1#include "builtin.h" 2#include "cache.h" 3#include "repository.h" 4#include "config.h" 5#include "attr.h" 6#include "object.h" 7#include "blob.h" 8#include "commit.h" 9#include "tag.h" 10#include "tree.h" 11#include "delta.h" 12#include "pack.h" 13#include "pack-revindex.h" 14#include "csum-file.h" 15#include "tree-walk.h" 16#include "diff.h" 17#include "revision.h" 18#include "list-objects.h" 19#include "list-objects-filter.h" 20#include "list-objects-filter-options.h" 21#include "pack-objects.h" 22#include "progress.h" 23#include "refs.h" 24#include "streaming.h" 25#include "thread-utils.h" 26#include "pack-bitmap.h" 27#include "delta-islands.h" 28#include "reachable.h" 29#include "sha1-array.h" 30#include "argv-array.h" 31#include "list.h" 32#include "packfile.h" 33#include "object-store.h" 34#include "dir.h" 35#include "midx.h" 36#include "trace2.h" 37 38#define IN_PACK(obj) oe_in_pack(&to_pack, obj) 39#define SIZE(obj) oe_size(&to_pack, obj) 40#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size) 41#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj) 42#define DELTA(obj) oe_delta(&to_pack, obj) 43#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj) 44#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj) 45#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val) 46#define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid) 47#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val) 48#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val) 49#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val) 50 51static const char *pack_usage[] = { 52 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), 53 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), 54 NULL 55}; 56 57/* 58 * Objects we are going to pack are collected in the `to_pack` structure. 59 * It contains an array (dynamically expanded) of the object data, and a map 60 * that can resolve SHA1s to their position in the array. 61 */ 62static struct packing_data to_pack; 63 64static struct pack_idx_entry **written_list; 65static uint32_t nr_result, nr_written, nr_seen; 66static struct bitmap_index *bitmap_git; 67static uint32_t write_layer; 68 69static int non_empty; 70static int reuse_delta = 1, reuse_object = 1; 71static int keep_unreachable, unpack_unreachable, include_tag; 72static timestamp_t unpack_unreachable_expiration; 73static int pack_loose_unreachable; 74static int local; 75static int have_non_local_packs; 76static int incremental; 77static int ignore_packed_keep_on_disk; 78static int ignore_packed_keep_in_core; 79static int allow_ofs_delta; 80static struct pack_idx_option pack_idx_opts; 81static const char *base_name; 82static int progress = 1; 83static int window = 10; 84static unsigned long pack_size_limit; 85static int depth = 50; 86static int delta_search_threads; 87static int pack_to_stdout; 88static int sparse; 89static int thin; 90static int num_preferred_base; 91static struct progress *progress_state; 92 93static struct packed_git *reuse_packfile; 94static uint32_t reuse_packfile_objects; 95static off_t reuse_packfile_offset; 96 97static int use_bitmap_index_default = 1; 98static int use_bitmap_index = -1; 99static int write_bitmap_index; 100static uint16_t write_bitmap_options; 101 102static int exclude_promisor_objects; 103 104static int use_delta_islands; 105 106static unsigned long delta_cache_size = 0; 107static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; 108static unsigned long cache_max_small_delta_size = 1000; 109 110static unsigned long window_memory_limit = 0; 111 112static struct list_objects_filter_options filter_options; 113 114enum missing_action { 115 MA_ERROR = 0, /* fail if any missing objects are encountered */ 116 MA_ALLOW_ANY, /* silently allow ALL missing objects */ 117 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */ 118}; 119static enum missing_action arg_missing_action; 120static show_object_fn fn_show_object; 121 122/* 123 * stats 124 */ 125static uint32_t written, written_delta; 126static uint32_t reused, reused_delta; 127 128/* 129 * Indexed commits 130 */ 131static struct commit **indexed_commits; 132static unsigned int indexed_commits_nr; 133static unsigned int indexed_commits_alloc; 134 135static void index_commit_for_bitmap(struct commit *commit) 136{ 137 if (indexed_commits_nr >= indexed_commits_alloc) { 138 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2; 139 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); 140 } 141 142 indexed_commits[indexed_commits_nr++] = commit; 143} 144 145static void *get_delta(struct object_entry *entry) 146{ 147 unsigned long size, base_size, delta_size; 148 void *buf, *base_buf, *delta_buf; 149 enum object_type type; 150 151 buf = read_object_file(&entry->idx.oid, &type, &size); 152 if (!buf) 153 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid)); 154 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type, 155 &base_size); 156 if (!base_buf) 157 die("unable to read %s", 158 oid_to_hex(&DELTA(entry)->idx.oid)); 159 delta_buf = diff_delta(base_buf, base_size, 160 buf, size, &delta_size, 0); 161 /* 162 * We succesfully computed this delta once but dropped it for 163 * memory reasons. Something is very wrong if this time we 164 * recompute and create a different delta. 165 */ 166 if (!delta_buf || delta_size != DELTA_SIZE(entry)) 167 BUG("delta size changed"); 168 free(buf); 169 free(base_buf); 170 return delta_buf; 171} 172 173static unsigned long do_compress(void **pptr, unsigned long size) 174{ 175 git_zstream stream; 176 void *in, *out; 177 unsigned long maxsize; 178 179 git_deflate_init(&stream, pack_compression_level); 180 maxsize = git_deflate_bound(&stream, size); 181 182 in = *pptr; 183 out = xmalloc(maxsize); 184 *pptr = out; 185 186 stream.next_in = in; 187 stream.avail_in = size; 188 stream.next_out = out; 189 stream.avail_out = maxsize; 190 while (git_deflate(&stream, Z_FINISH) == Z_OK) 191 ; /* nothing */ 192 git_deflate_end(&stream); 193 194 free(in); 195 return stream.total_out; 196} 197 198static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f, 199 const struct object_id *oid) 200{ 201 git_zstream stream; 202 unsigned char ibuf[1024 * 16]; 203 unsigned char obuf[1024 * 16]; 204 unsigned long olen = 0; 205 206 git_deflate_init(&stream, pack_compression_level); 207 208 for (;;) { 209 ssize_t readlen; 210 int zret = Z_OK; 211 readlen = read_istream(st, ibuf, sizeof(ibuf)); 212 if (readlen == -1) 213 die(_("unable to read %s"), oid_to_hex(oid)); 214 215 stream.next_in = ibuf; 216 stream.avail_in = readlen; 217 while ((stream.avail_in || readlen == 0) && 218 (zret == Z_OK || zret == Z_BUF_ERROR)) { 219 stream.next_out = obuf; 220 stream.avail_out = sizeof(obuf); 221 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH); 222 hashwrite(f, obuf, stream.next_out - obuf); 223 olen += stream.next_out - obuf; 224 } 225 if (stream.avail_in) 226 die(_("deflate error (%d)"), zret); 227 if (readlen == 0) { 228 if (zret != Z_STREAM_END) 229 die(_("deflate error (%d)"), zret); 230 break; 231 } 232 } 233 git_deflate_end(&stream); 234 return olen; 235} 236 237/* 238 * we are going to reuse the existing object data as is. make 239 * sure it is not corrupt. 240 */ 241static int check_pack_inflate(struct packed_git *p, 242 struct pack_window **w_curs, 243 off_t offset, 244 off_t len, 245 unsigned long expect) 246{ 247 git_zstream stream; 248 unsigned char fakebuf[4096], *in; 249 int st; 250 251 memset(&stream, 0, sizeof(stream)); 252 git_inflate_init(&stream); 253 do { 254 in = use_pack(p, w_curs, offset, &stream.avail_in); 255 stream.next_in = in; 256 stream.next_out = fakebuf; 257 stream.avail_out = sizeof(fakebuf); 258 st = git_inflate(&stream, Z_FINISH); 259 offset += stream.next_in - in; 260 } while (st == Z_OK || st == Z_BUF_ERROR); 261 git_inflate_end(&stream); 262 return (st == Z_STREAM_END && 263 stream.total_out == expect && 264 stream.total_in == len) ? 0 : -1; 265} 266 267static void copy_pack_data(struct hashfile *f, 268 struct packed_git *p, 269 struct pack_window **w_curs, 270 off_t offset, 271 off_t len) 272{ 273 unsigned char *in; 274 unsigned long avail; 275 276 while (len) { 277 in = use_pack(p, w_curs, offset, &avail); 278 if (avail > len) 279 avail = (unsigned long)len; 280 hashwrite(f, in, avail); 281 offset += avail; 282 len -= avail; 283 } 284} 285 286/* Return 0 if we will bust the pack-size limit */ 287static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry, 288 unsigned long limit, int usable_delta) 289{ 290 unsigned long size, datalen; 291 unsigned char header[MAX_PACK_OBJECT_HEADER], 292 dheader[MAX_PACK_OBJECT_HEADER]; 293 unsigned hdrlen; 294 enum object_type type; 295 void *buf; 296 struct git_istream *st = NULL; 297 const unsigned hashsz = the_hash_algo->rawsz; 298 299 if (!usable_delta) { 300 if (oe_type(entry) == OBJ_BLOB && 301 oe_size_greater_than(&to_pack, entry, big_file_threshold) && 302 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL) 303 buf = NULL; 304 else { 305 buf = read_object_file(&entry->idx.oid, &type, &size); 306 if (!buf) 307 die(_("unable to read %s"), 308 oid_to_hex(&entry->idx.oid)); 309 } 310 /* 311 * make sure no cached delta data remains from a 312 * previous attempt before a pack split occurred. 313 */ 314 FREE_AND_NULL(entry->delta_data); 315 entry->z_delta_size = 0; 316 } else if (entry->delta_data) { 317 size = DELTA_SIZE(entry); 318 buf = entry->delta_data; 319 entry->delta_data = NULL; 320 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 321 OBJ_OFS_DELTA : OBJ_REF_DELTA; 322 } else { 323 buf = get_delta(entry); 324 size = DELTA_SIZE(entry); 325 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 326 OBJ_OFS_DELTA : OBJ_REF_DELTA; 327 } 328 329 if (st) /* large blob case, just assume we don't compress well */ 330 datalen = size; 331 else if (entry->z_delta_size) 332 datalen = entry->z_delta_size; 333 else 334 datalen = do_compress(&buf, size); 335 336 /* 337 * The object header is a byte of 'type' followed by zero or 338 * more bytes of length. 339 */ 340 hdrlen = encode_in_pack_object_header(header, sizeof(header), 341 type, size); 342 343 if (type == OBJ_OFS_DELTA) { 344 /* 345 * Deltas with relative base contain an additional 346 * encoding of the relative offset for the delta 347 * base from this object's position in the pack. 348 */ 349 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 350 unsigned pos = sizeof(dheader) - 1; 351 dheader[pos] = ofs & 127; 352 while (ofs >>= 7) 353 dheader[--pos] = 128 | (--ofs & 127); 354 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 355 if (st) 356 close_istream(st); 357 free(buf); 358 return 0; 359 } 360 hashwrite(f, header, hdrlen); 361 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 362 hdrlen += sizeof(dheader) - pos; 363 } else if (type == OBJ_REF_DELTA) { 364 /* 365 * Deltas with a base reference contain 366 * additional bytes for the base object ID. 367 */ 368 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 369 if (st) 370 close_istream(st); 371 free(buf); 372 return 0; 373 } 374 hashwrite(f, header, hdrlen); 375 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 376 hdrlen += hashsz; 377 } else { 378 if (limit && hdrlen + datalen + hashsz >= limit) { 379 if (st) 380 close_istream(st); 381 free(buf); 382 return 0; 383 } 384 hashwrite(f, header, hdrlen); 385 } 386 if (st) { 387 datalen = write_large_blob_data(st, f, &entry->idx.oid); 388 close_istream(st); 389 } else { 390 hashwrite(f, buf, datalen); 391 free(buf); 392 } 393 394 return hdrlen + datalen; 395} 396 397/* Return 0 if we will bust the pack-size limit */ 398static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry, 399 unsigned long limit, int usable_delta) 400{ 401 struct packed_git *p = IN_PACK(entry); 402 struct pack_window *w_curs = NULL; 403 struct revindex_entry *revidx; 404 off_t offset; 405 enum object_type type = oe_type(entry); 406 off_t datalen; 407 unsigned char header[MAX_PACK_OBJECT_HEADER], 408 dheader[MAX_PACK_OBJECT_HEADER]; 409 unsigned hdrlen; 410 const unsigned hashsz = the_hash_algo->rawsz; 411 unsigned long entry_size = SIZE(entry); 412 413 if (DELTA(entry)) 414 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 415 OBJ_OFS_DELTA : OBJ_REF_DELTA; 416 hdrlen = encode_in_pack_object_header(header, sizeof(header), 417 type, entry_size); 418 419 offset = entry->in_pack_offset; 420 revidx = find_pack_revindex(p, offset); 421 datalen = revidx[1].offset - offset; 422 if (!pack_to_stdout && p->index_version > 1 && 423 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 424 error(_("bad packed object CRC for %s"), 425 oid_to_hex(&entry->idx.oid)); 426 unuse_pack(&w_curs); 427 return write_no_reuse_object(f, entry, limit, usable_delta); 428 } 429 430 offset += entry->in_pack_header_size; 431 datalen -= entry->in_pack_header_size; 432 433 if (!pack_to_stdout && p->index_version == 1 && 434 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { 435 error(_("corrupt packed object for %s"), 436 oid_to_hex(&entry->idx.oid)); 437 unuse_pack(&w_curs); 438 return write_no_reuse_object(f, entry, limit, usable_delta); 439 } 440 441 if (type == OBJ_OFS_DELTA) { 442 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 443 unsigned pos = sizeof(dheader) - 1; 444 dheader[pos] = ofs & 127; 445 while (ofs >>= 7) 446 dheader[--pos] = 128 | (--ofs & 127); 447 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 448 unuse_pack(&w_curs); 449 return 0; 450 } 451 hashwrite(f, header, hdrlen); 452 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 453 hdrlen += sizeof(dheader) - pos; 454 reused_delta++; 455 } else if (type == OBJ_REF_DELTA) { 456 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 457 unuse_pack(&w_curs); 458 return 0; 459 } 460 hashwrite(f, header, hdrlen); 461 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 462 hdrlen += hashsz; 463 reused_delta++; 464 } else { 465 if (limit && hdrlen + datalen + hashsz >= limit) { 466 unuse_pack(&w_curs); 467 return 0; 468 } 469 hashwrite(f, header, hdrlen); 470 } 471 copy_pack_data(f, p, &w_curs, offset, datalen); 472 unuse_pack(&w_curs); 473 reused++; 474 return hdrlen + datalen; 475} 476 477/* Return 0 if we will bust the pack-size limit */ 478static off_t write_object(struct hashfile *f, 479 struct object_entry *entry, 480 off_t write_offset) 481{ 482 unsigned long limit; 483 off_t len; 484 int usable_delta, to_reuse; 485 486 if (!pack_to_stdout) 487 crc32_begin(f); 488 489 /* apply size limit if limited packsize and not first object */ 490 if (!pack_size_limit || !nr_written) 491 limit = 0; 492 else if (pack_size_limit <= write_offset) 493 /* 494 * the earlier object did not fit the limit; avoid 495 * mistaking this with unlimited (i.e. limit = 0). 496 */ 497 limit = 1; 498 else 499 limit = pack_size_limit - write_offset; 500 501 if (!DELTA(entry)) 502 usable_delta = 0; /* no delta */ 503 else if (!pack_size_limit) 504 usable_delta = 1; /* unlimited packfile */ 505 else if (DELTA(entry)->idx.offset == (off_t)-1) 506 usable_delta = 0; /* base was written to another pack */ 507 else if (DELTA(entry)->idx.offset) 508 usable_delta = 1; /* base already exists in this pack */ 509 else 510 usable_delta = 0; /* base could end up in another pack */ 511 512 if (!reuse_object) 513 to_reuse = 0; /* explicit */ 514 else if (!IN_PACK(entry)) 515 to_reuse = 0; /* can't reuse what we don't have */ 516 else if (oe_type(entry) == OBJ_REF_DELTA || 517 oe_type(entry) == OBJ_OFS_DELTA) 518 /* check_object() decided it for us ... */ 519 to_reuse = usable_delta; 520 /* ... but pack split may override that */ 521 else if (oe_type(entry) != entry->in_pack_type) 522 to_reuse = 0; /* pack has delta which is unusable */ 523 else if (DELTA(entry)) 524 to_reuse = 0; /* we want to pack afresh */ 525 else 526 to_reuse = 1; /* we have it in-pack undeltified, 527 * and we do not need to deltify it. 528 */ 529 530 if (!to_reuse) 531 len = write_no_reuse_object(f, entry, limit, usable_delta); 532 else 533 len = write_reuse_object(f, entry, limit, usable_delta); 534 if (!len) 535 return 0; 536 537 if (usable_delta) 538 written_delta++; 539 written++; 540 if (!pack_to_stdout) 541 entry->idx.crc32 = crc32_end(f); 542 return len; 543} 544 545enum write_one_status { 546 WRITE_ONE_SKIP = -1, /* already written */ 547 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */ 548 WRITE_ONE_WRITTEN = 1, /* normal */ 549 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */ 550}; 551 552static enum write_one_status write_one(struct hashfile *f, 553 struct object_entry *e, 554 off_t *offset) 555{ 556 off_t size; 557 int recursing; 558 559 /* 560 * we set offset to 1 (which is an impossible value) to mark 561 * the fact that this object is involved in "write its base 562 * first before writing a deltified object" recursion. 563 */ 564 recursing = (e->idx.offset == 1); 565 if (recursing) { 566 warning(_("recursive delta detected for object %s"), 567 oid_to_hex(&e->idx.oid)); 568 return WRITE_ONE_RECURSIVE; 569 } else if (e->idx.offset || e->preferred_base) { 570 /* offset is non zero if object is written already. */ 571 return WRITE_ONE_SKIP; 572 } 573 574 /* if we are deltified, write out base object first. */ 575 if (DELTA(e)) { 576 e->idx.offset = 1; /* now recurse */ 577 switch (write_one(f, DELTA(e), offset)) { 578 case WRITE_ONE_RECURSIVE: 579 /* we cannot depend on this one */ 580 SET_DELTA(e, NULL); 581 break; 582 default: 583 break; 584 case WRITE_ONE_BREAK: 585 e->idx.offset = recursing; 586 return WRITE_ONE_BREAK; 587 } 588 } 589 590 e->idx.offset = *offset; 591 size = write_object(f, e, *offset); 592 if (!size) { 593 e->idx.offset = recursing; 594 return WRITE_ONE_BREAK; 595 } 596 written_list[nr_written++] = &e->idx; 597 598 /* make sure off_t is sufficiently large not to wrap */ 599 if (signed_add_overflows(*offset, size)) 600 die(_("pack too large for current definition of off_t")); 601 *offset += size; 602 return WRITE_ONE_WRITTEN; 603} 604 605static int mark_tagged(const char *path, const struct object_id *oid, int flag, 606 void *cb_data) 607{ 608 struct object_id peeled; 609 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL); 610 611 if (entry) 612 entry->tagged = 1; 613 if (!peel_ref(path, &peeled)) { 614 entry = packlist_find(&to_pack, peeled.hash, NULL); 615 if (entry) 616 entry->tagged = 1; 617 } 618 return 0; 619} 620 621static inline void add_to_write_order(struct object_entry **wo, 622 unsigned int *endp, 623 struct object_entry *e) 624{ 625 if (e->filled || oe_layer(&to_pack, e) != write_layer) 626 return; 627 wo[(*endp)++] = e; 628 e->filled = 1; 629} 630 631static void add_descendants_to_write_order(struct object_entry **wo, 632 unsigned int *endp, 633 struct object_entry *e) 634{ 635 int add_to_order = 1; 636 while (e) { 637 if (add_to_order) { 638 struct object_entry *s; 639 /* add this node... */ 640 add_to_write_order(wo, endp, e); 641 /* all its siblings... */ 642 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) { 643 add_to_write_order(wo, endp, s); 644 } 645 } 646 /* drop down a level to add left subtree nodes if possible */ 647 if (DELTA_CHILD(e)) { 648 add_to_order = 1; 649 e = DELTA_CHILD(e); 650 } else { 651 add_to_order = 0; 652 /* our sibling might have some children, it is next */ 653 if (DELTA_SIBLING(e)) { 654 e = DELTA_SIBLING(e); 655 continue; 656 } 657 /* go back to our parent node */ 658 e = DELTA(e); 659 while (e && !DELTA_SIBLING(e)) { 660 /* we're on the right side of a subtree, keep 661 * going up until we can go right again */ 662 e = DELTA(e); 663 } 664 if (!e) { 665 /* done- we hit our original root node */ 666 return; 667 } 668 /* pass it off to sibling at this level */ 669 e = DELTA_SIBLING(e); 670 } 671 }; 672} 673 674static void add_family_to_write_order(struct object_entry **wo, 675 unsigned int *endp, 676 struct object_entry *e) 677{ 678 struct object_entry *root; 679 680 for (root = e; DELTA(root); root = DELTA(root)) 681 ; /* nothing */ 682 add_descendants_to_write_order(wo, endp, root); 683} 684 685static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end) 686{ 687 unsigned int i, last_untagged; 688 struct object_entry *objects = to_pack.objects; 689 690 for (i = 0; i < to_pack.nr_objects; i++) { 691 if (objects[i].tagged) 692 break; 693 add_to_write_order(wo, wo_end, &objects[i]); 694 } 695 last_untagged = i; 696 697 /* 698 * Then fill all the tagged tips. 699 */ 700 for (; i < to_pack.nr_objects; i++) { 701 if (objects[i].tagged) 702 add_to_write_order(wo, wo_end, &objects[i]); 703 } 704 705 /* 706 * And then all remaining commits and tags. 707 */ 708 for (i = last_untagged; i < to_pack.nr_objects; i++) { 709 if (oe_type(&objects[i]) != OBJ_COMMIT && 710 oe_type(&objects[i]) != OBJ_TAG) 711 continue; 712 add_to_write_order(wo, wo_end, &objects[i]); 713 } 714 715 /* 716 * And then all the trees. 717 */ 718 for (i = last_untagged; i < to_pack.nr_objects; i++) { 719 if (oe_type(&objects[i]) != OBJ_TREE) 720 continue; 721 add_to_write_order(wo, wo_end, &objects[i]); 722 } 723 724 /* 725 * Finally all the rest in really tight order 726 */ 727 for (i = last_untagged; i < to_pack.nr_objects; i++) { 728 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer) 729 add_family_to_write_order(wo, wo_end, &objects[i]); 730 } 731} 732 733static struct object_entry **compute_write_order(void) 734{ 735 uint32_t max_layers = 1; 736 unsigned int i, wo_end; 737 738 struct object_entry **wo; 739 struct object_entry *objects = to_pack.objects; 740 741 for (i = 0; i < to_pack.nr_objects; i++) { 742 objects[i].tagged = 0; 743 objects[i].filled = 0; 744 SET_DELTA_CHILD(&objects[i], NULL); 745 SET_DELTA_SIBLING(&objects[i], NULL); 746 } 747 748 /* 749 * Fully connect delta_child/delta_sibling network. 750 * Make sure delta_sibling is sorted in the original 751 * recency order. 752 */ 753 for (i = to_pack.nr_objects; i > 0;) { 754 struct object_entry *e = &objects[--i]; 755 if (!DELTA(e)) 756 continue; 757 /* Mark me as the first child */ 758 e->delta_sibling_idx = DELTA(e)->delta_child_idx; 759 SET_DELTA_CHILD(DELTA(e), e); 760 } 761 762 /* 763 * Mark objects that are at the tip of tags. 764 */ 765 for_each_tag_ref(mark_tagged, NULL); 766 767 if (use_delta_islands) 768 max_layers = compute_pack_layers(&to_pack); 769 770 ALLOC_ARRAY(wo, to_pack.nr_objects); 771 wo_end = 0; 772 773 for (; write_layer < max_layers; ++write_layer) 774 compute_layer_order(wo, &wo_end); 775 776 if (wo_end != to_pack.nr_objects) 777 die(_("ordered %u objects, expected %"PRIu32), 778 wo_end, to_pack.nr_objects); 779 780 return wo; 781} 782 783static off_t write_reused_pack(struct hashfile *f) 784{ 785 unsigned char buffer[8192]; 786 off_t to_write, total; 787 int fd; 788 789 if (!is_pack_valid(reuse_packfile)) 790 die(_("packfile is invalid: %s"), reuse_packfile->pack_name); 791 792 fd = git_open(reuse_packfile->pack_name); 793 if (fd < 0) 794 die_errno(_("unable to open packfile for reuse: %s"), 795 reuse_packfile->pack_name); 796 797 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1) 798 die_errno(_("unable to seek in reused packfile")); 799 800 if (reuse_packfile_offset < 0) 801 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz; 802 803 total = to_write = reuse_packfile_offset - sizeof(struct pack_header); 804 805 while (to_write) { 806 int read_pack = xread(fd, buffer, sizeof(buffer)); 807 808 if (read_pack <= 0) 809 die_errno(_("unable to read from reused packfile")); 810 811 if (read_pack > to_write) 812 read_pack = to_write; 813 814 hashwrite(f, buffer, read_pack); 815 to_write -= read_pack; 816 817 /* 818 * We don't know the actual number of objects written, 819 * only how many bytes written, how many bytes total, and 820 * how many objects total. So we can fake it by pretending all 821 * objects we are writing are the same size. This gives us a 822 * smooth progress meter, and at the end it matches the true 823 * answer. 824 */ 825 written = reuse_packfile_objects * 826 (((double)(total - to_write)) / total); 827 display_progress(progress_state, written); 828 } 829 830 close(fd); 831 written = reuse_packfile_objects; 832 display_progress(progress_state, written); 833 return reuse_packfile_offset - sizeof(struct pack_header); 834} 835 836static const char no_split_warning[] = N_( 837"disabling bitmap writing, packs are split due to pack.packSizeLimit" 838); 839 840static void write_pack_file(void) 841{ 842 uint32_t i = 0, j; 843 struct hashfile *f; 844 off_t offset; 845 uint32_t nr_remaining = nr_result; 846 time_t last_mtime = 0; 847 struct object_entry **write_order; 848 849 if (progress > pack_to_stdout) 850 progress_state = start_progress(_("Writing objects"), nr_result); 851 ALLOC_ARRAY(written_list, to_pack.nr_objects); 852 write_order = compute_write_order(); 853 854 do { 855 struct object_id oid; 856 char *pack_tmp_name = NULL; 857 858 if (pack_to_stdout) 859 f = hashfd_throughput(1, "<stdout>", progress_state); 860 else 861 f = create_tmp_packfile(&pack_tmp_name); 862 863 offset = write_pack_header(f, nr_remaining); 864 865 if (reuse_packfile) { 866 off_t packfile_size; 867 assert(pack_to_stdout); 868 869 packfile_size = write_reused_pack(f); 870 offset += packfile_size; 871 } 872 873 nr_written = 0; 874 for (; i < to_pack.nr_objects; i++) { 875 struct object_entry *e = write_order[i]; 876 if (write_one(f, e, &offset) == WRITE_ONE_BREAK) 877 break; 878 display_progress(progress_state, written); 879 } 880 881 /* 882 * Did we write the wrong # entries in the header? 