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[] = { 51 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), 52 N_("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 void index_commit_for_bitmap(struct commit *commit) 135{ 136 if (indexed_commits_nr >= indexed_commits_alloc) { 137 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2; 138 REALLOC_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{ 146 unsigned long size, base_size, delta_size; 147 void *buf, *base_buf, *delta_buf; 148 enum object_type type; 149 150 buf = read_object_file(&entry->idx.oid, &type, &size); 151 if (!buf) 152 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid)); 153 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type, 154 &base_size); 155 if (!base_buf) 156 die("unable to read %s", 157 oid_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 */ 165 if (!delta_buf || delta_size != DELTA_SIZE(entry)) 166 BUG("delta size changed"); 167 free(buf); 168 free(base_buf); 169 return delta_buf; 170} 171 172static unsigned long do_compress(void **pptr, unsigned long size) 173{ 174 git_zstream stream; 175 void *in, *out; 176 unsigned long maxsize; 177 178 git_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; 189 while (git_deflate(&stream, Z_FINISH) == Z_OK) 190 ; /* nothing */ 191 git_deflate_end(&stream); 192 193 free(in); 194 return stream.total_out; 195} 196 197static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f, 198 const struct object_id *oid) 199{ 200 git_zstream stream; 201 unsigned char ibuf[1024 * 16]; 202 unsigned char obuf[1024 * 16]; 203 unsigned long olen = 0; 204 205 git_deflate_init(&stream, pack_compression_level); 206 207 for (;;) { 208 ssize_t readlen; 209 int zret = Z_OK; 210 readlen = read_istream(st, ibuf, sizeof(ibuf)); 211 if (readlen == -1) 212 die(_("unable to read %s"), oid_to_hex(oid)); 213 214 stream.next_in = ibuf; 215 stream.avail_in = readlen; 216 while ((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); 221 hashwrite(f, obuf, stream.next_out - obuf); 222 olen += stream.next_out - obuf; 223 } 224 if (stream.avail_in) 225 die(_("deflate error (%d)"), zret); 226 if (readlen == 0) { 227 if (zret != Z_STREAM_END) 228 die(_("deflate error (%d)"), zret); 229 break; 230 } 231 } 232 git_deflate_end(&stream); 233 return 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 int check_pack_inflate(struct packed_git *p, 241 struct pack_window **w_curs, 242 off_t offset, 243 off_t len, 244 unsigned long expect) 245{ 246 git_zstream stream; 247 unsigned char fakebuf[4096], *in; 248 int st; 249 250 memset(&stream, 0, sizeof(stream)); 251 git_inflate_init(&stream); 252 do { 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); 260 git_inflate_end(&stream); 261 return (st == Z_STREAM_END && 262 stream.total_out == expect && 263 stream.total_in == len) ? 0 : -1; 264} 265 266static void copy_pack_data(struct hashfile *f, 267 struct packed_git *p, 268 struct pack_window **w_curs, 269 off_t offset, 270 off_t len) 271{ 272 unsigned char *in; 273 unsigned long avail; 274 275 while (len) { 276 in = use_pack(p, w_curs, offset, &avail); 277 if (avail > len) 278 avail = (unsigned long)len; 279 hashwrite(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 long write_no_reuse_object(struct hashfile *f, struct object_entry *entry, 287 unsigned long limit, int usable_delta) 288{ 289 unsigned long size, datalen; 290 unsigned char header[MAX_PACK_OBJECT_HEADER], 291 dheader[MAX_PACK_OBJECT_HEADER]; 292 unsigned hdrlen; 293 enum object_type type; 294 void *buf; 295 struct git_istream *st = NULL; 296 const unsigned hashsz = the_hash_algo->rawsz; 297 298 if (!usable_delta) { 299 if (oe_type(entry) == OBJ_BLOB && 300 oe_size_greater_than(&to_pack, entry, big_file_threshold) && 301 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL) 302 buf = NULL; 303 else { 304 buf = read_object_file(&entry->idx.oid, &type, &size); 305 if (!buf) 306 die(_("unable to read %s"), 307 oid_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 */ 313 FREE_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 } 327 328 if (st) /* large blob case, just assume we don't compress well */ 329 datalen = size; 330 else if (entry->z_delta_size) 331 datalen = entry->z_delta_size; 332 else 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); 341 342 if (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; 349 unsigned pos = sizeof(dheader) - 1; 350 dheader[pos] = ofs & 127; 351 while (ofs >>= 7) 352 dheader[--pos] = 128 | (--ofs & 127); 353 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 354 if (st) 355 close_istream(st); 356 free(buf); 357 return 0; 358 } 359 hashwrite(f, header, hdrlen); 360 hashwrite(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 */ 367 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 368 if (st) 369 close_istream(st); 370 free(buf); 371 return 0; 372 } 373 hashwrite(f, header, hdrlen); 374 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 375 hdrlen += hashsz; 376 } else { 377 if (limit && hdrlen + datalen + hashsz >= limit) { 378 if (st) 379 close_istream(st); 380 free(buf); 381 return 0; 382 } 383 hashwrite(f, header, hdrlen); 384 } 385 if (st) { 386 datalen = write_large_blob_data(st, f, &entry->idx.oid); 387 close_istream(st); 388 } else { 389 hashwrite(f, buf, datalen); 390 free(buf); 391 } 392 393 return 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, 398 unsigned long limit, int usable_delta) 399{ 400 struct packed_git *p = IN_PACK(entry); 401 struct pack_window *w_curs = NULL; 402 struct revindex_entry *revidx; 403 off_t offset; 404 enum object_type type = oe_type(entry); 405 off_t datalen; 406 unsigned char header[MAX_PACK_OBJECT_HEADER], 407 dheader[MAX_PACK_OBJECT_HEADER]; 408 unsigned hdrlen; 409 const unsigned hashsz = the_hash_algo->rawsz; 410 unsigned long entry_size = SIZE(entry); 411 412 if (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; 421 if (!pack_to_stdout && p->index_version > 1 && 422 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 423 error(_("bad packed object CRC for %s"), 424 oid_to_hex(&entry->idx.oid)); 425 unuse_pack(&w_curs); 426 return write_no_reuse_object(f, entry, limit, usable_delta); 427 } 428 429 offset += entry->in_pack_header_size; 430 datalen -= entry->in_pack_header_size; 431 432 if (!pack_to_stdout && p->index_version == 1 && 433 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { 434 error(_("corrupt packed object for %s"), 435 oid_to_hex(&entry->idx.oid)); 436 unuse_pack(&w_curs); 437 return write_no_reuse_object(f, entry, limit, usable_delta); 438 } 439 440 if (type == OBJ_OFS_DELTA) { 441 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 442 unsigned pos = sizeof(dheader) - 1; 443 dheader[pos] = ofs & 127; 444 while (ofs >>= 7) 445 dheader[--pos] = 128 | (--ofs & 127); 446 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 447 unuse_pack(&w_curs); 448 return 0; 449 } 450 hashwrite(f, header, hdrlen); 451 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 452 hdrlen += sizeof(dheader) - pos; 453 reused_delta++; 454 } else if (type == OBJ_REF_DELTA) { 455 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 456 unuse_pack(&w_curs); 457 return 0; 458 } 459 hashwrite(f, header, hdrlen); 460 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 461 hdrlen += hashsz; 462 reused_delta++; 463 } else { 464 if (limit && hdrlen + datalen + hashsz >= limit) { 465 unuse_pack(&w_curs); 466 return 0; 467 } 468 hashwrite(f, header, hdrlen); 469 } 470 copy_pack_data(f, p, &w_curs, offset, datalen); 471 unuse_pack(&w_curs); 472 reused++; 473 return hdrlen + datalen; 474} 475 476/* Return 0 if we will bust the pack-size limit */ 477static off_t write_object(struct hashfile *f, 478 struct object_entry *entry, 479 off_t write_offset) 480{ 481 unsigned long limit; 482 off_t len; 483 int usable_delta, to_reuse; 484 485 if (!pack_to_stdout) 486 crc32_begin(f); 487 488 /* apply size limit if limited packsize and not first object */ 489 if (!pack_size_limit || !nr_written) 490 limit = 0; 491 else 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; 497 else 498 limit = pack_size_limit - write_offset; 499 500 if (!DELTA(entry)) 501 usable_delta = 0; /* no delta */ 502 else if (!pack_size_limit) 503 usable_delta = 1; /* unlimited packfile */ 504 else if (DELTA(entry)->idx.offset == (off_t)-1) 505 usable_delta = 0; /* base was written to another pack */ 506 else if (DELTA(entry)->idx.offset) 507 usable_delta = 1; /* base already exists in this pack */ 508 else 509 usable_delta = 0; /* base could end up in another pack */ 510 511 if (!reuse_object) 512 to_reuse = 0; /* explicit */ 513 else if (!IN_PACK(entry)) 514 to_reuse = 0; /* can't reuse what we don't have */ 515 else if (oe_type(entry) == OBJ_REF_DELTA || 516 oe_type(entry) == OBJ_OFS_DELTA) 517 /* check_object() decided it for us ... */ 518 to_reuse = usable_delta; 519 /* ... but pack split may override that */ 520 else if (oe_type(entry) != entry->in_pack_type) 521 to_reuse = 0; /* pack has delta which is unusable */ 522 else if (DELTA(entry)) 523 to_reuse = 0; /* we want to pack afresh */ 524 else 525 to_reuse = 1; /* we have it in-pack undeltified, 526 * and we do not need to deltify it. 527 */ 528 529 if (!to_reuse) 530 len = write_no_reuse_object(f, entry, limit, usable_delta); 531 else 532 len = write_reuse_object(f, entry, limit, usable_delta); 533 if (!len) 534 return 0; 535 536 if (usable_delta) 537 written_delta++; 538 written++; 539 if (!pack_to_stdout) 540 entry->idx.crc32 = crc32_end(f); 541 return 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, 552 struct object_entry *e, 553 off_t *offset) 554{ 555 off_t size; 556 int 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); 564 if (recursing) { 565 warning(_("recursive delta detected for object %s"), 566 oid_to_hex(&e->idx.oid)); 567 return WRITE_ONE_RECURSIVE; 568 } else if (e->idx.offset || e->preferred_base) { 569 /* offset is non zero if object is written already. */ 570 return WRITE_ONE_SKIP; 571 } 572 573 /* if we are deltified, write out base object first. */ 574 if (DELTA(e)) { 575 e->idx.offset = 1; /* now recurse */ 576 switch (write_one(f, DELTA(e), offset)) { 577 case WRITE_ONE_RECURSIVE: 578 /* we cannot depend on this one */ 579 SET_DELTA(e, NULL); 580 break; 581 default: 582 break; 583 case WRITE_ONE_BREAK: 584 e->idx.offset = recursing; 585 return WRITE_ONE_BREAK; 586 } 587 } 588 589 e->idx.offset = *offset; 590 size = write_object(f, e, *offset); 