1#include "builtin.h" 2#include "cache.h" 3#include "repository.h" 4#include "config.h" 5#include "attr.h" 6#include "object.h" 7#include "blob.h" 8#include "commit.h" 9#include "tag.h" 10#include "tree.h" 11#include "delta.h" 12#include "pack.h" 13#include "pack-revindex.h" 14#include "csum-file.h" 15#include "tree-walk.h" 16#include "diff.h" 17#include "revision.h" 18#include "list-objects.h" 19#include "list-objects-filter.h" 20#include "list-objects-filter-options.h" 21#include "pack-objects.h" 22#include "progress.h" 23#include "refs.h" 24#include "streaming.h" 25#include "thread-utils.h" 26#include "pack-bitmap.h" 27#include "delta-islands.h" 28#include "reachable.h" 29#include "sha1-array.h" 30#include "argv-array.h" 31#include "list.h" 32#include "packfile.h" 33#include "object-store.h" 34#include "dir.h" 35#include "midx.h" 36 37#define IN_PACK(obj) oe_in_pack(&to_pack, obj) 38#define SIZE(obj) oe_size(&to_pack, obj) 39#define SET_SIZE(obj,size) oe_set_size(&to_pack, obj, size) 40#define DELTA_SIZE(obj) oe_delta_size(&to_pack, obj) 41#define DELTA(obj) oe_delta(&to_pack, obj) 42#define DELTA_CHILD(obj) oe_delta_child(&to_pack, obj) 43#define DELTA_SIBLING(obj) oe_delta_sibling(&to_pack, obj) 44#define SET_DELTA(obj, val) oe_set_delta(&to_pack, obj, val) 45#define SET_DELTA_EXT(obj, oid) oe_set_delta_ext(&to_pack, obj, oid) 46#define SET_DELTA_SIZE(obj, val) oe_set_delta_size(&to_pack, obj, val) 47#define SET_DELTA_CHILD(obj, val) oe_set_delta_child(&to_pack, obj, val) 48#define SET_DELTA_SIBLING(obj, val) oe_set_delta_sibling(&to_pack, obj, val) 49 50static const char *pack_usage[] = { 51 N_("git pack-objects --stdout [<options>...] [< <ref-list> | < <object-list>]"), 52 N_("git pack-objects [<options>...] <base-name> [< <ref-list> | < <object-list>]"), 53 NULL 54}; 55 56/* 57 * Objects we are going to pack are collected in the `to_pack` structure. 58 * It contains an array (dynamically expanded) of the object data, and a map 59 * that can resolve SHA1s to their position in the array. 60 */ 61static struct packing_data to_pack; 62 63static struct pack_idx_entry **written_list; 64static uint32_t nr_result, nr_written, nr_seen; 65static struct bitmap_index *bitmap_git; 66static uint32_t write_layer; 67 68static int non_empty; 69static int reuse_delta = 1, reuse_object = 1; 70static int keep_unreachable, unpack_unreachable, include_tag; 71static timestamp_t unpack_unreachable_expiration; 72static int pack_loose_unreachable; 73static int local; 74static int have_non_local_packs; 75static int incremental; 76static int ignore_packed_keep_on_disk; 77static int ignore_packed_keep_in_core; 78static int allow_ofs_delta; 79static struct pack_idx_option pack_idx_opts; 80static const char *base_name; 81static int progress = 1; 82static int window = 10; 83static unsigned long pack_size_limit; 84static int depth = 50; 85static int delta_search_threads; 86static int pack_to_stdout; 87static int thin; 88static int num_preferred_base; 89static struct progress *progress_state; 90 91static struct packed_git *reuse_packfile; 92static uint32_t reuse_packfile_objects; 93static off_t reuse_packfile_offset; 94 95static int use_bitmap_index_default = 1; 96static int use_bitmap_index = -1; 97static int write_bitmap_index; 98static uint16_t write_bitmap_options; 99 100static int exclude_promisor_objects; 101 102static int use_delta_islands; 103 104static unsigned long delta_cache_size = 0; 105static unsigned long max_delta_cache_size = DEFAULT_DELTA_CACHE_SIZE; 106static unsigned long cache_max_small_delta_size = 1000; 107 108static unsigned long window_memory_limit = 0; 109 110static struct list_objects_filter_options filter_options; 111 112enum missing_action { 113 MA_ERROR = 0, /* fail if any missing objects are encountered */ 114 MA_ALLOW_ANY, /* silently allow ALL missing objects */ 115 MA_ALLOW_PROMISOR, /* silently allow all missing PROMISOR objects */ 116}; 117static enum missing_action arg_missing_action; 118static show_object_fn fn_show_object; 119 120/* 121 * stats 122 */ 123static uint32_t written, written_delta; 124static uint32_t reused, reused_delta; 125 126/* 127 * Indexed commits 128 */ 129static struct commit **indexed_commits; 130static unsigned int indexed_commits_nr; 131static unsigned int indexed_commits_alloc; 132 133static void index_commit_for_bitmap(struct commit *commit) 134{ 135 if (indexed_commits_nr >= indexed_commits_alloc) { 136 indexed_commits_alloc = (indexed_commits_alloc + 32) * 2; 137 REALLOC_ARRAY(indexed_commits, indexed_commits_alloc); 138 } 139 140 indexed_commits[indexed_commits_nr++] = commit; 141} 142 143static void *get_delta(struct object_entry *entry) 144{ 145 unsigned long size, base_size, delta_size; 146 void *buf, *base_buf, *delta_buf; 147 enum object_type type; 148 149 buf = read_object_file(&entry->idx.oid, &type, &size); 150 if (!buf) 151 die(_("unable to read %s"), oid_to_hex(&entry->idx.oid)); 152 base_buf = read_object_file(&DELTA(entry)->idx.oid, &type, 153 &base_size); 154 if (!base_buf) 155 die("unable to read %s", 156 oid_to_hex(&DELTA(entry)->idx.oid)); 157 delta_buf = diff_delta(base_buf, base_size, 158 buf, size, &delta_size, 0); 159 /* 160 * We succesfully computed this delta once but dropped it for 161 * memory reasons. Something is very wrong if this time we 162 * recompute and create a different delta. 163 */ 164 if (!delta_buf || delta_size != DELTA_SIZE(entry)) 165 BUG("delta size changed"); 166 free(buf); 167 free(base_buf); 168 return delta_buf; 169} 170 171static unsigned long do_compress(void **pptr, unsigned long size) 172{ 173 git_zstream stream; 174 void *in, *out; 175 unsigned long maxsize; 176 177 git_deflate_init(&stream, pack_compression_level); 178 maxsize = git_deflate_bound(&stream, size); 179 180 in = *pptr; 181 out = xmalloc(maxsize); 182 *pptr = out; 183 184 stream.next_in = in; 185 stream.avail_in = size; 186 stream.next_out = out; 187 stream.avail_out = maxsize; 188 while (git_deflate(&stream, Z_FINISH) == Z_OK) 189 ; /* nothing */ 190 git_deflate_end(&stream); 191 192 free(in); 193 return stream.total_out; 194} 195 196static unsigned long write_large_blob_data(struct git_istream *st, struct hashfile *f, 197 const struct object_id *oid) 198{ 199 git_zstream stream; 200 unsigned char ibuf[1024 * 16]; 201 unsigned char obuf[1024 * 16]; 202 unsigned long olen = 0; 203 204 git_deflate_init(&stream, pack_compression_level); 205 206 for (;;) { 207 ssize_t readlen; 208 int zret = Z_OK; 209 readlen = read_istream(st, ibuf, sizeof(ibuf)); 210 if (readlen == -1) 211 die(_("unable to read %s"), oid_to_hex(oid)); 212 213 stream.next_in = ibuf; 214 stream.avail_in = readlen; 215 while ((stream.avail_in || readlen == 0) && 216 (zret == Z_OK || zret == Z_BUF_ERROR)) { 217 stream.next_out = obuf; 218 stream.avail_out = sizeof(obuf); 219 zret = git_deflate(&stream, readlen ? 