Deltas with relative base contain an additional 346 * encoding of the relative offset for the delta 347 * base from this object's position in the pack. 348 */ 349 off_t ofs = entry->idx.offset -DELTA(entry)->idx.offset; 350unsigned pos =sizeof(dheader) -1; 351 dheader[pos] = ofs &127; 352while(ofs >>=7) 353 dheader[--pos] =128| (--ofs &127); 354if(limit && hdrlen +sizeof(dheader) - pos + datalen + hashsz >= limit) { 355if(st) 356close_istream(st); 357free(buf); 358return0; 359} 360hashwrite(f, header, hdrlen); 361hashwrite(f, dheader + pos,sizeof(dheader) - pos); 362 hdrlen +=sizeof(dheader) - pos; 363}else if(type == OBJ_REF_DELTA) { 364/* 365 * Deltas with a base reference contain 366 * additional bytes for the base object ID. 367 */ 368if(limit && hdrlen + hashsz + datalen + hashsz >= limit) { 369if(st) 370close_istream(st); 371free(buf); 372return0; 373} 374hashwrite(f, header, hdrlen); 375hashwrite(f,DELTA(entry)->idx.oid.hash, hashsz); 376 hdrlen += hashsz; 377}else{ 378if(limit && hdrlen + datalen + hashsz >= limit) { 379if(st) 380close_istream(st); 381free(buf); 382return0; 383} 384hashwrite(f, header, hdrlen); 385} 386if(st) { 387 datalen =write_large_blob_data(st, f, &entry->idx.oid); 388close_istream(st); 389}else{ 390hashwrite(f, buf, datalen); 391free(buf); 392} 393 394return hdrlen + datalen; 395} 396 397/* Return 0 if we will bust the pack-size limit */ 398static off_t write_reuse_object(struct hashfile *f,struct object_entry *entry, 399unsigned long limit,int usable_delta) 400{ 401struct packed_git *p =IN_PACK(entry); 402struct pack_window *w_curs = NULL; 403struct revindex_entry *revidx; 404 off_t offset; 405enum object_type type =oe_type(entry); 406 off_t datalen; 407unsigned char header[MAX_PACK_OBJECT_HEADER], 408 dheader[MAX_PACK_OBJECT_HEADER]; 409unsigned hdrlen; 410const unsigned hashsz = the_hash_algo->rawsz; 411unsigned long entry_size =SIZE(entry); 412 413if(DELTA(entry)) 414 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 415 OBJ_OFS_DELTA : OBJ_REF_DELTA; 416 hdrlen =encode_in_pack_object_header(header,sizeof(header), 417 type, entry_size); 418 419 offset = entry->in_pack_offset; 420 revidx =find_pack_revindex(p, offset); 421 datalen = revidx[1].offset - offset; 422if(!pack_to_stdout && p->index_version >1&& 423check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 424error(_("bad packed object CRC for%s"), 425oid_to_hex(&entry->idx.oid)); 426unuse_pack(&w_curs); 427returnwrite_no_reuse_object(f, entry, limit, usable_delta); 428} 429 430 offset += entry->in_pack_header_size; 431 datalen -= entry->in_pack_header_size; 432 433if(!pack_to_stdout && p->index_version ==1&& 434check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { 435error(_("corrupt packed object for%s"), 436oid_to_hex(&entry->idx.oid)); 437unuse_pack(&w_curs); 438returnwrite_no_reuse_object(f, entry, limit, usable_delta); 439} 440 441if(type == OBJ_OFS_DELTA) { 442 off_t ofs = entry->idx.offset -DELTA(entry)->idx.offset; 443unsigned pos =sizeof(dheader) -1; 444 dheader[pos] = ofs &127; 445while(ofs >>=7) 446 dheader[--pos] =128| (--ofs &127); 447if(limit && hdrlen +sizeof(dheader) - pos + datalen + hashsz >= limit) { 448unuse_pack(&w_curs); 449return0; 450} 451hashwrite(f, header, hdrlen); 452hashwrite(f, dheader + pos,sizeof(dheader) - pos); 453 hdrlen +=sizeof(dheader) - pos; 454 reused_delta++; 455}else if(type == OBJ_REF_DELTA) { 456if(limit && hdrlen + hashsz + datalen + hashsz >= limit) { 457unuse_pack(&w_curs); 458return0; 459} 460hashwrite(f, header, hdrlen); 461hashwrite(f,DELTA(entry)->idx.oid.hash, hashsz); 462 hdrlen += hashsz; 463 reused_delta++; 464}else{ 465if(limit && hdrlen + datalen + hashsz >= limit) { 466unuse_pack(&w_curs); 467return0; 468} 469hashwrite(f, header, hdrlen); 470} 471copy_pack_data(f, p, &w_curs, offset, datalen); 472unuse_pack(&w_curs); 473 reused++; 474return hdrlen + datalen; 475} 476 477/* Return 0 if we will bust the pack-size limit */ 478static off_t write_object(struct hashfile *f, 479struct object_entry *entry, 480 off_t write_offset) 481{ 482unsigned long limit; 483 off_t len; 484int usable_delta, to_reuse; 485 486if(!pack_to_stdout) 487crc32_begin(f); 488 489/* apply size limit if limited packsize and not first object */ 490if(!pack_size_limit || !nr_written) 491 limit =0; 492else if(pack_size_limit <= write_offset) 493/* 494 * the earlier object did not fit the limit; avoid 495 * mistaking this with unlimited (i.e. limit = 0). 496 */ 497 limit =1; 498else 499 limit = pack_size_limit - write_offset; 500 501if(!DELTA(entry)) 502 usable_delta =0;/* no delta */ 503else if(!pack_size_limit) 504 usable_delta =1;/* unlimited packfile */ 505else if(DELTA(entry)->idx.offset == (off_t)-1) 506 usable_delta =0;/* base was written to another pack */ 507else if(DELTA(entry)->idx.offset) 508 usable_delta =1;/* base already exists in this pack */ 509else 510 usable_delta =0;/* base could end up in another pack */ 511 512if(!reuse_object) 513 to_reuse =0;/* explicit */ 514else if(!IN_PACK(entry)) 515 to_reuse =0;/* can't reuse what we don't have */ 516else if(oe_type(entry) == OBJ_REF_DELTA || 517oe_type(entry) == OBJ_OFS_DELTA) 518/* check_object() decided it for us ... */ 519 to_reuse = usable_delta; 520/* ... but pack split may override that */ 521else if(oe_type(entry) != entry->in_pack_type) 522 to_reuse =0;/* pack has delta which is unusable */ 523else if(DELTA(entry)) 524 to_reuse =0;/* we want to pack afresh */ 525else 526 to_reuse =1;/* we have it in-pack undeltified, 527 * and we do not need to deltify it. 528 */ 529 530if(!to_reuse) 531 len =write_no_reuse_object(f, entry, limit, usable_delta); 532else 533 len =write_reuse_object(f, entry, limit, usable_delta); 534if(!len) 535return0; 536 537if(usable_delta) 538 written_delta++; 539 written++; 540if(!pack_to_stdout) 541 entry->idx.crc32 =crc32_end(f); 542return len; 543} 544 545enum write_one_status { 546 WRITE_ONE_SKIP = -1,/* already written */ 547 WRITE_ONE_BREAK =0,/* writing this will bust the limit; not written */ 548 WRITE_ONE_WRITTEN =1,/* normal */ 549 WRITE_ONE_RECURSIVE =2/* already scheduled to be written */ 550}; 551 552static enum write_one_status write_one(struct hashfile *f, 553struct object_entry *e, 554 off_t *offset) 555{ 556 off_t size; 557int recursing; 558 559/* 560 * we set offset to 1 (which is an impossible value) to mark 561 * the fact that this object is involved in "write its base 562 * first before writing a deltified object" recursion. 563 */ 564 recursing = (e->idx.offset ==1); 565if(recursing) { 566warning(_("recursive delta detected for object%s"), 567oid_to_hex(&e->idx.oid)); 568return WRITE_ONE_RECURSIVE; 569}else if(e->idx.offset || e->preferred_base) { 570/* offset is non zero if object is written already. */ 571return WRITE_ONE_SKIP; 572} 573 574/* if we are deltified, write out base object first. */ 575if(DELTA(e)) { 576 e->idx.offset =1;/* now recurse */ 577switch(write_one(f,DELTA(e), offset)) { 578case WRITE_ONE_RECURSIVE: 579/* we cannot depend on this one */ 580SET_DELTA(e, NULL); 581break; 582default: 583break; 584case WRITE_ONE_BREAK: 585 e->idx.offset = recursing; 586return WRITE_ONE_BREAK; 587} 588} 589 590 e->idx.offset = *offset; 591 size =write_object(f, e, *offset); 592if(!size) { 593 e->idx.offset = recursing; 594return WRITE_ONE_BREAK; 595} 596 written_list[nr_written++] = &e->idx; 597 598/* make sure off_t is sufficiently large not to wrap */ 599if(signed_add_overflows(*offset, size)) 600die(_("pack too large for current definition of off_t")); 601*offset += size; 602return WRITE_ONE_WRITTEN; 603} 604 605static intmark_tagged(const char*path,const struct object_id *oid,int flag, 606void*cb_data) 607{ 608struct object_id peeled; 609struct object_entry *entry =packlist_find(&to_pack, oid, NULL); 610 611if(entry) 612 entry->tagged =1; 613if(!peel_ref(path, &peeled)) { 614 entry =packlist_find(&to_pack, &peeled, NULL); 615if(entry) 616 entry->tagged =1; 617} 618return0; 619} 620 621staticinlinevoidadd_to_write_order(struct object_entry **wo, 622unsigned int*endp, 623struct object_entry *e) 624{ 625if(e->filled ||oe_layer(&to_pack, e) != write_layer) 626return; 627 wo[(*endp)++] = e; 628 e->filled =1; 629} 630 631static voidadd_descendants_to_write_order(struct object_entry **wo, 632unsigned int*endp, 633struct object_entry *e) 634{ 635int add_to_order =1; 636while(e) { 637if(add_to_order) { 638struct object_entry *s; 639/* add this node... */ 640add_to_write_order(wo, endp, e); 641/* all its siblings... */ 642for(s =DELTA_SIBLING(e); s; s =DELTA_SIBLING(s)) { 643add_to_write_order(wo, endp, s); 644} 645} 646/* drop down a level to add left subtree nodes if possible */ 647if(DELTA_CHILD(e)) { 648 add_to_order =1; 649 e =DELTA_CHILD(e); 650}else{ 651 add_to_order =0; 652/* our sibling might have some children, it is next */ 653if(DELTA_SIBLING(e)) { 654 e =DELTA_SIBLING(e); 655continue; 656} 657/* go back to our parent node */ 658 e =DELTA(e); 659while(e && !DELTA_SIBLING(e)) { 660/* we're on the right side of a subtree, keep 661 * going up until we can go right again */ 662 e =DELTA(e); 663} 664if(!e) { 665/* done- we hit our original root node */ 666return; 667} 668/* pass it off to sibling at this level */ 669 e =DELTA_SIBLING(e); 670} 671}; 672} 673 674static voidadd_family_to_write_order(struct object_entry **wo, 675unsigned int*endp, 676struct object_entry *e) 677{ 678struct object_entry *root; 679 680for(root = e;DELTA(root); root =DELTA(root)) 681;/* nothing */ 682add_descendants_to_write_order(wo, endp, root); 683} 684 685static voidcompute_layer_order(struct object_entry **wo,unsigned int*wo_end) 686{ 687unsigned int i, last_untagged; 688struct object_entry *objects = to_pack.objects; 689 690for(i =0; i < to_pack.nr_objects; i++) { 691if(objects[i].tagged) 692break; 693add_to_write_order(wo, wo_end, &objects[i]); 694} 695 last_untagged = i; 696 697/* 698 * Then fill all the tagged tips. 699 */ 700for(; i < to_pack.nr_objects; i++) { 701if(objects[i].tagged) 702add_to_write_order(wo, wo_end, &objects[i]); 703} 704 705/* 706 * And then all remaining commits and tags. 707 */ 708for(i = last_untagged; i < to_pack.nr_objects; i++) { 709if(oe_type(&objects[i]) != OBJ_COMMIT && 710oe_type(&objects[i]) != OBJ_TAG) 711continue; 712add_to_write_order(wo, wo_end, &objects[i]); 713} 714 715/* 716 * And then all the trees. 717 */ 718for(i = last_untagged; i < to_pack.nr_objects; i++) { 719if(oe_type(&objects[i]) != OBJ_TREE) 720continue; 721add_to_write_order(wo, wo_end, &objects[i]); 722} 723 724/* 725 * Finally all the rest in really tight order 726 */ 727for(i = last_untagged; i < to_pack.nr_objects; i++) { 728if(!objects[i].filled &&oe_layer(&to_pack, &objects[i]) == write_layer) 729add_family_to_write_order(wo, wo_end, &objects[i]); 730} 731} 732 733static struct object_entry **compute_write_order(void) 734{ 735uint32_t max_layers =1; 736unsigned int i, wo_end; 737 738struct object_entry **wo; 739struct object_entry *objects = to_pack.objects; 740 741for(i =0; i < to_pack.nr_objects; i++) { 742 objects[i].tagged =0; 743 objects[i].filled =0; 744SET_DELTA_CHILD(&objects[i], NULL); 745SET_DELTA_SIBLING(&objects[i], NULL); 746} 747 748/* 749 * Fully connect delta_child/delta_sibling network. 