883 * If so, rewrite it like in fast-import 884 */ 885 if (pack_to_stdout) { 886 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); 887 } else if (nr_written == nr_remaining) { 888 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); 889 } else { 890 int fd = finalize_hashfile(f, oid.hash, 0); 891 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, 892 nr_written, oid.hash, offset); 893 close(fd); 894 if (write_bitmap_index) { 895 warning(_(no_split_warning)); 896 write_bitmap_index = 0; 897 } 898 } 899 900 if (!pack_to_stdout) { 901 struct stat st; 902 struct strbuf tmpname = STRBUF_INIT; 903 904 /* 905 * Packs are runtime accessed in their mtime 906 * order since newer packs are more likely to contain 907 * younger objects. So if we are creating multiple 908 * packs then we should modify the mtime of later ones 909 * to preserve this property. 910 */ 911 if (stat(pack_tmp_name, &st) < 0) { 912 warning_errno(_("failed to stat %s"), pack_tmp_name); 913 } else if (!last_mtime) { 914 last_mtime = st.st_mtime; 915 } else { 916 struct utimbuf utb; 917 utb.actime = st.st_atime; 918 utb.modtime = --last_mtime; 919 if (utime(pack_tmp_name, &utb) < 0) 920 warning_errno(_("failed utime() on %s"), pack_tmp_name); 921 } 922 923 strbuf_addf(&tmpname, "%s-", base_name); 924 925 if (write_bitmap_index) { 926 bitmap_writer_set_checksum(oid.hash); 927 bitmap_writer_build_type_index( 928 &to_pack, written_list, nr_written); 929 } 930 931 finish_tmp_packfile(&tmpname, pack_tmp_name, 932 written_list, nr_written, 933 &pack_idx_opts, oid.hash); 934 935 if (write_bitmap_index) { 936 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid)); 937 938 stop_progress(&progress_state); 939 940 bitmap_writer_show_progress(progress); 941 bitmap_writer_reuse_bitmaps(&to_pack); 942 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 943 bitmap_writer_build(&to_pack); 944 bitmap_writer_finish(written_list, nr_written, 945 tmpname.buf, write_bitmap_options); 946 write_bitmap_index = 0; 947 } 948 949 strbuf_release(&tmpname); 950 free(pack_tmp_name); 951 puts(oid_to_hex(&oid)); 952 } 953 954 /* mark written objects as written to previous pack */ 955 for (j = 0; j < nr_written; j++) { 956 written_list[j]->offset = (off_t)-1; 957 } 958 nr_remaining -= nr_written; 959 } while (nr_remaining && i < to_pack.nr_objects); 960 961 free(written_list); 962 free(write_order); 963 stop_progress(&progress_state); 964 if (written != nr_result) 965 die(_("wrote %"PRIu32" objects while expecting %"PRIu32), 966 written, nr_result); 967} 968 969static int no_try_delta(const char *path) 970{ 971 static struct attr_check *check; 972 973 if (!check) 974 check = attr_check_initl("delta", NULL); 975 git_check_attr(the_repository->index, path, check); 976 if (ATTR_FALSE(check->items[0].value)) 977 return 1; 978 return 0; 979} 980 981/* 982 * When adding an object, check whether we have already added it 983 * to our packing list. If so, we can skip. However, if we are 984 * being asked to excludei t, but the previous mention was to include 985 * it, make sure to adjust its flags and tweak our numbers accordingly. 986 * 987 * As an optimization, we pass out the index position where we would have 988 * found the item, since that saves us from having to look it up again a 989 * few lines later when we want to add the new entry. 990 */ 991static int have_duplicate_entry(const struct object_id *oid, 992 int exclude, 993 uint32_t *index_pos) 994{ 995 struct object_entry *entry; 996 997 entry = packlist_find(&to_pack, oid->hash, index_pos); 998 if (!entry) 999 return 0;10001001 if (exclude) {1002 if (!entry->preferred_base)1003 nr_result--;1004 entry->preferred_base = 1;1005 }10061007 return 1;1008}10091010static int want_found_object(int exclude, struct packed_git *p)1011{1012 if (exclude)1013 return 1;1014 if (incremental)1015 return 0;10161017 /*1018 * When asked to do --local (do not include an object that appears in a1019 * pack we borrow from elsewhere) or --honor-pack-keep (do not include1020 * an object that appears in a pack marked with .keep), finding a pack1021 * that matches the criteria is sufficient for us to decide to omit it.1022 * However, even if this pack does not satisfy the criteria, we need to1023 * make sure no copy of this object appears in _any_ pack that makes us1024 * to omit the object, so we need to check all the packs.1025 *1026 * We can however first check whether these options can possible matter;1027 * if they do not matter we know we want the object in generated pack.1028 * Otherwise, we signal "-1" at the end to tell the caller that we do1029 * not know either way, and it needs to check more packs.1030 */1031 if (!ignore_packed_keep_on_disk &&1032 !ignore_packed_keep_in_core &&1033 (!local || !have_non_local_packs))1034 return 1;10351036 if (local && !p->pack_local)1037 return 0;1038 if (p->pack_local &&1039 ((ignore_packed_keep_on_disk && p->pack_keep) ||1040 (ignore_packed_keep_in_core && p->pack_keep_in_core)))1041 return 0;10421043 /* we don't know yet; keep looking for more packs */1044 return -1;1045}10461047/*1048 * Check whether we want the object in the pack (e.g., we do not want1049 * objects found in non-local stores if the "--local" option was used).1050 *1051 * If the caller already knows an existing pack it wants to take the object1052 * from, that is passed in *found_pack and *found_offset; otherwise this1053 * function finds if there is any pack that has the object and returns the pack1054 * and its offset in these variables.1055 */1056static int want_object_in_pack(const struct object_id *oid,1057 int exclude,1058 struct packed_git **found_pack,1059 off_t *found_offset)1060{1061 int want;1062 struct list_head *pos;1063 struct multi_pack_index *m;10641065 if (!exclude && local && has_loose_object_nonlocal(oid))1066 return 0;10671068 /*1069 * If we already know the pack object lives in, start checks from that1070 * pack - in the usual case when neither --local was given nor .keep files1071 * are present we will determine the answer right now.1072 */1073 if (*found_pack) {1074 want = want_found_object(exclude, *found_pack);1075 if (want != -1)1076 return want;1077 }10781079 for (m = get_multi_pack_index(the_repository); m; m = m->next) {1080 struct pack_entry e;1081 if (fill_midx_entry(oid, &e, m)) {1082 struct packed_git *p = e.p;1083 off_t offset;10841085 if (p == *found_pack)1086 offset = *found_offset;1087 else1088 offset = find_pack_entry_one(oid->hash, p);10891090 if (offset) {1091 if (!*found_pack) {1092 if (!is_pack_valid(p))1093 continue;1094 *found_offset = offset;1095 *found_pack = p;1096 }1097 want = want_found_object(exclude, p);1098 if (want != -1)1099 return want;1100 }1101 }1102 }11031104 list_for_each(pos, get_packed_git_mru(the_repository)) {1105 struct packed_git *p = list_entry(pos, struct packed_git, mru);1106 off_t offset;11071108 if (p == *found_pack)1109 offset = *found_offset;1110 else1111 offset = find_pack_entry_one(oid->hash, p);11121113 if (offset) {1114 if (!*found_pack) {1115 if (!is_pack_valid(p))1116 continue;1117 *found_offset = offset;1118 *found_pack = p;1119 }1120 want = want_found_object(exclude, p);1121 if (!exclude && want > 0)1122 list_move(&p->mru,1123 get_packed_git_mru(the_repository));1124 if (want != -1)1125 return want;1126 }1127 }11281129 return 1;1130}11311132static void create_object_entry(const struct object_id *oid,1133 enum object_type type,1134 uint32_t hash,1135 int exclude,1136 int no_try_delta,1137 uint32_t index_pos,1138 struct packed_git *found_pack,1139 off_t found_offset)1140{1141 struct object_entry *entry;11421143 entry = packlist_alloc(&to_pack, oid->hash, index_pos);1144 entry->hash = hash;1145 oe_set_type(entry, type);1146 if (exclude)1147 entry->preferred_base = 1;1148 else1149 nr_result++;1150 if (found_pack) {1151 oe_set_in_pack(&to_pack, entry, found_pack);1152 entry->in_pack_offset = found_offset;1153 }11541155 entry->no_try_delta = no_try_delta;1156}11571158static const char no_closure_warning[] = N_(1159"disabling bitmap writing, as some objects are not being packed"1160);11611162static int add_object_entry(const struct object_id *oid, enum object_type type,1163 const char *name, int exclude)1164{1165 struct packed_git *found_pack = NULL;1166 off_t found_offset = 0;1167 uint32_t index_pos;11681169 display_progress(progress_state, ++nr_seen);11701171 if (have_duplicate_entry(oid, exclude, &index_pos))1172 return 0;11731174 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1175 /* The pack is missing an object, so it will not have closure */1176 if (write_bitmap_index) {1177 warning(_(no_closure_warning));1178 write_bitmap_index = 0;1179 }1180 return 0;1181 }11821183 create_object_entry(oid, type, pack_name_hash(name),1184 exclude, name && no_try_delta(name),1185 index_pos, found_pack, found_offset);1186 return 1;1187}11881189static int add_object_entry_from_bitmap(const struct object_id *oid,1190 enum object_type type,1191 int flags, uint32_t name_hash,1192 struct packed_git *pack, off_t offset)1193{1194 uint32_t index_pos;11951196 display_progress(progress_state, ++nr_seen);11971198 if (have_duplicate_entry(oid, 0, &index_pos))1199 return 0;12001201 if (!want_object_in_pack(oid, 0, &pack, &offset))1202 return 0;12031204 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);1205 return 1;1206}12071208struct pbase_tree_cache {1209 struct object_id oid;1210 int ref;1211 int temporary;1212 void *tree_data;1213 unsigned long tree_size;1214};12151216static struct pbase_tree_cache *(pbase_tree_cache[256]);1217static int pbase_tree_cache_ix(const struct object_id *oid)1218{1219 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);1220}1221static int pbase_tree_cache_ix_incr(int ix)1222{1223 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);1224}12251226static struct pbase_tree {1227 struct pbase_tree *next;1228 /* This is a phony "cache" entry; we are not1229 * going to evict it or find it through _get()1230 * mechanism -- this is for the toplevel node that1231 * would almost always change with any commit.1232 */1233 struct pbase_tree_cache pcache;1234} *pbase_tree;12351236static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1237{1238 struct pbase_tree_cache *ent, *nent;1239 void *data;1240 unsigned long size;1241 enum object_type type;1242 int neigh;1243 int my_ix = pbase_tree_cache_ix(oid);1244 int available_ix = -1;12451246 /* pbase-tree-cache acts as a limited hashtable.1247 * your object will be found at your index or within a few1248 * slots after that slot if it is cached.1249 */1250 for (neigh = 0; neigh < 8; neigh++) {1251 ent = pbase_tree_cache[my_ix];1252 if (ent && oideq(&ent->oid, oid)) {1253 ent->ref++;1254 return ent;1255 }1256 else if (((available_ix < 0) && (!ent || !ent->ref)) ||1257 ((0 <= available_ix) &&1258 (!ent && pbase_tree_cache[available_ix])))1259 available_ix = my_ix;1260 if (!ent)1261 break;1262 my_ix = pbase_tree_cache_ix_incr(my_ix);1263 }12641265 /* Did not find one. Either we got a bogus request or1266 * we need to read and perhaps cache.1267 */1268 data = read_object_file(oid, &type, &size);1269 if (!data)1270 return NULL;1271 if (type != OBJ_TREE) {1272 free(data);1273 return NULL;1274 }12751276 /* We need to either cache or return a throwaway copy */12771278 if (available_ix < 0)1279 ent = NULL;1280 else {1281 ent = pbase_tree_cache[available_ix];1282 my_ix = available_ix;1283 }12841285 if (!ent) {1286 nent = xmalloc(sizeof(*nent));1287 nent->temporary = (available_ix < 0);1288 }1289 else {1290 /* evict and reuse */1291 free(ent->tree_data);1292 nent = ent;1293 }1294 oidcpy(&nent->oid, oid);1295 nent->tree_data = data;1296 nent->tree_size = size;1297 nent->ref = 1;1298 if (!nent->temporary)1299 pbase_tree_cache[my_ix] = nent;1300 return nent;1301}13021303static void pbase_tree_put(struct pbase_tree_cache *cache)1304{1305 if (!cache->temporary) {1306 cache->ref--;1307 return;1308 }1309 free(cache->tree_data);1310 free(cache);1311}13121313static int name_cmp_len(const char *name)1314{1315 int i;1316 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)1317 ;1318 return i;1319}13201321static void add_pbase_object(struct tree_desc *tree,1322 const char *name,1323 int cmplen,1324 const char *fullname)1325{1326 struct name_entry entry;1327 int cmp;13281329 while (tree_entry(tree,&entry)) {1330 if (S_ISGITLINK(entry.mode))1331 continue;1332 cmp = tree_entry_len(&entry) != cmplen ? 