0 : Z_FINISH); 220 hashwrite(f, obuf, stream.next_out - obuf); 221 olen += stream.next_out - obuf; 222 } 223 if (stream.avail_in) 224 die(_("deflate error (%d)"), zret); 225 if (readlen == 0) { 226 if (zret != Z_STREAM_END) 227 die(_("deflate error (%d)"), zret); 228 break; 229 } 230 } 231 git_deflate_end(&stream); 232 return olen; 233} 234 235/* 236 * we are going to reuse the existing object data as is. make 237 * sure it is not corrupt. 238 */ 239static int check_pack_inflate(struct packed_git *p, 240 struct pack_window **w_curs, 241 off_t offset, 242 off_t len, 243 unsigned long expect) 244{ 245 git_zstream stream; 246 unsigned char fakebuf[4096], *in; 247 int st; 248 249 memset(&stream, 0, sizeof(stream)); 250 git_inflate_init(&stream); 251 do { 252 in = use_pack(p, w_curs, offset, &stream.avail_in); 253 stream.next_in = in; 254 stream.next_out = fakebuf; 255 stream.avail_out = sizeof(fakebuf); 256 st = git_inflate(&stream, Z_FINISH); 257 offset += stream.next_in - in; 258 } while (st == Z_OK || st == Z_BUF_ERROR); 259 git_inflate_end(&stream); 260 return (st == Z_STREAM_END && 261 stream.total_out == expect && 262 stream.total_in == len) ? 0 : -1; 263} 264 265static void copy_pack_data(struct hashfile *f, 266 struct packed_git *p, 267 struct pack_window **w_curs, 268 off_t offset, 269 off_t len) 270{ 271 unsigned char *in; 272 unsigned long avail; 273 274 while (len) { 275 in = use_pack(p, w_curs, offset, &avail); 276 if (avail > len) 277 avail = (unsigned long)len; 278 hashwrite(f, in, avail); 279 offset += avail; 280 len -= avail; 281 } 282} 283 284/* Return 0 if we will bust the pack-size limit */ 285static unsigned long write_no_reuse_object(struct hashfile *f, struct object_entry *entry, 286 unsigned long limit, int usable_delta) 287{ 288 unsigned long size, datalen; 289 unsigned char header[MAX_PACK_OBJECT_HEADER], 290 dheader[MAX_PACK_OBJECT_HEADER]; 291 unsigned hdrlen; 292 enum object_type type; 293 void *buf; 294 struct git_istream *st = NULL; 295 const unsigned hashsz = the_hash_algo->rawsz; 296 297 if (!usable_delta) { 298 if (oe_type(entry) == OBJ_BLOB && 299 oe_size_greater_than(&to_pack, entry, big_file_threshold) && 300 (st = open_istream(&entry->idx.oid, &type, &size, NULL)) != NULL) 301 buf = NULL; 302 else { 303 buf = read_object_file(&entry->idx.oid, &type, &size); 304 if (!buf) 305 die(_("unable to read %s"), 306 oid_to_hex(&entry->idx.oid)); 307 } 308 /* 309 * make sure no cached delta data remains from a 310 * previous attempt before a pack split occurred. 311 */ 312 FREE_AND_NULL(entry->delta_data); 313 entry->z_delta_size = 0; 314 } else if (entry->delta_data) { 315 size = DELTA_SIZE(entry); 316 buf = entry->delta_data; 317 entry->delta_data = NULL; 318 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 319 OBJ_OFS_DELTA : OBJ_REF_DELTA; 320 } else { 321 buf = get_delta(entry); 322 size = DELTA_SIZE(entry); 323 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 324 OBJ_OFS_DELTA : OBJ_REF_DELTA; 325 } 326 327 if (st) /* large blob case, just assume we don't compress well */ 328 datalen = size; 329 else if (entry->z_delta_size) 330 datalen = entry->z_delta_size; 331 else 332 datalen = do_compress(&buf, size); 333 334 /* 335 * The object header is a byte of 'type' followed by zero or 336 * more bytes of length. 337 */ 338 hdrlen = encode_in_pack_object_header(header, sizeof(header), 339 type, size); 340 341 if (type == OBJ_OFS_DELTA) { 342 /* 343 * Deltas with relative base contain an additional 344 * encoding of the relative offset for the delta 345 * base from this object's position in the pack. 346 */ 347 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 348 unsigned pos = sizeof(dheader) - 1; 349 dheader[pos] = ofs & 127; 350 while (ofs >>= 7) 351 dheader[--pos] = 128 | (--ofs & 127); 352 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 353 if (st) 354 close_istream(st); 355 free(buf); 356 return 0; 357 } 358 hashwrite(f, header, hdrlen); 359 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 360 hdrlen += sizeof(dheader) - pos; 361 } else if (type == OBJ_REF_DELTA) { 362 /* 363 * Deltas with a base reference contain 364 * additional bytes for the base object ID. 365 */ 366 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 367 if (st) 368 close_istream(st); 369 free(buf); 370 return 0; 371 } 372 hashwrite(f, header, hdrlen); 373 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 374 hdrlen += hashsz; 375 } else { 376 if (limit && hdrlen + datalen + hashsz >= limit) { 377 if (st) 378 close_istream(st); 379 free(buf); 380 return 0; 381 } 382 hashwrite(f, header, hdrlen); 383 } 384 if (st) { 385 datalen = write_large_blob_data(st, f, &entry->idx.oid); 386 close_istream(st); 387 } else { 388 hashwrite(f, buf, datalen); 389 free(buf); 390 } 391 392 return hdrlen + datalen; 393} 394 395/* Return 0 if we will bust the pack-size limit */ 396static off_t write_reuse_object(struct hashfile *f, struct object_entry *entry, 397 unsigned long limit, int usable_delta) 398{ 399 struct packed_git *p = IN_PACK(entry); 400 struct pack_window *w_curs = NULL; 401 struct revindex_entry *revidx; 402 off_t offset; 403 enum object_type type = oe_type(entry); 404 off_t datalen; 405 unsigned char header[MAX_PACK_OBJECT_HEADER], 406 dheader[MAX_PACK_OBJECT_HEADER]; 407 unsigned hdrlen; 408 const unsigned hashsz = the_hash_algo->rawsz; 409 unsigned long entry_size = SIZE(entry); 410 411 if (DELTA(entry)) 412 type = (allow_ofs_delta && DELTA(entry)->idx.offset) ? 