750 * Make sure delta_sibling is sorted in the original 751 * recency order. 752 */ 753for(i = to_pack.nr_objects; i >0;) { 754struct object_entry *e = &objects[--i]; 755if(!DELTA(e)) 756continue; 757/* Mark me as the first child */ 758 e->delta_sibling_idx =DELTA(e)->delta_child_idx; 759SET_DELTA_CHILD(DELTA(e), e); 760} 761 762/* 763 * Mark objects that are at the tip of tags. 764 */ 765for_each_tag_ref(mark_tagged, NULL); 766 767if(use_delta_islands) 768 max_layers =compute_pack_layers(&to_pack); 769 770ALLOC_ARRAY(wo, to_pack.nr_objects); 771 wo_end =0; 772 773for(; write_layer < max_layers; ++write_layer) 774compute_layer_order(wo, &wo_end); 775 776if(wo_end != to_pack.nr_objects) 777die(_("ordered%uobjects, expected %"PRIu32), 778 wo_end, to_pack.nr_objects); 779 780return wo; 781} 782 783static off_t write_reused_pack(struct hashfile *f) 784{ 785unsigned char buffer[8192]; 786 off_t to_write, total; 787int fd; 788 789if(!is_pack_valid(reuse_packfile)) 790die(_("packfile is invalid:%s"), reuse_packfile->pack_name); 791 792 fd =git_open(reuse_packfile->pack_name); 793if(fd <0) 794die_errno(_("unable to open packfile for reuse:%s"), 795 reuse_packfile->pack_name); 796 797if(lseek(fd,sizeof(struct pack_header), SEEK_SET) == -1) 798die_errno(_("unable to seek in reused packfile")); 799 800if(reuse_packfile_offset <0) 801 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz; 802 803 total = to_write = reuse_packfile_offset -sizeof(struct pack_header); 804 805while(to_write) { 806int read_pack =xread(fd, buffer,sizeof(buffer)); 807 808if(read_pack <=0) 809die_errno(_("unable to read from reused packfile")); 810 811if(read_pack > to_write) 812 read_pack = to_write; 813 814hashwrite(f, buffer, read_pack); 815 to_write -= read_pack; 816 817/* 818 * We don't know the actual number of objects written, 819 * only how many bytes written, how many bytes total, and 820 * how many objects total. So we can fake it by pretending all 821 * objects we are writing are the same size. This gives us a 822 * smooth progress meter, and at the end it matches the true 823 * answer. 824 */ 825 written = reuse_packfile_objects * 826(((double)(total - to_write)) / total); 827display_progress(progress_state, written); 828} 829 830close(fd); 831 written = reuse_packfile_objects; 832display_progress(progress_state, written); 833return reuse_packfile_offset -sizeof(struct pack_header); 834} 835 836static const char no_split_warning[] =N_( 837"disabling bitmap writing, packs are split due to pack.packSizeLimit" 838); 839 840static voidwrite_pack_file(void) 841{ 842uint32_t i =0, j; 843struct hashfile *f; 844 off_t offset; 845uint32_t nr_remaining = nr_result; 846time_t last_mtime =0; 847struct object_entry **write_order; 848 849if(progress > pack_to_stdout) 850 progress_state =start_progress(_("Writing objects"), nr_result); 851ALLOC_ARRAY(written_list, to_pack.nr_objects); 852 write_order =compute_write_order(); 853 854do{ 855struct object_id oid; 856char*pack_tmp_name = NULL; 857 858if(pack_to_stdout) 859 f =hashfd_throughput(1,"<stdout>", progress_state); 860else 861 f =create_tmp_packfile(&pack_tmp_name); 862 863 offset =write_pack_header(f, nr_remaining); 864 865if(reuse_packfile) { 866 off_t packfile_size; 867assert(pack_to_stdout); 868 869 packfile_size =write_reused_pack(f); 870 offset += packfile_size; 871} 872 873 nr_written =0; 874for(; i < to_pack.nr_objects; i++) { 875struct object_entry *e = write_order[i]; 876if(write_one(f, e, &offset) == WRITE_ONE_BREAK) 877break; 878display_progress(progress_state, written); 879} 880 881/* 882 * Did we write the wrong # entries in the header? 883 * If so, rewrite it like in fast-import 884 */ 885if(pack_to_stdout) { 886finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); 887}else if(nr_written == nr_remaining) { 888finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); 889}else{ 890int fd =finalize_hashfile(f, oid.hash,0); 891fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, 892 nr_written, oid.hash, offset); 893close(fd); 894if(write_bitmap_index) { 895warning(_(no_split_warning)); 896 write_bitmap_index =0; 897} 898} 899 900if(!pack_to_stdout) { 901struct stat st; 902struct strbuf tmpname = STRBUF_INIT; 903 904/* 905 * Packs are runtime accessed in their mtime 906 * order since newer packs are more likely to contain 907 * younger objects. So if we are creating multiple 908 * packs then we should modify the mtime of later ones 909 * to preserve this property. 910 */ 911if(stat(pack_tmp_name, &st) <0) { 912warning_errno(_("failed to stat%s"), pack_tmp_name); 913}else if(!last_mtime) { 914 last_mtime = st.st_mtime; 915}else{ 916struct utimbuf utb; 917 utb.actime = st.st_atime; 918 utb.modtime = --last_mtime; 919if(utime(pack_tmp_name, &utb) <0) 920warning_errno(_("failed utime() on%s"), pack_tmp_name); 921} 922 923strbuf_addf(&tmpname,"%s-", base_name); 924 925if(write_bitmap_index) { 926bitmap_writer_set_checksum(oid.hash); 927bitmap_writer_build_type_index( 928&to_pack, written_list, nr_written); 929} 930 931finish_tmp_packfile(&tmpname, pack_tmp_name, 932 written_list, nr_written, 933&pack_idx_opts, oid.hash); 934 935if(write_bitmap_index) { 936strbuf_addf(&tmpname,"%s.bitmap",oid_to_hex(&oid)); 937 938stop_progress(&progress_state); 939 940bitmap_writer_show_progress(progress); 941bitmap_writer_reuse_bitmaps(&to_pack); 942bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 943bitmap_writer_build(&to_pack); 944bitmap_writer_finish(written_list, nr_written, 945 tmpname.buf, write_bitmap_options); 946 write_bitmap_index =0; 947} 948 949strbuf_release(&tmpname); 950free(pack_tmp_name); 951puts(oid_to_hex(&oid)); 952} 953 954/* mark written objects as written to previous pack */ 955for(j =0; j < nr_written; j++) { 956 written_list[j]->offset = (off_t)-1; 957} 958 nr_remaining -= nr_written; 959}while(nr_remaining && i < to_pack.nr_objects); 960 961free(written_list); 962free(write_order); 963stop_progress(&progress_state); 964if(written != nr_result) 965die(_("wrote %"PRIu32" objects while expecting %"PRIu32), 966 written, nr_result); 967trace2_data_intmax("pack-objects", the_repository, 968"write_pack_file/wrote", nr_result); 969} 970 971static intno_try_delta(const char*path) 972{ 973static struct attr_check *check; 974 975if(!check) 976 check =attr_check_initl("delta", NULL); 977git_check_attr(the_repository->index, path, check); 978if(ATTR_FALSE(check->items[0].value)) 979return1; 980return0; 981} 982 983/* 984 * When adding an object, check whether we have already added it 985 * to our packing list. If so, we can skip. However, if we are 986 * being asked to excludei t, but the previous mention was to include 987 * it, make sure to adjust its flags and tweak our numbers accordingly. 988 * 989 * As an optimization, we pass out the index position where we would have 990 * found the item, since that saves us from having to look it up again a 991 * few lines later when we want to add the new entry. 992 */ 993static inthave_duplicate_entry(const struct object_id *oid, 994int exclude, 995uint32_t*index_pos) 996{ 997struct object_entry *entry; 998 999 entry =packlist_find(&to_pack, oid, index_pos);1000if(!entry)1001return0;10021003if(exclude) {1004if(!entry->preferred_base)1005 nr_result--;1006 entry->preferred_base =1;1007}10081009return1;1010}10111012static intwant_found_object(int exclude,struct packed_git *p)1013{1014if(exclude)1015return1;1016if(incremental)1017return0;10181019/*1020 * When asked to do --local (do not include an object that appears in a1021 * pack we borrow from elsewhere) or --honor-pack-keep (do not include1022 * an object that appears in a pack marked with .keep), finding a pack1023 * that matches the criteria is sufficient for us to decide to omit it.1024 * However, even if this pack does not satisfy the criteria, we need to1025 * make sure no copy of this object appears in _any_ pack that makes us1026 * to omit the object, so we need to check all the packs.1027 *1028 * We can however first check whether these options can possible matter;1029 * if they do not matter we know we want the object in generated pack.1030 * Otherwise, we signal "-1" at the end to tell the caller that we do1031 * not know either way, and it needs to check more packs.1032 */1033if(!ignore_packed_keep_on_disk &&1034!ignore_packed_keep_in_core &&1035(!local || !have_non_local_packs))1036return1;10371038if(local && !p->pack_local)1039return0;1040if(p->pack_local &&1041((ignore_packed_keep_on_disk && p->pack_keep) ||1042(ignore_packed_keep_in_core && p->pack_keep_in_core)))1043return0;10441045/* we don't know yet; keep looking for more packs */1046return-1;1047}10481049/*1050 * Check whether we want the object in the pack (e.g., we do not want1051 * objects found in non-local stores if the "--local" option was used).1052 *1053 * If the caller already knows an existing pack it wants to take the object1054 * from, that is passed in *found_pack and *found_offset; otherwise this1055 * function finds if there is any pack that has the object and returns the pack1056 * and its offset in these variables.1057 */1058static intwant_object_in_pack(const struct object_id *oid,1059int exclude,1060struct packed_git **found_pack,1061 off_t *found_offset)1062{1063int want;1064struct list_head *pos;1065struct multi_pack_index *m;10661067if(!exclude && local &&has_loose_object_nonlocal(oid))1068return0;10691070/*1071 * If we already know the pack object lives in, start checks from that1072 * pack - in the usual case when neither --local was given nor .keep files1073 * are present we will determine the answer right now.1074 */1075if(*found_pack) {1076 want =want_found_object(exclude, *found_pack);1077if(want != -1)1078return want;1079}10801081for(m =get_multi_pack_index(the_repository); m; m = m->next) {1082struct pack_entry e;1083if(fill_midx_entry(the_repository, oid, &e, m)) {1084struct packed_git *p = e.p;1085 off_t offset;10861087if(p == *found_pack)1088 offset = *found_offset;1089else1090 offset =find_pack_entry_one(oid->hash, p);10911092if(offset) {1093if(!*found_pack) {1094if(!is_pack_valid(p))1095continue;1096*found_offset = offset;1097*found_pack = p;1098}1099 want =want_found_object(exclude, p);1100if(want != -1)1101return want;1102}1103}1104}11051106list_for_each(pos,get_packed_git_mru(the_repository)) {1107struct packed_git *p =list_entry(pos,struct packed_git, mru);1108 off_t offset;11091110if(p == *found_pack)1111 offset = *found_offset;1112else1113 offset =find_pack_entry_one(oid->hash, p);11141115if(offset) {1116if(!