591 if (!size) { 592 e->idx.offset = recursing; 593 return WRITE_ONE_BREAK; 594 } 595 written_list[nr_written++] = &e->idx; 596 597 /* make sure off_t is sufficiently large not to wrap */ 598 if (signed_add_overflows(*offset, size)) 599 die(_("pack too large for current definition of off_t")); 600 *offset += size; 601 return WRITE_ONE_WRITTEN; 602} 603 604static int mark_tagged(const char *path, const struct object_id *oid, int flag, 605 void *cb_data) 606{ 607 struct object_id peeled; 608 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL); 609 610 if (entry) 611 entry->tagged = 1; 612 if (!peel_ref(path, &peeled)) { 613 entry = packlist_find(&to_pack, peeled.hash, NULL); 614 if (entry) 615 entry->tagged = 1; 616 } 617 return 0; 618} 619 620static inline void add_to_write_order(struct object_entry **wo, 621 unsigned int *endp, 622 struct object_entry *e) 623{ 624 if (e->filled || oe_layer(&to_pack, e) != write_layer) 625 return; 626 wo[(*endp)++] = e; 627 e->filled = 1; 628} 629 630static void add_descendants_to_write_order(struct object_entry **wo, 631 unsigned int *endp, 632 struct object_entry *e) 633{ 634 int add_to_order = 1; 635 while (e) { 636 if (add_to_order) { 637 struct object_entry *s; 638 /* add this node... */ 639 add_to_write_order(wo, endp, e); 640 /* all its siblings... */ 641 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) { 642 add_to_write_order(wo, endp, s); 643 } 644 } 645 /* drop down a level to add left subtree nodes if possible */ 646 if (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 */ 652 if (DELTA_SIBLING(e)) { 653 e = DELTA_SIBLING(e); 654 continue; 655 } 656 /* go back to our parent node */ 657 e = DELTA(e); 658 while (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 } 663 if (!e) { 664 /* done- we hit our original root node */ 665 return; 666 } 667 /* pass it off to sibling at this level */ 668 e = DELTA_SIBLING(e); 669 } 670 }; 671} 672 673static void add_family_to_write_order(struct object_entry **wo, 674 unsigned int *endp, 675 struct object_entry *e) 676{ 677 struct object_entry *root; 678 679 for (root = e; DELTA(root); root = DELTA(root)) 680 ; /* nothing */ 681 add_descendants_to_write_order(wo, endp, root); 682} 683 684static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end) 685{ 686 unsigned int i, last_untagged; 687 struct object_entry *objects = to_pack.objects; 688 689 for (i = 0; i < to_pack.nr_objects; i++) { 690 if (objects[i].tagged) 691 break; 692 add_to_write_order(wo, wo_end, &objects[i]); 693 } 694 last_untagged = i; 695 696 /* 697 * Then fill all the tagged tips. 698 */ 699 for (; i < to_pack.nr_objects; i++) { 700 if (objects[i].tagged) 701 add_to_write_order(wo, wo_end, &objects[i]); 702 } 703 704 /* 705 * And then all remaining commits and tags. 706 */ 707 for (i = last_untagged; i < to_pack.nr_objects; i++) { 708 if (oe_type(&objects[i]) != OBJ_COMMIT && 709 oe_type(&objects[i]) != OBJ_TAG) 710 continue; 711 add_to_write_order(wo, wo_end, &objects[i]); 712 } 713 714 /* 715 * And then all the trees. 716 */ 717 for (i = last_untagged; i < to_pack.nr_objects; i++) { 718 if (oe_type(&objects[i]) != OBJ_TREE) 719 continue; 720 add_to_write_order(wo, wo_end, &objects[i]); 721 } 722 723 /* 724 * Finally all the rest in really tight order 725 */ 726 for (i = last_untagged; i < to_pack.nr_objects; i++) { 727 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer) 728 add_family_to_write_order(wo, wo_end, &objects[i]); 729 } 730} 731 732static struct object_entry **compute_write_order(void) 733{ 734 uint32_t max_layers = 1; 735 unsigned int i, wo_end; 736 737 struct object_entry **wo; 738 struct object_entry *objects = to_pack.objects; 739 740 for (i = 0; i < to_pack.nr_objects; i++) { 741 objects[i].tagged = 0; 742 objects[i].filled = 0; 743 SET_DELTA_CHILD(&objects[i], NULL); 744 SET_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 */ 752 for (i = to_pack.nr_objects; i > 0;) { 753 struct object_entry *e = &objects[--i]; 754 if (!DELTA(e)) 755 continue; 756 /* Mark me as the first child */ 757 e->delta_sibling_idx = DELTA(e)->delta_child_idx; 758 SET_DELTA_CHILD(DELTA(e), e); 759 } 760 761 /* 762 * Mark objects that are at the tip of tags. 763 */ 764 for_each_tag_ref(mark_tagged, NULL); 765 766 if (use_delta_islands) 767 max_layers = compute_pack_layers(&to_pack); 768 769 ALLOC_ARRAY(wo, to_pack.nr_objects); 770 wo_end = 0; 771 772 for (; write_layer < max_layers; ++write_layer) 773 compute_layer_order(wo, &wo_end); 774 775 if (wo_end != to_pack.nr_objects) 776 die(_("ordered %u objects, expected %"PRIu32), 777 wo_end, to_pack.nr_objects); 778 779 return wo; 780} 781 782static off_t write_reused_pack(struct hashfile *f) 783{ 784 unsigned char buffer[8192]; 785 off_t to_write, total; 786 int fd; 787 788 if (!is_pack_valid(reuse_packfile)) 789 die(_("packfile is invalid: %s"), reuse_packfile->pack_name); 790 791 fd = git_open(reuse_packfile->pack_name); 792 if (fd < 0) 793 die_errno(_("unable to open packfile for reuse: %s"), 794 reuse_packfile->pack_name); 795 796 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1) 797 die_errno(_("unable to seek in reused packfile")); 798 799 if (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); 803 804 while (to_write) { 805 int read_pack = xread(fd, buffer, sizeof(buffer)); 806 807 if (read_pack <= 0) 808 die_errno(_("unable to read from reused packfile")); 809 810 if (read_pack > to_write) 811 read_pack = to_write; 812 813 hashwrite(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); 826 display_progress(progress_state, written); 827 } 828 829 close(fd); 830 written = reuse_packfile_objects; 831 display_progress(progress_state, written); 832 return 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 void write_pack_file(void) 840{ 841 uint32_t i = 0, j; 842 struct hashfile *f; 843 off_t offset; 844 uint32_t nr_remaining = nr_result; 845 time_t last_mtime = 0; 846 struct object_entry **write_order; 847 848 if (progress > pack_to_stdout) 849 progress_state = start_progress(_("Writing objects"), nr_result); 850 ALLOC_ARRAY(written_list, to_pack.nr_objects); 851 write_order = compute_write_order(); 852 853 do { 854 struct object_id oid; 855 char *pack_tmp_name = NULL; 856 857 if (pack_to_stdout) 858 f = hashfd_throughput(1, "<stdout>", progress_state); 859 else 860 f = create_tmp_packfile(&pack_tmp_name); 861 862 offset = write_pack_header(f, nr_remaining); 863 864 if (reuse_packfile) { 865 off_t packfile_size; 866 assert(pack_to_stdout); 867 868 packfile_size = write_reused_pack(f); 869 offset += packfile_size; 870 } 871 872 nr_written = 0; 873 for (; i < to_pack.nr_objects; i++) { 874 struct object_entry *e = write_order[i]; 875 if (write_one(f, e, &offset) == WRITE_ONE_BREAK) 876 break; 877 display_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 */ 884 if (pack_to_stdout) { 885 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); 886 } else if (nr_written == nr_remaining) { 887 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); 888 } else { 889 int fd = finalize_hashfile(f, oid.hash, 0); 890 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, 891 nr_written, oid.hash, offset); 892 close(fd); 893 if (write_bitmap_index) { 894 warning(_(no_split_warning)); 895 write_bitmap_index = 0; 896 } 897 } 898 899 if (!pack_to_stdout) { 900 struct stat st; 901 struct 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 */ 910 if (stat(pack_tmp_name, &st) < 0) { 911 warning_errno(_("failed to stat %s"), pack_tmp_name); 912 } else if (!last_mtime) { 913 last_mtime = st.st_mtime; 914 } else { 915 struct utimbuf utb; 916 utb.actime = st.st_atime; 917 utb.modtime = --last_mtime; 918 if (utime(pack_tmp_name, &utb) < 0) 919 warning_errno(_("failed utime() on %s"), pack_tmp_name); 920 } 921 922 strbuf_addf(&tmpname, "%s-", base_name); 923 924 if (write_bitmap_index) { 925 bitmap_writer_set_checksum(oid.hash); 926 bitmap_writer_build_type_index( 927 &to_pack, written_list, nr_written); 928 } 929 930 finish_tmp_packfile(&tmpname, pack_tmp_name, 931 written_list, nr_written, 932 &pack_idx_opts, oid.hash); 933 934 if (write_bitmap_index) { 935 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid)); 936 937 stop_progress(&progress_state); 938 939 bitmap_writer_show_progress(progress); 940 bitmap_writer_reuse_bitmaps(&to_pack); 941 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 942 bitmap_writer_build(&to_pack); 943 bitmap_writer_finish(written_list, nr_written, 944 tmpname.buf, write_bitmap_options); 945 write_bitmap_index = 0; 946 } 947 948 strbuf_release(&tmpname); 949 free(pack_tmp_name); 950 puts(oid_to_hex(&oid)); 951 } 952 953 /* mark written objects as written to previous pack */ 954 for (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); 959 960 free(written_list); 961 free(write_order); 962 stop_progress(&progress_state); 963 if (written != nr_result) 964 die(_("wrote %"PRIu32" objects while expecting %"PRIu32), 965 written, nr_result); 966} 967 968static int no_try_delta(const char *path) 969{ 970 static struct attr_check *check; 971 972 if (!check) 973 check = attr_check_initl("delta", NULL); 974 git_check_attr(the_repository->index, path, check); 975 if (ATTR_FALSE(check->items[0].value)) 976 return 1; 977 return 0; 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 int have_duplicate_entry(const struct object_id *oid, 991 int exclude, 992 uint32_t *index_pos) 993{ 994 struct object_entry *entry; 995 996 entry = packlist_find(&to_pack, oid->hash, index_pos); 997 if (!entry) 998 return 0; 9991000 if (exclude) {1001 if (!entry->preferred_base)1002 nr_result--;1003 entry->preferred_base = 1;1004 }10051006 return 1;1007}10081009static int want_found_object(int exclude, struct packed_git *p)1010{1011 if (exclude)1012 return 1;1013 if (incremental)1014 return 0;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 */1030 if (!ignore_packed_keep_on_disk &&1031 !ignore_packed_keep_in_core &&1032 (!local || !have_non_local_packs))1033 return 1;10341035 if (local && !p->pack_local)1036 return 0;1037 if (p->pack_local &&1038 ((ignore_packed_keep_on_disk && p->pack_keep) ||1039 (ignore_packed_keep_in_core && p->pack_keep_in_core)))1040 return 0;10411042 /* we don't know yet; keep looking for more packs */1043 return -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 int want_object_in_pack(const struct object_id *oid,1056 int exclude,1057 struct packed_git **found_pack,1058 off_t *found_offset)1059{1060 int want;1061 struct list_head *pos;1062 struct multi_pack_index *m;10631064 if (!exclude && local && has_loose_object_nonlocal(oid))1065 return 0;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 */1072 if (*found_pack) {1073 want = want_found_object(exclude, *found_pack);1074 if (want != -1)1075 return want;1076 }10771078 for (m = get_multi_pack_index(the_repository); m; m = m->next) {1079 struct pack_entry e;1080 if (fill_midx_entry(oid, &e, m)) {1081 struct packed_git *p = e.p;1082 off_t offset;10831084 if (p == *found_pack)1085 offset = *found_offset;1086 else1087 offset = find_pack_entry_one(oid->hash, p);10881089 if (offset) {1090 if (!