413 OBJ_OFS_DELTA : OBJ_REF_DELTA; 414 hdrlen = encode_in_pack_object_header(header, sizeof(header), 415 type, entry_size); 416 417 offset = entry->in_pack_offset; 418 revidx = find_pack_revindex(p, offset); 419 datalen = revidx[1].offset - offset; 420 if (!pack_to_stdout && p->index_version > 1 && 421 check_pack_crc(p, &w_curs, offset, datalen, revidx->nr)) { 422 error(_("bad packed object CRC for %s"), 423 oid_to_hex(&entry->idx.oid)); 424 unuse_pack(&w_curs); 425 return write_no_reuse_object(f, entry, limit, usable_delta); 426 } 427 428 offset += entry->in_pack_header_size; 429 datalen -= entry->in_pack_header_size; 430 431 if (!pack_to_stdout && p->index_version == 1 && 432 check_pack_inflate(p, &w_curs, offset, datalen, entry_size)) { 433 error(_("corrupt packed object for %s"), 434 oid_to_hex(&entry->idx.oid)); 435 unuse_pack(&w_curs); 436 return write_no_reuse_object(f, entry, limit, usable_delta); 437 } 438 439 if (type == OBJ_OFS_DELTA) { 440 off_t ofs = entry->idx.offset - DELTA(entry)->idx.offset; 441 unsigned pos = sizeof(dheader) - 1; 442 dheader[pos] = ofs & 127; 443 while (ofs >>= 7) 444 dheader[--pos] = 128 | (--ofs & 127); 445 if (limit && hdrlen + sizeof(dheader) - pos + datalen + hashsz >= limit) { 446 unuse_pack(&w_curs); 447 return 0; 448 } 449 hashwrite(f, header, hdrlen); 450 hashwrite(f, dheader + pos, sizeof(dheader) - pos); 451 hdrlen += sizeof(dheader) - pos; 452 reused_delta++; 453 } else if (type == OBJ_REF_DELTA) { 454 if (limit && hdrlen + hashsz + datalen + hashsz >= limit) { 455 unuse_pack(&w_curs); 456 return 0; 457 } 458 hashwrite(f, header, hdrlen); 459 hashwrite(f, DELTA(entry)->idx.oid.hash, hashsz); 460 hdrlen += hashsz; 461 reused_delta++; 462 } else { 463 if (limit && hdrlen + datalen + hashsz >= limit) { 464 unuse_pack(&w_curs); 465 return 0; 466 } 467 hashwrite(f, header, hdrlen); 468 } 469 copy_pack_data(f, p, &w_curs, offset, datalen); 470 unuse_pack(&w_curs); 471 reused++; 472 return hdrlen + datalen; 473} 474 475/* Return 0 if we will bust the pack-size limit */ 476static off_t write_object(struct hashfile *f, 477 struct object_entry *entry, 478 off_t write_offset) 479{ 480 unsigned long limit; 481 off_t len; 482 int usable_delta, to_reuse; 483 484 if (!pack_to_stdout) 485 crc32_begin(f); 486 487 /* apply size limit if limited packsize and not first object */ 488 if (!pack_size_limit || !nr_written) 489 limit = 0; 490 else if (pack_size_limit <= write_offset) 491 /* 492 * the earlier object did not fit the limit; avoid 493 * mistaking this with unlimited (i.e. limit = 0). 494 */ 495 limit = 1; 496 else 497 limit = pack_size_limit - write_offset; 498 499 if (!DELTA(entry)) 500 usable_delta = 0; /* no delta */ 501 else if (!pack_size_limit) 502 usable_delta = 1; /* unlimited packfile */ 503 else if (DELTA(entry)->idx.offset == (off_t)-1) 504 usable_delta = 0; /* base was written to another pack */ 505 else if (DELTA(entry)->idx.offset) 506 usable_delta = 1; /* base already exists in this pack */ 507 else 508 usable_delta = 0; /* base could end up in another pack */ 509 510 if (!reuse_object) 511 to_reuse = 0; /* explicit */ 512 else if (!IN_PACK(entry)) 513 to_reuse = 0; /* can't reuse what we don't have */ 514 else if (oe_type(entry) == OBJ_REF_DELTA || 515 oe_type(entry) == OBJ_OFS_DELTA) 516 /* check_object() decided it for us ... */ 517 to_reuse = usable_delta; 518 /* ... but pack split may override that */ 519 else if (oe_type(entry) != entry->in_pack_type) 520 to_reuse = 0; /* pack has delta which is unusable */ 521 else if (DELTA(entry)) 522 to_reuse = 0; /* we want to pack afresh */ 523 else 524 to_reuse = 1; /* we have it in-pack undeltified, 525 * and we do not need to deltify it. 526 */ 527 528 if (!to_reuse) 529 len = write_no_reuse_object(f, entry, limit, usable_delta); 530 else 531 len = write_reuse_object(f, entry, limit, usable_delta); 532 if (!len) 533 return 0; 534 535 if (usable_delta) 536 written_delta++; 537 written++; 538 if (!pack_to_stdout) 539 entry->idx.crc32 = crc32_end(f); 540 return len; 541} 542 543enum write_one_status { 544 WRITE_ONE_SKIP = -1, /* already written */ 545 WRITE_ONE_BREAK = 0, /* writing this will bust the limit; not written */ 546 WRITE_ONE_WRITTEN = 1, /* normal */ 547 WRITE_ONE_RECURSIVE = 2 /* already scheduled to be written */ 548}; 549 550static enum write_one_status write_one(struct hashfile *f, 551 struct object_entry *e, 552 off_t *offset) 553{ 554 off_t size; 555 int recursing; 556 557 /* 558 * we set offset to 1 (which is an impossible value) to mark 559 * the fact that this object is involved in "write its base 560 * first before writing a deltified object" recursion. 561 */ 562 recursing = (e->idx.offset == 1); 563 if (recursing) { 564 warning(_("recursive delta detected for object %s"), 565 oid_to_hex(&e->idx.oid)); 566 return WRITE_ONE_RECURSIVE; 567 } else if (e->idx.offset || e->preferred_base) { 568 /* offset is non zero if object is written already. */ 569 return WRITE_ONE_SKIP; 570 } 571 572 /* if we are deltified, write out base object first. */ 573 if (DELTA(e)) { 574 e->idx.offset = 1; /* now recurse */ 575 switch (write_one(f, DELTA(e), offset)) { 576 case WRITE_ONE_RECURSIVE: 577 /* we cannot depend on this one */ 578 SET_DELTA(e, NULL); 579 break; 580 default: 581 break; 582 case WRITE_ONE_BREAK: 583 e->idx.offset = recursing; 584 return WRITE_ONE_BREAK; 585 } 586 } 587 588 e->idx.offset = *offset; 589 size = write_object(f, e, *offset); 590 if (!size) { 591 e->idx.offset = recursing; 592 return WRITE_ONE_BREAK; 593 } 594 written_list[nr_written++] = &e->idx; 595 596 /* make sure off_t is sufficiently large not to wrap */ 597 if (signed_add_overflows(*offset, size)) 598 die(_("pack too large for current definition of off_t")); 599 *offset += size; 600 return WRITE_ONE_WRITTEN; 601} 602 603static int mark_tagged(const char *path, const struct object_id *oid, int flag, 604 void *cb_data) 605{ 606 struct object_id peeled; 607 struct object_entry *entry = packlist_find(&to_pack, oid->hash, NULL); 608 609 if (entry) 610 entry->tagged = 1; 611 if (!peel_ref(path, &peeled)) { 612 entry = packlist_find(&to_pack, peeled.hash, NULL); 613 if (entry) 614 entry->tagged = 1; 615 } 616 return 0; 617} 618 619static inline void add_to_write_order(struct object_entry **wo, 620 unsigned int *endp, 621 struct object_entry *e) 622{ 623 if (e->filled || oe_layer(&to_pack, e) != write_layer) 624 return; 625 wo[(*endp)++] = e; 626 e->filled = 1; 627} 628 629static void add_descendants_to_write_order(struct object_entry **wo, 