*found_pack) {1117if(!is_pack_valid(p))1118continue;1119*found_offset = offset;1120*found_pack = p;1121}1122 want =want_found_object(exclude, p);1123if(!exclude && want >0)1124list_move(&p->mru,1125get_packed_git_mru(the_repository));1126if(want != -1)1127return want;1128}1129}11301131return1;1132}11331134static voidcreate_object_entry(const struct object_id *oid,1135enum object_type type,1136uint32_t hash,1137int exclude,1138int no_try_delta,1139uint32_t index_pos,1140struct packed_git *found_pack,1141 off_t found_offset)1142{1143struct object_entry *entry;11441145 entry =packlist_alloc(&to_pack, oid->hash, index_pos);1146 entry->hash = hash;1147oe_set_type(entry, type);1148if(exclude)1149 entry->preferred_base =1;1150else1151 nr_result++;1152if(found_pack) {1153oe_set_in_pack(&to_pack, entry, found_pack);1154 entry->in_pack_offset = found_offset;1155}11561157 entry->no_try_delta = no_try_delta;1158}11591160static const char no_closure_warning[] =N_(1161"disabling bitmap writing, as some objects are not being packed"1162);11631164static intadd_object_entry(const struct object_id *oid,enum object_type type,1165const char*name,int exclude)1166{1167struct packed_git *found_pack = NULL;1168 off_t found_offset =0;1169uint32_t index_pos;11701171display_progress(progress_state, ++nr_seen);11721173if(have_duplicate_entry(oid, exclude, &index_pos))1174return0;11751176if(!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1177/* The pack is missing an object, so it will not have closure */1178if(write_bitmap_index) {1179warning(_(no_closure_warning));1180 write_bitmap_index =0;1181}1182return0;1183}11841185create_object_entry(oid, type,pack_name_hash(name),1186 exclude, name &&no_try_delta(name),1187 index_pos, found_pack, found_offset);1188return1;1189}11901191static intadd_object_entry_from_bitmap(const struct object_id *oid,1192enum object_type type,1193int flags,uint32_t name_hash,1194struct packed_git *pack, off_t offset)1195{1196uint32_t index_pos;11971198display_progress(progress_state, ++nr_seen);11991200if(have_duplicate_entry(oid,0, &index_pos))1201return0;12021203if(!want_object_in_pack(oid,0, &pack, &offset))1204return0;12051206create_object_entry(oid, type, name_hash,0,0, index_pos, pack, offset);1207return1;1208}12091210struct pbase_tree_cache {1211struct object_id oid;1212int ref;1213int temporary;1214void*tree_data;1215unsigned long tree_size;1216};12171218static struct pbase_tree_cache *(pbase_tree_cache[256]);1219static intpbase_tree_cache_ix(const struct object_id *oid)1220{1221return oid->hash[0] %ARRAY_SIZE(pbase_tree_cache);1222}1223static intpbase_tree_cache_ix_incr(int ix)1224{1225return(ix+1) %ARRAY_SIZE(pbase_tree_cache);1226}12271228static struct pbase_tree {1229struct pbase_tree *next;1230/* This is a phony "cache" entry; we are not1231 * going to evict it or find it through _get()1232 * mechanism -- this is for the toplevel node that1233 * would almost always change with any commit.1234 */1235struct pbase_tree_cache pcache;1236} *pbase_tree;12371238static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1239{1240struct pbase_tree_cache *ent, *nent;1241void*data;1242unsigned long size;1243enum object_type type;1244int neigh;1245int my_ix =pbase_tree_cache_ix(oid);1246int available_ix = -1;12471248/* pbase-tree-cache acts as a limited hashtable.1249 * your object will be found at your index or within a few1250 * slots after that slot if it is cached.1251 */1252for(neigh =0; neigh <8; neigh++) {1253 ent = pbase_tree_cache[my_ix];1254if(ent &&oideq(&ent->oid, oid)) {1255 ent->ref++;1256return ent;1257}1258else if(((available_ix <0) && (!ent || !ent->ref)) ||1259((0<= available_ix) &&1260(!ent && pbase_tree_cache[available_ix])))1261 available_ix = my_ix;1262if(!ent)1263break;1264 my_ix =pbase_tree_cache_ix_incr(my_ix);1265}12661267/* Did not find one. Either we got a bogus request or1268 * we need to read and perhaps cache.1269 */1270 data =read_object_file(oid, &type, &size);1271if(!data)1272return NULL;1273if(type != OBJ_TREE) {1274free(data);1275return NULL;1276}12771278/* We need to either cache or return a throwaway copy */12791280if(available_ix <0)1281 ent = NULL;1282else{1283 ent = pbase_tree_cache[available_ix];1284 my_ix = available_ix;1285}12861287if(!ent) {1288 nent =xmalloc(sizeof(*nent));1289 nent->temporary = (available_ix <0);1290}1291else{1292/* evict and reuse */1293free(ent->tree_data);1294 nent = ent;1295}1296oidcpy(&nent->oid, oid);1297 nent->tree_data = data;1298 nent->tree_size = size;1299 nent->ref =1;1300if(!nent->temporary)1301 pbase_tree_cache[my_ix] = nent;1302return nent;1303}13041305static voidpbase_tree_put(struct pbase_tree_cache *cache)1306{1307if(!cache->temporary) {1308 cache->ref--;1309return;1310}1311free(cache->tree_data);1312free(cache);1313}13141315static intname_cmp_len(const char*name)1316{1317int i;1318for(i =0; name[i] && name[i] !='\n'&& name[i] !='/'; i++)1319;1320return i;1321}13221323static voidadd_pbase_object(struct tree_desc *tree,1324const char*name,1325int cmplen,1326const char*fullname)1327{1328struct name_entry entry;1329int cmp;13301331while(tree_entry(tree,&entry)) {1332if(S_ISGITLINK(entry.mode))1333continue;1334 cmp =tree_entry_len(&entry) != cmplen ?1:1335memcmp(name, entry.path, cmplen);1336if(cmp >0)1337continue;1338if(cmp <0)1339return;1340if(name[cmplen] !='/') {1341add_object_entry(&entry.oid,1342object_type(entry.mode),1343 fullname,1);1344return;1345}1346if(S_ISDIR(entry.mode)) {1347struct tree_desc sub;1348struct pbase_tree_cache *tree;1349const char*down = name+cmplen+1;1350int downlen =name_cmp_len(down);13511352 tree =pbase_tree_get(&entry.oid);1353if(!tree)1354return;1355init_tree_desc(&sub, tree->tree_data, tree->tree_size);13561357add_pbase_object(&sub, down, downlen, fullname);1358pbase_tree_put(tree);1359}1360}1361}13621363static unsigned*done_pbase_paths;1364static int done_pbase_paths_num;1365static int done_pbase_paths_alloc;1366static intdone_pbase_path_pos(unsigned hash)1367{1368int lo =0;1369int hi = done_pbase_paths_num;1370while(lo < hi) {1371int mi = lo + (hi - lo) /2;1372if(done_pbase_paths[mi] == hash)1373return mi;1374if(done_pbase_paths[mi] < hash)1375 hi = mi;1376else1377 lo = mi +1;1378}1379return-lo-1;1380}13811382static intcheck_pbase_path(unsigned hash)1383{1384int pos =done_pbase_path_pos(hash);1385if(0<= pos)1386return1;1387 pos = -pos -1;1388ALLOC_GROW(done_pbase_paths,1389 done_pbase_paths_num +1,1390 done_pbase_paths_alloc);1391 done_pbase_paths_num++;1392if(pos < done_pbase_paths_num)1393MOVE_ARRAY(done_pbase_paths + pos +1, done_pbase_paths + pos,1394 done_pbase_paths_num - pos -1);1395 done_pbase_paths[pos] = hash;1396return0;1397}13981399static voidadd_preferred_base_object(const char*name)1400{1401struct pbase_tree *it;1402int cmplen;1403unsigned hash =pack_name_hash(name);14041405if(!num_preferred_base ||check_pbase_path(hash))1406return;14071408 cmplen =name_cmp_len(name);1409for(it = pbase_tree; it; it = it->next) {1410if(cmplen ==0) {1411add_object_entry(&it->pcache.oid, OBJ_TREE, NULL,1);1412}1413else{1414struct tree_desc tree;1415init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1416add_pbase_object(&tree, name, cmplen, name);1417}1418}1419}14201421static voidadd_preferred_base(struct object_id *oid)1422{1423struct pbase_tree *it;1424void*data;1425unsigned long size;1426struct object_id tree_oid;14271428if(window <= num_preferred_base++)1429return;14301431 data =read_object_with_reference(oid, tree_type, &size, &tree_oid);1432if(!data)1433return;14341435for(it = pbase_tree; it; it = it->next) {1436if(oideq(&it->pcache.oid, &tree_oid)) {1437free(data);1438return;1439}1440}14411442 it =xcalloc(1,sizeof(*it));1443 it->next = pbase_tree;1444 pbase_tree = it;14451446oidcpy(&it->pcache.oid, &tree_oid);1447 it->pcache.tree_data = data;1448 it->pcache.tree_size = size;1449}14501451static voidcleanup_preferred_base(void)1452{1453struct pbase_tree *it;1454unsigned i;14551456 it = pbase_tree;1457 pbase_tree = NULL;1458while(it) {1459struct pbase_tree *tmp = it;1460 it = tmp->next;1461free(tmp->pcache.tree_data);1462free(tmp);1463}14641465for(i =0; i <ARRAY_SIZE(pbase_tree_cache); i++) {1466if(!pbase_tree_cache[i])1467continue;1468free(pbase_tree_cache[i]->tree_data);1469FREE_AND_NULL(pbase_tree_cache[i]);1470}14711472FREE_AND_NULL(done_pbase_paths);1473 done_pbase_paths_num = done_pbase_paths_alloc =0;1474}14751476/*1477 * Return 1 iff the object specified by "delta" can be sent1478 * literally as a delta against the base in "base_sha1". If1479 * so, then *base_out will point to the entry in our packing1480 * list, or NULL if we must use the external-base list.1481 *1482 * Depth value does not matter - find_deltas() will1483 * never consider reused delta as the base object to1484 * deltify other objects against, in order to avoid1485 * circular deltas.1486 */1487static intcan_reuse_delta(const unsigned char*base_sha1,1488struct object_entry *delta,1489struct object_entry **base_out)1490{1491struct object_entry *base;1492struct object_id base_oid;14931494if(!base_sha1)1495return0;14961497oidread(&base_oid, base_sha1);14981499/*1500 * First see if we're already sending the base (or it's explicitly in1501 * our "excluded" list).1502 */1503 base =packlist_find(&to_pack, &base_oid, NULL);1504if(base) {1505if(!in_same_island(&delta->idx.oid, &base->idx.oid))1506return0;1507*base_out = base;1508return1;1509}15101511/*1512 * Otherwise, reachability bitmaps may tell us if the receiver has it,1513 * even if it was buried too deep in history to make it into the1514 * packing list.1515 */1516if(thin &&bitmap_has_oid_in_uninteresting(bitmap_git, &base_oid)) {1517if(use_delta_islands) {1518if(!in_same_island(&delta->idx.oid, &base_oid))1519return0;1520}1521*base_out = NULL;1522return1;1523}15241525return0;1526}15271528static voidcheck_object(struct object_entry *entry)1529{1530unsigned long canonical_size;15311532if(IN_PACK(entry)) {1533struct packed_git *p =IN_PACK(entry);1534struct pack_window *w_curs = NULL;1535const unsigned char*base_ref = NULL;1536struct object_entry *base_entry;1537unsigned long used, used_0;1538unsigned long avail;1539 off_t ofs;1540unsigned char*buf, c;1541enum object_type type;1542unsigned long in_pack_size;15431544 buf =use_pack(p, &w_curs, entry->in_pack_offset, &avail);15451546/*1547 * We want in_pack_type even if we do not reuse delta1548 * since non-delta representations could still be reused.1549 */1550 used =unpack_object_header_buffer(buf, avail,1551&type,1552&in_pack_size);1553if(used ==0)1554goto give_up;15551556if(type <0)1557BUG("invalid type%d", type);1558 entry->in_pack_type = type;15591560/*1561 * Determine if this is a delta and if so whether we can1562 * reuse it or not. Otherwise let's find out as cheaply as1563 * possible what the actual type and size for this object is.1564 */1565switch(entry->in_pack_type) {1566default:1567/* Not a delta hence we've already got all we need. */1568oe_set_type(entry, entry->in_pack_type);1569SET_SIZE(entry, in_pack_size);1570 entry->in_pack_header_size = used;1571if(oe_type(entry) < OBJ_COMMIT ||oe_type(entry) > OBJ_BLOB)1572goto give_up;1573unuse_pack(&w_curs);1574return;1575case OBJ_REF_DELTA:1576if(reuse_delta && !entry->preferred_base)1577 base_ref =use_pack(p, &w_curs,1578 entry->in_pack_offset + used, NULL);1579 entry->in_pack_header_size = used + the_hash_algo->rawsz;1580break;1581case OBJ_OFS_DELTA:1582 buf =use_pack(p, &w_curs,1583 entry->in_pack_offset + used, NULL);1584 used_0 =0;1585 c = buf[used_0++];1586 ofs = c &127;1587while(c &128) {1588 ofs +=1;1589if(!ofs ||MSB(ofs,7)) {1590error(_("delta base offset overflow in pack for%s"),1591oid_to_hex(&entry->idx.oid));1592goto give_up;1593}1594 c = buf[used_0++];1595 ofs = (ofs <<7) + (c &127);1596}1597 