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 37#define IN_PACK(obj) oe_in_pack(&to_pack, obj) 38#define SIZE(obj) oe_size(&to_pack, obj) 39#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size) 40#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj) 41#define DELTA(obj) oe_delta(&to_pack, obj) 42#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj) 43#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj) 44#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val) 45#define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid) 46#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val) 47#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val) 48#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val) 49 50static const char*pack_usage[] = { 51N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), 52N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), 53 NULL 54}; 55 56/* 57 * Objects we are going to pack are collected in the `to_pack` structure. 58 * It contains an array (dynamically expanded) of the object data, and a map 59 * that can resolve SHA1s to their position in the array. 60 */ 61static struct packing_data to_pack; 62 63static struct pack_idx_entry **written_list; 64static uint32_t nr_result, nr_written, nr_seen; 65static struct bitmap_index *bitmap_git; 66static uint32_t write_layer; 67 68static int non_empty; 69static int reuse_delta =1, reuse_object =1; 70static int keep_unreachable, unpack_unreachable, include_tag; 71static timestamp_t unpack_unreachable_expiration; 72static int pack_loose_unreachable; 73static int local; 74static int have_non_local_packs; 75static int incremental; 76static int ignore_packed_keep_on_disk; 77static int ignore_packed_keep_in_core; 78static int allow_ofs_delta; 79static struct pack_idx_option pack_idx_opts; 80static const char*base_name; 81static int progress =1; 82static int window =10; 83static unsigned long pack_size_limit; 84static int depth =50; 85static int delta_search_threads; 86static int pack_to_stdout; 87static int sparse; 88static int thin; 89static int num_preferred_base; 90static struct progress *progress_state; 91 92static struct packed_git *reuse_packfile; 93static uint32_t reuse_packfile_objects; 94static off_t reuse_packfile_offset; 95 96static int use_bitmap_index_default =1; 97static int use_bitmap_index = -1; 98static int write_bitmap_index; 99static uint16_t write_bitmap_options; 100 101static int exclude_promisor_objects; 102 103static int use_delta_islands; 104 105static unsigned long delta_cache_size =0; 106static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; 107static unsigned long cache_max_small_delta_size =1000; 108 109static unsigned long window_memory_limit =0; 110 111static struct list_objects_filter_options filter_options; 112 113enum missing_action { 114 MA_ERROR =0,/* fail if any missing objects are encountered */ 115 MA_ALLOW_ANY,/* silently allow ALL missing objects */ 116 MA_ALLOW_PROMISOR,/* silently allow all missing PROMISOR objects */ 117}; 118static enum missing_action arg_missing_action; 119static show_object_fn fn_show_object; 120 121/* 122 * stats 123 */ 124static uint32_t written, written_delta; 125static uint32_t reused, reused_delta; 126 127/* 128 * Indexed commits 129 */ 130static struct commit **indexed_commits; 131static unsigned int indexed_commits_nr; 132static unsigned int indexed_commits_alloc; 133 134static voidindex_commit_for_bitmap(struct commit *commit) 135{ 136if(indexed_commits_nr >= indexed_commits_alloc) { 137 indexed_commits_alloc = (indexed_commits_alloc +32) *2; 138REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); 139} 140 141 indexed_commits[indexed_commits_nr++] = commit; 142} 143 144static void*get_delta(struct object_entry *entry) 145{ 146unsigned long size, base_size, delta_size; 147void*buf, *base_buf, *delta_buf; 148enum object_type type; 149 150 buf =read_object_file(&entry->idx.oid, &type, &size); 151if(!buf) 152die(_("unable to read%s"),oid_to_hex(&entry->idx.oid)); 153 base_buf =read_object_file(&DELTA(entry)->idx.oid, &type, 154&base_size); 155if(!base_buf) 156die("unable to read%s", 157oid_to_hex(&DELTA(entry)->idx.oid)); 158 delta_buf =diff_delta(base_buf, base_size, 159 buf, size, &delta_size,0); 160/* 161 * We succesfully computed this delta once but dropped it for 162 * memory reasons. Something is very wrong if this time we 163 * recompute and create a different delta. 164 */ 165if(!delta_buf || delta_size !=DELTA_SIZE(entry)) 166BUG("delta size changed"); 167free(buf); 168free(base_buf); 169return delta_buf; 170} 171 172static unsigned longdo_compress(void**pptr,unsigned long size) 173{ 174 git_zstream stream; 175void*in, *out; 176unsigned long maxsize; 177 178git_deflate_init(&stream, pack_compression_level); 179 maxsize =git_deflate_bound(&stream, size); 180 181 in = *pptr; 182 out =xmalloc(maxsize); 183*pptr = out; 184 185 stream.next_in = in; 186 stream.avail_in = size; 187 stream.next_out = out; 188 stream.avail_out = maxsize; 189while(git_deflate(&stream, Z_FINISH) == Z_OK) 190;/* nothing */ 191git_deflate_end(&stream); 192 193free(in); 194return stream.total_out; 195} 196 197static unsigned longwrite_large_blob_data(struct git_istream *st,struct hashfile *f, 198const struct object_id *oid) 199{ 200 git_zstream stream; 201unsigned char ibuf[1024*16]; 202unsigned char obuf[1024*16]; 203unsigned long olen =0; 204 205git_deflate_init(&stream, pack_compression_level); 206 207for(;;) { 208 ssize_t readlen; 209int zret = Z_OK; 210 readlen =read_istream(st, ibuf,sizeof(ibuf)); 211if(readlen == -1) 212die(_("unable to read%s"),oid_to_hex(oid)); 213 214 stream.next_in = ibuf; 215 stream.avail_in = readlen; 216while((stream.avail_in || readlen ==0) && 217(zret == Z_OK || zret == Z_BUF_ERROR)) { 218 stream.next_out = obuf; 219 stream.avail_out =sizeof(obuf); 220 zret =git_deflate(&stream, readlen ?0: Z_FINISH); 221hashwrite(f, obuf, stream.next_out - obuf); 222 olen += stream.next_out - obuf; 223} 224if(stream.avail_in) 225die(_("deflate error (%d)"), zret); 226if(readlen ==0) { 227if(zret != Z_STREAM_END) 228die(_("deflate error (%d)"), zret); 229break; 230} 231} 232git_deflate_end(&stream); 233return olen; 234} 235 236/* 237 * we are going to reuse the existing object data as is. make 238 * sure it is not corrupt. 239 */ 240static intcheck_pack_inflate(struct packed_git *p, 241struct pack_window **w_curs, 242 off_t offset, 243 off_t len, 244unsigned long expect) 245{ 246 git_zstream stream; 247unsigned char fakebuf[4096], *in; 248int st; 249 250memset(&stream,0,sizeof(stream)); 251git_inflate_init(&stream); 252do{ 253 in =use_pack(p, w_curs, offset, &stream.avail_in); 254 stream.next_in = in; 255 stream.next_out = fakebuf; 256 stream.avail_out =sizeof(fakebuf); 257 st =git_inflate(&stream, Z_FINISH); 258 offset += stream.next_in - in; 259}while(st == Z_OK || st == Z_BUF_ERROR); 260git_inflate_end(&stream); 261return(st == Z_STREAM_END && 262 stream.total_out == expect && 263 stream.total_in == len) ?0: -1; 264} 265 266static voidcopy_pack_data(struct hashfile *f, 267struct packed_git *p, 268struct pack_window **w_curs, 269 off_t offset, 270 off_t len) 271{ 272unsigned char*in; 273unsigned long avail; 274 275while(len) { 276 in =use_pack(p, w_curs, offset, &avail); 277if(avail > len) 278 avail = (unsigned long)len; 279hashwrite(f, in, avail); 280 offset += avail; 281 len -= avail; 282} 283} 284 285/* Return 0 if we will bust the pack-size limit */ 286static unsigned longwrite_no_reuse_object(struct hashfile *f,struct object_entry *entry, 287unsigned long limit,int usable_delta) 288{ 289unsigned long size, datalen; 290unsigned char header[MAX_PACK_OBJECT_HEADER], 291 dheader[MAX_PACK_OBJECT_HEADER]; 292unsigned hdrlen; 293enum object_type type; 294void*buf; 295struct git_istream *st = NULL; 296const unsigned hashsz = the_hash_algo->rawsz; 297 298if(!usable_delta) { 299if(oe_type(entry) == OBJ_BLOB && 300oe_size_greater_than(&to_pack, entry, big_file_threshold) && 301(st =open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL) 302 buf = NULL; 303else{ 304 buf =read_object_file(&entry->idx.oid, &type, &size); 305if(!buf) 306die(_("unable to read%s"), 307oid_to_hex(&entry->idx.oid)); 308} 309/* 310 * make sure no cached delta data remains from a 311 * previous attempt before a pack split occurred. 312 */ 313FREE_AND_NULL(entry->delta_data); 314 entry->z_delta_size =0; 315}else if(entry->delta_data) { 316 size =DELTA_SIZE(entry); 317 buf = entry->delta_data; 318 entry->delta_data = NULL; 319 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 320 OBJ_OFS_DELTA : OBJ_REF_DELTA; 321}else{ 322 buf =get_delta(entry); 323 size =DELTA_SIZE(entry); 324 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 325 OBJ_OFS_DELTA : OBJ_REF_DELTA; 326} 327 328if(st)/* large blob case, just assume we don't compress well */ 329 datalen = size; 330else if(entry->z_delta_size) 331 datalen = entry->z_delta_size; 332else 333 datalen =do_compress(&buf, size); 334 335/* 336 * The object header is a byte of 'type' followed by zero or 337 * more bytes of length. 338 */ 339 hdrlen =encode_in_pack_object_header(header,sizeof(header), 340 type, size); 341 342if(type == OBJ_OFS_DELTA) { 343/* 344 * Deltas with relative base contain an additional 345 * encoding of the relative offset for the delta 346 * base from this object's position in the pack. 347 */ 348 off_t ofs = entry->idx.offset -DELTA(entry)->idx.offset; 349unsigned pos =sizeof(dheader) -1; 350 dheader[pos] = ofs &127; 351while(ofs >>=7) 352 dheader[--pos] =128| (--ofs &127); 353if(limit && hdrlen +sizeof(dheader) - pos + datalen + hashsz >= limit) { 354if(st) 355close_istream(st); 356free(buf); 357return0; 358} 359hashwrite(f, header, hdrlen); 360hashwrite(f, dheader + pos,sizeof(dheader) - pos); 361 hdrlen +=sizeof(dheader) - pos; 362}else if(type == OBJ_REF_DELTA) { 363/* 364 * Deltas with a base reference contain 365 * additional bytes for the base object ID. 366 */ 367if(limit && hdrlen + hashsz + datalen + hashsz >= limit) { 368if(st) 369close_istream(st); 370free(buf); 371return0; 372} 373hashwrite(f, header, hdrlen); 374hashwrite(f,DELTA(entry)->idx.oid.hash, hashsz); 375 hdrlen += hashsz; 376}else{ 377if(limit && hdrlen + datalen + hashsz >= limit) { 378if(st) 379close_istream(st); 380free(buf); 381return0; 382} 383hashwrite(f, header, hdrlen); 384} 385if(st) { 386 datalen =write_large_blob_data(st, f, &entry->idx.oid); 387close_istream(st); 388}else{ 389hashwrite(f, buf, datalen); 390free(buf); 391} 392 393return hdrlen + datalen; 394} 395 396/* Return 0 if we will bust the pack-size limit */ 397static off_t write_reuse_object(struct hashfile *f,struct object_entry *entry, 398unsigned long limit,int usable_delta) 399{ 400struct packed_git *p =IN_PACK(entry); 401struct pack_window *w_curs = NULL; 402struct revindex_entry *revidx; 403 off_t offset; 404enum object_type type =oe_type(entry); 405 off_t datalen; 406unsigned char header[MAX_PACK_OBJECT_HEADER], 407 dheader[MAX_PACK_OBJECT_HEADER]; 408unsigned hdrlen; 409const unsigned hashsz = the_hash_algo->rawsz; 410unsigned long entry_size =SIZE(entry); 411 412if(DELTA(entry)) 413 type = (allow_ofs_delta &&DELTA(entry)->idx.offset) ? 