630 unsigned int *endp, 631 struct object_entry *e) 632{ 633 int add_to_order = 1; 634 while (e) { 635 if (add_to_order) { 636 struct object_entry *s; 637 /* add this node... */ 638 add_to_write_order(wo, endp, e); 639 /* all its siblings... */ 640 for (s = DELTA_SIBLING(e); s; s = DELTA_SIBLING(s)) { 641 add_to_write_order(wo, endp, s); 642 } 643 } 644 /* drop down a level to add left subtree nodes if possible */ 645 if (DELTA_CHILD(e)) { 646 add_to_order = 1; 647 e = DELTA_CHILD(e); 648 } else { 649 add_to_order = 0; 650 /* our sibling might have some children, it is next */ 651 if (DELTA_SIBLING(e)) { 652 e = DELTA_SIBLING(e); 653 continue; 654 } 655 /* go back to our parent node */ 656 e = DELTA(e); 657 while (e && !DELTA_SIBLING(e)) { 658 /* we're on the right side of a subtree, keep 659 * going up until we can go right again */ 660 e = DELTA(e); 661 } 662 if (!e) { 663 /* done- we hit our original root node */ 664 return; 665 } 666 /* pass it off to sibling at this level */ 667 e = DELTA_SIBLING(e); 668 } 669 }; 670} 671 672static void add_family_to_write_order(struct object_entry **wo, 673 unsigned int *endp, 674 struct object_entry *e) 675{ 676 struct object_entry *root; 677 678 for (root = e; DELTA(root); root = DELTA(root)) 679 ; /* nothing */ 680 add_descendants_to_write_order(wo, endp, root); 681} 682 683static void compute_layer_order(struct object_entry **wo, unsigned int *wo_end) 684{ 685 unsigned int i, last_untagged; 686 struct object_entry *objects = to_pack.objects; 687 688 for (i = 0; i < to_pack.nr_objects; i++) { 689 if (objects[i].tagged) 690 break; 691 add_to_write_order(wo, wo_end, &objects[i]); 692 } 693 last_untagged = i; 694 695 /* 696 * Then fill all the tagged tips. 697 */ 698 for (; i < to_pack.nr_objects; i++) { 699 if (objects[i].tagged) 700 add_to_write_order(wo, wo_end, &objects[i]); 701 } 702 703 /* 704 * And then all remaining commits and tags. 705 */ 706 for (i = last_untagged; i < to_pack.nr_objects; i++) { 707 if (oe_type(&objects[i]) != OBJ_COMMIT && 708 oe_type(&objects[i]) != OBJ_TAG) 709 continue; 710 add_to_write_order(wo, wo_end, &objects[i]); 711 } 712 713 /* 714 * And then all the trees. 715 */ 716 for (i = last_untagged; i < to_pack.nr_objects; i++) { 717 if (oe_type(&objects[i]) != OBJ_TREE) 718 continue; 719 add_to_write_order(wo, wo_end, &objects[i]); 720 } 721 722 /* 723 * Finally all the rest in really tight order 724 */ 725 for (i = last_untagged; i < to_pack.nr_objects; i++) { 726 if (!objects[i].filled && oe_layer(&to_pack, &objects[i]) == write_layer) 727 add_family_to_write_order(wo, wo_end, &objects[i]); 728 } 729} 730 731static struct object_entry **compute_write_order(void) 732{ 733 uint32_t max_layers = 1; 734 unsigned int i, wo_end; 735 736 struct object_entry **wo; 737 struct object_entry *objects = to_pack.objects; 738 739 for (i = 0; i < to_pack.nr_objects; i++) { 740 objects[i].tagged = 0; 741 objects[i].filled = 0; 742 SET_DELTA_CHILD(&objects[i], NULL); 743 SET_DELTA_SIBLING(&objects[i], NULL); 744 } 745 746 /* 747 * Fully connect delta_child/delta_sibling network. 748 * Make sure delta_sibling is sorted in the original 749 * recency order. 750 */ 751 for (i = to_pack.nr_objects; i > 0;) { 752 struct object_entry *e = &objects[--i]; 753 if (!DELTA(e)) 754 continue; 755 /* Mark me as the first child */ 756 e->delta_sibling_idx = DELTA(e)->delta_child_idx; 757 SET_DELTA_CHILD(DELTA(e), e); 758 } 759 760 /* 761 * Mark objects that are at the tip of tags. 762 */ 763 for_each_tag_ref(mark_tagged, NULL); 764 765 if (use_delta_islands) 766 max_layers = compute_pack_layers(&to_pack); 767 768 ALLOC_ARRAY(wo, to_pack.nr_objects); 769 wo_end = 0; 770 771 for (; write_layer < max_layers; ++write_layer) 772 compute_layer_order(wo, &wo_end); 773 774 if (wo_end != to_pack.nr_objects) 775 die(_("ordered %u objects, expected %"PRIu32), 776 wo_end, to_pack.nr_objects); 777 778 return wo; 779} 780 781static off_t write_reused_pack(struct hashfile *f) 782{ 783 unsigned char buffer[8192]; 784 off_t to_write, total; 785 int fd; 786 787 if (!is_pack_valid(reuse_packfile)) 788 die(_("packfile is invalid: %s"), reuse_packfile->pack_name); 789 790 fd = git_open(reuse_packfile->pack_name); 791 if (fd < 0) 792 die_errno(_("unable to open packfile for reuse: %s"), 793 reuse_packfile->pack_name); 794 795 if (lseek(fd, sizeof(struct pack_header), SEEK_SET) == -1) 796 die_errno(_("unable to seek in reused packfile")); 797 798 if (reuse_packfile_offset < 0) 799 reuse_packfile_offset = reuse_packfile->pack_size - the_hash_algo->rawsz; 800 801 total = to_write = reuse_packfile_offset - sizeof(struct pack_header); 802 803 while (to_write) { 804 int read_pack = xread(fd, buffer, sizeof(buffer)); 805 806 if (read_pack <= 0) 807 die_errno(_("unable to read from reused packfile")); 808 809 if (read_pack > to_write) 810 read_pack = to_write; 811 812 hashwrite(f, buffer, read_pack); 813 to_write -= read_pack; 814 815 /* 816 * We don't know the actual number of objects written, 817 * only how many bytes written, how many bytes total, and 818 * how many objects total. So we can fake it by pretending all 819 * objects we are writing are the same size. This gives us a 820 * smooth progress meter, and at the end it matches the true 821 * answer. 822 */ 823 written = reuse_packfile_objects * 824 (((double)(total - to_write)) / total); 825 display_progress(progress_state, written); 826 } 827 828 close(fd); 829 written = reuse_packfile_objects; 830 display_progress(progress_state, written); 831 return reuse_packfile_offset - sizeof(struct pack_header); 832} 833 834static const char no_split_warning[] = N_( 835"disabling bitmap writing, packs are split due to pack.packSizeLimit" 836); 837 838static void write_pack_file(void) 839{ 840 uint32_t i = 0, j; 841 struct hashfile *f; 842 off_t offset; 843 uint32_t nr_remaining = nr_result; 844 time_t last_mtime = 0; 845 struct object_entry **write_order; 846 847 if (progress > pack_to_stdout) 848 progress_state = start_progress(_("Writing objects"), nr_result); 849 ALLOC_ARRAY(written_list, to_pack.nr_objects); 850 write_order = compute_write_order(); 851 852 do { 853 struct object_id oid; 854 char *pack_tmp_name = NULL; 855 856 if (pack_to_stdout) 857 f = hashfd_throughput(1, "<stdout>", progress_state); 858 else 859 f = create_tmp_packfile(&pack_tmp_name); 860 861 offset = write_pack_header(f, nr_remaining); 862 863 if (reuse_packfile) { 864 off_t packfile_size; 865 assert(pack_to_stdout); 866 867 packfile_size = write_reused_pack(f); 868 offset += packfile_size; 869 } 870 871 nr_written = 0; 872 for (; i < to_pack.nr_objects; i++) { 873 struct object_entry *e = write_order[i]; 874 if (write_one(f, e, &offset) == WRITE_ONE_BREAK) 875 break; 876 display_progress(progress_state, written); 877 } 878 879 /* 880 * Did we write the wrong # entries in the header? 