*found_pack) {1091 if (!is_pack_valid(p))1092 continue;1093 *found_offset = offset;1094 *found_pack = p;1095 }1096 want = want_found_object(exclude, p);1097 if (want != -1)1098 return want;1099 }1100 }1101 }11021103 list_for_each(pos, get_packed_git_mru(the_repository)) {1104 struct packed_git *p = list_entry(pos, struct packed_git, mru);1105 off_t offset;11061107 if (p == *found_pack)1108 offset = *found_offset;1109 else1110 offset = find_pack_entry_one(oid->hash, p);11111112 if (offset) {1113 if (!*found_pack) {1114 if (!is_pack_valid(p))1115 continue;1116 *found_offset = offset;1117 *found_pack = p;1118 }1119 want = want_found_object(exclude, p);1120 if (!exclude && want > 0)1121 list_move(&p->mru,1122 get_packed_git_mru(the_repository));1123 if (want != -1)1124 return want;1125 }1126 }11271128 return 1;1129}11301131static void create_object_entry(const struct object_id *oid,1132 enum object_type type,1133 uint32_t hash,1134 int exclude,1135 int no_try_delta,1136 uint32_t index_pos,1137 struct packed_git *found_pack,1138 off_t found_offset)1139{1140 struct object_entry *entry;11411142 entry = packlist_alloc(&to_pack, oid->hash, index_pos);1143 entry->hash = hash;1144 oe_set_type(entry, type);1145 if (exclude)1146 entry->preferred_base = 1;1147 else1148 nr_result++;1149 if (found_pack) {1150 oe_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 int add_object_entry(const struct object_id *oid, enum object_type type,1162 const char *name, int exclude)1163{1164 struct packed_git *found_pack = NULL;1165 off_t found_offset = 0;1166 uint32_t index_pos;11671168 display_progress(progress_state, ++nr_seen);11691170 if (have_duplicate_entry(oid, exclude, &index_pos))1171 return 0;11721173 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1174 /* The pack is missing an object, so it will not have closure */1175 if (write_bitmap_index) {1176 warning(_(no_closure_warning));1177 write_bitmap_index = 0;1178 }1179 return 0;1180 }11811182 create_object_entry(oid, type, pack_name_hash(name),1183 exclude, name && no_try_delta(name),1184 index_pos, found_pack, found_offset);1185 return 1;1186}11871188static int add_object_entry_from_bitmap(const struct object_id *oid,1189 enum object_type type,1190 int flags, uint32_t name_hash,1191 struct packed_git *pack, off_t offset)1192{1193 uint32_t index_pos;11941195 display_progress(progress_state, ++nr_seen);11961197 if (have_duplicate_entry(oid, 0, &index_pos))1198 return 0;11991200 if (!want_object_in_pack(oid, 0, &pack, &offset))1201 return 0;12021203 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);1204 return 1;1205}12061207struct pbase_tree_cache {1208 struct object_id oid;1209 int ref;1210 int temporary;1211 void *tree_data;1212 unsigned long tree_size;1213};12141215static struct pbase_tree_cache *(pbase_tree_cache[256]);1216static int pbase_tree_cache_ix(const struct object_id *oid)1217{1218 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);1219}1220static int pbase_tree_cache_ix_incr(int ix)1221{1222 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);1223}12241225static struct pbase_tree {1226 struct 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 */1232 struct pbase_tree_cache pcache;1233} *pbase_tree;12341235static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1236{1237 struct pbase_tree_cache *ent, *nent;1238 void *data;1239 unsigned long size;1240 enum object_type type;1241 int neigh;1242 int my_ix = pbase_tree_cache_ix(oid);1243 int 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 */1249 for (neigh = 0; neigh < 8; neigh++) {1250 ent = pbase_tree_cache[my_ix];1251 if (ent && oideq(&ent->oid, oid)) {1252 ent->ref++;1253 return ent;1254 }1255 else if (((available_ix < 0) && (!ent || !ent->ref)) ||1256 ((0 <= available_ix) &&1257 (!ent && pbase_tree_cache[available_ix])))1258 available_ix = my_ix;1259 if (!ent)1260 break;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);1268 if (!data)1269 return NULL;1270 if (type != OBJ_TREE) {1271 free(data);1272 return NULL;1273 }12741275 /* We need to either cache or return a throwaway copy */12761277 if (available_ix < 0)1278 ent = NULL;1279 else {1280 ent = pbase_tree_cache[available_ix];1281 my_ix = available_ix;1282 }12831284 if (!ent) {1285 nent = xmalloc(sizeof(*nent));1286 nent->temporary = (available_ix < 0);1287 }1288 else {1289 /* evict and reuse */1290 free(ent->tree_data);1291 nent = ent;1292 }1293 oidcpy(&nent->oid, oid);1294 nent->tree_data = data;1295 nent->tree_size = size;1296 nent->ref = 1;1297 if (!nent->temporary)1298 pbase_tree_cache[my_ix] = nent;1299 return nent;1300}13011302static void pbase_tree_put(struct pbase_tree_cache *cache)1303{1304 if (!cache->temporary) {1305 cache->ref--;1306 return;1307 }1308 free(cache->tree_data);1309 free(cache);1310}13111312static int name_cmp_len(const char *name)1313{1314 int i;1315 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)1316 ;1317 return i;1318}13191320static void add_pbase_object(struct tree_desc *tree,1321 const char *name,1322 int cmplen,1323 const char *fullname)1324{1325 struct name_entry entry;1326 int cmp;13271328 while (tree_entry(tree,&entry)) {1329 if (S_ISGITLINK(entry.mode))1330 continue;1331 cmp = tree_entry_len(&entry) != cmplen ? 1 :1332 memcmp(name, entry.path, cmplen);1333 if (cmp > 0)1334 continue;1335 if (cmp < 0)1336 return;1337 if (name[cmplen] != '/') {1338 add_object_entry(&entry.oid,1339 object_type(entry.mode),1340 fullname, 1);1341 return;1342 }1343 if (S_ISDIR(entry.mode)) {1344 struct tree_desc sub;1345 struct pbase_tree_cache *tree;1346 const char *down = name+cmplen+1;1347 int downlen = name_cmp_len(down);13481349 tree = pbase_tree_get(&entry.oid);1350 if (!tree)1351 return;1352 init_tree_desc(&sub, tree->tree_data, tree->tree_size);13531354 add_pbase_object(&sub, down, downlen, fullname);1355 pbase_tree_put(tree);1356 }1357 }1358}13591360static unsigned *done_pbase_paths;1361static int done_pbase_paths_num;1362static int done_pbase_paths_alloc;1363static int done_pbase_path_pos(unsigned hash)1364{1365 int lo = 0;1366 int hi = done_pbase_paths_num;1367 while (lo < hi) {1368 int mi = lo + (hi - lo) / 2;1369 if (done_pbase_paths[mi] == hash)1370 return mi;1371 if (done_pbase_paths[mi] < hash)1372 hi = mi;1373 else1374 lo = mi + 1;1375 }1376 return -lo-1;1377}13781379static int check_pbase_path(unsigned hash)1380{1381 int pos = done_pbase_path_pos(hash);1382 if (0 <= pos)1383 return 1;1384 pos = -pos - 1;1385 ALLOC_GROW(done_pbase_paths,1386 done_pbase_paths_num + 1,1387 done_pbase_paths_alloc);1388 done_pbase_paths_num++;1389 if (pos < done_pbase_paths_num)1390 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,1391 done_pbase_paths_num - pos - 1);1392 done_pbase_paths[pos] = hash;1393 return 0;1394}13951396static void add_preferred_base_object(const char *name)1397{1398 struct pbase_tree *it;1399 int cmplen;1400 unsigned hash = pack_name_hash(name);14011402 if (!num_preferred_base || check_pbase_path(hash))1403 return;14041405 cmplen = name_cmp_len(name);1406 for (it = pbase_tree; it; it = it->next) {1407 if (cmplen == 0) {1408 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);1409 }1410 else {1411 struct tree_desc tree;1412 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1413 add_pbase_object(&tree, name, cmplen, name);1414 }1415 }1416}14171418static void add_preferred_base(struct object_id *oid)1419{1420 struct pbase_tree *it;1421 void *data;1422 unsigned long size;1423 struct object_id tree_oid;14241425 if (window <= num_preferred_base++)1426 return;14271428 data = read_object_with_reference(oid, tree_type, &size, &tree_oid);1429 if (!data)1430 return;14311432 for (it = pbase_tree; it; it = it->next) {1433 if (oideq(&it->pcache.oid, &tree_oid)) {1434 free(data);1435 return;1436 }1437 }14381439 it = xcalloc(1, sizeof(*it));1440 it->next = pbase_tree;1441 pbase_tree = it;14421443 oidcpy(&it->pcache.oid, &tree_oid);1444 it->pcache.tree_data = data;1445 it->pcache.tree_size = size;1446}14471448static void cleanup_preferred_base(void)1449{1450 struct pbase_tree *it;1451 unsigned i;14521453 it = pbase_tree;1454 pbase_tree = NULL;1455 while (it) {1456 struct pbase_tree *tmp = it;1457 it = tmp->next;1458 free(tmp->pcache.tree_data);1459 free(tmp);1460 }14611462 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {1463 if (!pbase_tree_cache[i])1464 continue;1465 free(pbase_tree_cache[i]->tree_data);1466 FREE_AND_NULL(pbase_tree_cache[i]);1467 }14681469 FREE_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 int can_reuse_delta(const unsigned char *base_sha1,1485 struct object_entry *delta,1486 struct object_entry **base_out)1487{1488 struct object_entry *base;14891490 if (!base_sha1)1491 return 0;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);1498 if (base) {1499 if (!in_same_island(&delta->idx.oid, &base->idx.oid))1500 return 0;1501 *base_out = base;1502 return 1;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 */1510 if (thin && bitmap_has_sha1_in_uninteresting(bitmap_git, base_sha1)) {1511 if (use_delta_islands) {1512 struct object_id base_oid;1513 hashcpy(base_oid.hash, base_sha1);1514 if (!in_same_island(&delta->idx.oid, &base_oid))1515 return 0;1516 }1517 *base_out = NULL;1518 return 1;1519 }15201521 return 0;1522}15231524static void check_object(struct object_entry *entry)1525{1526 unsigned long canonical_size;15271528 if (IN_PACK(entry)) {1529 struct packed_git *p = IN_PACK(entry);1530 struct pack_window *w_curs = NULL;1531 const unsigned char *base_ref = NULL;1532 struct object_entry *base_entry;1533 unsigned long used, used_0;1534 unsigned long avail;1535 off_t ofs;1536 unsigned char *buf, c;1537 enum object_type type;1538 unsigned 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);1549 if (used == 0)1550 goto give_up;15511552 if (type < 0)1553 