ofs = entry->in_pack_offset - ofs;1598if(ofs <=0|| ofs >= entry->in_pack_offset) {1599error(_("delta base offset out of bound for%s"),1600oid_to_hex(&entry->idx.oid));1601goto give_up;1602}1603if(reuse_delta && !entry->preferred_base) {1604struct revindex_entry *revidx;1605 revidx =find_pack_revindex(p, ofs);1606if(!revidx)1607goto give_up;1608 base_ref =nth_packed_object_sha1(p, revidx->nr);1609}1610 entry->in_pack_header_size = used + used_0;1611break;1612}16131614if(can_reuse_delta(base_ref, entry, &base_entry)) {1615oe_set_type(entry, entry->in_pack_type);1616SET_SIZE(entry, in_pack_size);/* delta size */1617SET_DELTA_SIZE(entry, in_pack_size);16181619if(base_entry) {1620SET_DELTA(entry, base_entry);1621 entry->delta_sibling_idx = base_entry->delta_child_idx;1622SET_DELTA_CHILD(base_entry, entry);1623}else{1624SET_DELTA_EXT(entry, base_ref);1625}16261627unuse_pack(&w_curs);1628return;1629}16301631if(oe_type(entry)) {1632 off_t delta_pos;16331634/*1635 * This must be a delta and we already know what the1636 * final object type is. Let's extract the actual1637 * object size from the delta header.1638 */1639 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;1640 canonical_size =get_size_from_delta(p, &w_curs, delta_pos);1641if(canonical_size ==0)1642goto give_up;1643SET_SIZE(entry, canonical_size);1644unuse_pack(&w_curs);1645return;1646}16471648/*1649 * No choice but to fall back to the recursive delta walk1650 * with oid_object_info() to find about the object type1651 * at this point...1652 */1653 give_up:1654unuse_pack(&w_curs);1655}16561657oe_set_type(entry,1658oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1659if(entry->type_valid) {1660SET_SIZE(entry, canonical_size);1661}else{1662/*1663 * Bad object type is checked in prepare_pack(). This is1664 * to permit a missing preferred base object to be ignored1665 * as a preferred base. Doing so can result in a larger1666 * pack file, but the transfer will still take place.1667 */1668}1669}16701671static intpack_offset_sort(const void*_a,const void*_b)1672{1673const struct object_entry *a = *(struct object_entry **)_a;1674const struct object_entry *b = *(struct object_entry **)_b;1675const struct packed_git *a_in_pack =IN_PACK(a);1676const struct packed_git *b_in_pack =IN_PACK(b);16771678/* avoid filesystem trashing with loose objects */1679if(!a_in_pack && !b_in_pack)1680returnoidcmp(&a->idx.oid, &b->idx.oid);16811682if(a_in_pack < b_in_pack)1683return-1;1684if(a_in_pack > b_in_pack)1685return1;1686return a->in_pack_offset < b->in_pack_offset ? -1:1687(a->in_pack_offset > b->in_pack_offset);1688}16891690/*1691 * Drop an on-disk delta we were planning to reuse. Naively, this would1692 * just involve blanking out the "delta" field, but we have to deal1693 * with some extra book-keeping:1694 *1695 * 1. Removing ourselves from the delta_sibling linked list.1696 *1697 * 2. Updating our size/type to the non-delta representation. These were1698 * either not recorded initially (size) or overwritten with the delta type1699 * (type) when check_object() decided to reuse the delta.1700 *1701 * 3. Resetting our delta depth, as we are now a base object.1702 */1703static voiddrop_reused_delta(struct object_entry *entry)1704{1705unsigned*idx = &to_pack.objects[entry->delta_idx -1].delta_child_idx;1706struct object_info oi = OBJECT_INFO_INIT;1707enum object_type type;1708unsigned long size;17091710while(*idx) {1711struct object_entry *oe = &to_pack.objects[*idx -1];17121713if(oe == entry)1714*idx = oe->delta_sibling_idx;1715else1716 idx = &oe->delta_sibling_idx;1717}1718SET_DELTA(entry, NULL);1719 entry->depth =0;17201721 oi.sizep = &size;1722 oi.typep = &type;1723if(packed_object_info(the_repository,IN_PACK(entry), entry->in_pack_offset, &oi) <0) {1724/*1725 * We failed to get the info from this pack for some reason;1726 * fall back to oid_object_info, which may find another copy.1727 * And if that fails, the error will be recorded in oe_type(entry)1728 * and dealt with in prepare_pack().1729 */1730oe_set_type(entry,1731oid_object_info(the_repository, &entry->idx.oid, &size));1732}else{1733oe_set_type(entry, type);1734}1735SET_SIZE(entry, size);1736}17371738/*1739 * Follow the chain of deltas from this entry onward, throwing away any links1740 * that cause us to hit a cycle (as determined by the DFS state flags in1741 * the entries).1742 *1743 * We also detect too-long reused chains that would violate our --depth1744 * limit.1745 */1746static voidbreak_delta_chains(struct object_entry *entry)1747{1748/*1749 * The actual depth of each object we will write is stored as an int,1750 * as it cannot exceed our int "depth" limit. But before we break1751 * changes based no that limit, we may potentially go as deep as the1752 * number of objects, which is elsewhere bounded to a uint32_t.1753 */1754uint32_t total_depth;1755struct object_entry *cur, *next;17561757for(cur = entry, total_depth =0;1758 cur;1759 cur =DELTA(cur), total_depth++) {1760if(cur->dfs_state == DFS_DONE) {1761/*1762 * We've already seen this object and know it isn't1763 * part of a cycle. We do need to append its depth1764 * to our count.1765 */1766 total_depth += cur->depth;1767break;1768}17691770/*1771 * We break cycles before looping, so an ACTIVE state (or any1772 * other cruft which made its way into the state variable)1773 * is a bug.1774 */1775if(cur->dfs_state != DFS_NONE)1776BUG("confusing delta dfs state in first pass:%d",1777 cur->dfs_state);17781779/*1780 * Now we know this is the first time we've seen the object. If1781 * it's not a delta, we're done traversing, but we'll mark it1782 * done to save time on future traversals.1783 */1784if(!DELTA(cur)) {1785 cur->dfs_state = DFS_DONE;1786break;1787}17881789/*1790 * Mark ourselves as active and see if the next step causes1791 * us to cycle to another active object. It's important to do1792 * this _before_ we loop, because it impacts where we make the1793 * cut, and thus how our total_depth counter works.1794 * E.g., We may see a partial loop like:1795 *1796 * A -> B -> C -> D -> B1797 *1798 * Cutting B->C breaks the cycle. But now the depth of A is1799 * only 1, and our total_depth counter is at 3. The size of the1800 * error is always one less than the size of the cycle we1801 * broke. Commits C and D were "lost" from A's chain.1802 *1803 * If we instead cut D->B, then the depth of A is correct at 3.1804 * We keep all commits in the chain that we examined.1805 */1806 cur->dfs_state = DFS_ACTIVE;1807if(DELTA(cur)->dfs_state == DFS_ACTIVE) {1808drop_reused_delta(cur);1809 cur->dfs_state = DFS_DONE;1810break;1811}1812}18131814/*1815 * And now that we've gone all the way to the bottom of the chain, we1816 * need to clear the active flags and set the depth fields as1817 * appropriate. Unlike the loop above, which can quit when it drops a1818 * delta, we need to keep going to look for more depth cuts. So we need1819 * an extra "next" pointer to keep going after we reset cur->delta.1820 */1821for(cur = entry; cur; cur = next) {1822 next =DELTA(cur);18231824/*1825 * We should have a chain of zero or more ACTIVE states down to1826 * a final DONE. We can quit after the DONE, because either it1827 * has no bases, or we've already handled them in a previous1828 * call.1829 */1830if(cur->dfs_state == DFS_DONE)1831break;1832else if(cur->dfs_state != DFS_ACTIVE)1833BUG("confusing delta dfs state in second pass:%d",1834 cur->dfs_state);18351836/*1837 * If the total_depth is more than depth, then we need to snip1838 * the chain into two or more smaller chains that don't exceed1839 * the maximum depth. Most of the resulting chains will contain1840 * (depth + 1) entries (i.e., depth deltas plus one base), and1841 * the last chain (i.e., the one containing entry) will contain1842 * whatever entries are left over, namely1843 * (total_depth % (depth + 1)) of them.1844 *1845 * Since we are iterating towards decreasing depth, we need to1846 * decrement total_depth as we go, and we need to write to the1847 * entry what its final depth will be after all of the1848 * snipping. Since we're snipping into chains of length (depth1849 * + 1) entries, the final depth of an entry will be its1850 * original depth modulo (depth + 1). Any time we encounter an1851 * entry whose final depth is supposed to be zero, we snip it1852 * from its delta base, thereby making it so.1853 */1854 cur->depth = (total_depth--) % (depth +1);1855if(!cur->depth)1856drop_reused_delta(cur);18571858 cur->dfs_state = DFS_DONE;1859}1860}18611862static voidget_object_details(void)1863{1864uint32_t i;1865struct object_entry **sorted_by_offset;18661867if(progress)1868 progress_state =start_progress(_("Counting objects"),1869 to_pack.nr_objects);18701871 sorted_by_offset =xcalloc(to_pack.nr_objects,sizeof(struct object_entry *));1872for(i =0; i < to_pack.nr_objects; i++)1873 sorted_by_offset[i] = to_pack.objects + i;1874QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);18751876for(i =0; i < to_pack.nr_objects; i++) {1877struct object_entry *entry = sorted_by_offset[i];1878check_object(entry);1879if(entry->type_valid &&1880oe_size_greater_than(&to_pack, entry, big_file_threshold))1881 entry->no_try_delta =1;1882display_progress(progress_state, i +1);1883}1884stop_progress(&progress_state);18851886/*1887 * This must happen in a second pass, since we rely on the delta1888 * information for the whole list being completed.1889 */1890for(i =0; i < to_pack.nr_objects; i++)1891break_delta_chains(&to_pack.objects[i]);18921893free(sorted_by_offset);1894}18951896/*1897 * We search for deltas in a list sorted by type, by filename hash, and then1898 * by size, so that we see progressively smaller and smaller files.1899 * That's because we prefer deltas to be from the bigger file1900 * to the smaller -- deletes are potentially cheaper, but perhaps1901 * more importantly, the bigger file is likely the more recent1902 * one. The deepest deltas are therefore the oldest objects which are1903 * less susceptible to be accessed often.1904 */1905static inttype_size_sort(const void*_a,const void*_b)1906{1907const struct object_entry *a = *(struct object_entry **)_a;1908const struct object_entry *b = *(struct object_entry **)_b;1909const enum object_type a_type =oe_type(a);1910const enum object_type b_type =oe_type(b);1911const unsigned long a_size =SIZE(a);1912const unsigned long b_size =SIZE(b);19131914if(a_type > b_type)1915return-1;1916if(a_type < b_type)1917return1;1918if(a->hash > b->hash)1919return-1;1920if(a->hash < b->hash)1921return1;1922if(a->preferred_base > b->preferred_base)1923return-1;1924if(a->preferred_base < b->preferred_base)1925return1;1926if(use_delta_islands) {1927const int island_cmp =island_delta_cmp(&a->idx.oid, &b->idx.oid);1928if(island_cmp)1929return island_cmp;1930}1931if(a_size > b_size)1932return-1;1933if(a_size < b_size)1934return1;1935return a < b ? -1: (a > b);/* newest first */1936}19371938struct unpacked {1939struct object_entry *entry;1940void*data;1941struct delta_index *index;1942unsigned depth;1943};19441945static intdelta_cacheable(unsigned long src_size,unsigned long trg_size,1946unsigned long delta_size)1947{1948if(max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1949return0;19501951if(delta_size < cache_max_small_delta_size)1952return1;19531954/* cache delta, if objects are large enough compared to delta size */1955if((src_size >>20) + (trg_size >>21) > (delta_size >>10))1956return1;19571958return0;1959}19601961/* Protect delta_cache_size */1962static pthread_mutex_t cache_mutex;1963#define cache_lock() pthread_mutex_lock(&cache_mutex)1964#define cache_unlock() pthread_mutex_unlock(&cache_mutex)19651966/*1967 * Protect object list partitioning (e.g. struct thread_param) and1968 * progress_state1969 */1970static pthread_mutex_t progress_mutex;1971#define progress_lock() pthread_mutex_lock(&progress_mutex)1972#define progress_unlock() pthread_mutex_unlock(&progress_mutex)19731974/*1975 * Access to struct object_entry is unprotected since each thread owns1976 * a portion of the main object list. Just don't access object entries1977 * ahead in the list because they can be stolen and would need1978 * progress_mutex for protection.1979 */19801981/*1982 * Return the size of the object without doing any delta1983 * reconstruction (so non-deltas are true object sizes, but deltas1984 * return the size of the delta data).1985 */1986unsigned longoe_get_size_slow(struct packing_data *pack,1987const struct object_entry *e)1988{1989struct packed_git *p;1990struct pack_window *w_curs;1991unsigned char*buf;1992enum object_type type;1993unsigned long used, avail, size;19941995if(e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {1996packing_data_lock(&to_pack);1997if(oid_object_info(the_repository, &e->idx.oid, &size) <0)1998die(_("unable to get size of%s"),1999oid_to_hex(&e->idx.oid));2000packing_data_unlock(&to_pack);2001return size;2002}20032004 p =oe_in_pack(pack, e);2005if(!p)2006BUG("when e->type is a delta, it must belong to a pack");20072008packing_data_lock(&to_pack);2009 w_curs = NULL;2010 buf =use_pack(p, &w_curs, e->in_pack_offset, &avail);2011 used =unpack_object_header_buffer(buf, avail, &type, &size);2012if(used ==0)2013die(_("unable to parse object header of%s"),2014oid_to_hex(&e->idx.oid));20152016unuse_pack(&w_curs);2017packing_data_unlock(&to_pack);2018return size;2019}20202021static inttry_delta(struct unpacked *trg,struct unpacked *src,2022unsigned max_depth,unsigned long*mem_usage)2023{2024struct object_entry *trg_entry = trg->entry;2025struct object_entry *src_entry = src->entry;2026unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;2027unsigned ref_depth;2028enum object_type type;2029void*delta_buf;20302031/* Don't bother doing diffs between different types */2032if(oe_type(trg_entry) !=oe_type(src_entry))2033return-1;20342035/*2036 * We do not bother to try a delta that we discarded on an2037 * earlier try, but only when reusing delta data. Note that2038 * src_entry that is marked as the preferred_base should always2039 * be considered, as even if we produce a suboptimal delta against2040 * it, we will still save the transfer cost, as we already know2041 * the other side has it and we won't send src_entry at all.2042 */2043if(reuse_delta &&IN_PACK(trg_entry) &&2044IN_PACK(trg_entry) ==IN_PACK(src_entry) &&2045!src_entry->preferred_base &&2046 trg_entry->in_pack_type != OBJ_REF_DELTA &&2047 trg_entry->in_pack_type != OBJ_OFS_DELTA)2048return0;20492050/* Let's not bust the allowed depth. */2051if(src->depth >= max_depth)2052return0;20532054/* Now some size filtering heuristics. */2055 trg_size =SIZE(trg_entry);2056if(!DELTA(trg_entry)) {2057 max_size = trg_size/2- the_hash_algo->rawsz;2058 ref_depth =1;2059}else{2060 max_size =DELTA_SIZE(trg_entry);2061 ref_depth = trg->depth;2062}2063 max_size = (uint64_t)max_size * (max_depth - src->depth) /2064(max_depth - ref_depth +1);2065if(max_size ==0)2066return0;2067 src_size =SIZE(src_entry);2068 sizediff = src_size < trg_size ? trg_size - src_size :0;2069if(sizediff >= max_size)2070return0;2071if(trg_size < src_size /32)2072return0;20732074if(!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))2075return0;20762077/* Load data if not already done */2078if(!trg->data) {2079packing_data_lock(&to_pack);2080 trg->data =read_object_file(&trg_entry->idx.oid, &type, &sz);2081packing_data_unlock(&to_pack);2082if(!trg->data)2083die(_("object%scannot be read"),2084oid_to_hex(&trg_entry->idx.oid));2085if(sz != trg_size)2086die(_("object%sinconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2087oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,2088(uintmax_t)trg_size);2089*mem_usage += sz;2090}2091if(!src->data) {2092packing_data_lock(&to_pack);2093 src->data =read_object_file(&src_entry->idx.oid, &type, &sz);2094packing_data_unlock(&to_pack);2095if(!src->data) {2096if(src_entry->preferred_base) {2097static int warned =0;2098if(!warned++)2099warning(_("object%scannot be read"),2100oid_to_hex(&src_entry->idx.oid));2101/*2102 * Those objects are not included in the2103 * resulting pack. Be resilient and ignore2104 * them if they can't be read, in case the2105 * pack could be created nevertheless.2106 */2107return0;2108}2109die(_("object%scannot be read"),2110oid_to_hex(&src_entry->idx.oid));2111}2112if(sz != src_size)2113die(_("object%sinconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2114oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,2115(uintmax_t)src_size);2116*mem_usage += sz;2117}2118if(!src->index) {2119 src->index =create_delta_index(src->data, src_size);2120if(!src->index) {2121static int warned =0;2122if(!warned++)2123warning(_("suboptimal pack - out of memory"));2124return0;2125}2126*mem_usage +=sizeof_delta_index(src->index);2127}21282129 delta_buf =create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2130if(!delta_buf)2131return0;21322133if(DELTA(trg_entry)) {2134/* Prefer only shallower same-sized deltas. */2135if(delta_size ==DELTA_SIZE(trg_entry) &&2136 src->depth +1>= trg->depth) {2137free(delta_buf);2138return0;2139}2140}21412142/*2143 * Handle memory allocation outside of the cache2144 * accounting lock. Compiler will optimize the strangeness2145 * away when NO_PTHREADS is defined.2146 */2147free(trg_entry->delta_data);2148cache_lock();2149if(trg_entry->delta_data) {2150 delta_cache_size -=DELTA_SIZE(trg_entry);2151 trg_entry->delta_data = NULL;2152}2153if(delta_cacheable(src_size, trg_size, delta_size)) {2154 delta_cache_size += delta_size;2155cache_unlock();2156 trg_entry->delta_data =xrealloc(delta_buf, delta_size);2157}else{2158cache_unlock();2159free(delta_buf);2160}21612162SET_DELTA(trg_entry, src_entry);2163SET_DELTA_SIZE(trg_entry, delta_size);2164 trg->depth = src->depth +1;21652166return1;2167}21682169static unsigned intcheck_delta_limit(struct object_entry *me,unsigned int n)2170{2171struct object_entry *child =DELTA_CHILD(me);2172unsigned int m = n;2173while(child) {2174const unsigned int c =check_delta_limit(child, n +1);2175if(m < c)2176 m = c;2177 child =DELTA_SIBLING(child);2178}2179return m;2180}21812182static unsigned longfree_unpacked(struct unpacked *n)2183{2184unsigned long freed_mem =sizeof_delta_index(n->index);2185free_delta_index(n->index);2186 n->index = NULL;2187if(n->data) {2188 freed_mem +=SIZE(n->entry);2189FREE_AND_NULL(n->data);2190}2191 n->entry = NULL;2192 n->depth =0;2193return freed_mem;2194}21952196static voidfind_deltas(struct object_entry **list,unsigned*list_size,2197int window,int depth,unsigned*processed)2198{2199uint32_t i, idx =0, count =0;2200struct unpacked *array;2201unsigned long mem_usage =0;22022203 array =xcalloc(window,sizeof(struct unpacked));22042205for(;;) {2206struct object_entry *entry;2207struct unpacked *n = array + idx;2208int j, max_depth, best_base = -1;22092210progress_lock();2211if(!*list_size) {2212progress_unlock();2213break;2214}2215 entry = *list++;2216(*list_size)--;2217if(!entry->preferred_base) {2218(*processed)++;2219display_progress(progress_state, *processed);2220}2221progress_unlock();22222223 mem_usage -=free_unpacked(n);2224 n->entry = entry;22252226while(window_memory_limit &&2227 mem_usage > window_memory_limit &&2228 count >1) {2229const uint32_t tail = (idx + window - count) % window;2230 mem_usage -=free_unpacked(array + tail);2231 count--;2232}22332234/* We do not compute delta to *create* objects we are not2235 * going to pack.2236 */2237if(entry->preferred_base)2238goto next;22392240/*2241 * If the current object is at pack edge, take the depth the2242 * objects that depend on the current object into account2243 * otherwise they would become too deep.2244 */2245 max_depth = depth;2246if(DELTA_CHILD(entry)) {2247 max_depth -=check_delta_limit(entry,0);2248if(max_depth <=0)2249goto next;2250}22512252 j = window;2253while(--j >0) {2254int ret;2255uint32_t other_idx = idx + j;2256struct unpacked *m;2257if(other_idx >= window)2258 other_idx -= window;2259 m = array + other_idx;2260if(!m->entry)2261break;2262 ret =try_delta(n, m, max_depth, &mem_usage);2263if(ret <0)2264break;2265else if(ret >0)2266 best_base = other_idx;2267}22682269/*2270 * If we decided to cache the delta data, then it is best2271 * to compress it right away. First because we have to do2272 * it anyway, and doing it here while we're threaded will2273 * save a lot of time in the non threaded write phase,2274 * as well as allow for caching more deltas within2275 * the same cache size limit.2276 * ...2277 * But only if not writing to stdout, since in that case2278 * the network is most likely throttling writes anyway,2279 * and therefore it is best to go to the write phase ASAP2280 * instead, as we can afford spending more time compressing2281 * between writes at that moment.2282 */2283if(entry->delta_data && !pack_to_stdout) {2284unsigned long size;22852286 size =do_compress(&entry->delta_data,DELTA_SIZE(entry));2287if(size < (1U<< OE_Z_DELTA_BITS)) {2288 entry->z_delta_size = size;2289cache_lock();2290 delta_cache_size -=DELTA_SIZE(entry);2291 delta_cache_size += entry->z_delta_size;2292cache_unlock();2293}else{2294FREE_AND_NULL(entry->delta_data);2295 entry->z_delta_size =0;2296}2297}22982299/* if we made n a delta, and if n is already at max2300 * depth, leaving it in the window is pointless. we2301 * should evict it first.2302 */2303if(DELTA(entry) && max_depth <= n->depth)2304continue;23052306/*2307 * Move the best delta base up in the window, after the2308 * currently deltified object, to keep it longer. It will2309 * be the first base object to be attempted next.2310 */2311if(DELTA(entry)) {2312struct unpacked swap = array[best_base];2313int dist = (window + idx - best_base) % window;2314int dst = best_base;2315while(dist--) {2316int src = (dst +1) % window;2317 array[dst] = array[src];2318 dst = src;2319}2320 array[dst] = swap;2321}23222323 next:2324 idx++;2325if(count +1< window)2326 count++;2327if(idx >= window)2328 idx =0;2329}23302331for(i =0; i < window; ++i) {2332free_delta_index(array[i].index);2333free(array[i].data);2334}2335free(array);2336}23372338static voidtry_to_free_from_threads(size_t size)2339{2340packing_data_lock(&to_pack);2341release_pack_memory(size);2342packing_data_unlock(&to_pack);2343}23442345static try_to_free_t old_try_to_free_routine;23462347/*2348 * The main object list is split into smaller lists, each is handed to2349 * one worker.2350 *2351 * The main thread waits on the condition that (at least) one of the workers2352 * has stopped working (which is indicated in the .working member of2353 * struct thread_params).2354 *2355 * When a work thread has completed its work, it sets .working to 0 and2356 * signals the main thread and waits on the condition that .data_ready2357 * becomes 1.2358 *2359 * The main thread steals half of the work from the worker that has2360 * most work left to hand it to the idle worker.2361 */23622363struct thread_params {2364 pthread_t thread;2365struct object_entry **list;2366unsigned list_size;2367unsigned remaining;2368int window;2369int depth;2370int working;2371int data_ready;2372 pthread_mutex_t mutex;2373 pthread_cond_t cond;2374unsigned*processed;2375};23762377static pthread_cond_t progress_cond;23782379/*2380 * Mutex and conditional variable can't be statically-initialized on Windows.2381 */2382static voidinit_threaded_search(void)2383{2384pthread_mutex_init(&cache_mutex, NULL);2385pthread_mutex_init(&progress_mutex, NULL);2386pthread_cond_init(&progress_cond, NULL);2387 old_try_to_free_routine =set_try_to_free_routine(try_to_free_from_threads);2388}23892390static voidcleanup_threaded_search(void)2391{2392set_try_to_free_routine(old_try_to_free_routine);2393pthread_cond_destroy(&progress_cond);2394pthread_mutex_destroy(&cache_mutex);2395pthread_mutex_destroy(&progress_mutex);2396}23972398static void*threaded_find_deltas(void*arg)2399{2400struct thread_params *me = arg;24012402progress_lock();2403while(me->remaining) {2404progress_unlock();24052406find_deltas(me->list, &me->remaining,2407 me->window, me->depth, me->processed);24082409progress_lock();2410 me->working =0;2411pthread_cond_signal(&progress_cond);2412progress_unlock();24132414/*2415 * We must not set ->data_ready before we wait on the2416 * condition because the main thread may have set it to 12417 * before we get here. In order to be sure that new2418 * work is available if we see 1 in ->data_ready, it2419 * was initialized to 0 before this thread was spawned2420 * and we reset it to 0 right away.2421 */2422pthread_mutex_lock(&me->mutex);2423while(!me->data_ready)2424pthread_cond_wait(&me->cond, &me->mutex);2425 me->data_ready =0;2426pthread_mutex_unlock(&me->mutex);24272428progress_lock();2429}2430progress_unlock();2431/* leave ->working 1 so that this doesn't get more work assigned */2432return NULL;2433}24342435static voidll_find_deltas(struct object_entry **list,unsigned list_size,2436int window,int depth,unsigned*processed)2437{2438struct thread_params *p;2439int i, ret, active_threads =0;24402441init_threaded_search();24422443if(delta_search_threads <=1) {2444find_deltas(list, &list_size, window, depth, processed);2445cleanup_threaded_search();2446return;2447}2448if(progress > pack_to_stdout)2449fprintf_ln(stderr,_("Delta compression using up to%dthreads"),2450 delta_search_threads);2451 p =xcalloc(delta_search_threads,sizeof(*p));24522453/* Partition the work amongst work threads. */2454for(i =0; i < delta_search_threads; i++) {2455unsigned sub_size = list_size / (delta_search_threads - i);24562457/* don't use too small segments or no deltas will be found */2458if(sub_size <2*window && i+1< delta_search_threads)2459 sub_size =0;24602461 p[i].window = window;2462 p[i].depth = depth;2463 p[i].processed = processed;2464 p[i].working =1;2465 p[i].data_ready =0;24662467/* try to split chunks on "path" boundaries */2468while(sub_size && sub_size < list_size &&2469 list[sub_size]->hash &&2470 list[sub_size]->hash == list[sub_size-1]->hash)2471 sub_size++;24722473 p[i].list = list;2474 p[i].list_size = sub_size;2475 p[i].remaining = sub_size;24762477 list += sub_size;2478 list_size -= sub_size;2479}24802481/* Start work threads. */2482for(i =0; i < delta_search_threads; i++) {2483if(!p[i].list_size)2484continue;2485pthread_mutex_init(&p[i].mutex, NULL);2486pthread_cond_init(&p[i].cond, NULL);2487 ret =pthread_create(&p[i].thread, NULL,2488 threaded_find_deltas, &p[i]);2489if(ret)2490die(_("unable to create thread:%s"),strerror(ret));2491 active_threads++;2492}24932494/*2495 * Now let's wait for work completion. Each time a thread is done2496 * with its work, we steal half of the remaining work from the2497 * thread with the largest number of unprocessed objects and give2498 * it to that newly idle thread. This ensure good load balancing2499 * until the remaining object list segments are simply too short2500 * to be worth splitting anymore.2501 */2502while(active_threads) {2503struct thread_params *target = NULL;2504struct thread_params *victim = NULL;2505unsigned sub_size =0;25062507progress_lock();2508for(;;) {2509for(i =0; !target && i < delta_search_threads; i++)2510if(!p[i].working)2511 target = &p[i];2512if(target)2513break;2514pthread_cond_wait(&progress_cond, &progress_mutex);2515}25162517for(i =0; i < delta_search_threads; i++)2518if(p[i].remaining >2*window &&2519(!victim || victim->remaining < p[i].remaining))2520 victim = &p[i];2521if(victim) {2522 sub_size = victim->remaining /2;2523 list = victim->list + victim->list_size - sub_size;2524while(sub_size && list[0]->hash &&2525 list[0]->hash == list[-1]->hash) {2526 list++;2527 sub_size--;2528}2529if(!sub_size) {2530/*2531 * It is possible for some "paths" to have2532 * so many objects that no hash boundary2533 * might be found. Let's just steal the2534 * exact half in that case.2535 */2536 sub_size = victim->remaining /2;2537 list -= sub_size;2538}2539 target->list = list;2540 victim->list_size -= sub_size;2541 victim->remaining -= sub_size;2542}2543 target->list_size = sub_size;2544 target->remaining = sub_size;2545 target->working =1;2546progress_unlock();25472548pthread_mutex_lock(&target->mutex);2549 target->data_ready =1;2550pthread_cond_signal(&target->cond);2551pthread_mutex_unlock(&target->mutex);25522553if(!sub_size) {2554pthread_join(target->thread, NULL);2555pthread_cond_destroy(&target->cond);2556pthread_mutex_destroy(&target->mutex);2557 active_threads--;2558}2559}2560cleanup_threaded_search();2561free(p);2562}25632564static voidadd_tag_chain(const struct object_id *oid)2565{2566struct tag *tag;25672568/*2569 * We catch duplicates already in add_object_entry(), but we'd2570 * prefer to do this extra check to avoid having to parse the2571 * tag at all if we already know that it's being packed (e.g., if2572 * it was included via bitmaps, we would not have parsed it2573 * previously).2574 */2575if(packlist_find(&to_pack, oid, NULL))2576return;25772578 tag =lookup_tag(the_repository, oid);2579while(1) {2580if(!tag ||parse_tag(tag) || !tag->tagged)2581die(_("unable to pack objects reachable from tag%s"),2582oid_to_hex(oid));25832584add_object_entry(&tag->object.oid, OBJ_TAG, NULL,0);25852586if(tag->tagged->type != OBJ_TAG)2587return;25882589 tag = (struct tag *)tag->tagged;2590}2591}25922593static intadd_ref_tag(const char*path,const struct object_id *oid,int flag,void*cb_data)2594{2595struct object_id peeled;25962597if(starts_with(path,"refs/tags/") &&/* is a tag? */2598!peel_ref(path, &peeled) &&/* peelable? */2599packlist_find(&to_pack, &peeled, NULL))/* object packed? */2600add_tag_chain(oid);2601return0;2602}26032604static voidprepare_pack(int window,int depth)2605{2606struct object_entry **delta_list;2607uint32_t i, nr_deltas;2608unsigned n;26092610if(use_delta_islands)2611resolve_tree_islands(the_repository, progress, &to_pack);26122613get_object_details();26142615/*2616 * If we're locally repacking then we need to be doubly careful2617 * from now on in order to make sure no stealth corruption gets2618 * propagated to the new pack. Clients receiving streamed packs2619 * should validate everything they get anyway so no need to incur2620 * the additional cost here in that case.2621 */2622if(!pack_to_stdout)2623 do_check_packed_object_crc =1;26242625if(!to_pack.nr_objects || !window || !depth)2626return;26272628ALLOC_ARRAY(delta_list, to_pack.nr_objects);2629 nr_deltas = n =0;26302631for(i =0; i < to_pack.nr_objects; i++) {2632struct object_entry *entry = to_pack.objects + i;26332634if(DELTA(entry))2635/* This happens if we decided to reuse existing2636 * delta from a pack. "reuse_delta &&" is implied.2637 */2638continue;26392640if(!entry->type_valid ||2641oe_size_less_than(&to_pack, entry,50))2642continue;26432644if(entry->no_try_delta)2645continue;26462647if(!entry->preferred_base) {2648 nr_deltas++;2649if(oe_type(entry) <0)2650die(_("unable to get type of object%s"),2651oid_to_hex(&entry->idx.oid));2652}else{2653if(oe_type(entry) <0) {2654/*2655 * This object is not found, but we2656 * don't have to include it anyway.2657 */2658continue;2659}2660}26612662 delta_list[n++] = entry;2663}26642665if(nr_deltas && n >1) {2666unsigned nr_done =0;2667if(progress)2668 progress_state =start_progress(_("Compressing objects"),2669 nr_deltas);2670QSORT(delta_list, n, type_size_sort);2671ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2672stop_progress(&progress_state);2673if(nr_done != nr_deltas)2674die(_("inconsistency with delta count"));2675}2676free(delta_list);2677}26782679static intgit_pack_config(const char*k,const char*v,void*cb)2680{2681if(!strcmp(k,"pack.window")) {2682 window =git_config_int(k, v);2683return0;2684}2685if(!strcmp(k,"pack.windowmemory")) {2686 window_memory_limit =git_config_ulong(k, v);2687return0;2688}2689if(!strcmp(k,"pack.depth")) {2690 depth =git_config_int(k, v);2691return0;2692}2693if(!strcmp(k,"pack.deltacachesize")) {2694 max_delta_cache_size =git_config_int(k, v);2695return0;2696}2697if(!strcmp(k,"pack.deltacachelimit")) {2698 cache_max_small_delta_size =git_config_int(k, v);2699return0;2700}2701if(!strcmp(k,"pack.writebitmaphashcache")) {2702if(git_config_bool(k, v))2703 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2704else2705 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2706}2707if(!strcmp(k,"pack.usebitmaps")) {2708 use_bitmap_index_default =git_config_bool(k, v);2709return0;2710}2711if(!strcmp(k,"pack.usesparse")) {2712 sparse =git_config_bool(k, v);2713return0;2714}2715if(!strcmp(k,"pack.threads")) {2716 delta_search_threads =git_config_int(k, v);2717if(delta_search_threads <0)2718die(_("invalid number of threads specified (%d)"),2719 delta_search_threads);2720if(!HAVE_THREADS && delta_search_threads !