1 :1333 memcmp(name, entry.path, cmplen);1334 if (cmp > 0)1335 continue;1336 if (cmp < 0)1337 return;1338 if (name[cmplen] != '/') {1339 add_object_entry(&entry.oid,1340 object_type(entry.mode),1341 fullname, 1);1342 return;1343 }1344 if (S_ISDIR(entry.mode)) {1345 struct tree_desc sub;1346 struct pbase_tree_cache *tree;1347 const char *down = name+cmplen+1;1348 int downlen = name_cmp_len(down);13491350 tree = pbase_tree_get(&entry.oid);1351 if (!tree)1352 return;1353 init_tree_desc(&sub, tree->tree_data, tree->tree_size);13541355 add_pbase_object(&sub, down, downlen, fullname);1356 pbase_tree_put(tree);1357 }1358 }1359}13601361static unsigned *done_pbase_paths;1362static int done_pbase_paths_num;1363static int done_pbase_paths_alloc;1364static int done_pbase_path_pos(unsigned hash)1365{1366 int lo = 0;1367 int hi = done_pbase_paths_num;1368 while (lo < hi) {1369 int mi = lo + (hi - lo) / 2;1370 if (done_pbase_paths[mi] == hash)1371 return mi;1372 if (done_pbase_paths[mi] < hash)1373 hi = mi;1374 else1375 lo = mi + 1;1376 }1377 return -lo-1;1378}13791380static int check_pbase_path(unsigned hash)1381{1382 int pos = done_pbase_path_pos(hash);1383 if (0 <= pos)1384 return 1;1385 pos = -pos - 1;1386 ALLOC_GROW(done_pbase_paths,1387 done_pbase_paths_num + 1,1388 done_pbase_paths_alloc);1389 done_pbase_paths_num++;1390 if (pos < done_pbase_paths_num)1391 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,1392 done_pbase_paths_num - pos - 1);1393 done_pbase_paths[pos] = hash;1394 return 0;1395}13961397static void add_preferred_base_object(const char *name)1398{1399 struct pbase_tree *it;1400 int cmplen;1401 unsigned hash = pack_name_hash(name);14021403 if (!num_preferred_base || check_pbase_path(hash))1404 return;14051406 cmplen = name_cmp_len(name);1407 for (it = pbase_tree; it; it = it->next) {1408 if (cmplen == 0) {1409 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);1410 }1411 else {1412 struct tree_desc tree;1413 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1414 add_pbase_object(&tree, name, cmplen, name);1415 }1416 }1417}14181419static void add_preferred_base(struct object_id *oid)1420{1421 struct pbase_tree *it;1422 void *data;1423 unsigned long size;1424 struct object_id tree_oid;14251426 if (window <= num_preferred_base++)1427 return;14281429 data = read_object_with_reference(oid, tree_type, &size, &tree_oid);1430 if (!data)1431 return;14321433 for (it = pbase_tree; it; it = it->next) {1434 if (oideq(&it->pcache.oid, &tree_oid)) {1435 free(data);1436 return;1437 }1438 }14391440 it = xcalloc(1, sizeof(*it));1441 it->next = pbase_tree;1442 pbase_tree = it;14431444 oidcpy(&it->pcache.oid, &tree_oid);1445 it->pcache.tree_data = data;1446 it->pcache.tree_size = size;1447}14481449static void cleanup_preferred_base(void)1450{1451 struct pbase_tree *it;1452 unsigned i;14531454 it = pbase_tree;1455 pbase_tree = NULL;1456 while (it) {1457 struct pbase_tree *tmp = it;1458 it = tmp->next;1459 free(tmp->pcache.tree_data);1460 free(tmp);1461 }14621463 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {1464 if (!pbase_tree_cache[i])1465 continue;1466 free(pbase_tree_cache[i]->tree_data);1467 FREE_AND_NULL(pbase_tree_cache[i]);1468 }14691470 FREE_AND_NULL(done_pbase_paths);1471 done_pbase_paths_num = done_pbase_paths_alloc = 0;1472}14731474/*1475 * Return 1 iff the object specified by "delta" can be sent1476 * literally as a delta against the base in "base_sha1". If1477 * so, then *base_out will point to the entry in our packing1478 * list, or NULL if we must use the external-base list.1479 *1480 * Depth value does not matter - find_deltas() will1481 * never consider reused delta as the base object to1482 * deltify other objects against, in order to avoid1483 * circular deltas.1484 */1485static int can_reuse_delta(const unsigned char *base_sha1,1486 struct object_entry *delta,1487 struct object_entry **base_out)1488{1489 struct object_entry *base;14901491 if (!base_sha1)1492 return 0;14931494 /*1495 * First see if we're already sending the base (or it's explicitly in1496 * our "excluded" list).1497 */1498 base = packlist_find(&to_pack, base_sha1, NULL);1499 if (base) {1500 if (!in_same_island(&delta->idx.oid, &base->idx.oid))1501 return 0;1502 *base_out = base;1503 return 1;1504 }15051506 /*1507 * Otherwise, reachability bitmaps may tell us if the receiver has it,1508 * even if it was buried too deep in history to make it into the1509 * packing list.1510 */1511 if (thin && bitmap_has_sha1_in_uninteresting(bitmap_git, base_sha1)) {1512 if (use_delta_islands) {1513 struct object_id base_oid;1514 hashcpy(base_oid.hash, base_sha1);1515 if (!in_same_island(&delta->idx.oid, &base_oid))1516 return 0;1517 }1518 *base_out = NULL;1519 return 1;1520 }15211522 return 0;1523}15241525static void check_object(struct object_entry *entry)1526{1527 unsigned long canonical_size;15281529 if (IN_PACK(entry)) {1530 struct packed_git *p = IN_PACK(entry);1531 struct pack_window *w_curs = NULL;1532 const unsigned char *base_ref = NULL;1533 struct object_entry *base_entry;1534 unsigned long used, used_0;1535 unsigned long avail;1536 off_t ofs;1537 unsigned char *buf, c;1538 enum object_type type;1539 unsigned long in_pack_size;15401541 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);15421543 /*1544 * We want in_pack_type even if we do not reuse delta1545 * since non-delta representations could still be reused.1546 */1547 used = unpack_object_header_buffer(buf, avail,1548 &type,1549 &in_pack_size);1550 if (used == 0)1551 goto give_up;15521553 if (type < 0)1554 BUG("invalid type %d", type);1555 entry->in_pack_type = type;15561557 /*1558 * Determine if this is a delta and if so whether we can1559 * reuse it or not. Otherwise let's find out as cheaply as1560 * possible what the actual type and size for this object is.1561 */1562 switch (entry->in_pack_type) {1563 default:1564 /* Not a delta hence we've already got all we need. */1565 oe_set_type(entry, entry->in_pack_type);1566 SET_SIZE(entry, in_pack_size);1567 entry->in_pack_header_size = used;1568 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)1569 goto give_up;1570 unuse_pack(&w_curs);1571 return;1572 case OBJ_REF_DELTA:1573 if (reuse_delta && !entry->preferred_base)1574 base_ref = use_pack(p, &w_curs,1575 entry->in_pack_offset + used, NULL);1576 entry->in_pack_header_size = used + the_hash_algo->rawsz;1577 break;1578 case OBJ_OFS_DELTA:1579 buf = use_pack(p, &w_curs,1580 entry->in_pack_offset + used, NULL);1581 used_0 = 0;1582 c = buf[used_0++];1583 ofs = c & 127;1584 while (c & 128) {1585 ofs += 1;1586 if (!ofs || MSB(ofs, 7)) {1587 error(_("delta base offset overflow in pack for %s"),1588 oid_to_hex(&entry->idx.oid));1589 goto give_up;1590 }1591 c = buf[used_0++];1592 ofs = (ofs << 7) + (c & 127);1593 }1594 ofs = entry->in_pack_offset - ofs;1595 if (ofs <= 0 || ofs >= entry->in_pack_offset) {1596 error(_("delta base offset out of bound for %s"),1597 oid_to_hex(&entry->idx.oid));1598 goto give_up;1599 }1600 if (reuse_delta && !entry->preferred_base) {1601 struct revindex_entry *revidx;1602 revidx = find_pack_revindex(p, ofs);1603 if (!revidx)1604 goto give_up;1605 base_ref = nth_packed_object_sha1(p, revidx->nr);1606 }1607 entry->in_pack_header_size = used + used_0;1608 break;1609 }16101611 if (can_reuse_delta(base_ref, entry, &base_entry)) {1612 oe_set_type(entry, entry->in_pack_type);1613 SET_SIZE(entry, in_pack_size); /* delta size */1614 SET_DELTA_SIZE(entry, in_pack_size);16151616 if (base_entry) {1617 SET_DELTA(entry, base_entry);1618 entry->delta_sibling_idx = base_entry->delta_child_idx;1619 SET_DELTA_CHILD(base_entry, entry);1620 } else {1621 SET_DELTA_EXT(entry, base_ref);1622 }16231624 unuse_pack(&w_curs);1625 return;1626 }16271628 if (oe_type(entry)) {1629 off_t delta_pos;16301631 /*1632 * This must be a delta and we already know what the1633 * final object type is. Let's extract the actual1634 * object size from the delta header.1635 */1636 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;1637 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);1638 if (canonical_size == 0)1639 goto give_up;1640 SET_SIZE(entry, canonical_size);1641 unuse_pack(&w_curs);1642 return;1643 }16441645 /*1646 * No choice but to fall back to the recursive delta walk1647 * with oid_object_info() to find about the object type1648 * at this point...1649 */1650 give_up:1651 unuse_pack(&w_curs);1652 }16531654 oe_set_type(entry,1655 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1656 if (entry->type_valid) {1657 SET_SIZE(entry, canonical_size);1658 } else {1659 /*1660 * Bad object type is checked in prepare_pack(). This is1661 * to permit a missing preferred base object to be ignored1662 * as a preferred base. Doing so can result in a larger1663 * pack file, but the transfer will still take place.1664 */1665 }1666}16671668static int pack_offset_sort(const void *_a, const void *_b)1669{1670 const struct object_entry *a = *(struct object_entry **)_a;1671 const struct object_entry *b = *(struct object_entry **)_b;1672 const struct packed_git *a_in_pack = IN_PACK(a);1673 const struct packed_git *b_in_pack = IN_PACK(b);16741675 /* avoid filesystem trashing with loose objects */1676 if (!a_in_pack && !b_in_pack)1677 return oidcmp(&a->idx.oid, &b->idx.oid);16781679 if (a_in_pack < b_in_pack)1680 return -1;1681 if (a_in_pack > b_in_pack)1682 return 1;1683 return a->in_pack_offset < b->in_pack_offset ? -1 :1684 (a->in_pack_offset > b->in_pack_offset);1685}16861687/*1688 * Drop an on-disk delta we were planning to reuse. Naively, this would1689 * just involve blanking out the "delta" field, but we have to deal1690 * with some extra book-keeping:1691 *1692 * 1. Removing ourselves from the delta_sibling linked list.1693 *1694 * 2. Updating our size/type to the non-delta representation. These were1695 * either not recorded initially (size) or overwritten with the delta type1696 * (type) when check_object() decided to reuse the delta.1697 *1698 * 3. Resetting our delta depth, as we are now a base object.1699 */1700static void drop_reused_delta(struct object_entry *entry)1701{1702 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;1703 struct object_info oi = OBJECT_INFO_INIT;1704 enum object_type type;1705 unsigned long size;17061707 while (*idx) {1708 struct object_entry *oe = &to_pack.objects[*idx - 1];17091710 if (oe == entry)1711 *idx = oe->delta_sibling_idx;1712 else1713 idx = &oe->delta_sibling_idx;1714 }1715 SET_DELTA(entry, NULL);1716 entry->depth = 0;17171718 oi.sizep = &size;1719 oi.typep = &type;1720 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {1721 /*1722 * We failed to get the info from this pack for some reason;1723 * fall back to oid_object_info, which may find another copy.1724 * And if that fails, the error will be recorded in oe_type(entry)1725 * and dealt with in prepare_pack().1726 */1727 oe_set_type(entry,1728 oid_object_info(the_repository, &entry->idx.oid, &size));1729 } else {1730 oe_set_type(entry, type);1731 }1732 SET_SIZE(entry, size);1733}17341735/*1736 * Follow the chain of deltas from this entry onward, throwing away any links1737 * that cause us to hit a cycle (as determined by the DFS state flags in1738 * the entries).1739 *1740 * We also detect too-long reused chains that would violate our --depth1741 * limit.1742 */1743static void break_delta_chains(struct object_entry *entry)1744{1745 /*1746 * The actual depth of each object we will write is stored as an int,1747 * as it cannot exceed our int "depth" limit. But before we break1748 * changes based no that limit, we may potentially go as deep as the1749 * number of objects, which is elsewhere bounded to a uint32_t.1750 */1751 uint32_t total_depth;1752 struct object_entry *cur, *next;17531754 for (cur = entry, total_depth = 0;1755 cur;1756 cur = DELTA(cur), total_depth++) {1757 if (cur->dfs_state == DFS_DONE) {1758 /*1759 * We've already seen this object and know it isn't1760 * part of a cycle. We do need to append its depth1761 * to our count.1762 */1763 total_depth += cur->depth;1764 break;1765 }17661767 /*1768 * We break cycles before looping, so an ACTIVE state (or any1769 * other cruft which made its way into the state variable)1770 * is a bug.1771 */1772 if (cur->dfs_state != DFS_NONE)1773 BUG("confusing delta dfs state in first pass: %d",1774 cur->dfs_state);17751776 /*1777 * Now we know this is the first time we've seen the object. If1778 * it's not a delta, we're done traversing, but we'll mark it1779 * done to save time on future traversals.1780 */1781 if (!DELTA(cur)) {1782 cur->dfs_state = DFS_DONE;1783 break;1784 }17851786 /*1787 * Mark ourselves as active and see if the next step causes1788 * us to cycle to another active object. It's important to do1789 * this _before_ we loop, because it impacts where we make the1790 * cut, and thus how our total_depth counter works.1791 * E.g., We may see a partial loop like:1792 *1793 * A -> B -> C -> D -> B1794 *1795 * Cutting B->C breaks the cycle. But now the depth of A is1796 * only 1, and our total_depth counter is at 3. The size of the1797 * error is always one less than the size of the