BUG("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 */1561 switch (entry->in_pack_type) {1562 default:1563 /* Not a delta hence we've already got all we need. */1564 oe_set_type(entry, entry->in_pack_type);1565 SET_SIZE(entry, in_pack_size);1566 entry->in_pack_header_size = used;1567 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)1568 goto give_up;1569 unuse_pack(&w_curs);1570 return;1571 case OBJ_REF_DELTA:1572 if (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;1576 break;1577 case 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;1583 while (c & 128) {1584 ofs += 1;1585 if (!ofs || MSB(ofs, 7)) {1586 error(_("delta base offset overflow in pack for %s"),1587 oid_to_hex(&entry->idx.oid));1588 goto give_up;1589 }1590 c = buf[used_0++];1591 ofs = (ofs << 7) + (c & 127);1592 }1593 ofs = entry->in_pack_offset - ofs;1594 if (ofs <= 0 || ofs >= entry->in_pack_offset) {1595 error(_("delta base offset out of bound for %s"),1596 oid_to_hex(&entry->idx.oid));1597 goto give_up;1598 }1599 if (reuse_delta && !entry->preferred_base) {1600 struct revindex_entry *revidx;1601 revidx = find_pack_revindex(p, ofs);1602 if (!revidx)1603 goto give_up;1604 base_ref = nth_packed_object_sha1(p, revidx->nr);1605 }1606 entry->in_pack_header_size = used + used_0;1607 break;1608 }16091610 if (can_reuse_delta(base_ref, entry, &base_entry)) {1611 oe_set_type(entry, entry->in_pack_type);1612 SET_SIZE(entry, in_pack_size); /* delta size */1613 SET_DELTA_SIZE(entry, in_pack_size);16141615 if (base_entry) {1616 SET_DELTA(entry, base_entry);1617 entry->delta_sibling_idx = base_entry->delta_child_idx;1618 SET_DELTA_CHILD(base_entry, entry);1619 } else {1620 SET_DELTA_EXT(entry, base_ref);1621 }16221623 unuse_pack(&w_curs);1624 return;1625 }16261627 if (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);1637 if (canonical_size == 0)1638 goto give_up;1639 SET_SIZE(entry, canonical_size);1640 unuse_pack(&w_curs);1641 return;1642 }16431644 /*1645 * No choice but to fall back to the recursive delta walk1646 * with oid_object_info() to find about the object type1647 * at this point...1648 */1649 give_up:1650 unuse_pack(&w_curs);1651 }16521653 oe_set_type(entry,1654 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1655 if (entry->type_valid) {1656 SET_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 int pack_offset_sort(const void *_a, const void *_b)1668{1669 const struct object_entry *a = *(struct object_entry **)_a;1670 const struct object_entry *b = *(struct object_entry **)_b;1671 const struct packed_git *a_in_pack = IN_PACK(a);1672 const struct packed_git *b_in_pack = IN_PACK(b);16731674 /* avoid filesystem trashing with loose objects */1675 if (!a_in_pack && !b_in_pack)1676 return oidcmp(&a->idx.oid, &b->idx.oid);16771678 if (a_in_pack < b_in_pack)1679 return -1;1680 if (a_in_pack > b_in_pack)1681 return 1;1682 return 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 void drop_reused_delta(struct object_entry *entry)1700{1701 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;1702 struct object_info oi = OBJECT_INFO_INIT;1703 enum object_type type;1704 unsigned long size;17051706 while (*idx) {1707 struct object_entry *oe = &to_pack.objects[*idx - 1];17081709 if (oe == entry)1710 *idx = oe->delta_sibling_idx;1711 else1712 idx = &oe->delta_sibling_idx;1713 }1714 SET_DELTA(entry, NULL);1715 entry->depth = 0;17161717 oi.sizep = &size;1718 oi.typep = &type;1719 if (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 oid_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 */1726 oe_set_type(entry,1727 oid_object_info(the_repository, &entry->idx.oid, &size));1728 } else {1729 oe_set_type(entry, type);1730 }1731 SET_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 void break_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 */1750 uint32_t total_depth;1751 struct object_entry *cur, *next;17521753 for (cur = entry, total_depth = 0;1754 cur;1755 cur = DELTA(cur), total_depth++) {1756 if (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;1763 break;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 */1771 if (cur->dfs_state != DFS_NONE)1772 BUG("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 */1780 if (!DELTA(cur)) {1781 cur->dfs_state = DFS_DONE;1782 break;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;1803 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {1804 drop_reused_delta(cur);1805 cur->dfs_state = DFS_DONE;1806 break;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 */1817 for (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 */1826 if (cur->dfs_state == DFS_DONE)1827 break;1828 else if (cur->dfs_state != DFS_ACTIVE)1829 BUG("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);1851 if (!cur->depth)1852 drop_reused_delta(cur);18531854 cur->dfs_state = DFS_DONE;1855 }1856}18571858static void get_object_details(void)1859{1860 uint32_t i;1861 struct object_entry **sorted_by_offset;18621863 if (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 *));1868 for (i = 0; i < to_pack.nr_objects; i++)1869 sorted_by_offset[i] = to_pack.objects + i;1870 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);18711872 for (i = 0; i < to_pack.nr_objects; i++) {1873 struct object_entry *entry = sorted_by_offset[i];1874 check_object(entry);1875 if (entry->type_valid &&1876 oe_size_greater_than(&to_pack, entry, big_file_threshold))1877 entry->no_try_delta = 1;1878 display_progress(progress_state, i + 1);1879 }1880 stop_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 */1886 for (i = 0; i < to_pack.nr_objects; i++)1887 break_delta_chains(&to_pack.objects[i]);18881889 free(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 int type_size_sort(const void *_a, const void *_b)1902{1903 const struct object_entry *a = *(struct object_entry **)_a;1904 const struct object_entry *b = *(struct object_entry **)_b;1905 const enum object_type a_type = oe_type(a);1906 const enum object_type b_type = oe_type(b);1907 const unsigned long a_size = SIZE(a);1908 const unsigned long b_size = SIZE(b);19091910 if (a_type > b_type)1911 return -1;1912 if (a_type < b_type)1913 return 1;1914 if (a->hash > b->hash)1915 return -1;1916 if (a->hash < b->hash)1917 return 1;1918 if (a->preferred_base > b->preferred_base)1919 return -1;1920 if (a->preferred_base < b->preferred_base)1921 return 1;1922 if (use_delta_islands) {1923 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);1924 if (island_cmp)1925 return island_cmp;1926 }1927 if (a_size > b_size)1928 return -1;1929 if (a_size < b_size)1930 return 1;1931 return a < b ? -1 : (a > b); /* newest first */1932}19331934struct unpacked {1935 struct object_entry *entry;1936 void *data;1937 struct delta_index *index;1938 unsigned depth;1939};19401941static int delta_cacheable(unsigned long src_size, unsigned long trg_size,1942 unsigned long delta_size)1943{1944 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1945 return 0;19461947 if (delta_size < cache_max_small_delta_size)1948 return 1;19491950 /* cache delta, if objects are large enough compared to delta size */1951 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))1952 return 1;19531954 return 0;1955}19561957/* Protect delta_cache_size */1958static pthread_mutex_t cache_mutex;1959#define cache_lock() pthread_mutex_lock(&cache_mutex)1960#define cache_unlock() pthread_mutex_unlock(&cache_mutex)19611962/*1963 * Protect object list partitioning (e.g. struct thread_param) and1964 * progress_state1965 */1966static pthread_mutex_t progress_mutex;1967#define progress_lock() pthread_mutex_lock(&progress_mutex)1968#define progress_unlock() pthread_mutex_unlock(&progress_mutex)19691970/*1971 * Access to struct object_entry is unprotected since each thread owns1972 * a portion of the main object list. Just don't access object entries1973 * ahead in the list because they can be stolen and would need1974 * progress_mutex for protection.1975 */19761977/*1978 * Return the size of the object without doing any delta1979 * reconstruction (so non-deltas are true object sizes, but deltas1980 * return the size of the delta data).1981 */1982unsigned long oe_get_size_slow(struct packing_data *pack,1983 const struct object_entry *e)1984{1985 struct packed_git *p;1986 struct pack_window *w_curs;1987 unsigned char *buf;1988 enum object_type type;1989 unsigned long used, avail, size;19901991 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {1992 packing_data_lock(&to_pack);1993 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)1994 die(_("unable to get size of %s"),1995 oid_to_hex(&e->idx.oid));1996 packing_data_unlock(&to_pack);1997 return size;1998 }19992000 p = oe_in_pack(pack, e);2001 if (!p)2002 BUG("when e->type is a delta, it must belong to a pack");20032004 packing_data_lock(&to_pack);2005 w_curs = NULL;2006 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);2007 used = unpack_object_header_buffer(buf, avail, &type, &size);2008 if (used == 0)2009 die(_("unable to parse object header of %s"),2010 oid_to_hex(&e->idx.oid));20112012 unuse_pack(&w_curs);2013 packing_data_unlock(&to_pack);2014 return size;2015}20162017static int try_delta(struct unpacked *trg, struct unpacked *src,2018 unsigned max_depth, unsigned long *mem_usage)2019{2020 struct object_entry *trg_entry = trg->entry;2021 struct object_entry *src_entry = src->entry;2022 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;2023 unsigned ref_depth;2024 enum object_type type;2025 void *delta_buf;20262027 /* Don't bother doing diffs between different types */2028 if (oe_type(trg_entry) != oe_type(src_entry))2029 return -1;20302031 /*2032 * We do not bother to try a delta that we discarded on an2033 * earlier try, but only when reusing delta data. Note that2034 * src_entry that is marked as the preferred_base should always2035 * be considered, as even if we produce a suboptimal delta against2036 * it, we will still save the transfer cost, as we already know2037 * the other side has it and we won't send src_entry at all.2038 */2039 if (reuse_delta && IN_PACK(trg_entry) &&2040 IN_PACK(trg_entry) == IN_PACK(src_entry) &&2041 !src_entry->preferred_base &&2042 trg_entry->in_pack_type != OBJ_REF_DELTA &&2043 trg_entry->in_pack_type != OBJ_OFS_DELTA)2044 return 0;20452046 /* Let's not bust the allowed depth. */2047 if (src->depth >= max_depth)2048 return 0;20492050 /* Now some size filtering heuristics. */2051 trg_size = SIZE(trg_entry);2052 if (!DELTA(trg_entry)) {2053 max_size = trg_size/2 - the_hash_algo->rawsz;2054 ref_depth = 1;2055 } else {2056 max_size = DELTA_SIZE(trg_entry);2057 ref_depth = trg->depth;2058 }2059 max_size = (uint64_t)max_size * (max_depth - src->depth) /2060 (max_depth - ref_depth + 1);2061 if (max_size == 0)2062 return 0;2063 src_size = SIZE(src_entry);2064 sizediff = src_size < trg_size ? trg_size - src_size : 0;2065 if (sizediff >= max_size)2066 return 0;2067 if (trg_size < src_size / 32)2068 return 0;20692070 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))2071 return 0;20722073 /* Load data if not already done */2074 if (!trg->data) {2075 packing_data_lock(&to_pack);2076 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);2077 packing_data_unlock(&to_pack);2078 if (!trg->data)2079 die(_("object %s cannot be read"),2080 oid_to_hex(&trg_entry->idx.oid));2081 if (sz != trg_size)2082 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2083 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,2084 (uintmax_t)trg_size);2085 *mem_usage += sz;2086 }2087 if (!src->data) {2088 packing_data_lock(&to_pack);2089 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);2090 packing_data_unlock(&to_pack);2091 if (!src->data) {2092 if (src_entry->preferred_base) {2093 static int warned = 0;2094 if (!warned++)2095 warning(_("object %s cannot be read"),2096 oid_to_hex(&src_entry->idx.oid));2097 /*2098 * Those objects are not included in the2099 * resulting pack. Be resilient and ignore2100 * them if they can't be read, in case the2101 * pack could be created nevertheless.2102 */2103 return 0;2104 }2105 die(_("object %s cannot be read"),2106 oid_to_hex(&src_entry->idx.oid));2107 }2108 if (sz != src_size)2109 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2110 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,2111 (uintmax_t)src_size);2112 *mem_usage += sz;2113 }2114 if (!src->index) {2115 src->index = create_delta_index(src->data, src_size);2116 if (!src->index) {2117 static int warned = 0;2118 if (!warned++)2119 warning(_("suboptimal pack - out of memory"));2120 return 0;2121 }2122 *mem_usage += sizeof_delta_index(src->index);2123 }21242125 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2126 if (!delta_buf)2127 return 0;21282129 if (DELTA(trg_entry)) {2130 /* Prefer only shallower same-sized deltas. */2131 if (delta_size == DELTA_SIZE(trg_entry) &&2132 src->depth + 1 >= trg->depth) {2133 free(delta_buf);2134 return 0;2135 }2136 }21372138 /*2139 * Handle memory allocation outside of the cache2140 * accounting lock. Compiler will optimize the strangeness2141 * away when NO_PTHREADS is defined.2142 */2143 free(trg_entry->delta_data);2144 cache_lock();2145 if (trg_entry->delta_data) {2146 delta_cache_size -= DELTA_SIZE(trg_entry);2147 trg_entry->delta_data = NULL;2148 }2149 if (delta_cacheable(src_size, trg_size, delta_size)) {2150 delta_cache_size += delta_size;2151 cache_unlock();2152 trg_entry->delta_data = xrealloc(delta_buf, delta_size);2153 } else {2154 cache_unlock();2155 free(delta_buf);2156 }21572158 SET_DELTA(trg_entry, src_entry);2159 SET_DELTA_SIZE(trg_entry, delta_size);2160 trg->depth = src->depth + 1;21612162 return 1;2163}21642165static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)2166{2167 struct object_entry *child = DELTA_CHILD(me);2168 unsigned int m = n;2169 while (child) {2170 const unsigned int c = check_delta_limit(child, n + 1);2171 if (m < c)2172 m = c;2173 child = DELTA_SIBLING(child);2174 }2175 return m;2176}21772178static unsigned long free_unpacked(struct unpacked *n)2179{2180 unsigned long freed_mem = sizeof_delta_index(n->index);2181 free_delta_index(n->index);2182 n->index = NULL;2183 if (n->data) {2184 freed_mem += SIZE(n->entry);2185 FREE_AND_NULL(n->data);2186 }2187 n->entry = NULL;2188 n->depth = 0;2189 return freed_mem;2190}21912192static void find_deltas(struct object_entry **list, unsigned *list_size,2193 int window, int depth, unsigned *processed)2194{2195 uint32_t i, idx = 0, count = 0;2196 struct unpacked *array;2197 unsigned long mem_usage = 0;21982199 array = xcalloc(window, sizeof(struct unpacked));22002201 for (;;) {2202 struct object_entry *entry;2203 struct unpacked *n = array + idx;2204 int j, max_depth, best_base = -1;22052206 progress_lock();2207 if (!*list_size) {2208 progress_unlock();2209 break;2210 }2211 entry = *list++;2212 (*list_size)--;2213 if (!entry->preferred_base) {2214 (*processed)++;2215 display_progress(progress_state, *processed);2216 }2217 progress_unlock();22182219 mem_usage -= free_unpacked(n);2220 n->entry = entry;22212222 while (window_memory_limit &&2223 mem_usage > window_memory_limit &&2224 count > 1) {2225 const uint32_t tail = (idx + window - count) % window;2226 mem_usage -= free_unpacked(array + tail);2227 count--;2228 }22292230 /* We do not compute delta to *create* objects we are not2231 * going to pack.2232 */2233 if (entry->preferred_base)2234 goto next;22352236 /*2237 * If the current object is at pack edge, take the depth the2238 * objects that depend on the current object into account2239 * otherwise they would become too deep.2240 */2241 max_depth = depth;2242 if (DELTA_CHILD(entry)) {2243 max_depth -= check_delta_limit(entry, 0);2244 if (max_depth <= 0)2245 goto next;2246 }22472248 j = window;2249 while (--j > 0) {2250 int ret;2251 uint32_t other_idx = idx + j;2252 struct unpacked *m;2253 if (other_idx >= window)2254 other_idx -= window;2255 m = array + other_idx;2256 if (!m->entry)2257 break;2258 ret = try_delta(n, m, max_depth, &mem_usage);2259 if (ret < 0)2260 break;2261 else if (ret > 0)2262 best_base = other_idx;2263 }22642265 /*2266 * If we decided to cache the delta data, then it is best2267 * to compress it right away. First because we have to do2268 * it anyway, and doing it here while we're threaded will2269 * save a lot of time in the non threaded write phase,2270 * as well as allow for caching more deltas within2271 * the same cache size limit.2272 * ...2273 * But only if not writing to stdout, since in that case2274 * the network is most likely throttling writes anyway,2275 * and therefore it is best to go to the write phase ASAP2276 * instead, as we can afford spending more time compressing2277 * between writes at that moment.2278 */2279 if (entry->delta_data && !pack_to_stdout) {2280 unsigned long size;22812282 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));2283 if (size < (1U << OE_Z_DELTA_BITS)) {2284 entry->z_delta_size = size;2285 cache_lock();2286 delta_cache_size -= DELTA_SIZE(entry);2287 delta_cache_size += entry->z_delta_size;2288 cache_unlock();2289 } else {2290 FREE_AND_NULL(entry->delta_data);2291 entry->z_delta_size = 0;2292 }2293 }22942295 /* if we made n a delta, and if n is already at max2296 * depth, leaving it in the window is pointless. we2297 * should evict it first.2298 */2299 if (DELTA(entry) && max_depth <= n->depth)2300 continue;23012302 /*2303 * Move the best delta base up in the window, after the2304 * currently deltified object, to keep it longer. It will2305 * be the first base object to be attempted next.2306 */2307 if (DELTA(entry)) {2308 struct unpacked swap = array[best_base];2309 int dist = (window + idx - best_base) % window;2310 int dst = best_base;2311 while (dist--) {2312 int src = (dst + 1) % window;2313 array[dst] = array[src];2314 dst = src;2315 }2316 array[dst] = swap;2317 }23182319 next:2320 idx++;2321 if (count + 1 < window)2322 count++;2323 if (idx >= window)2324 idx = 0;2325 }23262327 for (i = 0; i < window; ++i) {2328 free_delta_index(array[i].index);2329 free(array[i].data);2330 }2331 free(array);2332}23332334static void try_to_free_from_threads(size_t size)2335{2336 packing_data_lock(&to_pack);2337 release_pack_memory(size);2338 packing_data_unlock(&to_pack);2339}23402341static try_to_free_t old_try_to_free_routine;23422343/*2344 * The main object list is split into smaller lists, each is handed to2345 * one worker.2346 *2347 * The main thread waits on the condition that (at least) one of the workers2348 * has stopped working (which is indicated in the .working member of2349 * struct thread_params).2350 *2351 * When a work thread has completed its work, it sets .working to 0 and2352 * signals the main thread and waits on the condition that .data_ready2353 * becomes 1.2354 *2355 * The main thread steals half of the work from the worker that has2356 * most work left to hand it to the idle worker.2357 */23582359struct thread_params {2360 pthread_t thread;2361 struct object_entry **list;2362 unsigned list_size;2363 unsigned remaining;2364 int window;2365 int depth;2366 int working;2367 int data_ready;2368 pthread_mutex_t mutex;2369 pthread_cond_t cond;2370 unsigned *processed;2371};23722373static pthread_cond_t progress_cond;23742375/*2376 * Mutex and conditional variable can't be statically-initialized on Windows.2377 */2378static void init_threaded_search(void)2379{2380 pthread_mutex_init(&cache_mutex, NULL);2381 pthread_mutex_init(&progress_mutex, NULL);2382 pthread_cond_init(&progress_cond, NULL);2383 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);2384}23852386static void cleanup_threaded_search(void)2387{2388 set_try_to_free_routine(old_try_to_free_routine);2389 pthread_cond_destroy(&progress_cond);2390 pthread_mutex_destroy(&cache_mutex);2391 pthread_mutex_destroy(&progress_mutex);2392}23932394static void *threaded_find_deltas(void *arg)2395{2396 struct thread_params *me = arg;23972398 progress_lock();2399 while (me->remaining) {2400 progress_unlock();24012402 find_deltas(me->list, &me->remaining,2403 me->window, me->depth, me->processed);24042405 progress_lock();2406 me->working = 0;2407 pthread_cond_signal(&progress_cond);2408 progress_unlock();24092410 /*2411 * We must not set ->data_ready before we wait on the2412 * condition because the main thread may have set it to 12413 * before we get here. In order to be sure that new2414 * work is available if we see 1 in ->data_ready, it2415 * was initialized to 0 before this thread was spawned2416 * and we reset it to 0 right away.2417 */2418 pthread_mutex_lock(&me->mutex);2419 while (!me->data_ready)2420 pthread_cond_wait(&me->cond, &me->mutex);2421 me->data_ready = 0;2422 pthread_mutex_unlock(&me->mutex);24232424 progress_lock();2425 }2426 progress_unlock();2427 /* leave ->working 1 so that this doesn't get more work assigned */2428 return NULL;2429}24302431static void ll_find_deltas(struct object_entry **list, unsigned list_size,2432 int window, int depth, unsigned *processed)2433{2434 struct thread_params *p;2435 int i, ret, active_threads = 0;24362437 init_threaded_search();24382439 if (delta_search_threads <= 1) {2440 find_deltas(list, &list_size, window, depth, processed);2441 cleanup_threaded_search();2442 return;2443 }2444 if (progress > pack_to_stdout)2445 fprintf_ln(stderr, _("Delta compression using up to %d threads"),2446 delta_search_threads);2447 p = xcalloc(delta_search_threads, sizeof(*p));24482449 /* Partition the work amongst work threads. */2450 for (i = 0; i < delta_search_threads; i++) {2451 unsigned sub_size = list_size / (delta_search_threads - i);24522453 /* don't use too small segments or no deltas will be found */2454 if (sub_size < 2*window && i+1 < delta_search_threads)2455 sub_size = 0;24562457 p[i].window = window;2458 p[i].depth = depth;2459 p[i].processed = processed;2460 p[i].working = 1;2461 p[i].data_ready = 0;24622463 /* try to split chunks on "path" boundaries */2464 while (sub_size && sub_size < list_size &&2465 list[sub_size]->hash &&2466 list[sub_size]->hash == list[sub_size-1]->hash)2467 sub_size++;24682469 p[i].list = list;2470 p[i].list_size = sub_size;2471 p[i].remaining = sub_size;24722473 list += sub_size;2474 list_size -= sub_size;2475 }24762477 /* Start work threads. */2478 for (i = 0; i < delta_search_threads; i++) {2479 if (!p[i].list_size)2480 continue;2481 pthread_mutex_init(&p[i].mutex, NULL);2482 pthread_cond_init(&p[i].cond, NULL);2483 ret = pthread_create(&p[i].thread, NULL,2484 threaded_find_deltas, &p[i]);2485 if (ret)2486 die(_("unable to create thread: %s"), strerror(ret));2487 active_threads++;2488 }24892490 /*2491 * Now let's wait for work completion. Each time a thread is done2492 * with its work, we steal half of the remaining work from the2493 * thread with the largest number of unprocessed objects and give2494 * it to that newly idle thread. This ensure good load balancing2495 * until the remaining object list segments are simply too short2496 * to be worth splitting anymore.2497 */2498 while (active_threads) {2499 struct thread_params *target = NULL;2500 struct thread_params *victim = NULL;2501 unsigned sub_size = 0;25022503 progress_lock();2504 for (;;) {2505 for (i = 0; !target && i < delta_search_threads; i++)2506 if (!p[i].working)2507 target = &p[i];2508 if (target)2509 break;2510 pthread_cond_wait(&progress_cond, &progress_mutex);2511 }25122513 for (i = 0; i < delta_search_threads; i++)2514 if (p[i].remaining > 2*window &&2515 (!victim || victim->remaining < p[i].remaining))2516 victim = &p[i];2517 if (victim) {2518 sub_size = victim->remaining / 2;2519 list = victim->list + victim->list_size - sub_size;2520 while (sub_size && list[0]->hash &&2521 list[0]->hash == list[-1]->hash) {2522 list++;2523 sub_size--;2524 }2525 if (!sub_size) {2526 /*2527 * It is possible for some "paths" to have2528 * so many objects that no hash boundary2529 * might be found. Let's just steal the2530 * exact half in that case.2531 */2532 sub_size = victim->remaining / 2;2533 list -= sub_size;2534 }2535 target->list = list;2536 victim->list_size -= sub_size;2537 victim->remaining -= sub_size;2538 }2539 target->list_size = sub_size;2540 target->remaining = sub_size;2541 target->working = 1;2542 progress_unlock();25432544 pthread_mutex_lock(&target->mutex);2545 target->data_ready = 1;2546 pthread_cond_signal(&target->cond);2547 pthread_mutex_unlock(&target->mutex);25482549 if (!sub_size) {2550 pthread_join(target->thread, NULL);2551 pthread_cond_destroy(&target->cond);2552 pthread_mutex_destroy(&target->mutex);2553 active_threads--;2554 }2555 }2556 cleanup_threaded_search();2557 free(p);2558}25592560static void add_tag_chain(const struct object_id *oid)2561{2562 struct tag *tag;25632564 /*2565 * We catch duplicates already in add_object_entry(), but we'd2566 * prefer to do this extra check to avoid having to parse the2567 * tag at all if we already know that it's being packed (e.g., if2568 * it was included via bitmaps, we would not have parsed it2569 * previously).2570 */2571 if (packlist_find(&to_pack, oid->hash, NULL))2572 return;25732574 tag = lookup_tag(the_repository, oid);2575 while (1) {2576 if (!tag || parse_tag(tag) || !tag->tagged)2577 die(_("unable to pack objects reachable from tag %s"),2578 oid_to_hex(oid));25792580 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);25812582 if (tag->tagged->type != OBJ_TAG)2583 return;25842585 tag = (struct tag *)tag->tagged;2586 }2587}25882589static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)2590{2591 struct object_id peeled;25922593 if (starts_with(path, "refs/tags/") && /* is a tag? */2594 !peel_ref(path, &peeled) && /* peelable? */2595 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */2596 add_tag_chain(oid);2597 return 0;2598}25992600static void prepare_pack(int window, int depth)2601{2602 struct object_entry **delta_list;2603 uint32_t i, nr_deltas;2604 unsigned n;26052606 if (use_delta_islands)2607 resolve_tree_islands(the_repository, progress, &to_pack);26082609 get_object_details();26102611 /*2612 * If we're locally repacking then we need to be doubly careful2613 * from now on in order to make sure no stealth corruption gets2614 * propagated to the new pack. Clients receiving streamed packs2615 * should validate everything they get anyway so no need to incur2616 * the additional cost here in that case.2617 */2618 if (!pack_to_stdout)2619 do_check_packed_object_crc = 1;26202621 if (!to_pack.nr_objects || !window || !depth)2622 return;26232624 ALLOC_ARRAY(delta_list, to_pack.nr_objects);2625 nr_deltas = n = 0;26262627 for (i = 0; i < to_pack.nr_objects; i++) {2628 struct object_entry *entry = to_pack.objects + i;26292630 if (DELTA(entry))2631 /* This happens if we decided to reuse existing2632 * delta from a pack. "reuse_delta &&" is implied.2633 */2634 continue;26352636 if (!entry->type_valid ||2637 oe_size_less_than(&to_pack, entry, 50))2638 continue;26392640 if (entry->no_try_delta)2641 continue;26422643 if (!entry->preferred_base) {2644 nr_deltas++;2645 if (oe_type(entry) < 0)2646 die(_("unable to get type of object %s"),2647 oid_to_hex(&entry->idx.oid));2648 } else {2649 if (oe_type(entry) < 0) {2650 /*2651 * This object is not found, but we2652 * don't have to include it anyway.2653 */2654 continue;2655 }2656 }26572658 delta_list[n++] = entry;2659 }26602661 if (nr_deltas && n > 1) {2662 unsigned nr_done = 0;2663 if (progress)2664 progress_state = start_progress(_("Compressing objects"),2665 nr_deltas);2666 QSORT(delta_list, n, type_size_sort);2667 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2668 stop_progress(&progress_state);2669 if (nr_done != nr_deltas)2670 die(_("inconsistency with delta count"));2671 }2672 free(delta_list);2673}26742675static int git_pack_config(const char *k, const char *v, void *cb)2676{2677 if (!strcmp(k, "pack.window")) {2678 window = git_config_int(k, v);2679 return 0;2680 }2681 if (!strcmp(k, "pack.windowmemory")) {2682 window_memory_limit = git_config_ulong(k, v);2683 return 0;2684 }2685 if (!strcmp(k, "pack.depth")) {2686 depth = git_config_int(k, v);2687 return 0;2688 }2689 if (!strcmp(k, "pack.deltacachesize")) {2690 max_delta_cache_size = git_config_int(k, v);2691 return 0;2692 }2693 if (!strcmp(k, "pack.deltacachelimit")) {2694 cache_max_small_delta_size = git_config_int(k, v);2695 return 0;2696 }2697 if (!strcmp(k, "pack.writebitmaphashcache")) {2698 if (git_config_bool(k, v))2699 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2700 else2701 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2702 }2703 if (!strcmp(k, "pack.usebitmaps")) {2704 use_bitmap_index_default = git_config_bool(k, v);2705 return 0;2706 }2707 if (!strcmp(k, "pack.usesparse")) {2708 sparse = git_config_bool(k, v);2709 return 0;2710 }2711 if (!strcmp(k, "pack.threads")) {2712 delta_search_threads = git_config_int(k, v);2713 if (delta_search_threads < 0)2714 die(_("invalid number of threads specified (%d)"),2715 delta_search_threads);2716 if (!HAVE_THREADS && delta_search_threads != 1) {2717 warning(_("no threads support, ignoring %s"), k);2718 delta_search_threads = 0;2719 }2720 return 0;2721 }2722 if (!strcmp(k, "pack.indexversion")) {2723 pack_idx_opts.version = git_config_int(k, v);2724 if (pack_idx_opts.version > 2)2725 die(_("bad pack.indexversion=%"PRIu32),2726 pack_idx_opts.version);2727 return 0;2728 }2729 return git_default_config(k, v, cb);2730}27312732static void read_object_list_from_stdin(void)2733{2734 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];2735 struct object_id oid;2736 const char *p;27372738 for (;;) {2739 if (!fgets(line, sizeof(line), stdin)) {2740 if (feof(stdin))2741 break;2742 if (!ferror(stdin))2743 die("BUG: fgets returned NULL, not EOF, not error!");2744 if (errno != EINTR)2745 die_errno("fgets");2746 clearerr(stdin);2747 continue;2748 }2749 if (line[0] == '-') {2750 if (get_oid_hex(line+1, &oid))2751 die(_("expected edge object ID, got garbage:\n %s"),2752 line);2753 add_preferred_base(&oid);2754 continue;2755 }2756 if (parse_oid_hex(line, &oid, &p))2757 die(_("expected object ID, got garbage:\n %s"), line);27582759 add_preferred_base_object(p + 1);2760 add_object_entry(&oid, OBJ_NONE, p + 1, 0);2761 }2762}27632764/* Remember to update object flag allocation in object.h */2765#define OBJECT_ADDED (1u<<20)27662767static void show_commit(struct commit *commit, void *data)2768{2769 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);2770 commit->object.flags |= OBJECT_ADDED;27712772 if (write_bitmap_index)2773 index_commit_for_bitmap(commit);27742775 if (use_delta_islands)2776 propagate_island_marks(commit);2777}27782779static void show_object(struct object *obj, const char *name, void *data)2780{2781 add_preferred_base_object(name);2782 add_object_entry(&obj->oid, obj->type, name, 0);2783 obj->flags |= OBJECT_ADDED;27842785 if (use_delta_islands) {2786 const char *p;2787 unsigned depth;2788 struct object_entry *ent;27892790 /* the empty string is a root tree, which is depth 0 */2791 depth = *name ? 