414 OBJ_OFS_DELTA : OBJ_REF_DELTA; 415 hdrlen =encode_in_pack_object_header(header,sizeof(header), 416 type, entry_size); 417 418 offset = entry->in_pack_offset; 419 revidx =find_pack_revindex(p, offset); 420 datalen = revidx[1].offset - offset; 421if(!pack_to_stdout && p->index_version >1&& 422check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 423error(_("bad packed object CRC for%s"), 424oid_to_hex(&entry->idx.oid)); 425unuse_pack(&w_curs); 426returnwrite_no_reuse_object(f, entry, limit, usable_delta); 427} 428 429 offset += entry->in_pack_header_size; 430 datalen -= entry->in_pack_header_size; 431 432if(!pack_to_stdout && p->index_version ==1&& 433check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { 434error(_("corrupt packed object for%s"), 435oid_to_hex(&entry->idx.oid)); 436unuse_pack(&w_curs); 437returnwrite_no_reuse_object(f, entry, limit, usable_delta); 438} 439 440if(type == OBJ_OFS_DELTA) { 441 off_t ofs = entry->idx.offset -DELTA(entry)->idx.offset; 442unsigned pos =sizeof(dheader) -1; 443 dheader[pos] = ofs &127; 444while(ofs >>=7) 445 dheader[--pos] =128| (--ofs &127); 446if(limit && hdrlen +sizeof(dheader) - pos + datalen + hashsz >= limit) { 447unuse_pack(&w_curs); 448return0; 449} 450hashwrite(f, header, hdrlen); 451hashwrite(f, dheader + pos,sizeof(dheader) - pos); 452 hdrlen +=sizeof(dheader) - pos; 453 reused_delta++; 454}else if(type == OBJ_REF_DELTA) { 455if(limit && hdrlen + hashsz + datalen + hashsz >= limit) { 456unuse_pack(&w_curs); 457return0; 458} 459hashwrite(f, header, hdrlen); 460hashwrite(f,DELTA(entry)->idx.oid.hash, hashsz); 461 hdrlen += hashsz; 462 reused_delta++; 463}else{ 464if(limit && hdrlen + datalen + hashsz >= limit) { 465unuse_pack(&w_curs); 466return0; 467} 468hashwrite(f, header, hdrlen); 469} 470copy_pack_data(f, p, &w_curs, offset, datalen); 471unuse_pack(&w_curs); 472 reused++; 473return hdrlen + datalen; 474} 475 476/* Return 0 if we will bust the pack-size limit */ 477static off_t write_object(struct hashfile *f, 478struct object_entry *entry, 479 off_t write_offset) 480{ 481unsigned long limit; 482 off_t len; 483int usable_delta, to_reuse; 484 485if(!pack_to_stdout) 486crc32_begin(f); 487 488/* apply size limit if limited packsize and not first object */ 489if(!pack_size_limit || !nr_written) 490 limit =0; 491else if(pack_size_limit <= write_offset) 492/* 493 * the earlier object did not fit the limit; avoid 494 * mistaking this with unlimited (i.e. limit = 0). 495 */ 496 limit =1; 497else 498 limit = pack_size_limit - write_offset; 499 500if(!DELTA(entry)) 501 usable_delta =0;/* no delta */ 502else if(!pack_size_limit) 503 usable_delta =1;/* unlimited packfile */ 504else if(DELTA(entry)->idx.offset == (off_t)-1) 505 usable_delta =0;/* base was written to another pack */ 506else if(DELTA(entry)->idx.offset) 507 usable_delta =1;/* base already exists in this pack */ 508else 509 usable_delta =0;/* base could end up in another pack */ 510 511if(!reuse_object) 512 to_reuse =0;/* explicit */ 513else if(!IN_PACK(entry)) 514 to_reuse =0;/* can't reuse what we don't have */ 515else if(oe_type(entry) == OBJ_REF_DELTA || 516oe_type(entry) == OBJ_OFS_DELTA) 517/* check_object() decided it for us ... */ 518 to_reuse = usable_delta; 519/* ... but pack split may override that */ 520else if(oe_type(entry) != entry->in_pack_type) 521 to_reuse =0;/* pack has delta which is unusable */ 522else if(DELTA(entry)) 523 to_reuse =0;/* we want to pack afresh */ 524else 525 to_reuse =1;/* we have it in-pack undeltified, 526 * and we do not need to deltify it. 527 */ 528 529if(!to_reuse) 530 len =write_no_reuse_object(f, entry, limit, usable_delta); 531else 532 len =write_reuse_object(f, entry, limit, usable_delta); 533if(!len) 534return0; 535 536if(usable_delta) 537 written_delta++; 538 written++; 539if(!pack_to_stdout) 540 entry->idx.crc32 =crc32_end(f); 541return len; 542} 543 544enum write_one_status { 545 WRITE_ONE_SKIP = -1,/* already written */ 546 WRITE_ONE_BREAK =0,/* writing this will bust the limit; not written */ 547 WRITE_ONE_WRITTEN =1,/* normal */ 548 WRITE_ONE_RECURSIVE =2/* already scheduled to be written */ 549}; 550 551static enum write_one_status write_one(struct hashfile *f, 552struct object_entry *e, 553 off_t *offset) 554{ 555 off_t size; 556int recursing; 557 558/* 559 * we set offset to 1 (which is an impossible value) to mark 560 * the fact that this object is involved in "write its base 561 * first before writing a deltified object" recursion. 562 */ 563 recursing = (e->idx.offset ==1); 564if(recursing) { 565warning(_("recursive delta detected for object%s"), 566oid_to_hex(&e->idx.oid)); 567return WRITE_ONE_RECURSIVE; 568}else if(e->idx.offset || e->preferred_base) { 569/* offset is non zero if object is written already. */ 570return WRITE_ONE_SKIP; 571} 572 573/* if we are deltified, write out base object first. */ 574if(DELTA(e)) { 575 e->idx.offset =1;/* now recurse */ 576switch(write_one(f,DELTA(e), offset)) { 577case WRITE_ONE_RECURSIVE: 578/* we cannot depend on this one */ 579SET_DELTA(e, NULL); 580break; 581default: 582break; 583case WRITE_ONE_BREAK: 584 e->idx.offset = recursing; 585return WRITE_ONE_BREAK; 586} 587} 588 589 e->idx.offset = *offset; 590 size =write_object(f, e, *offset); 591if(!size) { 592 e->idx.offset = recursing; 593return WRITE_ONE_BREAK; 594} 595 written_list[nr_written++] = &e->idx; 596 597/* make sure off_t is sufficiently large not to wrap */ 598if(signed_add_overflows(*offset, size)) 599die(_("pack too large for current definition of off_t")); 600*offset += size; 601return WRITE_ONE_WRITTEN; 602} 603 604static intmark_tagged(const char*path,const struct object_id *oid,int flag, 605void*cb_data) 606{ 607struct object_id peeled; 608struct object_entry *entry =packlist_find(&to_pack, oid->hash, NULL); 609 610if(entry) 611 entry->tagged =1; 612if(!peel_ref(path, &peeled)) { 613 entry =packlist_find(&to_pack, peeled.hash, NULL); 614if(entry) 615 entry->tagged =1; 616} 617return0; 618} 619 620staticinlinevoidadd_to_write_order(struct object_entry **wo, 621unsigned int*endp, 622struct object_entry *e) 623{ 624if(e->filled ||oe_layer(&to_pack, e) != write_layer) 625return; 626 wo[(*endp)++] = e; 627 e->filled =1; 628} 629 630static voidadd_descendants_to_write_order(struct object_entry **wo, 631unsigned int*endp, 632struct object_entry *e) 633{ 634int add_to_order =1; 635while(e) { 636if(add_to_order) { 637struct object_entry *s; 638/* add this node... */ 639add_to_write_order(wo, endp, e); 640/* all its siblings... */ 641for(s =DELTA_SIBLING(e); s; s =DELTA_SIBLING(s)) { 642add_to_write_order(wo, endp, s); 643} 644} 645/* drop down a level to add left subtree nodes if possible */ 646if(DELTA_CHILD(e)) { 647 add_to_order =1; 648 e =DELTA_CHILD(e); 649}else{ 650 add_to_order =0; 651/* our sibling might have some children, it is next */ 652if(DELTA_SIBLING(e)) { 653 e =DELTA_SIBLING(e); 654continue; 655} 656/* go back to our parent node */ 657 e =DELTA(e); 658while(e && !DELTA_SIBLING(e)) { 659/* we're on the right side of a subtree, keep 660 * going up until we can go right again */ 661 e =DELTA(e); 662} 663if(!e) { 664/* done- we hit our original root node */ 665return; 666} 667/* pass it off to sibling at this level */ 668 e =DELTA_SIBLING(e); 669} 670}; 671} 672 673static voidadd_family_to_write_order(struct object_entry **wo, 674unsigned int*endp, 675struct object_entry *e) 676{ 677struct object_entry *root; 678 679for(root = e;DELTA(root); root =DELTA(root)) 680;/* nothing */ 681add_descendants_to_write_order(wo, endp, root); 682} 683 684static voidcompute_layer_order(struct object_entry **wo,unsigned int*wo_end) 685{ 686unsigned int i, last_untagged; 687struct object_entry *objects = to_pack.objects; 688 689for(i =0; i < to_pack.nr_objects; i++) { 690if(objects[i].tagged) 691break; 692add_to_write_order(wo, wo_end, &objects[i]); 693} 694 last_untagged = i; 695 696/* 697 * Then fill all the tagged tips. 698 */ 699for(; i < to_pack.nr_objects; i++) { 700if(objects[i].tagged) 701add_to_write_order(wo, wo_end, &objects[i]); 702} 703 704/* 705 * And then all remaining commits and tags. 706 */ 707for(i = last_untagged; i < to_pack.nr_objects; i++) { 708if(oe_type(&objects[i]) != OBJ_COMMIT && 709oe_type(&objects[i]) != OBJ_TAG) 710continue; 711add_to_write_order(wo, wo_end, &objects[i]); 712} 713 714/* 715 * And then all the trees. 716 */ 717for(i = last_untagged; i < to_pack.nr_objects; i++) { 718if(oe_type(&objects[i]) != OBJ_TREE) 719continue; 720add_to_write_order(wo, wo_end, &objects[i]); 721} 722 723/* 724 * Finally all the rest in really tight order 725 */ 726for(i = last_untagged; i < to_pack.nr_objects; i++) { 727if(!objects[i].filled &&oe_layer(&to_pack, &objects[i]) == write_layer) 728add_family_to_write_order(wo, wo_end, &objects[i]); 729} 730} 731 732static struct object_entry **compute_write_order(void) 733{ 734uint32_t max_layers =1; 735unsigned int i, wo_end; 736 737struct object_entry **wo; 738struct object_entry *objects = to_pack.objects; 739 740for(i =0; i < to_pack.nr_objects; i++) { 741 objects[i].tagged =0; 742 objects[i].filled =0; 743SET_DELTA_CHILD(&objects[i], NULL); 744SET_DELTA_SIBLING(&objects[i], NULL); 745} 746 747/* 748 * Fully connect delta_child/delta_sibling network. 749 * Make sure delta_sibling is sorted in the original 750 * recency order. 751 */ 752for(i = to_pack.nr_objects; i >0;) { 753struct object_entry *e = &objects[--i]; 754if(!DELTA(e)) 755continue; 756/* Mark me as the first child */ 757 e->delta_sibling_idx =DELTA(e)->delta_child_idx; 758SET_DELTA_CHILD(DELTA(e), e); 759} 760 761/* 762 * Mark objects that are at the tip of tags. 763 */ 764for_each_tag_ref(mark_tagged, NULL); 765 766if(use_delta_islands) 767 max_layers =compute_pack_layers(&to_pack); 768 769ALLOC_ARRAY(wo, to_pack.nr_objects); 770 wo_end =0; 771 772for(; write_layer < max_layers; ++write_layer) 773compute_layer_order(wo, &wo_end); 774 775if(wo_end != to_pack.nr_objects) 776die(_("ordered%uobjects, expected %"PRIu32), 777 wo_end, to_pack.nr_objects); 778 779return wo; 780} 781 782static off_t write_reused_pack(struct hashfile *f) 783{ 784unsigned char buffer[8192]; 785 off_t to_write, total; 786int fd; 787 788if(!is_pack_valid(reuse_packfile)) 789die(_("packfile is invalid:%s"), reuse_packfile->pack_name); 790 791 fd =git_open(reuse_packfile->pack_name); 792if(fd <0) 793die_errno(_("unable to open packfile for reuse:%s"), 794 reuse_packfile->pack_name); 795 796if(lseek(fd,sizeof(struct pack_header), SEEK_SET) == -1) 797die_errno(_("unable to seek in reused packfile")); 798 799if(reuse_packfile_offset <0) 800 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz; 801 802 total = to_write = reuse_packfile_offset -sizeof(struct pack_header); 803 804while(to_write) { 805int read_pack =xread(fd, buffer,sizeof(buffer)); 806 807if(read_pack <=0) 808die_errno(_("unable to read from reused packfile")); 809 810if(read_pack > to_write) 811 read_pack = to_write; 812 813hashwrite(f, buffer, read_pack); 814 to_write -= read_pack; 815 816/* 817 * We don't know the actual number of objects written, 818 * only how many bytes written, how many bytes total, and 819 * how many objects total. So we can fake it by pretending all 820 * objects we are writing are the same size. This gives us a 821 * smooth progress meter, and at the end it matches the true 822 * answer. 823 */ 824 written = reuse_packfile_objects * 825(((double)(total - to_write)) / total); 826display_progress(progress_state, written); 827} 828 829close(fd); 830 written = reuse_packfile_objects; 831display_progress(progress_state, written); 832return reuse_packfile_offset -sizeof(struct pack_header); 833} 834 835static const char no_split_warning[] =N_( 836"disabling bitmap writing, packs are split due to pack.packSizeLimit" 837); 838 839static voidwrite_pack_file(void) 840{ 841uint32_t i =0, j; 842struct hashfile *f; 843 off_t offset; 844uint32_t nr_remaining = nr_result; 845time_t last_mtime =0; 846struct object_entry **write_order; 847 848if(progress > pack_to_stdout) 849 progress_state =start_progress(_("Writing objects"), nr_result); 850ALLOC_ARRAY(written_list, to_pack.nr_objects); 851 write_order =compute_write_order(); 852 853do{ 854struct object_id oid; 855char*pack_tmp_name = NULL; 856 857if(pack_to_stdout) 858 f =hashfd_throughput(1,"<stdout>", progress_state); 859else 860 f =create_tmp_packfile(&pack_tmp_name); 861 862 offset =write_pack_header(f, nr_remaining); 863 864if(reuse_packfile) { 865 off_t packfile_size; 866assert(pack_to_stdout); 867 868 packfile_size =write_reused_pack(f); 869 offset += packfile_size; 870} 871 872 nr_written =0; 873for(; i < to_pack.nr_objects; i++) { 874struct object_entry *e = write_order[i]; 875if(write_one(f, e, &offset) == WRITE_ONE_BREAK) 876break; 877display_progress(progress_state, written); 878} 879 880/* 881 * Did we write the wrong # entries in the header? 