881 * If so, rewrite it like in fast-import 882 */ 883 if (pack_to_stdout) { 884 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_CLOSE); 885 } else if (nr_written == nr_remaining) { 886 finalize_hashfile(f, oid.hash, CSUM_HASH_IN_STREAM | CSUM_FSYNC | CSUM_CLOSE); 887 } else { 888 int fd = finalize_hashfile(f, oid.hash, 0); 889 fixup_pack_header_footer(fd, oid.hash, pack_tmp_name, 890 nr_written, oid.hash, offset); 891 close(fd); 892 if (write_bitmap_index) { 893 warning(_(no_split_warning)); 894 write_bitmap_index = 0; 895 } 896 } 897 898 if (!pack_to_stdout) { 899 struct stat st; 900 struct strbuf tmpname = STRBUF_INIT; 901 902 /* 903 * Packs are runtime accessed in their mtime 904 * order since newer packs are more likely to contain 905 * younger objects. So if we are creating multiple 906 * packs then we should modify the mtime of later ones 907 * to preserve this property. 908 */ 909 if (stat(pack_tmp_name, &st) < 0) { 910 warning_errno(_("failed to stat %s"), pack_tmp_name); 911 } else if (!last_mtime) { 912 last_mtime = st.st_mtime; 913 } else { 914 struct utimbuf utb; 915 utb.actime = st.st_atime; 916 utb.modtime = --last_mtime; 917 if (utime(pack_tmp_name, &utb) < 0) 918 warning_errno(_("failed utime() on %s"), pack_tmp_name); 919 } 920 921 strbuf_addf(&tmpname, "%s-", base_name); 922 923 if (write_bitmap_index) { 924 bitmap_writer_set_checksum(oid.hash); 925 bitmap_writer_build_type_index( 926 &to_pack, written_list, nr_written); 927 } 928 929 finish_tmp_packfile(&tmpname, pack_tmp_name, 930 written_list, nr_written, 931 &pack_idx_opts, oid.hash); 932 933 if (write_bitmap_index) { 934 strbuf_addf(&tmpname, "%s.bitmap", oid_to_hex(&oid)); 935 936 stop_progress(&progress_state); 937 938 bitmap_writer_show_progress(progress); 939 bitmap_writer_reuse_bitmaps(&to_pack); 940 bitmap_writer_select_commits(indexed_commits, indexed_commits_nr, -1); 941 bitmap_writer_build(&to_pack); 942 bitmap_writer_finish(written_list, nr_written, 943 tmpname.buf, write_bitmap_options); 944 write_bitmap_index = 0; 945 } 946 947 strbuf_release(&tmpname); 948 free(pack_tmp_name); 949 puts(oid_to_hex(&oid)); 950 } 951 952 /* mark written objects as written to previous pack */ 953 for (j = 0; j < nr_written; j++) { 954 written_list[j]->offset = (off_t)-1; 955 } 956 nr_remaining -= nr_written; 957 } while (nr_remaining && i < to_pack.nr_objects); 958 959 free(written_list); 960 free(write_order); 961 stop_progress(&progress_state); 962 if (written != nr_result) 963 die(_("wrote %"PRIu32" objects while expecting %"PRIu32), 964 written, nr_result); 965} 966 967static int no_try_delta(const char *path) 968{ 969 static struct attr_check *check; 970 971 if (!check) 972 check = attr_check_initl("delta", NULL); 973 git_check_attr(&the_index, path, check); 974 if (ATTR_FALSE(check->items[0].value)) 975 return 1; 976 return 0; 977} 978 979/* 980 * When adding an object, check whether we have already added it 981 * to our packing list. If so, we can skip. However, if we are 982 * being asked to excludei t, but the previous mention was to include 983 * it, make sure to adjust its flags and tweak our numbers accordingly. 984 * 985 * As an optimization, we pass out the index position where we would have 986 * found the item, since that saves us from having to look it up again a 987 * few lines later when we want to add the new entry. 988 */ 989static int have_duplicate_entry(const struct object_id *oid, 990 int exclude, 991 uint32_t *index_pos) 992{ 993 struct object_entry *entry; 994 995 entry = packlist_find(&to_pack, oid->hash, index_pos); 996 if (!entry) 997 return 0; 998 999 if (exclude) {1000 if (!entry->preferred_base)1001 nr_result--;1002 entry->preferred_base = 1;1003 }10041005 return 1;1006}10071008static int want_found_object(int exclude, struct packed_git *p)1009{1010 if (exclude)1011 return 1;1012 if (incremental)1013 return 0;10141015 /*1016 * When asked to do --local (do not include an object that appears in a1017 * pack we borrow from elsewhere) or --honor-pack-keep (do not include1018 * an object that appears in a pack marked with .keep), finding a pack1019 * that matches the criteria is sufficient for us to decide to omit it.1020 * However, even if this pack does not satisfy the criteria, we need to1021 * make sure no copy of this object appears in _any_ pack that makes us1022 * to omit the object, so we need to check all the packs.1023 *1024 * We can however first check whether these options can possible matter;1025 * if they do not matter we know we want the object in generated pack.1026 * Otherwise, we signal "-1" at the end to tell the caller that we do1027 * not know either way, and it needs to check more packs.1028 */1029 if (!ignore_packed_keep_on_disk &&1030 !ignore_packed_keep_in_core &&1031 (!local || !have_non_local_packs))1032 return 1;10331034 if (local && !p->pack_local)1035 return 0;1036 if (p->pack_local &&1037 ((ignore_packed_keep_on_disk && p->pack_keep) ||1038 (ignore_packed_keep_in_core && p->pack_keep_in_core)))1039 return 0;10401041 /* we don't know yet; keep looking for more packs */1042 return -1;1043}10441045/*1046 * Check whether we want the object in the pack (e.g., we do not want1047 * objects found in non-local stores if the "--local" option was used).1048 *1049 * If the caller already knows an existing pack it wants to take the object1050 * from, that is passed in *found_pack and *found_offset; otherwise this1051 * function finds if there is any pack that has the object and returns the pack1052 * and its offset in these variables.1053 */1054static int want_object_in_pack(const struct object_id *oid,1055 int exclude,1056 struct packed_git **found_pack,1057 off_t *found_offset)1058{1059 int want;1060 struct list_head *pos;1061 struct multi_pack_index *m;10621063 if (!exclude && local && has_loose_object_nonlocal(oid))1064 return 0;10651066 /*1067 * If we already know the pack object lives in, start checks from that1068 * pack - in the usual case when neither --local was given nor .keep files1069 * are present we will determine the answer right now.1070 */1071 if (*found_pack) {1072 want = want_found_object(exclude, *found_pack);1073 if (want != -1)1074 return want;1075 }10761077 for (m = get_multi_pack_index(the_repository); m; m = m->next) {1078 struct pack_entry e;1079 if (fill_midx_entry(oid, &e, m)) {1080 struct packed_git *p = e.p;1081 off_t offset;10821083 if (p == *found_pack)1084 offset = *found_offset;1085 else1086 offset = find_pack_entry_one(oid->hash, p);10871088 if (offset) {1089 if (!