cycle we1798 * broke. Commits C and D were "lost" from A's chain.1799 *1800 * If we instead cut D->B, then the depth of A is correct at 3.1801 * We keep all commits in the chain that we examined.1802 */1803 cur->dfs_state = DFS_ACTIVE;1804 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {1805 drop_reused_delta(cur);1806 cur->dfs_state = DFS_DONE;1807 break;1808 }1809 }18101811 /*1812 * And now that we've gone all the way to the bottom of the chain, we1813 * need to clear the active flags and set the depth fields as1814 * appropriate. Unlike the loop above, which can quit when it drops a1815 * delta, we need to keep going to look for more depth cuts. So we need1816 * an extra "next" pointer to keep going after we reset cur->delta.1817 */1818 for (cur = entry; cur; cur = next) {1819 next = DELTA(cur);18201821 /*1822 * We should have a chain of zero or more ACTIVE states down to1823 * a final DONE. We can quit after the DONE, because either it1824 * has no bases, or we've already handled them in a previous1825 * call.1826 */1827 if (cur->dfs_state == DFS_DONE)1828 break;1829 else if (cur->dfs_state != DFS_ACTIVE)1830 BUG("confusing delta dfs state in second pass: %d",1831 cur->dfs_state);18321833 /*1834 * If the total_depth is more than depth, then we need to snip1835 * the chain into two or more smaller chains that don't exceed1836 * the maximum depth. Most of the resulting chains will contain1837 * (depth + 1) entries (i.e., depth deltas plus one base), and1838 * the last chain (i.e., the one containing entry) will contain1839 * whatever entries are left over, namely1840 * (total_depth % (depth + 1)) of them.1841 *1842 * Since we are iterating towards decreasing depth, we need to1843 * decrement total_depth as we go, and we need to write to the1844 * entry what its final depth will be after all of the1845 * snipping. Since we're snipping into chains of length (depth1846 * + 1) entries, the final depth of an entry will be its1847 * original depth modulo (depth + 1). Any time we encounter an1848 * entry whose final depth is supposed to be zero, we snip it1849 * from its delta base, thereby making it so.1850 */1851 cur->depth = (total_depth--) % (depth + 1);1852 if (!cur->depth)1853 drop_reused_delta(cur);18541855 cur->dfs_state = DFS_DONE;1856 }1857}18581859static void get_object_details(void)1860{1861 uint32_t i;1862 struct object_entry **sorted_by_offset;18631864 if (progress)1865 progress_state = start_progress(_("Counting objects"),1866 to_pack.nr_objects);18671868 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));1869 for (i = 0; i < to_pack.nr_objects; i++)1870 sorted_by_offset[i] = to_pack.objects + i;1871 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);18721873 for (i = 0; i < to_pack.nr_objects; i++) {1874 struct object_entry *entry = sorted_by_offset[i];1875 check_object(entry);1876 if (entry->type_valid &&1877 oe_size_greater_than(&to_pack, entry, big_file_threshold))1878 entry->no_try_delta = 1;1879 display_progress(progress_state, i + 1);1880 }1881 stop_progress(&progress_state);18821883 /*1884 * This must happen in a second pass, since we rely on the delta1885 * information for the whole list being completed.1886 */1887 for (i = 0; i < to_pack.nr_objects; i++)1888 break_delta_chains(&to_pack.objects[i]);18891890 free(sorted_by_offset);1891}18921893/*1894 * We search for deltas in a list sorted by type, by filename hash, and then1895 * by size, so that we see progressively smaller and smaller files.1896 * That's because we prefer deltas to be from the bigger file1897 * to the smaller -- deletes are potentially cheaper, but perhaps1898 * more importantly, the bigger file is likely the more recent1899 * one. The deepest deltas are therefore the oldest objects which are1900 * less susceptible to be accessed often.1901 */1902static int type_size_sort(const void *_a, const void *_b)1903{1904 const struct object_entry *a = *(struct object_entry **)_a;1905 const struct object_entry *b = *(struct object_entry **)_b;1906 const enum object_type a_type = oe_type(a);1907 const enum object_type b_type = oe_type(b);1908 const unsigned long a_size = SIZE(a);1909 const unsigned long b_size = SIZE(b);19101911 if (a_type > b_type)1912 return -1;1913 if (a_type < b_type)1914 return 1;1915 if (a->hash > b->hash)1916 return -1;1917 if (a->hash < b->hash)1918 return 1;1919 if (a->preferred_base > b->preferred_base)1920 return -1;1921 if (a->preferred_base < b->preferred_base)1922 return 1;1923 if (use_delta_islands) {1924 const int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);1925 if (island_cmp)1926 return island_cmp;1927 }1928 if (a_size > b_size)1929 return -1;1930 if (a_size < b_size)1931 return 1;1932 return a < b ? -1 : (a > b); /* newest first */1933}19341935struct unpacked {1936 struct object_entry *entry;1937 void *data;1938 struct delta_index *index;1939 unsigned depth;1940};19411942static int delta_cacheable(unsigned long src_size, unsigned long trg_size,1943 unsigned long delta_size)1944{1945 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1946 return 0;19471948 if (delta_size < cache_max_small_delta_size)1949 return 1;19501951 /* cache delta, if objects are large enough compared to delta size */1952 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))1953 return 1;19541955 return 0;1956}19571958/* Protect delta_cache_size */1959static pthread_mutex_t cache_mutex;1960#define cache_lock() pthread_mutex_lock(&cache_mutex)1961#define cache_unlock() pthread_mutex_unlock(&cache_mutex)19621963/*1964 * Protect object list partitioning (e.g. struct thread_param) and1965 * progress_state1966 */1967static pthread_mutex_t progress_mutex;1968#define progress_lock() pthread_mutex_lock(&progress_mutex)1969#define progress_unlock() pthread_mutex_unlock(&progress_mutex)19701971/*1972 * Access to struct object_entry is unprotected since each thread owns1973 * a portion of the main object list. Just don't access object entries1974 * ahead in the list because they can be stolen and would need1975 * progress_mutex for protection.1976 */19771978/*1979 * Return the size of the object without doing any delta1980 * reconstruction (so non-deltas are true object sizes, but deltas1981 * return the size of the delta data).1982 */1983unsigned long oe_get_size_slow(struct packing_data *pack,1984 const struct object_entry *e)1985{1986 struct packed_git *p;1987 struct pack_window *w_curs;1988 unsigned char *buf;1989 enum object_type type;1990 unsigned long used, avail, size;19911992 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {1993 packing_data_lock(&to_pack);1994 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)1995 die(_("unable to get size of %s"),1996 oid_to_hex(&e->idx.oid));1997 packing_data_unlock(&to_pack);1998 return size;1999 }20002001 p = oe_in_pack(pack, e);2002 if (!p)2003 BUG("when e->type is a delta, it must belong to a pack");20042005 packing_data_lock(&to_pack);2006 w_curs = NULL;2007 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);2008 used = unpack_object_header_buffer(buf, avail, &type, &size);2009 if (used == 0)2010 die(_("unable to parse object header of %s"),2011 oid_to_hex(&e->idx.oid));20122013 unuse_pack(&w_curs);2014 packing_data_unlock(&to_pack);2015 return size;2016}20172018static int try_delta(struct unpacked *trg, struct unpacked *src,2019 unsigned max_depth, unsigned long *mem_usage)2020{2021 struct object_entry *trg_entry = trg->entry;2022 struct object_entry *src_entry = src->entry;2023 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;2024 unsigned ref_depth;2025 enum object_type type;2026 void *delta_buf;20272028 /* Don't bother doing diffs between different types */2029 if (oe_type(trg_entry) != oe_type(src_entry))2030 return -1;20312032 /*2033 * We do not bother to try a delta that we discarded on an2034 * earlier try, but only when reusing delta data. Note that2035 * src_entry that is marked as the preferred_base should always2036 * be considered, as even if we produce a suboptimal delta against2037 * it, we will still save the transfer cost, as we already know2038 * the other side has it and we won't send src_entry at all.2039 */2040 if (reuse_delta && IN_PACK(trg_entry) &&2041 IN_PACK(trg_entry) == IN_PACK(src_entry) &&2042 !src_entry->preferred_base &&2043 trg_entry->in_pack_type != OBJ_REF_DELTA &&2044 trg_entry->in_pack_type != OBJ_OFS_DELTA)2045 return 0;20462047 /* Let's not bust the allowed depth. */2048 if (src->depth >= max_depth)2049 return 0;20502051 /* Now some size filtering heuristics. */2052 trg_size = SIZE(trg_entry);2053 if (!DELTA(trg_entry)) {2054 max_size = trg_size/2 - the_hash_algo->rawsz;2055 ref_depth = 1;2056 } else {2057 max_size = DELTA_SIZE(trg_entry);2058 ref_depth = trg->depth;2059 }2060 max_size = (uint64_t)max_size * (max_depth - src->depth) /2061 (max_depth - ref_depth + 1);2062 if (max_size == 0)2063 return 0;2064 src_size = SIZE(src_entry);2065 sizediff = src_size < trg_size ? trg_size - src_size : 0;2066 if (sizediff >= max_size)2067 return 0;2068 if (trg_size < src_size / 32)2069 return 0;20702071 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))2072 return 0;20732074 /* Load data if not already done */2075 if (!trg->data) {2076 packing_data_lock(&to_pack);2077 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);2078 packing_data_unlock(&to_pack);2079 if (!trg->data)2080 die(_("object %s cannot be read"),2081 oid_to_hex(&trg_entry->idx.oid));2082 if (sz != trg_size)2083 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2084 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,2085 (uintmax_t)trg_size);2086 *mem_usage += sz;2087 }2088 if (!src->data) {2089 packing_data_lock(&to_pack);2090 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);2091 packing_data_unlock(&to_pack);2092 if (!src->data) {2093 if (src_entry->preferred_base) {2094 static int warned = 0;2095 if (!warned++)2096 warning(_("object %s cannot be read"),2097 oid_to_hex(&src_entry->idx.oid));2098 /*2099 * Those objects are not included in the2100 * resulting pack. Be resilient and ignore2101 * them if they can't be read, in case the2102 * pack could be created nevertheless.2103 */2104 return 0;2105 }2106 die(_("object %s cannot be read"),2107 oid_to_hex(&src_entry->idx.oid));2108 }2109 if (sz != src_size)2110 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2111 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,2112 (uintmax_t)src_size);2113 *mem_usage += sz;2114 }2115 if (!src->index) {2116 src->index = create_delta_index(src->data, src_size);2117 if (!src->index) {2118 static int warned = 0;2119 if (!warned++)2120 warning(_("suboptimal pack - out of memory"));2121 return 0;2122 }2123 *mem_usage += sizeof_delta_index(src->index);2124 }21252126 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2127 if (!delta_buf)2128 return 0;21292130 if (DELTA(trg_entry)) {2131 /* Prefer only shallower same-sized deltas. */2132 if (delta_size == DELTA_SIZE(trg_entry) &&2133 src->depth + 1 >= trg->depth) {2134 free(delta_buf);2135 return 0;2136 }2137 }21382139 /*2140 * Handle memory allocation outside of the cache2141 * accounting lock. Compiler will optimize the strangeness2142 * away when NO_PTHREADS is defined.2143 */2144 free(trg_entry->delta_data);2145 cache_lock();2146 if (trg_entry->delta_data) {2147 delta_cache_size -= DELTA_SIZE(trg_entry);2148 trg_entry->delta_data = NULL;2149 }2150 if (delta_cacheable(src_size, trg_size, delta_size)) {2151 delta_cache_size += delta_size;2152 cache_unlock();2153 trg_entry->delta_data = xrealloc(delta_buf, delta_size);2154 } else {2155 cache_unlock();2156 free(delta_buf);2157 }21582159 SET_DELTA(trg_entry, src_entry);2160 SET_DELTA_SIZE(trg_entry, delta_size);2161 trg->depth = src->depth + 1;21622163 return 1;2164}21652166static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)2167{2168 struct object_entry *child = DELTA_CHILD(me);2169 unsigned int m = n;2170 while (child) {2171 const unsigned int c = check_delta_limit(child, n + 1);2172 if (m < c)2173 m = c;2174 child = DELTA_SIBLING(child);2175 }2176 return m;2177}21782179static unsigned long free_unpacked(struct unpacked *n)2180{2181 unsigned long freed_mem = sizeof_delta_index(n->index);2182 free_delta_index(n->index);2183 n->index = NULL;2184 if (n->data) {2185 freed_mem += SIZE(n->entry);2186 FREE_AND_NULL(n->data);2187 }2188 n->entry = NULL;2189 n->depth = 0;2190 return freed_mem;2191}21922193static void find_deltas(struct object_entry **list, unsigned *list_size,2194 int window, int depth, unsigned *processed)2195{2196 uint32_t i, idx = 0, count = 0;2197 struct unpacked *array;2198 unsigned long mem_usage = 0;21992200 array = xcalloc(window, sizeof(struct unpacked));22012202 for (;;) {2203 struct object_entry *entry;2204 struct unpacked *n = array + idx;2205 int j, max_depth, best_base = -1;22062207 progress_lock();2208 if (!