882 * If so, rewrite it like in fast-import 883 */ 884if(pack_to_stdout) { 885finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); 886}else if(nr_written == nr_remaining) { 887finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); 888}else{ 889int fd =finalize_hashfile(f, oid.hash,0); 890fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, 891 nr_written, oid.hash, offset); 892close(fd); 893if(write_bitmap_index) { 894warning(_(no_split_warning)); 895 write_bitmap_index =0; 896} 897} 898 899if(!pack_to_stdout) { 900struct stat st; 901struct strbuf tmpname = STRBUF_INIT; 902 903/* 904 * Packs are runtime accessed in their mtime 905 * order since newer packs are more likely to contain 906 * younger objects. So if we are creating multiple 907 * packs then we should modify the mtime of later ones 908 * to preserve this property. 909 */ 910if(stat(pack_tmp_name, &st) <0) { 911warning_errno(_("failed to stat%s"), pack_tmp_name); 912}else if(!last_mtime) { 913 last_mtime = st.st_mtime; 914}else{ 915struct utimbuf utb; 916 utb.actime = st.st_atime; 917 utb.modtime = --last_mtime; 918if(utime(pack_tmp_name, &utb) <0) 919warning_errno(_("failed utime() on%s"), pack_tmp_name); 920} 921 922strbuf_addf(&tmpname,"%s-", base_name); 923 924if(write_bitmap_index) { 925bitmap_writer_set_checksum(oid.hash); 926bitmap_writer_build_type_index( 927&to_pack, written_list, nr_written); 928} 929 930finish_tmp_packfile(&tmpname, pack_tmp_name, 931 written_list, nr_written, 932&pack_idx_opts, oid.hash); 933 934if(write_bitmap_index) { 935strbuf_addf(&tmpname,"%s.bitmap",oid_to_hex(&oid)); 936 937stop_progress(&progress_state); 938 939bitmap_writer_show_progress(progress); 940bitmap_writer_reuse_bitmaps(&to_pack); 941bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 942bitmap_writer_build(&to_pack); 943bitmap_writer_finish(written_list, nr_written, 944 tmpname.buf, write_bitmap_options); 945 write_bitmap_index =0; 946} 947 948strbuf_release(&tmpname); 949free(pack_tmp_name); 950puts(oid_to_hex(&oid)); 951} 952 953/* mark written objects as written to previous pack */ 954for(j =0; j < nr_written; j++) { 955 written_list[j]->offset = (off_t)-1; 956} 957 nr_remaining -= nr_written; 958}while(nr_remaining && i < to_pack.nr_objects); 959 960free(written_list); 961free(write_order); 962stop_progress(&progress_state); 963if(written != nr_result) 964die(_("wrote %"PRIu32" objects while expecting %"PRIu32), 965 written, nr_result); 966} 967 968static intno_try_delta(const char*path) 969{ 970static struct attr_check *check; 971 972if(!check) 973 check =attr_check_initl("delta", NULL); 974git_check_attr(&the_index, path, check); 975if(ATTR_FALSE(check->items[0].value)) 976return1; 977return0; 978} 979 980/* 981 * When adding an object, check whether we have already added it 982 * to our packing list. If so, we can skip. However, if we are 983 * being asked to excludei t, but the previous mention was to include 984 * it, make sure to adjust its flags and tweak our numbers accordingly. 985 * 986 * As an optimization, we pass out the index position where we would have 987 * found the item, since that saves us from having to look it up again a 988 * few lines later when we want to add the new entry. 989 */ 990static inthave_duplicate_entry(const struct object_id *oid, 991int exclude, 992uint32_t*index_pos) 993{ 994struct object_entry *entry; 995 996 entry =packlist_find(&to_pack, oid->hash, index_pos); 997if(!entry) 998return0; 9991000if(exclude) {1001if(!entry->preferred_base)1002 nr_result--;1003 entry->preferred_base =1;1004}10051006return1;1007}10081009static intwant_found_object(int exclude,struct packed_git *p)1010{1011if(exclude)1012return1;1013if(incremental)1014return0;10151016/*1017 * When asked to do --local (do not include an object that appears in a1018 * pack we borrow from elsewhere) or --honor-pack-keep (do not include1019 * an object that appears in a pack marked with .keep), finding a pack1020 * that matches the criteria is sufficient for us to decide to omit it.1021 * However, even if this pack does not satisfy the criteria, we need to1022 * make sure no copy of this object appears in _any_ pack that makes us1023 * to omit the object, so we need to check all the packs.1024 *1025 * We can however first check whether these options can possible matter;1026 * if they do not matter we know we want the object in generated pack.1027 * Otherwise, we signal "-1" at the end to tell the caller that we do1028 * not know either way, and it needs to check more packs.1029 */1030if(!ignore_packed_keep_on_disk &&1031!ignore_packed_keep_in_core &&1032(!local || !have_non_local_packs))1033return1;10341035if(local && !p->pack_local)1036return0;1037if(p->pack_local &&1038((ignore_packed_keep_on_disk && p->pack_keep) ||1039(ignore_packed_keep_in_core && p->pack_keep_in_core)))1040return0;10411042/* we don't know yet; keep looking for more packs */1043return-1;1044}10451046/*1047 * Check whether we want the object in the pack (e.g., we do not want1048 * objects found in non-local stores if the "--local" option was used).1049 *1050 * If the caller already knows an existing pack it wants to take the object1051 * from, that is passed in *found_pack and *found_offset; otherwise this1052 * function finds if there is any pack that has the object and returns the pack1053 * and its offset in these variables.1054 */1055static intwant_object_in_pack(const struct object_id *oid,1056int exclude,1057struct packed_git **found_pack,1058 off_t *found_offset)1059{1060int want;1061struct list_head *pos;1062struct multi_pack_index *m;10631064if(!exclude && local &&has_loose_object_nonlocal(oid))1065return0;10661067/*1068 * If we already know the pack object lives in, start checks from that1069 * pack - in the usual case when neither --local was given nor .keep files1070 * are present we will determine the answer right now.1071 */1072if(*found_pack) {1073 want =want_found_object(exclude, *found_pack);1074if(want != -1)1075return want;1076}10771078for(m =get_multi_pack_index(the_repository); m; m = m->next) {1079struct pack_entry e;1080if(fill_midx_entry(oid, &e, m)) {1081struct packed_git *p = e.p;1082 off_t offset;10831084if(p == *found_pack)1085 offset = *found_offset;1086else1087 offset =find_pack_entry_one(oid->hash, p);10881089if(offset) {1090if(!*found_pack) {1091if(!is_pack_valid(p))1092continue;1093*found_offset = offset;1094*found_pack = p;1095}1096 want =want_found_object(exclude, p);1097if(want != -1)1098return want;1099}1100}1101}11021103list_for_each(pos,get_packed_git_mru(the_repository)) {1104struct packed_git *p =list_entry(pos,struct packed_git, mru);1105 off_t offset;11061107if(p == *found_pack)1108 offset = *found_offset;1109else1110 offset =find_pack_entry_one(oid->hash, p);11111112if(offset) {1113if(!*found_pack) {1114if(!is_pack_valid(p))1115continue;1116*found_offset = offset;1117*found_pack = p;1118}1119 want =want_found_object(exclude, p);1120if(!exclude && want >0)1121list_move(&p->mru,1122get_packed_git_mru(the_repository));1123if(want != -1)1124return want;1125}1126}11271128return1;1129}11301131static voidcreate_object_entry(const struct object_id *oid,1132enum object_type type,1133uint32_t hash,1134int exclude,1135int no_try_delta,1136uint32_t index_pos,1137struct packed_git *found_pack,1138 off_t found_offset)1139{1140struct object_entry *entry;11411142 entry =packlist_alloc(&to_pack, oid->hash, index_pos);1143 entry->hash = hash;1144oe_set_type(entry, type);1145if(exclude)1146 entry->preferred_base =1;1147else1148 nr_result++;1149if(found_pack) {1150oe_set_in_pack(&to_pack, entry, found_pack);1151 entry->in_pack_offset = found_offset;1152}11531154 entry->no_try_delta = no_try_delta;1155}11561157static const char no_closure_warning[] =N_(1158"disabling bitmap writing, as some objects are not being packed"1159);11601161static intadd_object_entry(const struct object_id *oid,enum object_type type,1162const char*name,int exclude)1163{1164struct packed_git *found_pack = NULL;1165 off_t found_offset =0;1166uint32_t index_pos;11671168display_progress(progress_state, ++nr_seen);11691170if(have_duplicate_entry(oid, exclude, &index_pos))1171return0;11721173if(!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1174/* The pack is missing an object, so it will not have closure */1175if(write_bitmap_index) {1176warning(_(no_closure_warning));1177 write_bitmap_index =0;1178}1179return0;1180}11811182create_object_entry(oid, type,pack_name_hash(name),1183 exclude, name &&no_try_delta(name),1184 index_pos, found_pack, found_offset);1185return1;1186}11871188static intadd_object_entry_from_bitmap(const struct object_id *oid,1189enum object_type type,1190int flags,uint32_t name_hash,1191struct packed_git *pack, off_t offset)1192{1193uint32_t index_pos;11941195display_progress(progress_state, ++nr_seen);11961197if(have_duplicate_entry(oid,0, &index_pos))1198return0;11991200if(!want_object_in_pack(oid,0, &pack, &offset))1201return0;12021203create_object_entry(oid, type, name_hash,0,0, index_pos, pack, offset);1204return1;1205}12061207struct pbase_tree_cache {1208struct object_id oid;1209int ref;1210int temporary;1211void*tree_data;1212unsigned long tree_size;1213};12141215static struct pbase_tree_cache *(pbase_tree_cache[256]);1216static intpbase_tree_cache_ix(const struct object_id *oid)1217{1218return oid->hash[0] %ARRAY_SIZE(pbase_tree_cache);1219}1220static intpbase_tree_cache_ix_incr(int ix)1221{1222return(ix+1) %ARRAY_SIZE(pbase_tree_cache);1223}12241225static struct pbase_tree {1226struct pbase_tree *next;1227/* This is a phony "cache" entry; we are not1228 * going to evict it or find it through _get()1229 * mechanism -- this is for the toplevel node that1230 * would almost always change with any commit.1231 */1232struct pbase_tree_cache pcache;1233} *pbase_tree;12341235static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1236{1237struct pbase_tree_cache *ent, *nent;1238void*data;1239unsigned long size;1240enum object_type type;1241int neigh;1242int my_ix =pbase_tree_cache_ix(oid);1243int available_ix = -1;12441245/* pbase-tree-cache acts as a limited hashtable.1246 * your object will be found at your index or within a few1247 * slots after that slot if it is cached.1248 */1249for(neigh =0; neigh <8; neigh++) {1250 ent = pbase_tree_cache[my_ix];1251if(ent &&oideq(&ent->oid, oid)) {1252 ent->ref++;1253return ent;1254}1255else if(((available_ix <0) && (!ent || !ent->ref)) ||1256((0<= available_ix) &&1257(!ent && pbase_tree_cache[available_ix])))1258 available_ix = my_ix;1259if(!ent)1260break;1261 my_ix =pbase_tree_cache_ix_incr(my_ix);1262}12631264/* Did not find one. Either we got a bogus request or1265 * we need to read and perhaps cache.1266 */1267 data =read_object_file(oid, &type, &size);1268if(!data)1269return NULL;1270if(type != OBJ_TREE) {1271free(data);1272return NULL;1273}12741275/* We need to either cache or return a throwaway copy */12761277if(available_ix <0)1278 ent = NULL;1279else{1280 ent = pbase_tree_cache[available_ix];1281 my_ix = available_ix;1282}12831284if(!ent) {1285 nent =xmalloc(sizeof(*nent));1286 nent->temporary = (available_ix <0);1287}1288else{1289/* evict and reuse */1290free(ent->tree_data);1291 nent = ent;1292}1293oidcpy(&nent->oid, oid);1294 nent->tree_data = data;1295 nent->tree_size = size;1296 nent->ref =1;1297if(!nent->temporary)1298 pbase_tree_cache[my_ix] = nent;1299return nent;1300}13011302static voidpbase_tree_put(struct pbase_tree_cache *cache)1303{1304if(!cache->temporary) {1305 cache->ref--;1306return;1307}1308free(cache->tree_data);1309free(cache);1310}13111312static intname_cmp_len(const char*name)1313{1314int i;1315for(i =0; name[i] && name[i] !