=1) {2721warning(_("no threads support, ignoring%s"), k);2722 delta_search_threads =0;2723}2724return0;2725}2726if(!strcmp(k,"pack.indexversion")) {2727 pack_idx_opts.version =git_config_int(k, v);2728if(pack_idx_opts.version >2)2729die(_("bad pack.indexversion=%"PRIu32),2730 pack_idx_opts.version);2731return0;2732}2733returngit_default_config(k, v, cb);2734}27352736static voidread_object_list_from_stdin(void)2737{2738char line[GIT_MAX_HEXSZ +1+ PATH_MAX +2];2739struct object_id oid;2740const char*p;27412742for(;;) {2743if(!fgets(line,sizeof(line), stdin)) {2744if(feof(stdin))2745break;2746if(!ferror(stdin))2747die("BUG: fgets returned NULL, not EOF, not error!");2748if(errno != EINTR)2749die_errno("fgets");2750clearerr(stdin);2751continue;2752}2753if(line[0] =='-') {2754if(get_oid_hex(line+1, &oid))2755die(_("expected edge object ID, got garbage:\n%s"),2756 line);2757add_preferred_base(&oid);2758continue;2759}2760if(parse_oid_hex(line, &oid, &p))2761die(_("expected object ID, got garbage:\n%s"), line);27622763add_preferred_base_object(p +1);2764add_object_entry(&oid, OBJ_NONE, p +1,0);2765}2766}27672768/* Remember to update object flag allocation in object.h */2769#define OBJECT_ADDED (1u<<20)27702771static voidshow_commit(struct commit *commit,void*data)2772{2773add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL,0);2774 commit->object.flags |= OBJECT_ADDED;27752776if(write_bitmap_index)2777index_commit_for_bitmap(commit);27782779if(use_delta_islands)2780propagate_island_marks(commit);2781}27822783static voidshow_object(struct object *obj,const char*name,void*data)2784{2785add_preferred_base_object(name);2786add_object_entry(&obj->oid, obj->type, name,0);2787 obj->flags |= OBJECT_ADDED;27882789if(use_delta_islands) {2790const char*p;2791unsigned depth;2792struct object_entry *ent;27932794/* the empty string is a root tree, which is depth 0 */2795 depth = *name ?1:0;2796for(p =strchr(name,'/'); p; p =strchr(p +1,'/'))2797 depth++;27982799 ent =packlist_find(&to_pack, &obj->oid, NULL);2800if(ent && depth >oe_tree_depth(&to_pack, ent))2801oe_set_tree_depth(&to_pack, ent, depth);2802}2803}28042805static voidshow_object__ma_allow_any(struct object *obj,const char*name,void*data)2806{2807assert(arg_missing_action == MA_ALLOW_ANY);28082809/*2810 * Quietly ignore ALL missing objects. This avoids problems with2811 * staging them now and getting an odd error later.2812 */2813if(!has_object_file(&obj->oid))2814return;28152816show_object(obj, name, data);2817}28182819static voidshow_object__ma_allow_promisor(struct object *obj,const char*name,void*data)2820{2821assert(arg_missing_action == MA_ALLOW_PROMISOR);28222823/*2824 * Quietly ignore EXPECTED missing objects. This avoids problems with2825 * staging them now and getting an odd error later.2826 */2827if(!has_object_file(&obj->oid) &&is_promisor_object(&obj->oid))2828return;28292830show_object(obj, name, data);2831}28322833static intoption_parse_missing_action(const struct option *opt,2834const char*arg,int unset)2835{2836assert(arg);2837assert(!unset);28382839if(!strcmp(arg,"error")) {2840 arg_missing_action = MA_ERROR;2841 fn_show_object = show_object;2842return0;2843}28442845if(!strcmp(arg,"allow-any")) {2846 arg_missing_action = MA_ALLOW_ANY;2847 fetch_if_missing =0;2848 fn_show_object = show_object__ma_allow_any;2849return0;2850}28512852if(!strcmp(arg,"allow-promisor")) {2853 arg_missing_action = MA_ALLOW_PROMISOR;2854 fetch_if_missing =0;2855 fn_show_object = show_object__ma_allow_promisor;2856return0;2857}28582859die(_("invalid value for --missing"));2860return0;2861}28622863static voidshow_edge(struct commit *commit)2864{2865add_preferred_base(&commit->object.oid);2866}28672868struct in_pack_object {2869 off_t offset;2870struct object *object;2871};28722873struct in_pack {2874unsigned int alloc;2875unsigned int nr;2876struct in_pack_object *array;2877};28782879static voidmark_in_pack_object(struct object *object,struct packed_git *p,struct in_pack *in_pack)2880{2881 in_pack->array[in_pack->nr].offset =find_pack_entry_one(object->oid.hash, p);2882 in_pack->array[in_pack->nr].object = object;2883 in_pack->nr++;2884}28852886/*2887 * Compare the objects in the offset order, in order to emulate the2888 * "git rev-list --objects" output that produced the pack originally.2889 */2890static intofscmp(const void*a_,const void*b_)2891{2892struct in_pack_object *a = (struct in_pack_object *)a_;2893struct in_pack_object *b = (struct in_pack_object *)b_;28942895if(a->offset < b->offset)2896return-1;2897else if(a->offset > b->offset)2898return1;2899else2900returnoidcmp(&a->object->oid, &b->object->oid);2901}29022903static voidadd_objects_in_unpacked_packs(void)2904{2905struct packed_git *p;2906struct in_pack in_pack;2907uint32_t i;29082909memset(&in_pack,0,sizeof(in_pack));29102911for(p =get_all_packs(the_repository); p; p = p->next) {2912struct object_id oid;2913struct object *o;29142915if(!p->pack_local || p->pack_keep || p->pack_keep_in_core)2916continue;2917if(open_pack_index(p))2918die(_("cannot open pack index"));29192920ALLOC_GROW(in_pack.array,2921 in_pack.nr + p->num_objects,2922 in_pack.alloc);29232924for(i =0; i < p->num_objects; i++) {2925nth_packed_object_oid(&oid, p, i);2926 o =lookup_unknown_object(&oid);2927if(!(o->flags & OBJECT_ADDED))2928mark_in_pack_object(o, p, &in_pack);2929 o->flags |= OBJECT_ADDED;2930}2931}29322933if(in_pack.nr) {2934QSORT(in_pack.array, in_pack.nr, ofscmp);2935for(i =0; i < in_pack.nr; i++) {2936struct object *o = in_pack.array[i].object;2937add_object_entry(&o->oid, o->type,"",0);2938}2939}2940free(in_pack.array);2941}29422943static intadd_loose_object(const struct object_id *oid,const char*path,2944void*data)2945{2946enum object_type type =oid_object_info(the_repository, oid, NULL);29472948if(type <0) {2949warning(_("loose object at%scould not be examined"), path);2950return0;2951}29522953add_object_entry(oid, type,"",0);2954return0;2955}29562957/*2958 * We actually don't even have to worry about reachability here.2959 * add_object_entry will weed out duplicates, so we just add every2960 * loose object we find.2961 */2962static voidadd_unreachable_loose_objects(void)2963{2964for_each_loose_file_in_objdir(get_object_directory(),2965 add_loose_object,2966 NULL, NULL, NULL);2967}29682969static inthas_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2970{2971static struct packed_git *last_found = (void*)1;2972struct packed_git *p;29732974 p = (last_found != (void*)1) ? last_found :2975get_all_packs(the_repository);29762977while(p) {2978if((!p->pack_local || p->pack_keep ||2979 p->pack_keep_in_core) &&2980find_pack_entry_one(oid->hash, p)) {2981 last_found = p;2982return1;2983}2984if(p == last_found)2985 p =get_all_packs(the_repository);2986else2987 p = p->next;2988if(p == last_found)2989 p = p->next;2990}2991return0;2992}29932994/*2995 * Store a list of sha1s that are should not be discarded2996 * because they are either written too recently, or are2997 * reachable from another object that was.2998 *2999 * This is filled by get_object_list.3000 */3001static struct oid_array recent_objects;30023003static intloosened_object_can_be_discarded(const struct object_id *oid,3004 timestamp_t mtime)3005{3006if(!unpack_unreachable_expiration)3007return0;3008if(mtime > unpack_unreachable_expiration)3009return0;3010if(oid_array_lookup(&recent_objects, oid) >=0)3011return0;3012return1;3013}30143015static voidloosen_unused_packed_objects(void)3016{3017struct packed_git *p;3018uint32_t i;3019struct object_id oid;30203021for(p =get_all_packs(the_repository); p; p = p->next) {3022if(!p->pack_local || p->pack_keep || p->pack_keep_in_core)3023continue;30243025if(open_pack_index(p))3026die(_("cannot open pack index"));30273028for(i =0; i < p->num_objects; i++) {3029nth_packed_object_oid(&oid, p, i);3030if(!packlist_find(&to_pack, &oid, NULL) &&3031!has_sha1_pack_kept_or_nonlocal(&oid) &&3032!loosened_object_can_be_discarded(&oid, p->mtime))3033if(force_object_loose(&oid, p->mtime))3034die(_("unable to force loose object"));3035}3036}3037}30383039/*3040 * This tracks any options which pack-reuse code expects to be on, or which a3041 * reader of the pack might not understand, and which would therefore prevent3042 * blind reuse of what we have on disk.3043 */3044static intpack_options_allow_reuse(void)3045{3046return pack_to_stdout &&3047 allow_ofs_delta &&3048!ignore_packed_keep_on_disk &&3049!ignore_packed_keep_in_core &&3050(!local || !have_non_local_packs) &&3051!incremental;3052}30533054static intget_object_list_from_bitmap(struct rev_info *revs)3055{3056if(!(bitmap_git =prepare_bitmap_walk(revs)))3057return-1;30583059if(pack_options_allow_reuse() &&3060!reuse_partial_packfile_from_bitmap(3061 bitmap_git,3062&reuse_packfile,3063&reuse_packfile_objects,3064&reuse_packfile_offset)) {3065assert(reuse_packfile_objects);3066 nr_result += reuse_packfile_objects;3067display_progress(progress_state, nr_result);3068}30693070traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);3071return0;3072}30733074static voidrecord_recent_object(struct object *obj,3075const char*name,3076void*data)3077{3078oid_array_append(&recent_objects, &obj->oid);3079}30803081static voidrecord_recent_commit(struct commit *commit,void*data)3082{3083oid_array_append(&recent_objects, &commit->object.oid);3084}30853086static voidget_object_list(int ac,const char**av)3087{3088struct rev_info revs;3089struct setup_revision_opt s_r_opt = {3090.allow_exclude_promisor_objects =1,3091};3092char line[1000];3093int flags =0;3094int save_warning;30953096repo_init_revisions(the_repository, &revs, NULL);3097 save_commit_buffer =0;3098setup_revisions(ac, av, &revs, &s_r_opt);30993100/* make sure shallows are read */3101is_repository_shallow(the_repository);31023103 save_warning = warn_on_object_refname_ambiguity;3104 warn_on_object_refname_ambiguity =0;31053106while(fgets(line,sizeof(line), stdin) != NULL) {3107int len =strlen(line);3108if(len && line[len -1] =='\n')3109 line[--len] =0;3110if(!len)3111break;3112if(*line =='-') {3113if(!strcmp(line,"--not")) {3114 flags ^= UNINTERESTING;3115 write_bitmap_index =0;3116continue;3117}3118if(starts_with(line,"--shallow ")) {3119struct object_id oid;3120if(get_oid_hex(line +10, &oid))3121die("not an SHA-1 '%s'", line +10);3122register_shallow(the_repository, &oid);3123 use_bitmap_index =0;3124continue;3125}3126die(_("not a rev '%s'"), line);3127}3128if(handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3129die(_("bad revision '%s'"), line);3130}31313132 warn_on_object_refname_ambiguity = save_warning;31333134if(use_bitmap_index && !get_object_list_from_bitmap(&revs))3135return;31363137if(use_delta_islands)3138load_delta_islands(the_repository, progress);31393140if(prepare_revision_walk(&revs))3141die(_("revision walk setup failed"));3142mark_edges_uninteresting(&revs, show_edge, sparse);31433144if(!fn_show_object)3145 fn_show_object = show_object;3146traverse_commit_list_filtered(&filter_options, &revs,3147 show_commit, fn_show_object, NULL,3148 NULL);31493150if(unpack_unreachable_expiration) {3151 revs.ignore_missing_links =1;3152if(add_unseen_recent_objects_to_traversal(&revs,3153 unpack_unreachable_expiration))3154die(_("unable to add recent objects"));3155if(prepare_revision_walk(&revs))3156die(_("revision walk setup failed"));3157traverse_commit_list(&revs, record_recent_commit,3158 record_recent_object, NULL);3159}31603161if(keep_unreachable)3162add_objects_in_unpacked_packs();3163if(pack_loose_unreachable)3164add_unreachable_loose_objects();3165if(unpack_unreachable)3166loosen_unused_packed_objects();31673168oid_array_clear(&recent_objects);3169}31703171static voidadd_extra_kept_packs(const struct string_list *names)3172{3173struct packed_git *p;31743175if(!names->nr)3176return;31773178for(p =get_all_packs(the_repository); p; p = p->next) {3179const char*name =basename(p->pack_name);3180int i;31813182if(!p->pack_local)3183continue;31843185for(i =0; i < names->nr; i++)3186if(!fspathcmp(name, names->items[i].string))3187break;31883189if(i < names->nr) {3190 p->pack_keep_in_core =1;3191 