*list_size) {2209 progress_unlock();2210 break;2211 }2212 entry = *list++;2213 (*list_size)--;2214 if (!entry->preferred_base) {2215 (*processed)++;2216 display_progress(progress_state, *processed);2217 }2218 progress_unlock();22192220 mem_usage -= free_unpacked(n);2221 n->entry = entry;22222223 while (window_memory_limit &&2224 mem_usage > window_memory_limit &&2225 count > 1) {2226 const uint32_t tail = (idx + window - count) % window;2227 mem_usage -= free_unpacked(array + tail);2228 count--;2229 }22302231 /* We do not compute delta to *create* objects we are not2232 * going to pack.2233 */2234 if (entry->preferred_base)2235 goto next;22362237 /*2238 * If the current object is at pack edge, take the depth the2239 * objects that depend on the current object into account2240 * otherwise they would become too deep.2241 */2242 max_depth = depth;2243 if (DELTA_CHILD(entry)) {2244 max_depth -= check_delta_limit(entry, 0);2245 if (max_depth <= 0)2246 goto next;2247 }22482249 j = window;2250 while (--j > 0) {2251 int ret;2252 uint32_t other_idx = idx + j;2253 struct unpacked *m;2254 if (other_idx >= window)2255 other_idx -= window;2256 m = array + other_idx;2257 if (!m->entry)2258 break;2259 ret = try_delta(n, m, max_depth, &mem_usage);2260 if (ret < 0)2261 break;2262 else if (ret > 0)2263 best_base = other_idx;2264 }22652266 /*2267 * If we decided to cache the delta data, then it is best2268 * to compress it right away. First because we have to do2269 * it anyway, and doing it here while we're threaded will2270 * save a lot of time in the non threaded write phase,2271 * as well as allow for caching more deltas within2272 * the same cache size limit.2273 * ...2274 * But only if not writing to stdout, since in that case2275 * the network is most likely throttling writes anyway,2276 * and therefore it is best to go to the write phase ASAP2277 * instead, as we can afford spending more time compressing2278 * between writes at that moment.2279 */2280 if (entry->delta_data && !pack_to_stdout) {2281 unsigned long size;22822283 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));2284 if (size < (1U << OE_Z_DELTA_BITS)) {2285 entry->z_delta_size = size;2286 cache_lock();2287 delta_cache_size -= DELTA_SIZE(entry);2288 delta_cache_size += entry->z_delta_size;2289 cache_unlock();2290 } else {2291 FREE_AND_NULL(entry->delta_data);2292 entry->z_delta_size = 0;2293 }2294 }22952296 /* if we made n a delta, and if n is already at max2297 * depth, leaving it in the window is pointless. we2298 * should evict it first.2299 */2300 if (DELTA(entry) && max_depth <= n->depth)2301 continue;23022303 /*2304 * Move the best delta base up in the window, after the2305 * currently deltified object, to keep it longer. It will2306 * be the first base object to be attempted next.2307 */2308 if (DELTA(entry)) {2309 struct unpacked swap = array[best_base];2310 int dist = (window + idx - best_base) % window;2311 int dst = best_base;2312 while (dist--) {2313 int src = (dst + 1) % window;2314 array[dst] = array[src];2315 dst = src;2316 }2317 array[dst] = swap;2318 }23192320 next:2321 idx++;2322 if (count + 1 < window)2323 count++;2324 if (idx >= window)2325 idx = 0;2326 }23272328 for (i = 0; i < window; ++i) {2329 free_delta_index(array[i].index);2330 free(array[i].data);2331 }2332 free(array);2333}23342335static void try_to_free_from_threads(size_t size)2336{2337 packing_data_lock(&to_pack);2338 release_pack_memory(size);2339 packing_data_unlock(&to_pack);2340}23412342static try_to_free_t old_try_to_free_routine;23432344/*2345 * The main object list is split into smaller lists, each is handed to2346 * one worker.2347 *2348 * The main thread waits on the condition that (at least) one of the workers2349 * has stopped working (which is indicated in the .working member of2350 * struct thread_params).2351 *2352 * When a work thread has completed its work, it sets .working to 0 and2353 * signals the main thread and waits on the condition that .data_ready2354 * becomes 1.2355 *2356 * The main thread steals half of the work from the worker that has2357 * most work left to hand it to the idle worker.2358 */23592360struct thread_params {2361 pthread_t thread;2362 struct object_entry **list;2363 unsigned list_size;2364 unsigned remaining;2365 int window;2366 int depth;2367 int working;2368 int data_ready;2369 pthread_mutex_t mutex;2370 pthread_cond_t cond;2371 unsigned *processed;2372};23732374static pthread_cond_t progress_cond;23752376/*2377 * Mutex and conditional variable can't be statically-initialized on Windows.2378 */2379static void init_threaded_search(void)2380{2381 pthread_mutex_init(&cache_mutex, NULL);2382 pthread_mutex_init(&progress_mutex, NULL);2383 pthread_cond_init(&progress_cond, NULL);2384 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);2385}23862387static void cleanup_threaded_search(void)2388{2389 set_try_to_free_routine(old_try_to_free_routine);2390 pthread_cond_destroy(&progress_cond);2391 pthread_mutex_destroy(&cache_mutex);2392 pthread_mutex_destroy(&progress_mutex);2393}23942395static void *threaded_find_deltas(void *arg)2396{2397 struct thread_params *me = arg;23982399 progress_lock();2400 while (me->remaining) {2401 progress_unlock();24022403 find_deltas(me->list, &me->remaining,2404 me->window, me->depth, me->processed);24052406 progress_lock();2407 me->working = 0;2408 pthread_cond_signal(&progress_cond);2409 progress_unlock();24102411 /*2412 * We must not set ->data_ready before we wait on the2413 * condition because the main thread may have set it to 12414 * before we get here. In order to be sure that new2415 * work is available if we see 1 in ->data_ready, it2416 * was initialized to 0 before this thread was spawned2417 * and we reset it to 0 right away.2418 */2419 pthread_mutex_lock(&me->mutex);2420 while (!me->data_ready)2421 pthread_cond_wait(&me->cond, &me->mutex);2422 me->data_ready = 0;2423 pthread_mutex_unlock(&me->mutex);24242425 progress_lock();2426 }2427 progress_unlock();2428 /* leave ->working 1 so that this doesn't get more work assigned */2429 return NULL;2430}24312432static void ll_find_deltas(struct object_entry **list, unsigned list_size,2433 int window, int depth, unsigned *processed)2434{2435 struct thread_params *p;2436 int i, ret, active_threads = 0;24372438 init_threaded_search();24392440 if (delta_search_threads <= 1) {2441 find_deltas(list, &list_size, window, depth, processed);2442 cleanup_threaded_search();2443 return;2444 }2445 if (progress > pack_to_stdout)2446 fprintf_ln(stderr, _("Delta compression using up to %d threads"),2447 delta_search_threads);2448 p = xcalloc(delta_search_threads, sizeof(*p));24492450 /* Partition the work amongst work threads. */2451 for (i = 0; i < delta_search_threads; i++) {2452 unsigned sub_size = list_size / (delta_search_threads - i);24532454 /* don't use too small segments or no deltas will be found */2455 if (sub_size < 2*window && i+1 < delta_search_threads)2456 sub_size = 0;24572458 p[i].window = window;2459 p[i].depth = depth;2460 p[i].processed = processed;2461 p[i].working = 1;2462 p[i].data_ready = 0;24632464 /* try to split chunks on "path" boundaries */2465 while (sub_size && sub_size < list_size &&2466 list[sub_size]->hash &&2467 list[sub_size]->hash == list[sub_size-1]->hash)2468 sub_size++;24692470 p[i].list = list;2471 p[i].list_size = sub_size;2472 p[i].remaining = sub_size;24732474 list += sub_size;2475 list_size -= sub_size;2476 }24772478 /* Start work threads. */2479 for (i = 0; i < delta_search_threads; i++) {2480 if (!p[i].list_size)2481 continue;2482 pthread_mutex_init(&p[i].mutex, NULL);2483 pthread_cond_init(&p[i].cond, NULL);2484 ret = pthread_create(&p[i].thread, NULL,2485 threaded_find_deltas, &p[i]);2486 if (ret)2487 die(_("unable to create thread: %s"), strerror(ret));2488 active_threads++;2489 }24902491 /*2492 * Now let's wait for work completion. Each time a thread is done2493 * with its work, we steal half of the remaining work from the2494 * thread with the largest number of unprocessed objects and give2495 * it to that newly idle thread. This ensure good load balancing2496 * until the remaining object list segments are simply too short2497 * to be worth splitting anymore.2498 */2499 while (active_threads) {2500 struct thread_params *target = NULL;2501 struct thread_params *victim = NULL;2502 unsigned sub_size = 0;25032504 progress_lock();2505 for (;;) {2506 for (i = 0; !target && i < delta_search_threads; i++)2507 if (!p[i].working)2508 target = &p[i];2509 if (target)2510 break;2511 pthread_cond_wait(&progress_cond, &progress_mutex);2512 }25132514 for (i = 0; i < delta_search_threads; i++)2515 if (p[i].remaining > 2*window &&2516 (!victim || victim->remaining < p[i].remaining))2517 victim = &p[i];2518 if (victim) {2519 sub_size = victim->remaining / 2;2520 list = victim->list + victim->list_size - sub_size;2521 while (sub_size && list[0]->hash &&2522 list[0]->hash == list[-1]->hash) {2523 list++;2524 sub_size--;2525 }2526 if (!sub_size) {2527 /*2528 * It is possible for some "paths" to have2529 * so many objects that no hash boundary2530 * might be found. Let's just steal the2531 * exact half in that case.2532 */2533 sub_size = victim->remaining / 2;2534 list -= sub_size;2535 }2536 target->list = list;2537 victim->list_size -= sub_size;2538 victim->remaining -= sub_size;2539 }2540 target->list_size = sub_size;2541 target->remaining = sub_size;2542 target->working = 1;2543 progress_unlock();25442545 pthread_mutex_lock(&target->mutex);2546 target->data_ready = 1;2547 pthread_cond_signal(&target->cond);2548 pthread_mutex_unlock(&target->mutex);25492550 if (!sub_size) {2551 pthread_join(target->thread, NULL);2552 pthread_cond_destroy(&target->cond);2553 pthread_mutex_destroy(&target->mutex);2554 active_threads--;2555 }2556 }2557 cleanup_threaded_search();2558 free(p);2559}25602561static void add_tag_chain(const struct object_id *oid)2562{2563 struct tag *tag;25642565 /*2566 * We catch duplicates already in add_object_entry(), but we'd2567 * prefer to do this extra check to avoid having to parse the2568 * tag at all if we already know that it's being packed (e.g., if2569 * it was included via bitmaps, we would not have parsed it2570 * previously).2571 */2572 if (packlist_find(&to_pack, oid->hash, NULL))2573 return;25742575 tag = lookup_tag(the_repository, oid);2576 while (1) {2577 if (!tag || parse_tag(tag) || !tag->tagged)2578 die(_("unable to pack objects reachable from tag %s"),2579 oid_to_hex(oid));25802581 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);25822583 if (tag->tagged->type != OBJ_TAG)2584 return;25852586 tag = (struct tag *)tag->tagged;2587 }2588}25892590static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)2591{2592 struct object_id peeled;25932594 if (starts_with(path, "refs/tags/") && /* is a tag? */2595 !peel_ref(path, &peeled) && /* peelable? */2596 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */2597 add_tag_chain(oid);2598 return 0;2599}26002601static void prepare_pack(int window, int depth)2602{2603 struct object_entry **delta_list;2604 uint32_t i, nr_deltas;2605 unsigned n;26062607 if (use_delta_islands)2608 resolve_tree_islands(the_repository, progress, &to_pack);26092610 get_object_details();26112612 /*2613 * If we're locally repacking then we need to be doubly careful2614 * from now on in order to make sure no stealth corruption gets2615 * propagated to the new pack. Clients receiving streamed packs2616 * should validate everything they get anyway so no need to incur2617 * the additional cost here in that case.2618 */2619 if (!pack_to_stdout)2620 do_check_packed_object_crc = 1;26212622 if (!to_pack.nr_objects || !window || !depth)2623 return;26242625 ALLOC_ARRAY(delta_list, to_pack.nr_objects);2626 nr_deltas = n = 0;26272628 for (i = 0; i < to_pack.nr_objects; i++) {2629 struct object_entry *entry = to_pack.objects + i;26302631 if (DELTA(entry))2632 /* This happens if we decided to reuse existing2633 * delta from a pack. "reuse_delta &&" is implied.2634 */2635 continue;26362637 if (!entry->type_valid ||2638 oe_size_less_than(&to_pack, entry, 50))2639 continue;26402641 if (entry->no_try_delta)2642 continue;26432644 if (!entry->preferred_base) {2645 nr_deltas++;2646 if (oe_type(entry) < 0)2647 die(_("unable to get type of object %s"),2648 oid_to_hex(&entry->idx.oid));2649 } else {2650 if (oe_type(entry) < 0) {2651 /*2652 * This object is not found, but we2653 * don't have to include it anyway.2654 */2655 continue;2656 }2657 }26582659 delta_list[n++] = entry;2660 }26612662 if (nr_deltas && n > 1) {2663 unsigned nr_done = 0;2664 if (progress)2665 progress_state = start_progress(_("Compressing