='\n'&& name[i] !='/'; i++)1316;1317return i;1318}13191320static voidadd_pbase_object(struct tree_desc *tree,1321const char*name,1322int cmplen,1323const char*fullname)1324{1325struct name_entry entry;1326int cmp;13271328while(tree_entry(tree,&entry)) {1329if(S_ISGITLINK(entry.mode))1330continue;1331 cmp =tree_entry_len(&entry) != cmplen ?1:1332memcmp(name, entry.path, cmplen);1333if(cmp >0)1334continue;1335if(cmp <0)1336return;1337if(name[cmplen] !='/') {1338add_object_entry(entry.oid,1339object_type(entry.mode),1340 fullname,1);1341return;1342}1343if(S_ISDIR(entry.mode)) {1344struct tree_desc sub;1345struct pbase_tree_cache *tree;1346const char*down = name+cmplen+1;1347int downlen =name_cmp_len(down);13481349 tree =pbase_tree_get(entry.oid);1350if(!tree)1351return;1352init_tree_desc(&sub, tree->tree_data, tree->tree_size);13531354add_pbase_object(&sub, down, downlen, fullname);1355pbase_tree_put(tree);1356}1357}1358}13591360static unsigned*done_pbase_paths;1361static int done_pbase_paths_num;1362static int done_pbase_paths_alloc;1363static intdone_pbase_path_pos(unsigned hash)1364{1365int lo =0;1366int hi = done_pbase_paths_num;1367while(lo < hi) {1368int mi = lo + (hi - lo) /2;1369if(done_pbase_paths[mi] == hash)1370return mi;1371if(done_pbase_paths[mi] < hash)1372 hi = mi;1373else1374 lo = mi +1;1375}1376return-lo-1;1377}13781379static intcheck_pbase_path(unsigned hash)1380{1381int pos =done_pbase_path_pos(hash);1382if(0<= pos)1383return1;1384 pos = -pos -1;1385ALLOC_GROW(done_pbase_paths,1386 done_pbase_paths_num +1,1387 done_pbase_paths_alloc);1388 done_pbase_paths_num++;1389if(pos < done_pbase_paths_num)1390MOVE_ARRAY(done_pbase_paths + pos +1, done_pbase_paths + pos,1391 done_pbase_paths_num - pos -1);1392 done_pbase_paths[pos] = hash;1393return0;1394}13951396static voidadd_preferred_base_object(const char*name)1397{1398struct pbase_tree *it;1399int cmplen;1400unsigned hash =pack_name_hash(name);14011402if(!num_preferred_base ||check_pbase_path(hash))1403return;14041405 cmplen =name_cmp_len(name);1406for(it = pbase_tree; it; it = it->next) {1407if(cmplen ==0) {1408add_object_entry(&it->pcache.oid, OBJ_TREE, NULL,1);1409}1410else{1411struct tree_desc tree;1412init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1413add_pbase_object(&tree, name, cmplen, name);1414}1415}1416}14171418static voidadd_preferred_base(struct object_id *oid)1419{1420struct pbase_tree *it;1421void*data;1422unsigned long size;1423struct object_id tree_oid;14241425if(window <= num_preferred_base++)1426return;14271428 data =read_object_with_reference(oid, tree_type, &size, &tree_oid);1429if(!data)1430return;14311432for(it = pbase_tree; it; it = it->next) {1433if(oideq(&it->pcache.oid, &tree_oid)) {1434free(data);1435return;1436}1437}14381439 it =xcalloc(1,sizeof(*it));1440 it->next = pbase_tree;1441 pbase_tree = it;14421443oidcpy(&it->pcache.oid, &tree_oid);1444 it->pcache.tree_data = data;1445 it->pcache.tree_size = size;1446}14471448static voidcleanup_preferred_base(void)1449{1450struct pbase_tree *it;1451unsigned i;14521453 it = pbase_tree;1454 pbase_tree = NULL;1455while(it) {1456struct pbase_tree *tmp = it;1457 it = tmp->next;1458free(tmp->pcache.tree_data);1459free(tmp);1460}14611462for(i =0; i <ARRAY_SIZE(pbase_tree_cache); i++) {1463if(!pbase_tree_cache[i])1464continue;1465free(pbase_tree_cache[i]->tree_data);1466FREE_AND_NULL(pbase_tree_cache[i]);1467}14681469FREE_AND_NULL(done_pbase_paths);1470 done_pbase_paths_num = done_pbase_paths_alloc =0;1471}14721473/*1474 * Return 1 iff the object specified by "delta" can be sent1475 * literally as a delta against the base in "base_sha1". If1476 * so, then *base_out will point to the entry in our packing1477 * list, or NULL if we must use the external-base list.1478 *1479 * Depth value does not matter - find_deltas() will1480 * never consider reused delta as the base object to1481 * deltify other objects against, in order to avoid1482 * circular deltas.1483 */1484static intcan_reuse_delta(const unsigned char*base_sha1,1485struct object_entry *delta,1486struct object_entry **base_out)1487{1488struct object_entry *base;14891490if(!base_sha1)1491return0;14921493/*1494 * First see if we're already sending the base (or it's explicitly in1495 * our "excluded" list).1496 */1497 base =packlist_find(&to_pack, base_sha1, NULL);1498if(base) {1499if(!in_same_island(&delta->idx.oid, &base->idx.oid))1500return0;1501*base_out = base;1502return1;1503}15041505/*1506 * Otherwise, reachability bitmaps may tell us if the receiver has it,1507 * even if it was buried too deep in history to make it into the1508 * packing list.1509 */1510if(thin &&bitmap_has_sha1_in_uninteresting(bitmap_git, base_sha1)) {1511if(use_delta_islands) {1512struct object_id base_oid;1513hashcpy(base_oid.hash, base_sha1);1514if(!in_same_island(&delta->idx.oid, &base_oid))1515return0;1516}1517*base_out = NULL;1518return1;1519}15201521return0;1522}15231524static voidcheck_object(struct object_entry *entry)1525{1526unsigned long canonical_size;15271528if(IN_PACK(entry)) {1529struct packed_git *p =IN_PACK(entry);1530struct pack_window *w_curs = NULL;1531const unsigned char*base_ref = NULL;1532struct object_entry *base_entry;1533unsigned long used, used_0;1534unsigned long avail;1535 off_t ofs;1536unsigned char*buf, c;1537enum object_type type;1538unsigned long in_pack_size;15391540 buf =use_pack(p, &w_curs, entry->in_pack_offset, &avail);15411542/*1543 * We want in_pack_type even if we do not reuse delta1544 * since non-delta representations could still be reused.1545 */1546 used =unpack_object_header_buffer(buf, avail,1547&type,1548&in_pack_size);1549if(used ==0)1550goto give_up;15511552if(type <0)1553BUG("invalid type%d", type);1554 entry->in_pack_type = type;15551556/*1557 * Determine if this is a delta and if so whether we can1558 * reuse it or not. Otherwise let's find out as cheaply as1559 * possible what the actual type and size for this object is.1560 */1561switch(entry->in_pack_type) {1562default:1563/* Not a delta hence we've already got all we need. */1564oe_set_type(entry, entry->in_pack_type);1565SET_SIZE(entry, in_pack_size);1566 entry->in_pack_header_size = used;1567if(oe_type(entry) < OBJ_COMMIT ||oe_type(entry) > OBJ_BLOB)1568goto give_up;1569unuse_pack(&w_curs);1570return;1571case OBJ_REF_DELTA:1572if(reuse_delta && !entry->preferred_base)1573 base_ref =use_pack(p, &w_curs,1574 entry->in_pack_offset + used, NULL);1575 entry->in_pack_header_size = used + the_hash_algo->rawsz;1576break;1577case OBJ_OFS_DELTA:1578 buf =use_pack(p, &w_curs,1579 entry->in_pack_offset + used, NULL);1580 used_0 =0;1581 c = buf[used_0++];1582 ofs = c &127;1583while(c &128) {1584 ofs +=1;1585if(!ofs ||MSB(ofs,7)) {1586error(_("delta base offset overflow in pack for%s"),1587oid_to_hex(&entry->idx.oid));1588goto give_up;1589}1590 c = buf[used_0++];1591 ofs = (ofs <<7) + (c &127);1592}1593 ofs = entry->in_pack_offset - ofs;1594if(ofs <=0|| ofs >= entry->in_pack_offset) {1595error(_("delta base offset out of bound for%s"),1596oid_to_hex(&entry->idx.oid));1597goto give_up;1598}1599if(reuse_delta && !entry->preferred_base) {1600struct revindex_entry *revidx;1601 revidx =find_pack_revindex(p, ofs);1602if(!revidx)1603goto give_up;1604 base_ref =nth_packed_object_sha1(p, revidx->nr);1605}1606 entry->in_pack_header_size = used + used_0;1607break;1608}16091610if(can_reuse_delta(base_ref, entry, &base_entry)) {1611oe_set_type(entry, entry->in_pack_type);1612SET_SIZE(entry, in_pack_size);/* delta size */1613SET_DELTA_SIZE(entry, in_pack_size);16141615if(base_entry) {1616SET_DELTA(entry, base_entry);1617 entry->delta_sibling_idx = base_entry->delta_child_idx;1618SET_DELTA_CHILD(base_entry, entry);1619}else{1620SET_DELTA_EXT(entry, base_ref);1621}16221623unuse_pack(&w_curs);1624return;1625}16261627if(oe_type(entry)) {1628 off_t delta_pos;16291630/*1631 * This must be a delta and we already know what the1632 * final object type is. Let's extract the actual1633 * object size from the delta header.1634 */1635 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;1636 canonical_size =get_size_from_delta(p, &w_curs, delta_pos);1637if(canonical_size ==0)1638goto give_up;1639SET_SIZE(entry, canonical_size);1640unuse_pack(&w_curs);1641return;1642}16431644/*1645 * No choice but to fall back to the recursive delta walk1646 * with sha1_object_info() to find about the object type1647 * at this point...1648 */1649 give_up:1650unuse_pack(&w_curs);1651}16521653oe_set_type(entry,1654oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1655if(entry->type_valid) {1656SET_SIZE(entry, canonical_size);1657}else{1658/*1659 * Bad object type is checked in prepare_pack(). This is1660 * to permit a missing preferred base object to be ignored1661 * as a preferred base. Doing so can result in a larger1662 * pack file, but the transfer will still take place.1663 */1664}1665}16661667static intpack_offset_sort(const void*_a,const void*_b)1668{1669const struct object_entry *a = *(struct object_entry **)_a;1670const struct object_entry *b = *(struct object_entry **)_b;1671const struct packed_git *a_in_pack =IN_PACK(a);1672const struct packed_git *b_in_pack =IN_PACK(b);16731674/* avoid filesystem trashing with loose objects */1675if(!a_in_pack && !b_in_pack)1676returnoidcmp(&a->idx.oid, &b->idx.oid);16771678if(a_in_pack < b_in_pack)1679return-1;1680if(a_in_pack > b_in_pack)1681return1;1682return a->in_pack_offset < b->in_pack_offset ? -1:1683(a->in_pack_offset > b->in_pack_offset);1684}16851686/*1687 * Drop an on-disk delta we were planning to reuse. Naively, this would1688 * just involve blanking out the "delta" field, but we have to deal1689 * with some extra book-keeping:1690 *1691 * 1. Removing ourselves from the delta_sibling linked list.1692 *1693 * 2. Updating our size/type to the non-delta representation. These were1694 * either not recorded initially (size) or overwritten with the delta type1695 * (type) when check_object() decided to reuse the delta.1696 *1697 * 3. Resetting our delta depth, as we are now a base object.1698 */1699static voiddrop_reused_delta(struct object_entry *entry)1700{1701unsigned*idx = &to_pack.objects[entry->delta_idx -1].delta_child_idx;1702struct object_info oi = OBJECT_INFO_INIT;1703enum object_type type;1704unsigned long size;17051706while(*idx) {1707struct object_entry *oe = &to_pack.objects[*idx -1];17081709if(oe == entry)1710*idx = oe->delta_sibling_idx;1711else1712 idx = &oe->delta_sibling_idx;1713}1714SET_DELTA(entry, NULL);1715 entry->depth =0;17161717 oi.sizep = &size;1718 oi.typep = &type;1719if(packed_object_info(the_repository,IN_PACK(entry), entry->in_pack_offset, &oi) <0) {1720/*1721 * We failed to get the info from this pack for some reason;1722 * fall back to sha1_object_info, which may find another copy.1723 * And if that fails, the error will be recorded in oe_type(entry)1724 * and dealt with in prepare_pack().1725 */1726oe_set_type(entry,1727oid_object_info(the_repository, &entry->idx.oid, &size));1728}else{1729oe_set_type(entry, type);1730}1731SET_SIZE(entry, size);1732}17331734/*1735 * Follow