1 : 0;2792 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))2793 depth++;27942795 ent = packlist_find(&to_pack, obj->oid.hash, NULL);2796 if (ent && depth > oe_tree_depth(&to_pack, ent))2797 oe_set_tree_depth(&to_pack, ent, depth);2798 }2799}28002801static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)2802{2803 assert(arg_missing_action == MA_ALLOW_ANY);28042805 /*2806 * Quietly ignore ALL missing objects. This avoids problems with2807 * staging them now and getting an odd error later.2808 */2809 if (!has_object_file(&obj->oid))2810 return;28112812 show_object(obj, name, data);2813}28142815static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)2816{2817 assert(arg_missing_action == MA_ALLOW_PROMISOR);28182819 /*2820 * Quietly ignore EXPECTED missing objects. This avoids problems with2821 * staging them now and getting an odd error later.2822 */2823 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))2824 return;28252826 show_object(obj, name, data);2827}28282829static int option_parse_missing_action(const struct option *opt,2830 const char *arg, int unset)2831{2832 assert(arg);2833 assert(!unset);28342835 if (!strcmp(arg, "error")) {2836 arg_missing_action = MA_ERROR;2837 fn_show_object = show_object;2838 return 0;2839 }28402841 if (!strcmp(arg, "allow-any")) {2842 arg_missing_action = MA_ALLOW_ANY;2843 fetch_if_missing = 0;2844 fn_show_object = show_object__ma_allow_any;2845 return 0;2846 }28472848 if (!strcmp(arg, "allow-promisor")) {2849 arg_missing_action = MA_ALLOW_PROMISOR;2850 fetch_if_missing = 0;2851 fn_show_object = show_object__ma_allow_promisor;2852 return 0;2853 }28542855 die(_("invalid value for --missing"));2856 return 0;2857}28582859static void show_edge(struct commit *commit)2860{2861 add_preferred_base(&commit->object.oid);2862}28632864struct in_pack_object {2865 off_t offset;2866 struct object *object;2867};28682869struct in_pack {2870 unsigned int alloc;2871 unsigned int nr;2872 struct in_pack_object *array;2873};28742875static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)2876{2877 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);2878 in_pack->array[in_pack->nr].object = object;2879 in_pack->nr++;2880}28812882/*2883 * Compare the objects in the offset order, in order to emulate the2884 * "git rev-list --objects" output that produced the pack originally.2885 */2886static int ofscmp(const void *a_, const void *b_)2887{2888 struct in_pack_object *a = (struct in_pack_object *)a_;2889 struct in_pack_object *b = (struct in_pack_object *)b_;28902891 if (a->offset < b->offset)2892 return -1;2893 else if (a->offset > b->offset)2894 return 1;2895 else2896 return oidcmp(&a->object->oid, &b->object->oid);2897}28982899static void add_objects_in_unpacked_packs(struct rev_info *revs)2900{2901 struct packed_git *p;2902 struct in_pack in_pack;2903 uint32_t i;29042905 memset(&in_pack, 0, sizeof(in_pack));29062907 for (p = get_all_packs(the_repository); p; p = p->next) {2908 struct object_id oid;2909 struct object *o;29102911 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)2912 continue;2913 if (open_pack_index(p))2914 die(_("cannot open pack index"));29152916 ALLOC_GROW(in_pack.array,2917 in_pack.nr + p->num_objects,2918 in_pack.alloc);29192920 for (i = 0; i < p->num_objects; i++) {2921 nth_packed_object_oid(&oid, p, i);2922 o = lookup_unknown_object(oid.hash);2923 if (!(o->flags & OBJECT_ADDED))2924 mark_in_pack_object(o, p, &in_pack);2925 o->flags |= OBJECT_ADDED;2926 }2927 }29282929 if (in_pack.nr) {2930 QSORT(in_pack.array, in_pack.nr, ofscmp);2931 for (i = 0; i < in_pack.nr; i++) {2932 struct object *o = in_pack.array[i].object;2933 add_object_entry(&o->oid, o->type, "", 0);2934 }2935 }2936 free(in_pack.array);2937}29382939static int add_loose_object(const struct object_id *oid, const char *path,2940 void *data)2941{2942 enum object_type type = oid_object_info(the_repository, oid, NULL);29432944 if (type < 0) {2945 warning(_("loose object at %s could not be examined"), path);2946 return 0;2947 }29482949 add_object_entry(oid, type, "", 0);2950 return 0;2951}29522953/*2954 * We actually don't even have to worry about reachability here.2955 * add_object_entry will weed out duplicates, so we just add every2956 * loose object we find.2957 */2958static void add_unreachable_loose_objects(void)2959{2960 for_each_loose_file_in_objdir(get_object_directory(),2961 add_loose_object,2962 NULL, NULL, NULL);2963}29642965static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2966{2967 static struct packed_git *last_found = (void *)1;2968 struct packed_git *p;29692970 p = (last_found != (void *)1) ? last_found :2971 get_all_packs(the_repository);29722973 while (p) {2974 if ((!p->pack_local || p->pack_keep ||2975 p->pack_keep_in_core) &&2976 find_pack_entry_one(oid->hash, p)) {2977 last_found = p;2978 return 1;2979 }2980 if (p == last_found)2981 p = get_all_packs(the_repository);2982 else2983 p = p->next;2984 if (p == last_found)2985 p = p->next;2986 }2987 return 0;2988}29892990/*2991 * Store a list of sha1s that are should not be discarded2992 * because they are either written too recently, or are2993 * reachable from another object that was.2994 *2995 * This is filled by get_object_list.2996 */2997static struct oid_array recent_objects;29982999static int loosened_object_can_be_discarded(const struct object_id *oid,3000 timestamp_t mtime)3001{3002 if (!unpack_unreachable_expiration)3003 return 0;3004 if (mtime > unpack_unreachable_expiration)3005 return 0;3006 if (oid_array_lookup(&recent_objects, oid) >= 0)3007 return 0;3008 return 1;3009}30103011static void loosen_unused_packed_objects(struct rev_info *revs)3012{3013 struct packed_git *p;3014 uint32_t i;3015 struct object_id oid;30163017 for (p = get_all_packs(the_repository); p; p = p->next) {3018 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)3019 continue;30203021 if (open_pack_index(p))3022 die(_("cannot open pack index"));30233024 for (i = 0; i < p->num_objects; i++) {3025 nth_packed_object_oid(&oid, p, i);3026 if (!packlist_find(&to_pack, oid.hash, NULL) &&3027 !has_sha1_pack_kept_or_nonlocal(&oid) &&3028 !loosened_object_can_be_discarded(&oid, p->mtime))3029 if (force_object_loose(&oid, p->mtime))3030 die(_("unable to force loose object"));3031 }3032 }3033}30343035/*3036 * This tracks any options which pack-reuse code expects to be on, or which a3037 * reader of the pack might not understand, and which would therefore prevent3038 * blind reuse of what we have on disk.3039 */3040static int pack_options_allow_reuse(void)3041{3042 return pack_to_stdout &&3043 allow_ofs_delta &&3044 !ignore_packed_keep_on_disk &&3045 !ignore_packed_keep_in_core &&3046 (!local || !have_non_local_packs) &&3047 !incremental;3048}30493050static int get_object_list_from_bitmap(struct rev_info *revs)3051{3052 if (!(bitmap_git = prepare_bitmap_walk(revs)))3053 return -1;30543055 if (pack_options_allow_reuse() &&3056 !reuse_partial_packfile_from_bitmap(3057 bitmap_git,3058 &reuse_packfile,3059 &reuse_packfile_objects,3060 &reuse_packfile_offset)) {3061 assert(reuse_packfile_objects);3062 nr_result += reuse_packfile_objects;3063 display_progress(progress_state, nr_result);3064 }30653066 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);3067 return 0;3068}30693070static void record_recent_object(struct object *obj,3071 const char *name,3072 void *data)3073{3074 oid_array_append(&recent_objects, &obj->oid);3075}30763077static void record_recent_commit(struct commit *commit, void *data)3078{3079 oid_array_append(&recent_objects, &commit->object.oid);3080}30813082static void get_object_list(int ac, const char **av)3083{3084 struct rev_info revs;3085 struct setup_revision_opt s_r_opt = {3086 .allow_exclude_promisor_objects = 1,3087 };3088 char line[1000];3089 int flags = 0;3090 int save_warning;30913092 repo_init_revisions(the_repository, &revs, NULL);3093 save_commit_buffer = 0;3094 setup_revisions(ac, av, &revs, &s_r_opt);30953096 /* make sure shallows are read */3097 is_repository_shallow(the_repository);30983099 save_warning = warn_on_object_refname_ambiguity;3100 warn_on_object_refname_ambiguity = 0;31013102 while (fgets(line, sizeof(line), stdin) != NULL) {3103 int len = strlen(line);3104 if (len && line[len - 1] == '\n')3105 line[--len] = 0;3106 if (!len)3107 break;3108 if (*line == '-') {3109 if (!strcmp(line, "--not")) {3110 flags ^= UNINTERESTING;3111 write_bitmap_index = 0;3112 continue;3113 }3114 if (starts_with(line, "--shallow ")) {3115 struct object_id oid;3116 if (get_oid_hex(line + 10, &oid))3117 die("not an SHA-1 '%s'", line + 10);3118 register_shallow(the_repository, &oid);3119 use_bitmap_index = 0;3120 continue;3121 }3122 die(_("not a rev '%s'"), line);3123 }3124 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3125 die(_("bad revision '%s'"), line);3126 }31273128 warn_on_object_refname_ambiguity = save_warning;31293130 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))3131 return;31323133 if (use_delta_islands)3134 load_delta_islands(the_repository);31353136 if (prepare_revision_walk(&revs))3137 die(_("revision walk setup failed"));3138 mark_edges_uninteresting(&revs, show_edge, sparse);31393140 if (!fn_show_object)3141 fn_show_object = show_object;3142 traverse_commit_list_filtered(&filter_options, &revs,3143 show_commit, fn_show_object, NULL,3144 NULL);31453146 if (unpack_unreachable_expiration) {3147 revs.ignore_missing_links = 1;3148 if (add_unseen_recent_objects_to_traversal(&revs,3149 unpack_unreachable_expiration))3150 die(_("unable to add recent objects"));3151 if (prepare_revision_walk(&revs))3152 die(_("revision walk setup failed"));3153 traverse_commit_list(&revs, record_recent_commit,3154 record_recent_object, NULL);3155 }31563157 if (keep_unreachable)3158 add_objects_in_unpacked_packs(&revs);3159 if (pack_loose_unreachable)3160 add_unreachable_loose_objects();3161 if (unpack_unreachable)3162 loosen_unused_packed_objects(&revs);31633164 oid_array_clear(&recent_objects);3165}31663167static void add_extra_kept_packs(const struct string_list *names)3168{3169 struct packed_git *p;31703171 if (!names->nr)3172 return;31733174 for (p = get_all_packs(the_repository); p; p = p->next) {3175 const char *name = basename(p->pack_name);3176 int i;31773178 if (!p->pack_local)3179 continue;31803181 for (i = 0; i < names->nr; i++)3182 if (!fspathcmp(name, names->items[i].string))3183 break;31843185 if (i < names->nr) {3186 p->pack_keep_in_core = 1;3187 ignore_packed_keep_in_core = 1;3188 continue;3189 }3190 }3191}31923193static int option_parse_index_version(const struct option *opt,3194 const char *arg, int unset)3195{3196 char *c;3197 const char *val = arg;31983199 BUG_ON_OPT_NEG(unset);32003201 pack_idx_opts.version = strtoul(val, &c, 10);3202 if (pack_idx_opts.version > 2)3203 die(_("unsupported index version %s"), val);3204 if (*c == ',' && c[1])3205 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);3206 if (*c || pack_idx_opts.off32_limit & 0x80000000)3207 die(_("bad index