*found_pack) {1090 if (!is_pack_valid(p))1091 continue;1092 *found_offset = offset;1093 *found_pack = p;1094 }1095 want = want_found_object(exclude, p);1096 if (want != -1)1097 return want;1098 }1099 }1100 }11011102 list_for_each(pos, get_packed_git_mru(the_repository)) {1103 struct packed_git *p = list_entry(pos, struct packed_git, mru);1104 off_t offset;11051106 if (p == *found_pack)1107 offset = *found_offset;1108 else1109 offset = find_pack_entry_one(oid->hash, p);11101111 if (offset) {1112 if (!*found_pack) {1113 if (!is_pack_valid(p))1114 continue;1115 *found_offset = offset;1116 *found_pack = p;1117 }1118 want = want_found_object(exclude, p);1119 if (!exclude && want > 0)1120 list_move(&p->mru,1121 get_packed_git_mru(the_repository));1122 if (want != -1)1123 return want;1124 }1125 }11261127 return 1;1128}11291130static void create_object_entry(const struct object_id *oid,1131 enum object_type type,1132 uint32_t hash,1133 int exclude,1134 int no_try_delta,1135 uint32_t index_pos,1136 struct packed_git *found_pack,1137 off_t found_offset)1138{1139 struct object_entry *entry;11401141 entry = packlist_alloc(&to_pack, oid->hash, index_pos);1142 entry->hash = hash;1143 oe_set_type(entry, type);1144 if (exclude)1145 entry->preferred_base = 1;1146 else1147 nr_result++;1148 if (found_pack) {1149 oe_set_in_pack(&to_pack, entry, found_pack);1150 entry->in_pack_offset = found_offset;1151 }11521153 entry->no_try_delta = no_try_delta;1154}11551156static const char no_closure_warning[] = N_(1157"disabling bitmap writing, as some objects are not being packed"1158);11591160static int add_object_entry(const struct object_id *oid, enum object_type type,1161 const char *name, int exclude)1162{1163 struct packed_git *found_pack = NULL;1164 off_t found_offset = 0;1165 uint32_t index_pos;11661167 display_progress(progress_state, ++nr_seen);11681169 if (have_duplicate_entry(oid, exclude, &index_pos))1170 return 0;11711172 if (!want_object_in_pack(oid, exclude, &found_pack, &found_offset)) {1173 /* The pack is missing an object, so it will not have closure */1174 if (write_bitmap_index) {1175 warning(_(no_closure_warning));1176 write_bitmap_index = 0;1177 }1178 return 0;1179 }11801181 create_object_entry(oid, type, pack_name_hash(name),1182 exclude, name && no_try_delta(name),1183 index_pos, found_pack, found_offset);1184 return 1;1185}11861187static int add_object_entry_from_bitmap(const struct object_id *oid,1188 enum object_type type,1189 int flags, uint32_t name_hash,1190 struct packed_git *pack, off_t offset)1191{1192 uint32_t index_pos;11931194 display_progress(progress_state, ++nr_seen);11951196 if (have_duplicate_entry(oid, 0, &index_pos))1197 return 0;11981199 if (!want_object_in_pack(oid, 0, &pack, &offset))1200 return 0;12011202 create_object_entry(oid, type, name_hash, 0, 0, index_pos, pack, offset);1203 return 1;1204}12051206struct pbase_tree_cache {1207 struct object_id oid;1208 int ref;1209 int temporary;1210 void *tree_data;1211 unsigned long tree_size;1212};12131214static struct pbase_tree_cache *(pbase_tree_cache[256]);1215static int pbase_tree_cache_ix(const struct object_id *oid)1216{1217 return oid->hash[0] % ARRAY_SIZE(pbase_tree_cache);1218}1219static int pbase_tree_cache_ix_incr(int ix)1220{1221 return (ix+1) % ARRAY_SIZE(pbase_tree_cache);1222}12231224static struct pbase_tree {1225 struct pbase_tree *next;1226 /* This is a phony "cache" entry; we are not1227 * going to evict it or find it through _get()1228 * mechanism -- this is for the toplevel node that1229 * would almost always change with any commit.1230 */1231 struct pbase_tree_cache pcache;1232} *pbase_tree;12331234static struct pbase_tree_cache *pbase_tree_get(const struct object_id *oid)1235{1236 struct pbase_tree_cache *ent, *nent;1237 void *data;1238 unsigned long size;1239 enum object_type type;1240 int neigh;1241 int my_ix = pbase_tree_cache_ix(oid);1242 int available_ix = -1;12431244 /* pbase-tree-cache acts as a limited hashtable.1245 * your object will be found at your index or within a few1246 * slots after that slot if it is cached.1247 */1248 for (neigh = 0; neigh < 8; neigh++) {1249 ent = pbase_tree_cache[my_ix];1250 if (ent && oideq(&ent->oid, oid)) {1251 ent->ref++;1252 return ent;1253 }1254 else if (((available_ix < 0) && (!ent || !ent->ref)) ||1255 ((0 <= available_ix) &&1256 (!ent && pbase_tree_cache[available_ix])))1257 available_ix = my_ix;1258 if (!ent)1259 break;1260 my_ix = pbase_tree_cache_ix_incr(my_ix);1261 }12621263 /* Did not find one. Either we got a bogus request or1264 * we need to read and perhaps cache.1265 */1266 data = read_object_file(oid, &type, &size);1267 if (!data)1268 return NULL;1269 if (type != OBJ_TREE) {1270 free(data);1271 return NULL;1272 }12731274 /* We need to either cache or return a throwaway copy */12751276 if (available_ix < 0)1277 ent = NULL;1278 else {1279 ent = pbase_tree_cache[available_ix];1280 my_ix = available_ix;1281 }12821283 if (!ent) {1284 nent = xmalloc(sizeof(*nent));1285 nent->temporary = (available_ix < 0);1286 }1287 else {1288 /* evict and reuse */1289 free(ent->tree_data);1290 nent = ent;1291 }1292 oidcpy(&nent->oid, oid);1293 nent->tree_data = data;1294 nent->tree_size = size;1295 nent->ref = 1;1296 if (!nent->temporary)1297 pbase_tree_cache[my_ix] = nent;1298 return nent;1299}13001301static void pbase_tree_put(struct pbase_tree_cache *cache)1302{1303 if (!cache->temporary) {1304 cache->ref--;1305 return;1306 }1307 free(cache->tree_data);1308 free(cache);1309}13101311static int name_cmp_len(const char *name)1312{1313 int i;1314 for (i = 0; name[i] && name[i] != '\n' && name[i] != '/'; i++)1315 ;1316 return i;1317}13181319static void add_pbase_object(struct tree_desc *tree,1320 const char *name,1321 int cmplen,1322 const char *fullname)1323{1324 struct name_entry entry;1325 int cmp;13261327 while (tree_entry(tree,&entry)) {1328 if (S_ISGITLINK(entry.mode))1329 continue;1330 cmp = tree_entry_len(&entry) != cmplen ? 