objects"),2666 nr_deltas);2667 QSORT(delta_list, n, type_size_sort);2668 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2669 stop_progress(&progress_state);2670 if (nr_done != nr_deltas)2671 die(_("inconsistency with delta count"));2672 }2673 free(delta_list);2674}26752676static int git_pack_config(const char *k, const char *v, void *cb)2677{2678 if (!strcmp(k, "pack.window")) {2679 window = git_config_int(k, v);2680 return 0;2681 }2682 if (!strcmp(k, "pack.windowmemory")) {2683 window_memory_limit = git_config_ulong(k, v);2684 return 0;2685 }2686 if (!strcmp(k, "pack.depth")) {2687 depth = git_config_int(k, v);2688 return 0;2689 }2690 if (!strcmp(k, "pack.deltacachesize")) {2691 max_delta_cache_size = git_config_int(k, v);2692 return 0;2693 }2694 if (!strcmp(k, "pack.deltacachelimit")) {2695 cache_max_small_delta_size = git_config_int(k, v);2696 return 0;2697 }2698 if (!strcmp(k, "pack.writebitmaphashcache")) {2699 if (git_config_bool(k, v))2700 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2701 else2702 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2703 }2704 if (!strcmp(k, "pack.usebitmaps")) {2705 use_bitmap_index_default = git_config_bool(k, v);2706 return 0;2707 }2708 if (!strcmp(k, "pack.usesparse")) {2709 sparse = git_config_bool(k, v);2710 return 0;2711 }2712 if (!strcmp(k, "pack.threads")) {2713 delta_search_threads = git_config_int(k, v);2714 if (delta_search_threads < 0)2715 die(_("invalid number of threads specified (%d)"),2716 delta_search_threads);2717 if (!HAVE_THREADS && delta_search_threads != 1) {2718 warning(_("no threads support, ignoring %s"), k);2719 delta_search_threads = 0;2720 }2721 return 0;2722 }2723 if (!strcmp(k, "pack.indexversion")) {2724 pack_idx_opts.version = git_config_int(k, v);2725 if (pack_idx_opts.version > 2)2726 die(_("bad pack.indexversion=%"PRIu32),2727 pack_idx_opts.version);2728 return 0;2729 }2730 return git_default_config(k, v, cb);2731}27322733static void read_object_list_from_stdin(void)2734{2735 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];2736 struct object_id oid;2737 const char *p;27382739 for (;;) {2740 if (!fgets(line, sizeof(line), stdin)) {2741 if (feof(stdin))2742 break;2743 if (!ferror(stdin))2744 die("BUG: fgets returned NULL, not EOF, not error!");2745 if (errno != EINTR)2746 die_errno("fgets");2747 clearerr(stdin);2748 continue;2749 }2750 if (line[0] == '-') {2751 if (get_oid_hex(line+1, &oid))2752 die(_("expected edge object ID, got garbage:\n %s"),2753 line);2754 add_preferred_base(&oid);2755 continue;2756 }2757 if (parse_oid_hex(line, &oid, &p))2758 die(_("expected object ID, got garbage:\n %s"), line);27592760 add_preferred_base_object(p + 1);2761 add_object_entry(&oid, OBJ_NONE, p + 1, 0);2762 }2763}27642765/* Remember to update object flag allocation in object.h */2766#define OBJECT_ADDED (1u<<20)27672768static void show_commit(struct commit *commit, void *data)2769{2770 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);2771 commit->object.flags |= OBJECT_ADDED;27722773 if (write_bitmap_index)2774 index_commit_for_bitmap(commit);27752776 if (use_delta_islands)2777 propagate_island_marks(commit);2778}27792780static void show_object(struct object *obj, const char *name, void *data)2781{2782 add_preferred_base_object(name);2783 add_object_entry(&obj->oid, obj->type, name, 0);2784 obj->flags |= OBJECT_ADDED;27852786 if (use_delta_islands) {2787 const char *p;2788 unsigned depth;2789 struct object_entry *ent;27902791 /* the empty string is a root tree, which is depth 0 */2792 depth = *name ? 1 : 0;2793 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))2794 depth++;27952796 ent = packlist_find(&to_pack, obj->oid.hash, NULL);2797 if (ent && depth > oe_tree_depth(&to_pack, ent))2798 oe_set_tree_depth(&to_pack, ent, depth);2799 }2800}28012802static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)2803{2804 assert(arg_missing_action == MA_ALLOW_ANY);28052806 /*2807 * Quietly ignore ALL missing objects. This avoids problems with2808 * staging them now and getting an odd error later.2809 */2810 if (!has_object_file(&obj->oid))2811 return;28122813 show_object(obj, name, data);2814}28152816static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)2817{2818 assert(arg_missing_action == MA_ALLOW_PROMISOR);28192820 /*2821 * Quietly ignore EXPECTED missing objects. This avoids problems with2822 * staging them now and getting an odd error later.2823 */2824 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))2825 return;28262827 show_object(obj, name, data);2828}28292830static int option_parse_missing_action(const struct option *opt,2831 const char *arg, int unset)2832{2833 assert(arg);2834 assert(!unset);28352836 if (!strcmp(arg, "error")) {2837 arg_missing_action = MA_ERROR;2838 fn_show_object = show_object;2839 return 0;2840 }28412842 if (!strcmp(arg, "allow-any")) {2843 arg_missing_action = MA_ALLOW_ANY;2844 fetch_if_missing = 0;2845 fn_show_object = show_object__ma_allow_any;2846 return 0;2847 }28482849 if (!strcmp(arg, "allow-promisor")) {2850 arg_missing_action = MA_ALLOW_PROMISOR;2851 fetch_if_missing = 0;2852 fn_show_object = show_object__ma_allow_promisor;2853 return 0;2854 }28552856 die(_("invalid value for --missing"));2857 return 0;2858}28592860static void show_edge(struct commit *commit)2861{2862 add_preferred_base(&commit->object.oid);2863}28642865struct in_pack_object {2866 off_t offset;2867 struct object *object;2868};28692870struct in_pack {2871 unsigned int alloc;2872 unsigned int nr;2873 struct in_pack_object *array;2874};28752876static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)2877{2878 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);2879 in_pack->array[in_pack->nr].object = object;2880 in_pack->nr++;2881}28822883/*2884 * Compare the objects in the offset order, in order to emulate the2885 * "git rev-list --objects" output that produced the pack originally.2886 */2887static int ofscmp(const void *a_, const void *b_)2888{2889 struct in_pack_object *a = (struct in_pack_object *)a_;2890 struct in_pack_object *b = (struct in_pack_object *)b_;28912892 if (a->offset < b->offset)2893 return -1;2894 else if (a->offset > b->offset)2895 return 1;2896 else2897 return oidcmp(&a->object->oid, &b->object->oid);2898}28992900static void add_objects_in_unpacked_packs(struct rev_info *revs)2901{2902 struct packed_git *p;2903 struct in_pack in_pack;2904 uint32_t i;29052906 memset(&in_pack, 0, sizeof(in_pack));29072908 for (p = get_all_packs(the_repository); p; p = p->next) {2909 struct object_id oid;2910 struct object *o;29112912 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)2913 continue;2914 if (open_pack_index(p))2915 die(_("cannot open pack index"));29162917 ALLOC_GROW(in_pack.array,2918 in_pack.nr + p->num_objects,2919 in_pack.alloc);29202921 for (i = 0; i < p->num_objects; i++) {2922 nth_packed_object_oid(&oid, p, i);2923 o = lookup_unknown_object(oid.hash);2924 if (!(o->flags & OBJECT_ADDED))2925 mark_in_pack_object(o, p, &in_pack);2926 o->flags |= OBJECT_ADDED;2927 }2928 }29292930 if (in_pack.nr) {2931 QSORT(in_pack.array, in_pack.nr, ofscmp);2932 for (i = 0; i < in_pack.nr; i++) {2933 struct object *o = in_pack.array[i].object;2934 add_object_entry(&o->oid, o->type, "", 0);2935 }2936 }2937 free(in_pack.array);2938}29392940static int add_loose_object(const struct object_id *oid, const char *path,2941 void *data)2942{2943 enum object_type type = oid_object_info(the_repository, oid, NULL);29442945 if (type < 0) {2946 warning(_("loose object at %s could not be examined"), path);2947 return 0;2948 }29492950 add_object_entry(oid, type, "", 0);2951 return 0;2952}29532954/*2955 * We actually don't even have to worry about reachability here.2956 * add_object_entry will weed out duplicates, so we just add every2957 * loose object we find.2958 */2959static void add_unreachable_loose_objects(void)2960{2961 for_each_loose_file_in_objdir(get_object_directory(),2962 add_loose_object,2963 NULL, NULL, NULL);2964}29652966static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2967{2968 static struct packed_git *last_found = (void *)1;2969 struct packed_git *p;29702971 p = (last_found != (void *)1) ? last_found :2972 get_all_packs(the_repository);29732974 while (p) {2975 if ((!p->pack_local || p->pack_keep ||2976 p->pack_keep_in_core) &&2977 find_pack_entry_one(oid->hash, p)) {2978 last_found = p;2979 return 1;2980 }2981 if (p == last_found)2982 p = get_all_packs(the_repository);2983 else2984 p = p->next;2985 if (p == last_found)2986 p = p->next;2987 }2988 return 0;2989}29902991/*2992 * Store a list of sha1s that are should not be discarded2993 * because they are either written too recently, or are2994 * reachable from another object that was.2995 *2996 * This is filled by get_object_list.2997 */2998static struct oid_array recent_objects;29993000static int loosened_object_can_be_discarded(const struct object_id *oid,3001 timestamp_t mtime)3002{3003 if (!unpack_unreachable_expiration)3004 return 0;3005 if (mtime > unpack_unreachable_expiration)3006 return 0;3007 if (oid_array_lookup(&recent_objects, oid) >= 0)3008 return 0;3009 return 1;3010}30113012static void loosen_unused_packed_objects(struct rev_info *revs)3013{3014 struct packed_git *p;3015 uint32_t i;3016 struct object_id oid;30173018 for (p = get_all_packs(the_repository); p; p = p->next) {3019 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)3020 continue;30213022 if (open_pack_index(p))3023 die(_("cannot open pack index"));30243025 for (i = 0; i < p->num_objects; i++) {3026 nth_packed_object_oid(&oid, p, i);3027 if (!packlist_find(&to_pack, oid.hash, NULL) &&3028 !has_sha1_pack_kept_or_nonlocal(&oid) &&3029 !loosened_object_can_be_discarded(&oid, p->mtime))3030 if (force_object_loose(&oid, p->mtime))3031 die(_("unable to force loose object"));3032 }3033 }3034}30353036/*3037 * This tracks any options which pack-reuse code expects to be on, or which a3038 * reader of the pack might not understand, and which would therefore prevent3039 * blind reuse of what we have on disk.3040 */3041static int pack_options_allow_reuse(void)3042{3043 return pack_to_stdout &&3044 allow_ofs_delta &&3045 !ignore_packed_keep_on_disk &&3046 !ignore_packed_keep_in_core &&3047 (!local || !have_non_local_packs) &&3048 !incremental;3049}30503051static int get_object_list_from_bitmap(struct rev_info *revs)3052{3053 if (!(bitmap_git = prepare_bitmap_walk(revs)))3054 return -1;30553056 if (pack_options_allow_reuse() &&3057 !reuse_partial_packfile_from_bitmap(3058 bitmap_git,3059 &reuse_packfile,3060 &reuse_packfile_objects,3061 &reuse_packfile_offset)) {3062 assert(reuse_packfile_objects);3063 nr_result += reuse_packfile_objects;3064 display_progress(progress_state, nr_result);3065 }30663067 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);3068 return 0;3069}30703071static void record_recent_object(struct object *obj,3072 const char *name,3073 void *data)3074{3075 oid_array_append(&recent_objects, &obj->oid);3076}30773078static void record_recent_commit(struct commit *commit, void *data)3079{3080 oid_array_append(&recent_objects, &commit->object.oid);3081}30823083static void get_object_list(int ac, const char **av)3084{3085 struct rev_info revs;3086 struct setup_revision_opt s_r_opt = {3087 .allow_exclude_promisor_objects = 1,3088 };3089 char line[1000];3090 int flags = 0;3091 int save_warning;30923093 repo_init_revisions(the_repository, &revs, NULL);3094 save_commit_buffer = 0;3095 setup_revisions(ac, av, &revs, &s_r_opt);30963097 /* make sure shallows are read */3098 is_repository_shallow(the_repository);30993100 save_warning = warn_on_object_refname_ambiguity;3101 warn_on_object_refname_ambiguity = 0;31023103 while (fgets(line, sizeof(line), stdin) != NULL) {3104 int len = strlen(line);3105 if (len && line[len - 1] == '\n')3106 line[--len] = 0;3107 if (!len)3108 break;3109 if (*line == '-') {3110 if (!strcmp(line, "--not")) {3111 flags ^= UNINTERESTING;3112 write_bitmap_index = 0;3113 continue;3114 }3115 if (starts_with(line, "--shallow ")) {3116 struct object_id oid;3117 if (get_oid_hex(line + 10, &oid))3118 die("not an SHA-1 '%s'", line + 10);3119 register_shallow(the_repository, &oid);3120 use_bitmap_index = 0;3121 continue;3122 }3123 die(_("not a rev '%s'"), line);3124 }3125 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3126 die(_("bad revision '%s'"), line);3127 }31283129 warn_on_object_refname_ambiguity = save_warning;31303131 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))3132 return;31333134 if (use_delta_islands)3135 load_delta_islands(the_repository);31363137 if (prepare_revision_walk(&revs))3138 die(_("revision walk setup failed"));3139 mark_edges_uninteresting(&revs, show_edge, sparse);31403141 if (!fn_show_object)3142 