the chain of deltas from this entry onward, throwing away any links1736 * that cause us to hit a cycle (as determined by the DFS state flags in1737 * the entries).1738 *1739 * We also detect too-long reused chains that would violate our --depth1740 * limit.1741 */1742static voidbreak_delta_chains(struct object_entry *entry)1743{1744/*1745 * The actual depth of each object we will write is stored as an int,1746 * as it cannot exceed our int "depth" limit. But before we break1747 * changes based no that limit, we may potentially go as deep as the1748 * number of objects, which is elsewhere bounded to a uint32_t.1749 */1750uint32_t total_depth;1751struct object_entry *cur, *next;17521753for(cur = entry, total_depth =0;1754 cur;1755 cur =DELTA(cur), total_depth++) {1756if(cur->dfs_state == DFS_DONE) {1757/*1758 * We've already seen this object and know it isn't1759 * part of a cycle. We do need to append its depth1760 * to our count.1761 */1762 total_depth += cur->depth;1763break;1764}17651766/*1767 * We break cycles before looping, so an ACTIVE state (or any1768 * other cruft which made its way into the state variable)1769 * is a bug.1770 */1771if(cur->dfs_state != DFS_NONE)1772BUG("confusing delta dfs state in first pass:%d",1773 cur->dfs_state);17741775/*1776 * Now we know this is the first time we've seen the object. If1777 * it's not a delta, we're done traversing, but we'll mark it1778 * done to save time on future traversals.1779 */1780if(!DELTA(cur)) {1781 cur->dfs_state = DFS_DONE;1782break;1783}17841785/*1786 * Mark ourselves as active and see if the next step causes1787 * us to cycle to another active object. It's important to do1788 * this _before_ we loop, because it impacts where we make the1789 * cut, and thus how our total_depth counter works.1790 * E.g., We may see a partial loop like:1791 *1792 * A -> B -> C -> D -> B1793 *1794 * Cutting B->C breaks the cycle. But now the depth of A is1795 * only 1, and our total_depth counter is at 3. The size of the1796 * error is always one less than the size of the cycle we1797 * broke. Commits C and D were "lost" from A's chain.1798 *1799 * If we instead cut D->B, then the depth of A is correct at 3.1800 * We keep all commits in the chain that we examined.1801 */1802 cur->dfs_state = DFS_ACTIVE;1803if(DELTA(cur)->dfs_state == DFS_ACTIVE) {1804drop_reused_delta(cur);1805 cur->dfs_state = DFS_DONE;1806break;1807}1808}18091810/*1811 * And now that we've gone all the way to the bottom of the chain, we1812 * need to clear the active flags and set the depth fields as1813 * appropriate. Unlike the loop above, which can quit when it drops a1814 * delta, we need to keep going to look for more depth cuts. So we need1815 * an extra "next" pointer to keep going after we reset cur->delta.1816 */1817for(cur = entry; cur; cur = next) {1818 next =DELTA(cur);18191820/*1821 * We should have a chain of zero or more ACTIVE states down to1822 * a final DONE. We can quit after the DONE, because either it1823 * has no bases, or we've already handled them in a previous1824 * call.1825 */1826if(cur->dfs_state == DFS_DONE)1827break;1828else if(cur->dfs_state != DFS_ACTIVE)1829BUG("confusing delta dfs state in second pass:%d",1830 cur->dfs_state);18311832/*1833 * If the total_depth is more than depth, then we need to snip1834 * the chain into two or more smaller chains that don't exceed1835 * the maximum depth. Most of the resulting chains will contain1836 * (depth + 1) entries (i.e., depth deltas plus one base), and1837 * the last chain (i.e., the one containing entry) will contain1838 * whatever entries are left over, namely1839 * (total_depth % (depth + 1)) of them.1840 *1841 * Since we are iterating towards decreasing depth, we need to1842 * decrement total_depth as we go, and we need to write to the1843 * entry what its final depth will be after all of the1844 * snipping. Since we're snipping into chains of length (depth1845 * + 1) entries, the final depth of an entry will be its1846 * original depth modulo (depth + 1). Any time we encounter an1847 * entry whose final depth is supposed to be zero, we snip it1848 * from its delta base, thereby making it so.1849 */1850 cur->depth = (total_depth--) % (depth +1);1851if(!cur->depth)1852drop_reused_delta(cur);18531854 cur->dfs_state = DFS_DONE;1855}1856}18571858static voidget_object_details(void)1859{1860uint32_t i;1861struct object_entry **sorted_by_offset;18621863if(progress)1864 progress_state =start_progress(_("Counting objects"),1865 to_pack.nr_objects);18661867 sorted_by_offset =xcalloc(to_pack.nr_objects,sizeof(struct object_entry *));1868for(i =0; i < to_pack.nr_objects; i++)1869 sorted_by_offset[i] = to_pack.objects + i;1870QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);18711872for(i =0; i < to_pack.nr_objects; i++) {1873struct object_entry *entry = sorted_by_offset[i];1874check_object(entry);1875if(entry->type_valid &&1876oe_size_greater_than(&to_pack, entry, big_file_threshold))1877 entry->no_try_delta =1;1878display_progress(progress_state, i +1);1879}1880stop_progress(&progress_state);18811882/*1883 * This must happen in a second pass, since we rely on the delta1884 * information for the whole list being completed.1885 */1886for(i =0; i < to_pack.nr_objects; i++)1887break_delta_chains(&to_pack.objects[i]);18881889free(sorted_by_offset);1890}18911892/*1893 * We search for deltas in a list sorted by type, by filename hash, and then1894 * by size, so that we see progressively smaller and smaller files.1895 * That's because we prefer deltas to be from the bigger file1896 * to the smaller -- deletes are potentially cheaper, but perhaps1897 * more importantly, the bigger file is likely the more recent1898 * one. The deepest deltas are therefore the oldest objects which are1899 * less susceptible to be accessed often.1900 */1901static inttype_size_sort(const void*_a,const void*_b)1902{1903const struct object_entry *a = *(struct object_entry **)_a;1904const struct object_entry *b = *(struct object_entry **)_b;1905enum object_type a_type =oe_type(a);1906enum object_type b_type =oe_type(b);1907unsigned long a_size =SIZE(a);1908unsigned long b_size =SIZE(b);19091910if(a_type > b_type)1911return-1;1912if(a_type < b_type)1913return1;1914if(a->hash > b->hash)1915return-1;1916if(a->hash < b->hash)1917return1;1918if(a->preferred_base > b->preferred_base)1919return-1;1920if(a->preferred_base < b->preferred_base)1921return1;1922if(use_delta_islands) {1923int island_cmp =island_delta_cmp(&a->idx.oid, &b->idx.oid);1924if(island_cmp)1925return island_cmp;1926}1927if(a_size > b_size)1928return-1;1929if(a_size < b_size)1930return1;1931return a < b ? -1: (a > b);/* newest first */1932}19331934struct unpacked {1935struct object_entry *entry;1936void*data;1937struct delta_index *index;1938unsigned depth;1939};19401941static intdelta_cacheable(unsigned long src_size,unsigned long trg_size,1942unsigned long delta_size)1943{1944if(max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1945return0;19461947if(delta_size < cache_max_small_delta_size)1948return1;19491950/* cache delta, if objects are large enough compared to delta size */1951if((src_size >>20) + (trg_size >>21) > (delta_size >>10))1952return1;19531954return0;1955}19561957/* Protect access to object database */1958static pthread_mutex_t read_mutex;1959#define read_lock() pthread_mutex_lock(&read_mutex)1960#define read_unlock() pthread_mutex_unlock(&read_mutex)19611962/* Protect delta_cache_size */1963static pthread_mutex_t cache_mutex;1964#define cache_lock() pthread_mutex_lock(&cache_mutex)1965#define cache_unlock() pthread_mutex_unlock(&cache_mutex)19661967/*1968 * Protect object list partitioning (e.g. struct thread_param) and1969 * progress_state1970 */1971static pthread_mutex_t progress_mutex;1972#define progress_lock() pthread_mutex_lock(&progress_mutex)1973#define progress_unlock() pthread_mutex_unlock(&progress_mutex)19741975/*1976 * Access to struct object_entry is unprotected since each thread owns1977 * a portion of the main object list. Just don't access object entries1978 * ahead in the list because they can be stolen and would need1979 * progress_mutex for protection.1980 */19811982/*1983 * Return the size of the object without doing any delta1984 * reconstruction (so non-deltas are true object sizes, but deltas1985 * return the size of the delta data).1986 */1987unsigned longoe_get_size_slow(struct packing_data *pack,1988const struct object_entry *e)1989{1990struct packed_git *p;1991struct pack_window *w_curs;1992unsigned char*buf;1993enum object_type type;1994unsigned long used, avail, size;19951996if(e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {1997read_lock();1998if(oid_object_info(the_repository, &e->idx.oid, &size) <0)1999die(_("unable to get size of%s"),2000oid_to_hex(&e->idx.oid));2001read_unlock();2002return size;2003}20042005 p =oe_in_pack(pack, e);2006if(!p)2007BUG("when e->type is a delta, it must belong to a pack");20082009read_lock();2010 w_curs = NULL;2011 buf =use_pack(p, &w_curs, e->in_pack_offset, &avail);2012 used =unpack_object_header_buffer(buf, avail, &type, &size);2013if(used ==0)2014die(_("unable to parse object header of%s"),2015oid_to_hex(&e->idx.oid));20162017unuse_pack(&w_curs);2018read_unlock();2019return size;2020}20212022static inttry_delta(struct unpacked *trg,struct unpacked *src,2023unsigned max_depth,unsigned long*mem_usage)2024{2025struct object_entry *trg_entry = trg->entry;2026struct object_entry *src_entry = src->entry;2027unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;2028unsigned ref_depth;2029enum object_type type;2030void*delta_buf;20312032/* Don't bother doing diffs between different types */2033if(oe_type(trg_entry) !=oe_type(src_entry))2034return-1;20352036/*2037 * We do not bother to try a delta that we discarded on an2038 * earlier try, but only when reusing delta data. Note that2039 * src_entry that is marked as the preferred_base should always2040 * be considered, as even if we produce a suboptimal delta against2041 * it, we will still save the transfer cost, as we already know2042 * the other side has it and we won't send src_entry at all.2043 */2044if(reuse_delta &&IN_PACK(trg_entry) &&2045IN_PACK(trg_entry) ==IN_PACK(src_entry) &&2046!src_entry->preferred_base &&2047 trg_entry->in_pack_type != OBJ_REF_DELTA &&2048 trg_entry->in_pack_type != OBJ_OFS_DELTA)2049return0;20502051/* Let's not bust the allowed depth. */2052if(src->depth >= max_depth)2053return0;20542055/* Now some size filtering heuristics. */2056 trg_size =SIZE(trg_entry);2057if(!DELTA(trg_entry)) {2058 max_size = trg_size/2- the_hash_algo->rawsz;2059 ref_depth =1;2060}else{2061 max_size =DELTA_SIZE(trg_entry);2062 ref_depth = trg->depth;2063}2064 max_size = (uint64_t)max_size * (max_depth - src->depth) /2065(max_depth - ref_depth +1);2066if(max_size ==0)2067return0;2068 src_size =SIZE(src_entry);2069 sizediff = src_size < trg_size ? trg_size - src_size :0;2070if(sizediff >= max_size)2071return0;2072if(trg_size < src_size /32)2073return0;20742075if(!