ignore_packed_keep_in_core =1;3192continue;3193}3194}3195}31963197static intoption_parse_index_version(const struct option *opt,3198const char*arg,int unset)3199{3200char*c;3201const char*val = arg;32023203BUG_ON_OPT_NEG(unset);32043205 pack_idx_opts.version =strtoul(val, &c,10);3206if(pack_idx_opts.version >2)3207die(_("unsupported index version%s"), val);3208if(*c ==','&& c[1])3209 pack_idx_opts.off32_limit =strtoul(c+1, &c,0);3210if(*c || pack_idx_opts.off32_limit &0x80000000)3211die(_("bad index version '%s'"), val);3212return0;3213}32143215static intoption_parse_unpack_unreachable(const struct option *opt,3216const char*arg,int unset)3217{3218if(unset) {3219 unpack_unreachable =0;3220 unpack_unreachable_expiration =0;3221}3222else{3223 unpack_unreachable =1;3224if(arg)3225 unpack_unreachable_expiration =approxidate(arg);3226}3227return0;3228}32293230intcmd_pack_objects(int argc,const char**argv,const char*prefix)3231{3232int use_internal_rev_list =0;3233int shallow =0;3234int all_progress_implied =0;3235struct argv_array rp = ARGV_ARRAY_INIT;3236int rev_list_unpacked =0, rev_list_all =0, rev_list_reflog =0;3237int rev_list_index =0;3238struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3239struct option pack_objects_options[] = {3240OPT_SET_INT('q',"quiet", &progress,3241N_("do not show progress meter"),0),3242OPT_SET_INT(0,"progress", &progress,3243N_("show progress meter"),1),3244OPT_SET_INT(0,"all-progress", &progress,3245N_("show progress meter during object writing phase"),2),3246OPT_BOOL(0,"all-progress-implied",3247&all_progress_implied,3248N_("similar to --all-progress when progress meter is shown")),3249{ OPTION_CALLBACK,0,"index-version", NULL,N_("<version>[,<offset>]"),3250N_("write the pack index file in the specified idx format version"),3251 PARSE_OPT_NONEG, option_parse_index_version },3252OPT_MAGNITUDE(0,"max-pack-size", &pack_size_limit,3253N_("maximum size of each output pack file")),3254OPT_BOOL(0,"local", &local,3255N_("ignore borrowed objects from alternate object store")),3256OPT_BOOL(0,"incremental", &incremental,3257N_("ignore packed objects")),3258OPT_INTEGER(0,"window", &window,3259N_("limit pack window by objects")),3260OPT_MAGNITUDE(0,"window-memory", &window_memory_limit,3261N_("limit pack window by memory in addition to object limit")),3262OPT_INTEGER(0,"depth", &depth,3263N_("maximum length of delta chain allowed in the resulting pack")),3264OPT_BOOL(0,"reuse-delta", &reuse_delta,3265N_("reuse existing deltas")),3266OPT_BOOL(0,"reuse-object", &reuse_object,3267N_("reuse existing objects")),3268OPT_BOOL(0,"delta-base-offset", &allow_ofs_delta,3269N_("use OFS_DELTA objects")),3270OPT_INTEGER(0,"threads", &delta_search_threads,3271N_("use threads when searching for best delta matches")),3272OPT_BOOL(0,"non-empty", &non_empty,3273N_("do not create an empty pack output")),3274OPT_BOOL(0,"revs", &use_internal_rev_list,3275N_("read revision arguments from standard input")),3276OPT_SET_INT_F(0,"unpacked", &rev_list_unpacked,3277N_("limit the objects to those that are not yet packed"),32781, PARSE_OPT_NONEG),3279OPT_SET_INT_F(0,"all", &rev_list_all,3280N_("include objects reachable from any reference"),32811, PARSE_OPT_NONEG),3282OPT_SET_INT_F(0,"reflog", &rev_list_reflog,3283N_("include objects referred by reflog entries"),32841, PARSE_OPT_NONEG),3285OPT_SET_INT_F(0,"indexed-objects", &rev_list_index,3286N_("include objects referred to by the index"),32871, PARSE_OPT_NONEG),3288OPT_BOOL(0,"stdout", &pack_to_stdout,3289N_("output pack to stdout")),3290OPT_BOOL(0,"include-tag", &include_tag,3291N_("include tag objects that refer to objects to be packed")),3292OPT_BOOL(0,"keep-unreachable", &keep_unreachable,3293N_("keep unreachable objects")),3294OPT_BOOL(0,"pack-loose-unreachable", &pack_loose_unreachable,3295N_("pack loose unreachable objects")),3296{ OPTION_CALLBACK,0,"unpack-unreachable", NULL,N_("time"),3297N_("unpack unreachable objects newer than <time>"),3298 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3299OPT_BOOL(0,"sparse", &sparse,3300N_("use the sparse reachability algorithm")),3301OPT_BOOL(0,"thin", &thin,3302N_("create thin packs")),3303OPT_BOOL(0,"shallow", &shallow,3304N_("create packs suitable for shallow fetches")),3305OPT_BOOL(0,"honor-pack-keep", &ignore_packed_keep_on_disk,3306N_("ignore packs that have companion .keep file")),3307OPT_STRING_LIST(0,"keep-pack", &keep_pack_list,N_("name"),3308N_("ignore this pack")),3309OPT_INTEGER(0,"compression", &pack_compression_level,3310N_("pack compression level")),3311OPT_SET_INT(0,"keep-true-parents", &grafts_replace_parents,3312N_("do not hide commits by grafts"),0),3313OPT_BOOL(0,"use-bitmap-index", &use_bitmap_index,3314N_("use a bitmap index if available to speed up counting objects")),3315OPT_BOOL(0,"write-bitmap-index", &write_bitmap_index,3316N_("write a bitmap index together with the pack index")),3317OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3318{ OPTION_CALLBACK,0,"missing", NULL,N_("action"),3319N_("handling for missing objects"), PARSE_OPT_NONEG,3320 option_parse_missing_action },3321OPT_BOOL(0,"exclude-promisor-objects", &exclude_promisor_objects,3322N_("do not pack objects in promisor packfiles")),3323OPT_BOOL(0,"delta-islands", &use_delta_islands,3324N_("respect islands during delta compression")),3325OPT_END(),3326};33273328if(DFS_NUM_STATES > (1<< OE_DFS_STATE_BITS))3329BUG("too many dfs states, increase OE_DFS_STATE_BITS");33303331 read_replace_refs =0;33323333 sparse =git_env_bool("GIT_TEST_PACK_SPARSE",0);3334reset_pack_idx_option(&pack_idx_opts);3335git_config(git_pack_config, NULL);33363337 progress =isatty(2);3338 argc =parse_options(argc, argv, prefix, pack_objects_options,3339 pack_usage,0);33403341if(argc) {3342 base_name = argv[0];3343 argc--;3344}3345if(pack_to_stdout != !base_name || argc)3346usage_with_options(pack_usage, pack_objects_options);33473348if(depth >= (1<< OE_DEPTH_BITS)) {3349warning(_("delta chain depth%dis too deep, forcing%d"),3350 depth, (1<< OE_DEPTH_BITS) -1);3351 depth = (1<< OE_DEPTH_BITS) -1;3352}3353if(cache_max_small_delta_size >= (1U<< OE_Z_DELTA_BITS)) {3354warning(_("pack.deltaCacheLimit is too high, forcing%d"),3355(1U<< OE_Z_DELTA_BITS) -1);3356 cache_max_small_delta_size = (1U<< OE_Z_DELTA_BITS) -1;3357}33583359argv_array_push(&rp,"pack-objects");3360if(thin) {3361 use_internal_rev_list =1;3362argv_array_push(&rp, shallow3363?"--objects-edge-aggressive"3364:"--objects-edge");3365}else3366argv_array_push(&rp,"--objects");33673368if(rev_list_all) {3369 use_internal_rev_list =1;3370argv_array_push(&rp,"--all");3371}3372if(rev_list_reflog) {3373 use_internal_rev_list =1;3374argv_array_push(&rp,"--reflog");3375}3376if(rev_list_index) {3377 use_internal_rev_list =1;3378argv_array_push(&rp,"--indexed-objects");3379}3380if(rev_list_unpacked) {3381 use_internal_rev_list =1;3382argv_array_push(&rp,"--unpacked");3383}33843385if(exclude_promisor_objects) {3386 use_internal_rev_list =1;3387 fetch_if_missing =0;3388argv_array_push(&rp,"--exclude-promisor-objects");3389}3390if(unpack_unreachable || keep_unreachable || pack_loose_unreachable)3391 use_internal_rev_list =1;33923393if(!reuse_object)3394 reuse_delta =0;3395if(pack_compression_level == -1)3396 pack_compression_level = Z_DEFAULT_COMPRESSION;3397else if(pack_compression_level <0|| pack_compression_level > Z_BEST_COMPRESSION)3398die(_("bad pack compression level%d"), pack_compression_level);33993400if(!delta_search_threads)/* --threads=0 means autodetect */3401 delta_search_threads =online_cpus();34023403if(!HAVE_THREADS && delta_search_threads !=1)3404warning(_("no threads support, ignoring --threads"));3405if(!pack_to_stdout && !pack_size_limit)3406 pack_size_limit = pack_size_limit_cfg;3407if(pack_to_stdout && pack_size_limit)3408die(_("--max-pack-size cannot be used to build a pack for transfer"));3409if(pack_size_limit && pack_size_limit <1024*1024) {3410warning(_("minimum pack size limit is 1 MiB"));3411 pack_size_limit =1024*1024;3412}34133414if(!pack_to_stdout && thin)3415die(_("--thin cannot be used to build an indexable pack"));34163417if(keep_unreachable && unpack_unreachable)3418die(_("--keep-unreachable and --unpack-unreachable are incompatible"));3419if(!rev_list_all || !rev_list_reflog || !rev_list_index)3420 unpack_unreachable_expiration =0;34213422if(filter_options.choice) {3423if(!pack_to_stdout)3424die(_("cannot use --filter without --stdout"));3425 use_bitmap_index =0;3426}34273428/*3429 * "soft" reasons not to use bitmaps - for on-disk repack by default we want3430 *3431 * - to produce good pack (with bitmap index not-yet-packed objects are3432 * packed in suboptimal order).3433 *3434 * - to use more robust pack-generation codepath (avoiding possible3435 * bugs in bitmap code and possible bitmap index corruption).3436 */3437if(!pack_to_stdout)3438 use_bitmap_index_default =0;34393440if(use_bitmap_index <0)3441 use_bitmap_index = use_bitmap_index_default;34423443/* "hard" reasons not to use bitmaps; these just won't work at all */3444if(!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) ||is_repository_shallow(the_repository))3445 use_bitmap_index =0;34463447if(pack_to_stdout || !rev_list_all)3448 write_bitmap_index =0;34493450if(use_delta_islands)3451argv_array_push(&rp,"--topo-order");34523453if(progress && all_progress_implied)3454 progress =2;34553456add_extra_kept_packs(&keep_pack_list);3457if(ignore_packed_keep_on_disk) {3458struct packed_git *p;3459for(p =get_all_packs(the_repository); p; p = p->next)3460if(p->pack_local && p->pack_keep)3461break;3462if(!p)/* no keep-able packs found */3463 ignore_packed_keep_on_disk =0;3464}3465if(local) {3466/*3467 * unlike ignore_packed_keep_on_disk above, we do not3468 * want to unset "local" based on looking at packs, as3469 * it also covers non-local objects3470 */3471struct packed_git *p;3472for(p =get_all_packs(the_repository); p; p = p->next) {3473if(!p->pack_local) {3474 have_non_local_packs =1;3475break;3476}3477}3478}34793480trace2_region_enter("pack-objects","enumerate-objects",3481 the_repository);3482prepare_packing_data(the_repository, &to_pack);34833484if(progress)3485 progress_state =start_progress(_("Enumerating objects"),0);3486if(!use_internal_rev_list)3487read_object_list_from_stdin();3488else{3489get_object_list(rp.argc, rp.argv);3490argv_array_clear(&rp);3491}3492cleanup_preferred_base();3493if(include_tag && nr_result)3494for_each_ref(add_ref_tag, NULL);3495stop_progress(&progress_state);3496trace2_region_leave("pack-objects","enumerate-objects",3497 the_repository);34983499if(non_empty && !nr_result)3500return0;3501if(nr_result) {3502trace2_region_enter("pack-objects","prepare-pack",3503 the_repository);3504prepare_pack(window, depth);3505trace2_region_leave("pack-objects","prepare-pack",3506 the_repository);3507}35083509trace2_region_enter("pack-objects","write-pack-file", the_repository);3510write_pack_file();3511trace2_region_leave("pack-objects","write-pack-file", the_repository);35123513if(progress)3514fprintf_ln(stderr,3515_("Total %"PRIu32" (delta %"PRIu32"),"3516" reused %"PRIu32" (delta %"PRIu32")"),3517 written, written_delta, reused, reused_delta);3518return0;3519}