version '%s'"), val);3208 return 0;3209}32103211static int option_parse_unpack_unreachable(const struct option *opt,3212 const char *arg, int unset)3213{3214 if (unset) {3215 unpack_unreachable = 0;3216 unpack_unreachable_expiration = 0;3217 }3218 else {3219 unpack_unreachable = 1;3220 if (arg)3221 unpack_unreachable_expiration = approxidate(arg);3222 }3223 return 0;3224}32253226int cmd_pack_objects(int argc, const char **argv, const char *prefix)3227{3228 int use_internal_rev_list = 0;3229 int shallow = 0;3230 int all_progress_implied = 0;3231 struct argv_array rp = ARGV_ARRAY_INIT;3232 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;3233 int rev_list_index = 0;3234 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3235 struct option pack_objects_options[] = {3236 OPT_SET_INT('q', "quiet", &progress,3237 N_("do not show progress meter"), 0),3238 OPT_SET_INT(0, "progress", &progress,3239 N_("show progress meter"), 1),3240 OPT_SET_INT(0, "all-progress", &progress,3241 N_("show progress meter during object writing phase"), 2),3242 OPT_BOOL(0, "all-progress-implied",3243 &all_progress_implied,3244 N_("similar to --all-progress when progress meter is shown")),3245 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),3246 N_("write the pack index file in the specified idx format version"),3247 PARSE_OPT_NONEG, option_parse_index_version },3248 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,3249 N_("maximum size of each output pack file")),3250 OPT_BOOL(0, "local", &local,3251 N_("ignore borrowed objects from alternate object store")),3252 OPT_BOOL(0, "incremental", &incremental,3253 N_("ignore packed objects")),3254 OPT_INTEGER(0, "window", &window,3255 N_("limit pack window by objects")),3256 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,3257 N_("limit pack window by memory in addition to object limit")),3258 OPT_INTEGER(0, "depth", &depth,3259 N_("maximum length of delta chain allowed in the resulting pack")),3260 OPT_BOOL(0, "reuse-delta", &reuse_delta,3261 N_("reuse existing deltas")),3262 OPT_BOOL(0, "reuse-object", &reuse_object,3263 N_("reuse existing objects")),3264 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,3265 N_("use OFS_DELTA objects")),3266 OPT_INTEGER(0, "threads", &delta_search_threads,3267 N_("use threads when searching for best delta matches")),3268 OPT_BOOL(0, "non-empty", &non_empty,3269 N_("do not create an empty pack output")),3270 OPT_BOOL(0, "revs", &use_internal_rev_list,3271 N_("read revision arguments from standard input")),3272 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,3273 N_("limit the objects to those that are not yet packed"),3274 1, PARSE_OPT_NONEG),3275 OPT_SET_INT_F(0, "all", &rev_list_all,3276 N_("include objects reachable from any reference"),3277 1, PARSE_OPT_NONEG),3278 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,3279 N_("include objects referred by reflog entries"),3280 1, PARSE_OPT_NONEG),3281 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,3282 N_("include objects referred to by the index"),3283 1, PARSE_OPT_NONEG),3284 OPT_BOOL(0, "stdout", &pack_to_stdout,3285 N_("output pack to stdout")),3286 OPT_BOOL(0, "include-tag", &include_tag,3287 N_("include tag objects that refer to objects to be packed")),3288 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,3289 N_("keep unreachable objects")),3290 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,3291 N_("pack loose unreachable objects")),3292 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),3293 N_("unpack unreachable objects newer than <time>"),3294 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3295 OPT_BOOL(0, "sparse", &sparse,3296 N_("use the sparse reachability algorithm")),3297 OPT_BOOL(0, "thin", &thin,3298 N_("create thin packs")),3299 OPT_BOOL(0, "shallow", &shallow,3300 N_("create packs suitable for shallow fetches")),3301 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,3302 N_("ignore packs that have companion .keep file")),3303 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),3304 N_("ignore this pack")),3305 OPT_INTEGER(0, "compression", &pack_compression_level,3306 N_("pack compression level")),3307 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,3308 N_("do not hide commits by grafts"), 0),3309 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,3310 N_("use a bitmap index if available to speed up counting objects")),3311 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,3312 N_("write a bitmap index together with the pack index")),3313 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3314 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),3315 N_("handling for missing objects"), PARSE_OPT_NONEG,3316 option_parse_missing_action },3317 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,3318 N_("do not pack objects in promisor packfiles")),3319 OPT_BOOL(0, "delta-islands", &use_delta_islands,3320 N_("respect islands during delta compression")),3321 OPT_END(),3322 };33233324 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))3325 BUG("too many dfs states, increase OE_DFS_STATE_BITS");33263327 read_replace_refs = 0;33283329 sparse = git_env_bool("GIT_TEST_PACK_SPARSE", 0);3330 reset_pack_idx_option(&pack_idx_opts);3331 git_config(git_pack_config, NULL);33323333 progress = isatty(2);3334 argc = parse_options(argc, argv, prefix, pack_objects_options,3335 pack_usage, 0);33363337 if (argc) {3338 base_name = argv[0];3339 argc--;3340 }3341 if (pack_to_stdout != !base_name || argc)3342 usage_with_options(pack_usage, pack_objects_options);33433344 if (depth >= (1 << OE_DEPTH_BITS)) {3345 warning(_("delta chain depth %d is too deep, forcing %d"),3346 depth, (1 << OE_DEPTH_BITS) - 1);3347 depth = (1 << OE_DEPTH_BITS) - 1;3348 }3349 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {3350 warning(_("pack.deltaCacheLimit is too high, forcing %d"),3351 (1U << OE_Z_DELTA_BITS) - 1);3352 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;3353 }33543355 argv_array_push(&rp, "pack-objects");3356 if (thin) {3357 use_internal_rev_list = 1;3358 argv_array_push(&rp, shallow3359 ? "--objects-edge-aggressive"3360 : "--objects-edge");3361 } else3362 argv_array_push(&rp, "--objects");33633364 if (rev_list_all) {3365 use_internal_rev_list = 1;3366 argv_array_push(&rp, "--all");3367 }3368 if (rev_list_reflog) {3369 use_internal_rev_list = 1;3370 argv_array_push(&rp, "--reflog");3371 }3372 if (rev_list_index) {3373 use_internal_rev_list = 1;3374 argv_array_push(&rp, "--indexed-objects");3375 }3376 if (rev_list_unpacked) {3377 use_internal_rev_list = 1;3378 argv_array_push(&rp, "--unpacked");3379 }33803381 if (exclude_promisor_objects) {3382 use_internal_rev_list = 1;3383 fetch_if_missing = 0;3384 argv_array_push(&rp, "--exclude-promisor-objects");3385 }3386 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)3387 use_internal_rev_list = 1;33883389 if (!reuse_object)3390 reuse_delta = 0;3391 if (pack_compression_level == -1)3392 pack_compression_level = Z_DEFAULT_COMPRESSION;3393 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)3394 die(_("bad pack compression level %d"), pack_compression_level);33953396 if (!delta_search_threads) /* --threads=0 means autodetect */3397 delta_search_threads = online_cpus();33983399 if (!HAVE_THREADS && delta_search_threads != 1)3400 warning(_("no threads support, ignoring --threads"));3401 if (!pack_to_stdout && !pack_size_limit)3402 pack_size_limit = pack_size_limit_cfg;3403 if (pack_to_stdout && pack_size_limit)3404 die(_("--max-pack-size cannot be used to build a pack for transfer"));3405 if (pack_size_limit && pack_size_limit < 1024*1024) {3406 warning(_("minimum pack size limit is 1 MiB"));3407 pack_size_limit = 1024*1024;3408 }34093410 if (!pack_to_stdout && thin)3411 die(_("--thin cannot be used to build an indexable pack"));34123413 if (keep_unreachable && unpack_unreachable)3414 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));3415 if (!rev_list_all || !rev_list_reflog || !rev_list_index)3416 unpack_unreachable_expiration = 0;34173418 if (filter_options.choice) {3419 if (!pack_to_stdout)3420 die(_("cannot use --filter without --stdout"));3421 use_bitmap_index = 0;3422 }34233424 /*3425 * "soft" reasons not to use bitmaps - for on-disk repack by default we want3426 *3427 * - to produce good pack (with bitmap index not-yet-packed objects are3428 * packed in suboptimal order).3429 *3430 * - to use more robust pack-generation codepath (avoiding possible3431 * bugs in bitmap code and possible bitmap index corruption).3432 */3433 if (!pack_to_stdout)3434 use_bitmap_index_default = 0;34353436 if (use_bitmap_index < 0)3437 use_bitmap_index = use_bitmap_index_default;34383439 /* "hard" reasons not to use bitmaps; these just won't work at all */3440 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))3441 use_bitmap_index = 0;34423443 if (pack_to_stdout || !rev_list_all)3444 write_bitmap_index = 0;34453446 if (use_delta_islands)3447 argv_array_push(&rp, "--topo-order");34483449 if (progress && all_progress_implied)3450 progress = 2;34513452 add_extra_kept_packs(&keep_pack_list);3453 if (ignore_packed_keep_on_disk) {3454 struct packed_git *p;3455 for (p = get_all_packs(the_repository); p; p = p->next)3456 if (p->pack_local && p->pack_keep)3457 break;3458 if (!p) /* no keep-able packs found */3459 ignore_packed_keep_on_disk = 0;3460 }3461 if (local) {3462 /*3463 * unlike ignore_packed_keep_on_disk above, we do not3464 * want to unset "local" based on looking at packs, as3465 * it also covers non-local objects3466 */3467 struct packed_git *p;3468 for (p = get_all_packs(the_repository); p; p = p->next) {3469 if (!p->pack_local) {3470 have_non_local_packs = 1;3471 break;3472 }3473 }3474 }34753476 prepare_packing_data(the_repository, &to_pack);34773478 if (progress)3479 progress_state = start_progress(_("Enumerating objects"), 0);3480 if (!use_internal_rev_list)3481 read_object_list_from_stdin();3482 else {3483 get_object_list(rp.argc, rp.argv);3484 argv_array_clear(&rp);3485 }3486 cleanup_preferred_base();3487 if (include_tag && nr_result)3488 for_each_ref(add_ref_tag, NULL);3489 stop_progress(&progress_state);34903491 if (non_empty && !nr_result)3492 return 0;3493 if (nr_result)3494 prepare_pack(window, depth);3495 write_pack_file();3496 if (progress)3497 fprintf_ln(stderr,3498 _("Total %"PRIu32" (delta %"PRIu32"),"3499 " reused %"PRIu32" (delta %"PRIu32")"),3500 written, written_delta, reused, reused_delta);3501 return 0;3502}