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))2076return0;20772078/* Load data if not already done */2079if(!trg->data) {2080read_lock();2081 trg->data =read_object_file(&trg_entry->idx.oid, &type, &sz);2082read_unlock();2083if(!trg->data)2084die(_("object%scannot be read"),2085oid_to_hex(&trg_entry->idx.oid));2086if(sz != trg_size)2087die(_("object%sinconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2088oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,2089(uintmax_t)trg_size);2090*mem_usage += sz;2091}2092if(!src->data) {2093read_lock();2094 src->data =read_object_file(&src_entry->idx.oid, &type, &sz);2095read_unlock();2096if(!src->data) {2097if(src_entry->preferred_base) {2098static int warned =0;2099if(!warned++)2100warning(_("object%scannot be read"),2101oid_to_hex(&src_entry->idx.oid));2102/*2103 * Those objects are not included in the2104 * resulting pack. Be resilient and ignore2105 * them if they can't be read, in case the2106 * pack could be created nevertheless.2107 */2108return0;2109}2110die(_("object%scannot be read"),2111oid_to_hex(&src_entry->idx.oid));2112}2113if(sz != src_size)2114die(_("object%sinconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2115oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,2116(uintmax_t)src_size);2117*mem_usage += sz;2118}2119if(!src->index) {2120 src->index =create_delta_index(src->data, src_size);2121if(!src->index) {2122static int warned =0;2123if(!warned++)2124warning(_("suboptimal pack - out of memory"));2125return0;2126}2127*mem_usage +=sizeof_delta_index(src->index);2128}21292130 delta_buf =create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2131if(!delta_buf)2132return0;21332134if(DELTA(trg_entry)) {2135/* Prefer only shallower same-sized deltas. */2136if(delta_size ==DELTA_SIZE(trg_entry) &&2137 src->depth +1>= trg->depth) {2138free(delta_buf);2139return0;2140}2141}21422143/*2144 * Handle memory allocation outside of the cache2145 * accounting lock. Compiler will optimize the strangeness2146 * away when NO_PTHREADS is defined.2147 */2148free(trg_entry->delta_data);2149cache_lock();2150if(trg_entry->delta_data) {2151 delta_cache_size -=DELTA_SIZE(trg_entry);2152 trg_entry->delta_data = NULL;2153}2154if(delta_cacheable(src_size, trg_size, delta_size)) {2155 delta_cache_size += delta_size;2156cache_unlock();2157 trg_entry->delta_data =xrealloc(delta_buf, delta_size);2158}else{2159cache_unlock();2160free(delta_buf);2161}21622163SET_DELTA(trg_entry, src_entry);2164SET_DELTA_SIZE(trg_entry, delta_size);2165 trg->depth = src->depth +1;21662167return1;2168}21692170static unsigned intcheck_delta_limit(struct object_entry *me,unsigned int n)2171{2172struct object_entry *child =DELTA_CHILD(me);2173unsigned int m = n;2174while(child) {2175unsigned int c =check_delta_limit(child, n +1);2176if(m < c)2177 m = c;2178 child =DELTA_SIBLING(child);2179}2180return m;2181}21822183static unsigned longfree_unpacked(struct unpacked *n)2184{2185unsigned long freed_mem =sizeof_delta_index(n->index);2186free_delta_index(n->index);2187 n->index = NULL;2188if(n->data) {2189 freed_mem +=SIZE(n->entry);2190FREE_AND_NULL(n->data);2191}2192 n->entry = NULL;2193 n->depth =0;2194return freed_mem;2195}21962197static voidfind_deltas(struct object_entry **list,unsigned*list_size,2198int window,int depth,unsigned*processed)2199{2200uint32_t i, idx =0, count =0;2201struct unpacked *array;2202unsigned long mem_usage =0;22032204 array =xcalloc(window,sizeof(struct unpacked));22052206for(;;) {2207struct object_entry *entry;2208struct unpacked *n = array + idx;2209int j, max_depth, best_base = -1;22102211progress_lock();2212if(!*list_size) {2213progress_unlock();2214break;2215}2216 entry = *list++;2217(*list_size)--;2218if(!entry->preferred_base) {2219(*processed)++;2220display_progress(progress_state, *processed);2221}2222progress_unlock();22232224 mem_usage -=free_unpacked(n);2225 n->entry = entry;22262227while(window_memory_limit &&2228 mem_usage > window_memory_limit &&2229 count >1) {2230uint32_t tail = (idx + window - count) % window;2231 mem_usage -=free_unpacked(array + tail);2232 count--;2233}22342235/* We do not compute delta to *create* objects we are not2236 * going to pack.2237 */2238if(entry->preferred_base)2239goto next;22402241/*2242 * If the current object is at pack edge, take the depth the2243 * objects that depend on the current object into account2244 * otherwise they would become too deep.2245 */2246 max_depth = depth;2247if(DELTA_CHILD(entry)) {2248 max_depth -=check_delta_limit(entry,0);2249if(max_depth <=0)2250goto next;2251}22522253 j = window;2254while(--j >0) {2255int ret;2256uint32_t other_idx = idx + j;2257struct unpacked *m;2258if(other_idx >= window)2259 other_idx -= window;2260 m = array + other_idx;2261if(!m->entry)2262break;2263 ret =try_delta(n, m, max_depth, &mem_usage);2264if(ret <0)2265break;2266else if(ret >0)2267 best_base = other_idx;2268}22692270/*2271 * If we decided to cache the delta data, then it is best2272 * to compress it right away. First because we have to do2273 * it anyway, and doing it here while we're threaded will2274 * save a lot of time in the non threaded write phase,2275 * as well as allow for caching more deltas within2276 * the same cache size limit.2277 * ...2278 * But only if not writing to stdout, since in that case2279 * the network is most likely throttling writes anyway,2280 * and therefore it is best to go to the write phase ASAP2281 * instead, as we can afford spending more time compressing2282 * between writes at that moment.2283 */2284if(entry->delta_data && !pack_to_stdout) {2285unsigned long size;22862287 size =do_compress(&entry->delta_data,DELTA_SIZE(entry));2288if(size < (1U<< OE_Z_DELTA_BITS)) {2289 entry->z_delta_size = size;2290cache_lock();2291 delta_cache_size -=DELTA_SIZE(entry);2292 delta_cache_size += entry->z_delta_size;2293cache_unlock();2294}else{2295FREE_AND_NULL(entry->delta_data);2296 entry->z_delta_size =0;2297}2298}22992300/* if we made n a delta, and if n is already at max2301 * depth, leaving it in the window is pointless. we2302 * should evict it first.2303 */2304if(DELTA(entry) && max_depth <= n->depth)2305continue;23062307/*2308 * Move the best delta base up in the window, after the2309 * currently deltified object, to keep it longer. It will2310 * be the first base object to be attempted next.2311 */2312if(DELTA(entry)) {2313struct unpacked swap = array[best_base];2314int dist = (window + idx - best_base) % window;2315int dst = best_base;2316while(dist--) {2317int src = (dst +1) % window;2318 array[dst] = array[src];2319 dst = src;2320}2321 array[dst] = swap;2322}23232324 next:2325 idx++;2326if(count +1< window)2327 count++;2328if(idx >= window)2329 idx =0;2330}23312332for(i =0; i < window; ++i) {2333free_delta_index(array[i].index);2334free(array[i].data);2335}2336free(array);2337}23382339static voidtry_to_free_from_threads(size_t size)2340{2341read_lock();2342release_pack_memory(size);2343read_unlock();2344}23452346static try_to_free_t old_try_to_free_routine;23472348/*2349 * The main object list is split into smaller lists, each is handed to2350 * one worker.2351 *2352 * The main thread waits on the condition that (at least) one of the workers2353 * has stopped working (which is indicated in the .working member of2354 * struct thread_params).2355 *2356 * When a work thread has completed its work, it sets .working to 0 and2357 * signals the main thread and waits on the condition that .data_ready2358 * becomes 1.2359 *2360 * The main thread steals half of the work from the worker that has2361 * most work left to hand it to the idle worker.2362 */23632364struct thread_params {2365 pthread_t thread;2366struct object_entry **list;2367unsigned list_size;2368unsigned remaining;2369int window;2370int depth;2371int working;2372int data_ready;2373 pthread_mutex_t mutex;2374 pthread_cond_t cond;2375unsigned*processed;2376};23772378static pthread_cond_t progress_cond;23792380/*2381 * Mutex and conditional variable can't be statically-initialized on Windows.2382 */2383static voidinit_threaded_search(void)2384{2385init_recursive_mutex(&read_mutex);2386pthread_mutex_init(&cache_mutex, NULL);2387pthread_mutex_init(&progress_mutex, NULL);2388pthread_cond_init(&progress_cond, NULL);2389 old_try_to_free_routine =set_try_to_free_routine(try_to_free_from_threads);2390}23912392static voidcleanup_threaded_search(void)2393{2394set_try_to_free_routine(old_try_to_free_routine);2395pthread_cond_destroy(&progress_cond);2396pthread_mutex_destroy(&read_mutex);2397pthread_mutex_destroy(&cache_mutex);2398pthread_mutex_destroy(&progress_mutex);2399}24002401static void*threaded_find_deltas(void*arg)2402{2403struct thread_params *me = arg;24042405progress_lock();2406while(me->remaining) {2407progress_unlock();24082409find_deltas(me->list, &me->remaining,2410 me->window, me->depth, me->processed);24112412progress_lock();2413 me->working =0;2414pthread_cond_signal(&progress_cond);2415progress_unlock();24162417/*2418 * We must not set ->data_ready before we wait on the2419 * condition because the main thread may have set it to 12420 * before we get here. In order to be sure that new2421 * work is available if we see 1 in ->data_ready, it2422 * was initialized to 0 before this thread was spawned2423 * and we reset it to 0 right away.2424 */2425pthread_mutex_lock(&me->mutex);2426while(!me->data_ready)2427pthread_cond_wait(&me->cond, &me->mutex);2428 me->data_ready =0;2429pthread_mutex_unlock(&me->mutex);24302431progress_lock();2432}2433progress_unlock();2434/* leave ->working 1 so that this doesn't get more work assigned */2435return NULL;2436}24372438static voidll_find_deltas(struct object_entry **list,unsigned list_size,2439int window,int depth,unsigned*processed)2440{2441struct thread_params *p;2442int i, ret, active_threads =0;24432444init_threaded_search();24452446if(delta_search_threads <=1) {2447find_deltas(list, &list_size, window, depth, processed);2448cleanup_threaded_search();2449return;2450}2451if(progress > pack_to_stdout)2452fprintf_ln(stderr,_("Delta compression using up to%dthreads"),2453 delta_search_threads);2454 p =xcalloc(delta_search_threads,sizeof(*p));24552456/* Partition the work amongst work threads. */2457for(i =0; i < delta_search_threads; i++) {2458unsigned sub_size = list_size / (delta_search_threads - i);24592460/* don't use too small segments or no deltas will be found */2461if(sub_size <2*window && i+1< delta_search_threads)2462 sub_size =0;24632464 p[i].window = window;2465 p[i].depth = depth;2466 p[i].processed = processed;2467 p[i].working =1;2468 p[i].data_ready =0;24692470/* try to split chunks on "path" boundaries */2471while(sub_size && sub_size < list_size &&2472 list[sub_size]->hash &&2473 list[sub_size]->hash == list[sub_size-1]->hash)2474 sub_size++;24752476 p[i].list = list;2477 p[i].list_size = sub_size;2478 p[i].remaining = sub_size;24792480 list += sub_size;2481 list_size -= sub_size;2482}24832484/* Start work threads. */2485for(i =0; i < delta_search_threads; i++) {2486if(!p[i].list_size)2487continue;2488pthread_mutex_init(&p[i].mutex, NULL);2489pthread_cond_init(&p[i].cond, NULL);2490 ret =pthread_create(&p[i].thread, NULL,2491 threaded_find_deltas, &p[i]);2492if(ret)2493die(_("unable to create thread:%s"),strerror(ret));2494 active_threads++;2495}24962497/*2498 * Now let's wait for work completion. Each time a thread is done2499 * with its work, we steal half of the remaining work from the2500 * thread with the largest number of unprocessed objects and give2501 * it to that newly idle thread. This ensure good load balancing2502 * until the remaining object list segments are simply too short2503 * to be worth splitting anymore.2504 */2505while(active_threads) {2506struct thread_params *target = NULL;2507struct thread_params *victim = NULL;2508unsigned sub_size =0;25092510progress_lock();2511for(;;) {2512for(i =0; !target && i < delta_search_threads; i++)2513if(!p[i].working)2514 target = &p[i];2515if(target)2516break;2517pthread_cond_wait(&progress_cond, &progress_mutex);2518}25192520for(i =0; i < delta_search_threads; i++)2521if(p[i].remaining >2*window &&2522(!victim || victim->remaining < p[i].remaining))2523 victim = &p[i];2524if(victim) {2525 sub_size = victim->remaining /2;2526 list = victim->list + victim->list_size - sub_size;2527while(sub_size && list[0]->hash &&2528 list[0]->hash == list[-1]->hash) {2529 