fn_show_object = show_object;3143 traverse_commit_list_filtered(&filter_options, &revs,3144 show_commit, fn_show_object, NULL,3145 NULL);31463147 if (unpack_unreachable_expiration) {3148 revs.ignore_missing_links = 1;3149 if (add_unseen_recent_objects_to_traversal(&revs,3150 unpack_unreachable_expiration))3151 die(_("unable to add recent objects"));3152 if (prepare_revision_walk(&revs))3153 die(_("revision walk setup failed"));3154 traverse_commit_list(&revs, record_recent_commit,3155 record_recent_object, NULL);3156 }31573158 if (keep_unreachable)3159 add_objects_in_unpacked_packs(&revs);3160 if (pack_loose_unreachable)3161 add_unreachable_loose_objects();3162 if (unpack_unreachable)3163 loosen_unused_packed_objects(&revs);31643165 oid_array_clear(&recent_objects);3166}31673168static void add_extra_kept_packs(const struct string_list *names)3169{3170 struct packed_git *p;31713172 if (!names->nr)3173 return;31743175 for (p = get_all_packs(the_repository); p; p = p->next) {3176 const char *name = basename(p->pack_name);3177 int i;31783179 if (!p->pack_local)3180 continue;31813182 for (i = 0; i < names->nr; i++)3183 if (!fspathcmp(name, names->items[i].string))3184 break;31853186 if (i < names->nr) {3187 p->pack_keep_in_core = 1;3188 ignore_packed_keep_in_core = 1;3189 continue;3190 }3191 }3192}31933194static int option_parse_index_version(const struct option *opt,3195 const char *arg, int unset)3196{3197 char *c;3198 const char *val = arg;31993200 BUG_ON_OPT_NEG(unset);32013202 pack_idx_opts.version = strtoul(val, &c, 10);3203 if (pack_idx_opts.version > 2)3204 die(_("unsupported index version %s"), val);3205 if (*c == ',' && c[1])3206 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);3207 if (*c || pack_idx_opts.off32_limit & 0x80000000)3208 die(_("bad index version '%s'"), val);3209 return 0;3210}32113212static int option_parse_unpack_unreachable(const struct option *opt,3213 const char *arg, int unset)3214{3215 if (unset) {3216 unpack_unreachable = 0;3217 unpack_unreachable_expiration = 0;3218 }3219 else {3220 unpack_unreachable = 1;3221 if (arg)3222 unpack_unreachable_expiration = approxidate(arg);3223 }3224 return 0;3225}32263227int cmd_pack_objects(int argc, const char **argv, const char *prefix)3228{3229 int use_internal_rev_list = 0;3230 int shallow = 0;3231 int all_progress_implied = 0;3232 struct argv_array rp = ARGV_ARRAY_INIT;3233 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;3234 int rev_list_index = 0;3235 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3236 struct option pack_objects_options[] = {3237 OPT_SET_INT('q', "quiet", &progress,3238 N_("do not show progress meter"), 0),3239 OPT_SET_INT(0, "progress", &progress,3240 N_("show progress meter"), 1),3241 OPT_SET_INT(0, "all-progress", &progress,3242 N_("show progress meter during object writing phase"), 2),3243 OPT_BOOL(0, "all-progress-implied",3244 &all_progress_implied,3245 N_("similar to --all-progress when progress meter is shown")),3246 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),3247 N_("write the pack index file in the specified idx format version"),3248 PARSE_OPT_NONEG, option_parse_index_version },3249 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,3250 N_("maximum size of each output pack file")),3251 OPT_BOOL(0, "local", &local,3252 N_("ignore borrowed objects from alternate object store")),3253 OPT_BOOL(0, "incremental", &incremental,3254 N_("ignore packed objects")),3255 OPT_INTEGER(0, "window", &window,3256 N_("limit pack window by objects")),3257 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,3258 N_("limit pack window by memory in addition to object limit")),3259 OPT_INTEGER(0, "depth", &depth,3260 N_("maximum length of delta chain allowed in the resulting pack")),3261 OPT_BOOL(0, "reuse-delta", &reuse_delta,3262 N_("reuse existing deltas")),3263 OPT_BOOL(0, "reuse-object", &reuse_object,3264 N_("reuse existing objects")),3265 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,3266 N_("use OFS_DELTA objects")),3267 OPT_INTEGER(0, "threads", &delta_search_threads,3268 N_("use threads when searching for best delta matches")),3269 OPT_BOOL(0, "non-empty", &non_empty,3270 N_("do not create an empty pack output")),3271 OPT_BOOL(0, "revs", &use_internal_rev_list,3272 N_("read revision arguments from standard input")),3273 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,3274 N_("limit the objects to those that are not yet packed"),3275 1, PARSE_OPT_NONEG),3276 OPT_SET_INT_F(0, "all", &rev_list_all,3277 N_("include objects reachable from any reference"),3278 1, PARSE_OPT_NONEG),3279 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,3280 N_("include objects referred by reflog entries"),3281 1, PARSE_OPT_NONEG),3282 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,3283 N_("include objects referred to by the index"),3284 1, PARSE_OPT_NONEG),3285 OPT_BOOL(0, "stdout", &pack_to_stdout,3286 N_("output pack to stdout")),3287 OPT_BOOL(0, "include-tag", &include_tag,3288 N_("include tag objects that refer to objects to be packed")),3289 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,3290 N_("keep unreachable objects")),3291 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,3292 N_("pack loose unreachable objects")),3293 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),3294 N_("unpack unreachable objects newer than <time>"),3295 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3296 OPT_BOOL(0, "sparse", &sparse,3297 N_("use the sparse reachability algorithm")),3298 OPT_BOOL(0, "thin", &thin,3299 N_("create thin packs")),3300 OPT_BOOL(0, "shallow", &shallow,3301 N_("create packs suitable for shallow fetches")),3302 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,3303 N_("ignore packs that have companion .keep file")),3304 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),3305 N_("ignore this pack")),3306 OPT_INTEGER(0, "compression", &pack_compression_level,3307 N_("pack compression level")),3308 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,3309 N_("do not hide commits by grafts"), 0),3310 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,3311 N_("use a bitmap index if available to speed up counting objects")),3312 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,3313 N_("write a bitmap index together with the pack index")),3314 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3315 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),3316 N_("handling for missing objects"), PARSE_OPT_NONEG,3317 option_parse_missing_action },3318 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,3319 N_("do not pack objects in promisor packfiles")),3320 OPT_BOOL(0, "delta-islands", &use_delta_islands,3321 N_("respect islands during delta compression")),3322 OPT_END(),3323 };33243325 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))3326 BUG("too many dfs states, increase OE_DFS_STATE_BITS");33273328 read_replace_refs = 0;33293330 sparse = git_env_bool("GIT_TEST_PACK_SPARSE", 0);3331 reset_pack_idx_option(&pack_idx_opts);3332 git_config(git_pack_config, NULL);33333334 progress = isatty(2);3335 argc = parse_options(argc, argv, prefix, pack_objects_options,3336 pack_usage, 0);33373338 if (argc) {3339 base_name = argv[0];3340 argc--;3341 }3342 if (pack_to_stdout != !base_name || argc)3343 usage_with_options(pack_usage, pack_objects_options);33443345 if (depth >= (1 << OE_DEPTH_BITS)) {3346 warning(_("delta chain depth %d is too deep, forcing %d"),3347 depth, (1 << OE_DEPTH_BITS) - 1);3348 depth = (1 << OE_DEPTH_BITS) - 1;3349 }3350 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {3351 warning(_("pack.deltaCacheLimit is too high, forcing %d"),3352 (1U << OE_Z_DELTA_BITS) - 1);3353 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;3354 }33553356 argv_array_push(&rp, "pack-objects");3357 if (thin) {3358 use_internal_rev_list = 1;3359 argv_array_push(&rp, shallow3360 ? "--objects-edge-aggressive"3361 : "--objects-edge");3362 } else3363 argv_array_push(&rp, "--objects");33643365 if (rev_list_all) {3366 use_internal_rev_list = 1;3367 argv_array_push(&rp, "--all");3368 }3369 if (rev_list_reflog) {3370 use_internal_rev_list = 1;3371 argv_array_push(&rp, "--reflog");3372 }3373 if (rev_list_index) {3374 use_internal_rev_list = 1;3375 argv_array_push(&rp, "--indexed-objects");3376 }3377 if (rev_list_unpacked) {3378 use_internal_rev_list = 1;3379 argv_array_push(&rp, "--unpacked");3380 }33813382 if (exclude_promisor_objects) {3383 use_internal_rev_list = 1;3384 fetch_if_missing = 0;3385 argv_array_push(&rp, "--exclude-promisor-objects");3386 }3387 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)3388 use_internal_rev_list = 1;33893390 if (!reuse_object)3391 reuse_delta = 0;3392 if (pack_compression_level == -1)3393 pack_compression_level = Z_DEFAULT_COMPRESSION;3394 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)3395 die(_("bad pack compression level %d"), pack_compression_level);33963397 if (!delta_search_threads) /* --threads=0 means autodetect */3398 delta_search_threads = online_cpus();33993400 if (!HAVE_THREADS && delta_search_threads != 1)3401 warning(_("no threads support, ignoring --threads"));3402 if (!pack_to_stdout && !pack_size_limit)3403 pack_size_limit = pack_size_limit_cfg;3404 if (pack_to_stdout && pack_size_limit)3405 die(_("--max-pack-size cannot be used to build a pack for transfer"));3406 if (pack_size_limit && pack_size_limit < 1024*1024) {3407 warning(_("minimum pack size limit is 1 MiB"));3408 pack_size_limit = 1024*1024;3409 }34103411 if (!pack_to_stdout && thin)3412 die(_("--thin cannot be used to build an indexable pack"));34133414 if (keep_unreachable && unpack_unreachable)3415 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));3416 if (!rev_list_all || !rev_list_reflog || !rev_list_index)3417 unpack_unreachable_expiration = 0;34183419 if (filter_options.choice) {3420 if (!pack_to_stdout)3421 die(_("cannot use --filter without --stdout"));3422 use_bitmap_index = 0;3423 }34243425 /*3426 * "soft" reasons not to use bitmaps - for on-disk repack by default we want3427 *3428 * - to produce good pack (with bitmap index not-yet-packed objects are3429 * packed in suboptimal order).3430 *3431 * - to use more robust pack-generation codepath (avoiding possible3432 * bugs in bitmap code and possible bitmap index corruption).3433 */3434 if (!pack_to_stdout)3435 use_bitmap_index_default = 0;34363437 if (use_bitmap_index < 0)3438 use_bitmap_index = use_bitmap_index_default;34393440 /* "hard" reasons not to use bitmaps; these just won't work at all */3441 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))3442 use_bitmap_index = 0;34433444 if (pack_to_stdout || !rev_list_all)3445 write_bitmap_index = 0;34463447 if (use_delta_islands)3448 argv_array_push(&rp, "--topo-order");34493450 if (progress && all_progress_implied)3451 progress = 2;34523453 add_extra_kept_packs(&keep_pack_list);3454 if (ignore_packed_keep_on_disk) {3455 struct packed_git *p;3456 for (p = get_all_packs(the_repository); p; p = p->next)3457 if (p->pack_local && p->pack_keep)3458 break;3459 if (!p) /* no keep-able packs found */3460 ignore_packed_keep_on_disk = 0;3461 }3462 if (local) {3463 /*3464 * unlike ignore_packed_keep_on_disk above, we do not3465 * want to unset "local" based on looking at packs, as3466 * it also covers non-local objects3467 */3468 struct packed_git *p;3469 for (p = get_all_packs(the_repository); p; p = p->next) {3470 if (!p->pack_local) {3471 have_non_local_packs = 1;3472 break;3473 }3474 }3475 }34763477 trace2_region_enter("pack-objects", "enumerate-objects",3478 the_repository);3479 prepare_packing_data(the_repository, &to_pack);34803481 if (progress)3482 progress_state = start_progress(_("Enumerating objects"), 0);3483 if (!use_internal_rev_list)3484 read_object_list_from_stdin();3485 else {3486 get_object_list(rp.argc, rp.argv);3487 argv_array_clear(&rp);3488 }3489 cleanup_preferred_base();3490 if (include_tag && nr_result)3491 for_each_ref(add_ref_tag, NULL);3492 stop_progress(&progress_state);3493 trace2_region_leave("pack-objects", "enumerate-objects",3494 the_repository);34953496 if (non_empty && !nr_result)3497 return 0;3498 if (nr_result) {3499 trace2_region_enter("pack-objects", "prepare-pack",3500 the_repository);3501 prepare_pack(window, depth);3502 trace2_region_leave("pack-objects", "prepare-pack",3503 the_repository);3504 }35053506 trace2_region_enter("pack-objects", "write-pack-file", the_repository);3507 write_pack_file();3508 trace2_region_leave("pack-objects", "write-pack-file", the_repository);35093510 if (progress)3511 fprintf_ln(stderr,3512 _("Total %"PRIu32" (delta %"PRIu32"),"3513 " reused %"PRIu32" (delta %"PRIu32")"),3514 written, written_delta, reused, reused_delta);3515 return 0;3516}