list++;2530 sub_size--;2531}2532if(!sub_size) {2533/*2534 * It is possible for some "paths" to have2535 * so many objects that no hash boundary2536 * might be found. Let's just steal the2537 * exact half in that case.2538 */2539 sub_size = victim->remaining /2;2540 list -= sub_size;2541}2542 target->list = list;2543 victim->list_size -= sub_size;2544 victim->remaining -= sub_size;2545}2546 target->list_size = sub_size;2547 target->remaining = sub_size;2548 target->working =1;2549progress_unlock();25502551pthread_mutex_lock(&target->mutex);2552 target->data_ready =1;2553pthread_cond_signal(&target->cond);2554pthread_mutex_unlock(&target->mutex);25552556if(!sub_size) {2557pthread_join(target->thread, NULL);2558pthread_cond_destroy(&target->cond);2559pthread_mutex_destroy(&target->mutex);2560 active_threads--;2561}2562}2563cleanup_threaded_search();2564free(p);2565}25662567static voidadd_tag_chain(const struct object_id *oid)2568{2569struct tag *tag;25702571/*2572 * We catch duplicates already in add_object_entry(), but we'd2573 * prefer to do this extra check to avoid having to parse the2574 * tag at all if we already know that it's being packed (e.g., if2575 * it was included via bitmaps, we would not have parsed it2576 * previously).2577 */2578if(packlist_find(&to_pack, oid->hash, NULL))2579return;25802581 tag =lookup_tag(the_repository, oid);2582while(1) {2583if(!tag ||parse_tag(tag) || !tag->tagged)2584die(_("unable to pack objects reachable from tag%s"),2585oid_to_hex(oid));25862587add_object_entry(&tag->object.oid, OBJ_TAG, NULL,0);25882589if(tag->tagged->type != OBJ_TAG)2590return;25912592 tag = (struct tag *)tag->tagged;2593}2594}25952596static intadd_ref_tag(const char*path,const struct object_id *oid,int flag,void*cb_data)2597{2598struct object_id peeled;25992600if(starts_with(path,"refs/tags/") &&/* is a tag? */2601!peel_ref(path, &peeled) &&/* peelable? */2602packlist_find(&to_pack, peeled.hash, NULL))/* object packed? */2603add_tag_chain(oid);2604return0;2605}26062607static voidprepare_pack(int window,int depth)2608{2609struct object_entry **delta_list;2610uint32_t i, nr_deltas;2611unsigned n;26122613if(use_delta_islands)2614resolve_tree_islands(progress, &to_pack);26152616get_object_details();26172618/*2619 * If we're locally repacking then we need to be doubly careful2620 * from now on in order to make sure no stealth corruption gets2621 * propagated to the new pack. Clients receiving streamed packs2622 * should validate everything they get anyway so no need to incur2623 * the additional cost here in that case.2624 */2625if(!pack_to_stdout)2626 do_check_packed_object_crc =1;26272628if(!to_pack.nr_objects || !window || !depth)2629return;26302631ALLOC_ARRAY(delta_list, to_pack.nr_objects);2632 nr_deltas = n =0;26332634for(i =0; i < to_pack.nr_objects; i++) {2635struct object_entry *entry = to_pack.objects + i;26362637if(DELTA(entry))2638/* This happens if we decided to reuse existing2639 * delta from a pack. "reuse_delta &&" is implied.2640 */2641continue;26422643if(!entry->type_valid ||2644oe_size_less_than(&to_pack, entry,50))2645continue;26462647if(entry->no_try_delta)2648continue;26492650if(!entry->preferred_base) {2651 nr_deltas++;2652if(oe_type(entry) <0)2653die(_("unable to get type of object%s"),2654oid_to_hex(&entry->idx.oid));2655}else{2656if(oe_type(entry) <0) {2657/*2658 * This object is not found, but we2659 * don't have to include it anyway.2660 */2661continue;2662}2663}26642665 delta_list[n++] = entry;2666}26672668if(nr_deltas && n >1) {2669unsigned nr_done =0;2670if(progress)2671 progress_state =start_progress(_("Compressing objects"),2672 nr_deltas);2673QSORT(delta_list, n, type_size_sort);2674ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2675stop_progress(&progress_state);2676if(nr_done != nr_deltas)2677die(_("inconsistency with delta count"));2678}2679free(delta_list);2680}26812682static intgit_pack_config(const char*k,const char*v,void*cb)2683{2684if(!strcmp(k,"pack.window")) {2685 window =git_config_int(k, v);2686return0;2687}2688if(!strcmp(k,"pack.windowmemory")) {2689 window_memory_limit =git_config_ulong(k, v);2690return0;2691}2692if(!strcmp(k,"pack.depth")) {2693 depth =git_config_int(k, v);2694return0;2695}2696if(!strcmp(k,"pack.deltacachesize")) {2697 max_delta_cache_size =git_config_int(k, v);2698return0;2699}2700if(!strcmp(k,"pack.deltacachelimit")) {2701 cache_max_small_delta_size =git_config_int(k, v);2702return0;2703}2704if(!strcmp(k,"pack.writebitmaphashcache")) {2705if(git_config_bool(k, v))2706 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2707else2708 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2709}2710if(!strcmp(k,"pack.usebitmaps")) {2711 use_bitmap_index_default =git_config_bool(k, v);2712return0;2713}2714if(!strcmp(k,"pack.usesparse")) {2715 sparse =git_config_bool(k, v);2716return0;2717}2718if(!strcmp(k,"pack.threads")) {2719 delta_search_threads =git_config_int(k, v);2720if(delta_search_threads <0)2721die(_("invalid number of threads specified (%d)"),2722 delta_search_threads);2723if(!HAVE_THREADS && delta_search_threads !=1) {2724warning(_("no threads support, ignoring%s"), k);2725 delta_search_threads =0;2726}2727return0;2728}2729if(!strcmp(k,"pack.indexversion")) {2730 pack_idx_opts.version =git_config_int(k, v);2731if(pack_idx_opts.version >2)2732die(_("bad pack.indexversion=%"PRIu32),2733 pack_idx_opts.version);2734return0;2735}2736returngit_default_config(k, v, cb);2737}27382739static voidread_object_list_from_stdin(void)2740{2741char line[GIT_MAX_HEXSZ +1+ PATH_MAX +2];2742struct object_id oid;2743const char*p;27442745for(;;) {2746if(!fgets(line,sizeof(line), stdin)) {2747if(feof(stdin))2748break;2749if(!ferror(stdin))2750die("BUG: fgets returned NULL, not EOF, not error!");2751if(errno != EINTR)2752die_errno("fgets");2753clearerr(stdin);2754continue;2755}2756if(line[0] =='-') {2757if(get_oid_hex(line+1, &oid))2758die(_("expected edge object ID, got garbage:\n%s"),2759 line);2760add_preferred_base(&oid);2761continue;2762}2763if(parse_oid_hex(line, &oid, &p))2764die(_("expected object ID, got garbage:\n%s"), line);27652766add_preferred_base_object(p +1);2767add_object_entry(&oid, OBJ_NONE, p +1,0);2768}2769}27702771/* Remember to update object flag allocation in object.h */2772#define OBJECT_ADDED (1u<<20)27732774static voidshow_commit(struct commit *commit,void*data)2775{2776add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL,0);2777 commit->object.flags |= OBJECT_ADDED;27782779if(write_bitmap_index)2780index_commit_for_bitmap(commit);27812782if(use_delta_islands)2783propagate_island_marks(commit);2784}27852786static voidshow_object(struct object *obj,const char*name,void*data)2787{2788add_preferred_base_object(name);2789add_object_entry(&obj->oid, obj->type, name,0);2790 obj->flags |= OBJECT_ADDED;27912792if(use_delta_islands) {2793const char*p;2794unsigned depth;2795struct object_entry *ent;27962797/* the empty string is a root tree, which is depth 0 */2798 depth = *name ?1:0;2799for(p =strchr(name,'/'); p; p =strchr(p +1,'/'))2800 depth++;28012802 ent =packlist_find(&to_pack, obj->oid.hash, NULL);2803if(ent && depth >oe_tree_depth(&to_pack, ent))2804oe_set_tree_depth(&to_pack, ent, depth);2805}2806}28072808static voidshow_object__ma_allow_any(struct object *obj,const char*name,void*data)2809{2810assert(arg_missing_action == MA_ALLOW_ANY);28112812/*2813 * Quietly ignore ALL missing objects. This avoids problems with2814 * staging them now and getting an odd error later.2815 */2816if(!has_object_file(&obj->oid))2817return;28182819show_object(obj, name, data);2820}28212822static voidshow_object__ma_allow_promisor(struct object *obj,const char*name,void*data)2823{2824assert(arg_missing_action == MA_ALLOW_PROMISOR);28252826/*2827 * Quietly ignore EXPECTED missing objects. This avoids problems with2828 * staging them now and getting an odd error later.2829 */2830if(!has_object_file(&obj->oid) &&is_promisor_object(&obj->oid))2831return;28322833show_object(obj, name, data);2834}28352836static intoption_parse_missing_action(const struct option *opt,2837const char*arg,int unset)2838{2839assert(arg);2840assert(!unset);28412842if(!strcmp(arg,"error")) {2843 arg_missing_action = MA_ERROR;2844 fn_show_object = show_object;2845return0;2846}28472848if(!strcmp(arg,"allow-any")) {2849 arg_missing_action = MA_ALLOW_ANY;2850 fetch_if_missing =0;2851 fn_show_object = show_object__ma_allow_any;2852return0;2853}28542855if(!strcmp(arg,"allow-promisor")) {2856 arg_missing_action = MA_ALLOW_PROMISOR;2857 fetch_if_missing =0;2858 fn_show_object = show_object__ma_allow_promisor;2859return0;2860}28612862die(_("invalid value for --missing"));2863return0;2864}28652866static voidshow_edge(struct commit *commit)2867{2868add_preferred_base(&commit->object.oid);2869}28702871struct in_pack_object {2872 off_t offset;2873struct object *object;2874};28752876struct in_pack {2877unsigned int alloc;2878unsigned int nr;2879struct in_pack_object *array;2880};28812882static voidmark_in_pack_object(struct object *object,struct packed_git *p,struct in_pack *in_pack)2883{2884 in_pack->array[in_pack->nr].offset =find_pack_entry_one(object->oid.hash, p);2885 in_pack->array[in_pack->nr].object = object;2886 in_pack->nr++;2887}28882889/*2890 * Compare the objects in the offset order, in order to emulate the2891 * "git rev-list --objects" output that produced the pack originally.2892 */2893static intofscmp(const void*a_,const void*b_)2894{2895struct in_pack_object *a = (struct in_pack_object *)a_;2896struct in_pack_object *b = (struct in_pack_object *)b_;28972898if(a->offset < b->offset)2899return-1;2900else if(a->offset > b->offset)2901return1;2902else2903returnoidcmp(&a->object->oid, &b->object->oid);2904}29052906static voidadd_objects_in_unpacked_packs(struct rev_info *revs)2907{2908struct packed_git *p;2909struct in_pack in_pack;2910uint32_t i;29112912memset(&in_pack,0,sizeof(in_pack));29132914for(p =get_all_packs(the_repository); p; p = p->next) {2915struct object_id oid;2916struct object *o;29172918if(!p->pack_local || p->pack_keep || p->pack_keep_in_core)2919continue;2920if(open_pack_index(p))2921die(_("cannot open pack index"));29222923ALLOC_GROW(in_pack.array,2924 in_pack.nr + p->num_objects,2925 in_pack.alloc);29262927for(i =0; i < p->num_objects; i++) {2928nth_packed_object_oid(&oid, p, i);2929 o =lookup_unknown_object(oid.hash);2930if(!(o->flags & OBJECT_ADDED))2931mark_in_pack_object(o, p, &in_pack);2932 o->flags |= OBJECT_ADDED;2933}2934}29352936if(in_pack.nr) {2937QSORT(in_pack.array, in_pack.nr, ofscmp);2938for(i =0; i < in_pack.nr; i++) {2939struct object *o = in_pack.array[i].object;2940add_object_entry(&o->oid, o->type,"",0);2941}2942}2943free(in_pack.array);2944}29452946static intadd_loose_object(const struct object_id *oid,const char*path,2947void*data)2948{2949enum object_type type =oid_object_info(the_repository, oid, NULL);29502951if(type <0) {2952warning(_("loose object at%scould not be examined"), path);2953return0;2954}29552956add_object_entry(oid, type,"",0);2957return0;2958}29592960/*2961 * We actually don't even have to worry about reachability here.2962 * add_object_entry will weed out duplicates, so we just add every2963 * loose object we find.2964 */2965static voidadd_unreachable_loose_objects(void)2966{2967for_each_loose_file_in_objdir(get_object_directory(),2968 add_loose_object,2969 NULL, NULL, NULL);2970}29712972static inthas_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2973{2974static struct packed_git *last_found = (void*)1;2975struct packed_git *p;29762977 p = (last_found != (void*)1) ? last_found :2978get_all_packs(the_repository);29792980while(p) {2981if((!p->pack_local || p->pack_keep ||2982 p->pack_keep_in_core) &&2983find_pack_entry_one(oid->hash, p)) {2984 last_found = p;2985return1;2986}2987if(p == last_found)2988 p =get_all_packs(the_repository);2989else2990 p = p->next;2991if(p == last_found)2992 p = p->next;2993}2994return0;2995}29962997/*2998 * Store a list of sha1s that are should not be discarded2999 * because they are either written too recently, or are3000 * reachable from another object that was.3001 *3002 * This is filled by get_object_list.3003 */3004static struct oid_array recent_objects;30053006static intloosened_object_can_be_discarded(const struct object_id *oid,3007 timestamp_t mtime)3008{3009if(!unpack_unreachable_expiration)3010return0;3011if(mtime > unpack_unreachable_expiration)3012return0;3013if(oid_array_lookup(&recent_objects, oid) >=0)3014return0;3015return1;3016}30173018static voidloosen_unused_packed_objects(struct rev_info *revs)3019{3020struct packed_git *p;3021uint32_t i;3022struct object_id oid;30233024for(p =get_all_packs(the_repository); p; p = p->next) {3025if(!p->pack_local || p->pack_keep || p->pack_keep_in_core)3026continue;30273028if(open_pack_index(p))3029die(_("cannot open pack index"));30303031for(i =0; i < p->num_objects; i++) {3032nth_packed_object_oid(&oid, p, i);3033if(!packlist_find(&to_pack, oid.hash, NULL) &&3034!has_sha1_pack_kept_or_nonlocal(&oid) &&3035!loosened_object_can_be_discarded(&oid, p->mtime))3036if(force_object_loose(&oid, p->mtime))3037die(_("unable to force loose object"));3038}3039}3040}30413042/*3043 * This tracks any options which pack-reuse code expects to be on, or which a3044 * reader of the pack might not understand, and which would therefore prevent3045 * blind reuse of what we have on disk.3046 */3047static intpack_options_allow_reuse(void)3048{3049return pack_to_stdout &&3050 allow_ofs_delta &&3051!ignore_packed_keep_on_disk &&3052!ignore_packed_keep_in_core &&3053(!local || !have_non_local_packs) &&3054!incremental;3055}30563057static intget_object_list_from_bitmap(struct rev_info *revs)3058{3059if(!(bitmap_git =prepare_bitmap_walk(revs)))3060return-1;30613062if(pack_options_allow_reuse() &&3063!reuse_partial_packfile_from_bitmap(3064 bitmap_git,3065&reuse_packfile,3066&reuse_packfile_objects,3067&reuse_packfile_offset)) {3068assert(reuse_packfile_objects);3069 nr_result += reuse_packfile_objects;3070display_progress(progress_state, nr_result);3071}30723073traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);3074return0;3075}30763077static voidrecord_recent_object(struct object *obj,3078const char*name,3079void*data)3080{3081oid_array_append(&recent_objects, &obj->oid);3082}30833084static voidrecord_recent_commit(struct commit *commit,void*data)3085{3086oid_array_append(&recent_objects, &commit->object.oid);3087}30883089static voidget_object_list(int ac,const char**av)3090{3091struct rev_info revs;3092char line[1000];3093int flags =0;3094int save_warning;30953096repo_init_revisions(the_repository, &revs, NULL);3097 save_commit_buffer =0;3098 revs.allow_exclude_promisor_objects_opt =1;3099setup_revisions(ac, av, &revs, NULL);31003101/* make sure shallows are read */3102is_repository_shallow(the_repository);31033104 save_warning = warn_on_object_refname_ambiguity;3105 warn_on_object_refname_ambiguity =0;31063107while(fgets(line,sizeof(line), stdin) != NULL) {3108int len =strlen(line);3109if(len && line[len -1] =='\n')3110 line[--len] =0;3111if(!len)3112break;3113if(*line =='-') {3114if(!strcmp(line,"--not")) {3115 flags ^= UNINTERESTING;3116 write_bitmap_index =0;3117continue;3118}3119if(starts_with(line,"--shallow ")) {3120struct object_id oid;3121if(get_oid_hex(line +10, &oid))3122die("not an SHA-1 '%s'", line +10);3123register_shallow(the_repository, &oid);3124 use_bitmap_index =0;3125continue;3126}3127die(_("not a rev '%s'"), line);3128}3129if(handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3130die(_("bad revision '%s'"), line);3131}31323133 warn_on_object_refname_ambiguity = save_warning;31343135if(use_bitmap_index && !get_object_list_from_bitmap(&revs))3136return;31373138if(use_delta_islands)3139load_delta_islands();31403141if(prepare_revision_walk(&revs))3142die(_("revision walk setup failed"));3143mark_edges_uninteresting(&revs, show_edge, sparse);31443145if(!fn_show_object)3146 fn_show_object = show_object;3147traverse_commit_list_filtered(&filter_options, &revs,3148 show_commit, fn_show_object, NULL,3149 NULL);31503151if(unpack_unreachable_expiration) {3152 revs.ignore_missing_links =1;3153if(add_unseen_recent_objects_to_traversal(&revs,3154 unpack_unreachable_expiration))3155die(_("unable to add recent objects"));3156if(prepare_revision_walk(&revs))3157die(_("revision walk setup failed"));3158traverse_commit_list(&revs, record_recent_commit,3159 record_recent_object, NULL);3160}31613162if(keep_unreachable)3163add_objects_in_unpacked_packs(&revs);3164if(pack_loose_unreachable)3165add_unreachable_loose_objects();3166if(unpack_unreachable)3167loosen_unused_packed_objects(&revs);31683169oid_array_clear(&recent_objects);3170}31713172static voidadd_extra_kept_packs(const struct string_list *names)3173{3174struct packed_git *p;31753176if(!names->nr)3177return;31783179for(p =get_all_packs(the_repository); p; p = p->next) {3180const char*name =basename(p->pack_name);3181int i;31823183if(!p->pack_local)3184continue;31853186for(i =0; i < names->nr; i++)3187if(!fspathcmp(name, names->items[i].string))3188break;31893190if(i < names->nr) {3191 p->pack_keep_in_core =1;3192 ignore_packed_keep_in_core =1;3193continue;3194}3195}3196}31973198static intoption_parse_index_version(const struct option *opt,3199const char*arg,int unset)3200{3201char*c;3202const char*val = arg;32033204BUG_ON_OPT_NEG(unset);32053206 pack_idx_opts.version =strtoul(val, &c,10);3207if(pack_idx_opts.version >2)3208die(_("unsupported index version%s"), val);3209if(*c ==','&& c[1])3210 pack_idx_opts.off32_limit =strtoul(c+1, &c,0);3211if(*c || pack_idx_opts.off32_limit &0x80000000)3212die(_("bad index version '%s'"), val);3213return0;3214}32153216static intoption_parse_unpack_unreachable(const struct option *opt,3217const char*arg,int unset)3218{3219if(unset) {3220 unpack_unreachable =0;3221 unpack_unreachable_expiration =0;3222}3223else{3224 unpack_unreachable =1;3225if(arg)3226 unpack_unreachable_expiration =approxidate(arg);3227}3228return0;3229}32303231intcmd_pack_objects(int argc,const char**argv,const char*prefix)3232{3233int use_internal_rev_list =0;3234int shallow =0;3235int all_progress_implied =0;3236struct argv_array rp = ARGV_ARRAY_INIT;3237int rev_list_unpacked =0, rev_list_all =0, rev_list_reflog =0;3238int rev_list_index =0;3239struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3240struct option pack_objects_options[] = {3241OPT_SET_INT('q',"quiet", &progress,3242N_("do not show progress meter"),0),3243OPT_SET_INT(0,"progress", &progress,3244N_("show progress meter"),1),3245OPT_SET_INT(0,"all-progress", &progress,3246N_("show progress meter during object writing phase"),2),3247OPT_BOOL(0,"all-progress-implied",3248&all_progress_implied,3249N_("similar to --all-progress when progress meter is shown")),3250{ OPTION_CALLBACK,0,"index-version", NULL,N_("<version>[,<offset>]"),3251N_("write the pack index file in the specified idx format version"),3252 PARSE_OPT_NONEG, option_parse_index_version },3253OPT_MAGNITUDE(0,"max-pack-size", &pack_size_limit,3254N_("maximum size of each output pack file")),3255OPT_BOOL(0,"local", &local,3256N_("ignore borrowed objects from alternate object store")),3257OPT_BOOL(0,"incremental", &incremental,3258N_("ignore packed objects")),3259OPT_INTEGER(0,"window", &window,3260N_("limit pack window by objects")),3261OPT_MAGNITUDE(0,"window-memory", &window_memory_limit,3262N_("limit pack window by memory in addition to object limit")),3263OPT_INTEGER(0,"depth", &depth,3264N_("maximum length of delta chain allowed in the resulting pack")),3265OPT_BOOL(0,"reuse-delta", &reuse_delta,3266N_("reuse existing deltas")),3267OPT_BOOL(0,"reuse-object", &reuse_object,3268N_("reuse existing objects")),3269OPT_BOOL(0,"delta-base-offset", &allow_ofs_delta,3270N_("use OFS_DELTA objects")),3271OPT_INTEGER(0,"threads", &delta_search_threads,3272N_("use threads when searching for best delta matches")),3273OPT_BOOL(0,"non-empty", &non_empty,3274N_("do not create an empty pack output")),3275OPT_BOOL(0,"revs", &use_internal_rev_list,3276N_("read revision arguments from standard input")),3277OPT_SET_INT_F(0,"unpacked", &rev_list_unpacked,3278N_("limit the objects to those that are not yet packed"),32791, PARSE_OPT_NONEG),3280OPT_SET_INT_F(0,"all", &rev_list_all,3281N_("include objects reachable from any reference"),32821, PARSE_OPT_NONEG),3283OPT_SET_INT_F(0,"reflog", &rev_list_reflog,3284N_("include objects referred by reflog entries"),32851, PARSE_OPT_NONEG),3286OPT_SET_INT_F(0,"indexed-objects", &rev_list_index,3287N_("include objects referred to by the index"),32881, PARSE_OPT_NONEG),3289OPT_BOOL(0,"stdout", &pack_to_stdout,3290N_("output pack to stdout")),3291OPT_BOOL(0,"include-tag", &include_tag,3292N_("include tag objects that refer to objects to be packed")),3293OPT_BOOL(0,"keep-unreachable", &keep_unreachable,3294N_("keep unreachable objects")),3295OPT_BOOL(0,"pack-loose-unreachable", &pack_loose_unreachable,3296N_("pack loose unreachable objects")),3297{ OPTION_CALLBACK,0,"unpack-unreachable", NULL,N_("time"),3298N_("unpack unreachable objects newer than <time>"),3299 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3300OPT_BOOL(0,"sparse", &sparse,3301N_("use the sparse reachability algorithm")),3302OPT_BOOL(0,"thin", &thin,3303N_("create thin packs")),3304OPT_BOOL(0,"shallow", &shallow,3305N_("create packs suitable for shallow fetches")),3306OPT_BOOL(0,"honor-pack-keep", &ignore_packed_keep_on_disk,3307N_("ignore packs that have companion .keep file")),3308OPT_STRING_LIST(0,"keep-pack", &keep_pack_list,N_("name"),3309N_("ignore this pack")),3310OPT_INTEGER(0,"compression", &pack_compression_level,3311N_("pack compression level")),3312OPT_SET_INT(0,"keep-true-parents", &grafts_replace_parents,3313N_("do not hide commits by grafts"),0),3314OPT_BOOL(0,"use-bitmap-index", &use_bitmap_index,3315N_("use a bitmap index if available to speed up counting objects")),3316OPT_BOOL(0,"write-bitmap-index", &write_bitmap_index,3317N_("write a bitmap index together with the pack index")),3318OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3319{ OPTION_CALLBACK,0,"missing", NULL,N_("action"),3320N_("handling for missing objects"), PARSE_OPT_NONEG,3321 option_parse_missing_action },3322OPT_BOOL(0,"exclude-promisor-objects", &exclude_promisor_objects,3323N_("do not pack objects in promisor packfiles")),3324OPT_BOOL(0,"delta-islands", &use_delta_islands,3325N_("respect islands during delta compression")),3326OPT_END(),3327};33283329if(DFS_NUM_STATES > (1<< OE_DFS_STATE_BITS))3330BUG("too many dfs states, increase OE_DFS_STATE_BITS");33313332 read_replace_refs =0;33333334reset_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}34793480prepare_packing_data(&to_pack);34813482if(progress)3483 progress_state =start_progress(_("Enumerating objects"),0);3484if(!use_internal_rev_list)3485read_object_list_from_stdin();3486else{3487get_object_list(rp.argc, rp.argv);3488argv_array_clear(&rp);3489}3490cleanup_preferred_base();3491if(include_tag && nr_result)3492for_each_ref(add_ref_tag, NULL);3493stop_progress(&progress_state);34943495if(non_empty && !nr_result)3496return0;3497if(nr_result)3498prepare_pack(window, depth);3499write_pack_file();3500if(progress)3501fprintf_ln(stderr,3502_("Total %"PRIu32" (delta %"PRIu32"),"3503" reused %"PRIu32" (delta %"PRIu32")"),3504 written, written_delta, reused, reused_delta);3505return0;3506}