1 :1331 memcmp(name, entry.path, cmplen);1332 if (cmp > 0)1333 continue;1334 if (cmp < 0)1335 return;1336 if (name[cmplen] != '/') {1337 add_object_entry(entry.oid,1338 object_type(entry.mode),1339 fullname, 1);1340 return;1341 }1342 if (S_ISDIR(entry.mode)) {1343 struct tree_desc sub;1344 struct pbase_tree_cache *tree;1345 const char *down = name+cmplen+1;1346 int downlen = name_cmp_len(down);13471348 tree = pbase_tree_get(entry.oid);1349 if (!tree)1350 return;1351 init_tree_desc(&sub, tree->tree_data, tree->tree_size);13521353 add_pbase_object(&sub, down, downlen, fullname);1354 pbase_tree_put(tree);1355 }1356 }1357}13581359static unsigned *done_pbase_paths;1360static int done_pbase_paths_num;1361static int done_pbase_paths_alloc;1362static int done_pbase_path_pos(unsigned hash)1363{1364 int lo = 0;1365 int hi = done_pbase_paths_num;1366 while (lo < hi) {1367 int mi = lo + (hi - lo) / 2;1368 if (done_pbase_paths[mi] == hash)1369 return mi;1370 if (done_pbase_paths[mi] < hash)1371 hi = mi;1372 else1373 lo = mi + 1;1374 }1375 return -lo-1;1376}13771378static int check_pbase_path(unsigned hash)1379{1380 int pos = done_pbase_path_pos(hash);1381 if (0 <= pos)1382 return 1;1383 pos = -pos - 1;1384 ALLOC_GROW(done_pbase_paths,1385 done_pbase_paths_num + 1,1386 done_pbase_paths_alloc);1387 done_pbase_paths_num++;1388 if (pos < done_pbase_paths_num)1389 MOVE_ARRAY(done_pbase_paths + pos + 1, done_pbase_paths + pos,1390 done_pbase_paths_num - pos - 1);1391 done_pbase_paths[pos] = hash;1392 return 0;1393}13941395static void add_preferred_base_object(const char *name)1396{1397 struct pbase_tree *it;1398 int cmplen;1399 unsigned hash = pack_name_hash(name);14001401 if (!num_preferred_base || check_pbase_path(hash))1402 return;14031404 cmplen = name_cmp_len(name);1405 for (it = pbase_tree; it; it = it->next) {1406 if (cmplen == 0) {1407 add_object_entry(&it->pcache.oid, OBJ_TREE, NULL, 1);1408 }1409 else {1410 struct tree_desc tree;1411 init_tree_desc(&tree, it->pcache.tree_data, it->pcache.tree_size);1412 add_pbase_object(&tree, name, cmplen, name);1413 }1414 }1415}14161417static void add_preferred_base(struct object_id *oid)1418{1419 struct pbase_tree *it;1420 void *data;1421 unsigned long size;1422 struct object_id tree_oid;14231424 if (window <= num_preferred_base++)1425 return;14261427 data = read_object_with_reference(oid, tree_type, &size, &tree_oid);1428 if (!data)1429 return;14301431 for (it = pbase_tree; it; it = it->next) {1432 if (oideq(&it->pcache.oid, &tree_oid)) {1433 free(data);1434 return;1435 }1436 }14371438 it = xcalloc(1, sizeof(*it));1439 it->next = pbase_tree;1440 pbase_tree = it;14411442 oidcpy(&it->pcache.oid, &tree_oid);1443 it->pcache.tree_data = data;1444 it->pcache.tree_size = size;1445}14461447static void cleanup_preferred_base(void)1448{1449 struct pbase_tree *it;1450 unsigned i;14511452 it = pbase_tree;1453 pbase_tree = NULL;1454 while (it) {1455 struct pbase_tree *tmp = it;1456 it = tmp->next;1457 free(tmp->pcache.tree_data);1458 free(tmp);1459 }14601461 for (i = 0; i < ARRAY_SIZE(pbase_tree_cache); i++) {1462 if (!pbase_tree_cache[i])1463 continue;1464 free(pbase_tree_cache[i]->tree_data);1465 FREE_AND_NULL(pbase_tree_cache[i]);1466 }14671468 FREE_AND_NULL(done_pbase_paths);1469 done_pbase_paths_num = done_pbase_paths_alloc = 0;1470}14711472/*1473 * Return 1 iff the object specified by "delta" can be sent1474 * literally as a delta against the base in "base_sha1". If1475 * so, then *base_out will point to the entry in our packing1476 * list, or NULL if we must use the external-base list.1477 *1478 * Depth value does not matter - find_deltas() will1479 * never consider reused delta as the base object to1480 * deltify other objects against, in order to avoid1481 * circular deltas.1482 */1483static int can_reuse_delta(const unsigned char *base_sha1,1484 struct object_entry *delta,1485 struct object_entry **base_out)1486{1487 struct object_entry *base;14881489 if (!base_sha1)1490 return 0;14911492 /*1493 * First see if we're already sending the base (or it's explicitly in1494 * our "excluded" list).1495 */1496 base = packlist_find(&to_pack, base_sha1, NULL);1497 if (base) {1498 if (!in_same_island(&delta->idx.oid, &base->idx.oid))1499 return 0;1500 *base_out = base;1501 return 1;1502 }15031504 /*1505 * Otherwise, reachability bitmaps may tell us if the receiver has it,1506 * even if it was buried too deep in history to make it into the1507 * packing list.1508 */1509 if (thin && bitmap_has_sha1_in_uninteresting(bitmap_git, base_sha1)) {1510 if (use_delta_islands) {1511 struct object_id base_oid;1512 hashcpy(base_oid.hash, base_sha1);1513 if (!in_same_island(&delta->idx.oid, &base_oid))1514 return 0;1515 }1516 *base_out = NULL;1517 return 1;1518 }15191520 return 0;1521}15221523static void check_object(struct object_entry *entry)1524{1525 unsigned long canonical_size;15261527 if (IN_PACK(entry)) {1528 struct packed_git *p = IN_PACK(entry);1529 struct pack_window *w_curs = NULL;1530 const unsigned char *base_ref = NULL;1531 struct object_entry *base_entry;1532 unsigned long used, used_0;1533 unsigned long avail;1534 off_t ofs;1535 unsigned char *buf, c;1536 enum object_type type;1537 unsigned long in_pack_size;15381539 buf = use_pack(p, &w_curs, entry->in_pack_offset, &avail);15401541 /*1542 * We want in_pack_type even if we do not reuse delta1543 * since non-delta representations could still be reused.1544 */1545 used = unpack_object_header_buffer(buf, avail,1546 &type,1547 &in_pack_size);1548 if (used == 0)1549 goto give_up;15501551 if (type < 0)1552 BUG("invalid type %d", type);1553 entry->in_pack_type = type;15541555 /*1556 * Determine if this is a delta and if so whether we can1557 * reuse it or not. Otherwise let's find out as cheaply as1558 * possible what the actual type and size for this object is.1559 */1560 switch (entry->in_pack_type) {1561 default:1562 /* Not a delta hence we've already got all we need. */1563 oe_set_type(entry, entry->in_pack_type);1564 SET_SIZE(entry, in_pack_size);1565 entry->in_pack_header_size = used;1566 if (oe_type(entry) < OBJ_COMMIT || oe_type(entry) > OBJ_BLOB)1567 goto give_up;1568 unuse_pack(&w_curs);1569 return;1570 case OBJ_REF_DELTA:1571 if (reuse_delta && !entry->preferred_base)1572 base_ref = use_pack(p, &w_curs,1573 entry->in_pack_offset + used, NULL);1574 entry->in_pack_header_size = used + the_hash_algo->rawsz;1575 break;1576 case OBJ_OFS_DELTA:1577 buf = use_pack(p, &w_curs,1578 entry->in_pack_offset + used, NULL);1579 used_0 = 0;1580 c = buf[used_0++];1581 ofs = c & 127;1582 while (c & 128) {1583 ofs += 1;1584 if (!ofs || MSB(ofs, 7)) {1585 error(_("delta base offset overflow in pack for %s"),1586 oid_to_hex(&entry->idx.oid));1587 goto give_up;1588 }1589 c = buf[used_0++];1590 ofs = (ofs << 7) + (c & 127);1591 }1592 ofs = entry->in_pack_offset - ofs;1593 if (ofs <= 0 || ofs >= entry->in_pack_offset) {1594 error(_("delta base offset out of bound for %s"),1595 oid_to_hex(&entry->idx.oid));1596 goto give_up;1597 }1598 if (reuse_delta && !entry->preferred_base) {1599 struct revindex_entry *revidx;1600 revidx = find_pack_revindex(p, ofs);1601 if (!revidx)1602 goto give_up;1603 base_ref = nth_packed_object_sha1(p, revidx->nr);1604 }1605 entry->in_pack_header_size = used + used_0;1606 break;1607 }16081609 if (can_reuse_delta(base_ref, entry, &base_entry)) {1610 oe_set_type(entry, entry->in_pack_type);1611 SET_SIZE(entry, in_pack_size); /* delta size */1612 SET_DELTA_SIZE(entry, in_pack_size);16131614 if (base_entry) {1615 SET_DELTA(entry, base_entry);1616 entry->delta_sibling_idx = base_entry->delta_child_idx;1617 SET_DELTA_CHILD(base_entry, entry);1618 } else {1619 SET_DELTA_EXT(entry, base_ref);1620 }16211622 unuse_pack(&w_curs);1623 return;1624 }16251626 if (oe_type(entry)) {1627 off_t delta_pos;16281629 /*1630 * This must be a delta and we already know what the1631 * final object type is. Let's extract the actual1632 * object size from the delta header.1633 */1634 delta_pos = entry->in_pack_offset + entry->in_pack_header_size;1635 canonical_size = get_size_from_delta(p, &w_curs, delta_pos);1636 if (canonical_size == 0)1637 goto give_up;1638 SET_SIZE(entry, canonical_size);1639 unuse_pack(&w_curs);1640 return;1641 }16421643 /*1644 * No choice but to fall back to the recursive delta walk1645 * with sha1_object_info() to find about the object type1646 * at this point...1647 */1648 give_up:1649 unuse_pack(&w_curs);1650 }16511652 oe_set_type(entry,1653 oid_object_info(the_repository, &entry->idx.oid, &canonical_size));1654 if (entry->type_valid) {1655 SET_SIZE(entry, canonical_size);1656 } else {1657 /*1658 * Bad object type is checked in prepare_pack(). This is1659 * to permit a missing preferred base object to be ignored1660 * as a preferred base. Doing so can result in a larger1661 * pack file, but the transfer will still take place.1662 */1663 }1664}16651666static int pack_offset_sort(const void *_a, const void *_b)1667{1668 const struct object_entry *a = *(struct object_entry **)_a;1669 const struct object_entry *b = *(struct object_entry **)_b;1670 const struct packed_git *a_in_pack = IN_PACK(a);1671 const struct packed_git *b_in_pack = IN_PACK(b);16721673 /* avoid filesystem trashing with loose objects */1674 if (!a_in_pack && !b_in_pack)1675 return oidcmp(&a->idx.oid, &b->idx.oid);16761677 if (a_in_pack < b_in_pack)1678 return -1;1679 if (a_in_pack > b_in_pack)1680 return 1;1681 return a->in_pack_offset < b->in_pack_offset ? -1 :1682 (a->in_pack_offset > b->in_pack_offset);1683}16841685/*1686 * Drop an on-disk delta we were planning to reuse. Naively, this would1687 * just involve blanking out the "delta" field, but we have to deal1688 * with some extra book-keeping:1689 *1690 * 1. Removing ourselves from the delta_sibling linked list.1691 *1692 * 2. Updating our size/type to the non-delta representation. These were1693 * either not recorded initially (size) or overwritten with the delta type1694 * (type) when check_object() decided to reuse the delta.1695 *1696 * 3. Resetting our delta depth, as we are now a base object.1697 */1698static void drop_reused_delta(struct object_entry *entry)1699{1700 unsigned *idx = &to_pack.objects[entry->delta_idx - 1].delta_child_idx;1701 struct object_info oi = OBJECT_INFO_INIT;1702 enum object_type type;1703 unsigned long size;17041705 while (*idx) {1706 struct object_entry *oe = &to_pack.objects[*idx - 1];17071708 if (oe == entry)1709 *idx = oe->delta_sibling_idx;1710 else1711 idx = &oe->delta_sibling_idx;1712 }1713 SET_DELTA(entry, NULL);1714 entry->depth = 0;17151716 oi.sizep = &size;1717 oi.typep = &type;1718 if (packed_object_info(the_repository, IN_PACK(entry), entry->in_pack_offset, &oi) < 0) {1719 /*1720 * We failed to get the info from this pack for some reason;1721 * fall back to sha1_object_info, which may find another copy.1722 * And if that fails, the error will be recorded in oe_type(entry)1723 * and dealt with in prepare_pack().1724 */1725 oe_set_type(entry,1726 oid_object_info(the_repository, &entry->idx.oid, &size));1727 } else {1728 oe_set_type(entry, type);1729 }1730 SET_SIZE(entry, size);1731}17321733/*1734 * Follow the chain of deltas from this entry onward, throwing away any links1735 * that cause us to hit a cycle (as determined by the DFS state flags in1736 * the entries).1737 *1738 * We also detect too-long reused chains that would violate our --depth1739 * limit.1740 */1741static void break_delta_chains(struct object_entry *entry)1742{1743 /*1744 * The actual depth of each object we will write is stored as an int,1745 * as it cannot exceed our int "depth" limit. But before we break1746 * changes based no that limit, we may potentially go as deep as the1747 * number of objects, which is elsewhere bounded to a uint32_t.1748 */1749 uint32_t total_depth;1750 struct object_entry *cur, *next;17511752 for (cur = entry, total_depth = 0;1753 cur;1754 cur = DELTA(cur), total_depth++) {1755 if (cur->dfs_state == DFS_DONE) {1756 /*1757 * We've already seen this object and know it isn't1758 * part of a cycle. We do need to append its depth1759 * to our count.1760 */1761 total_depth += cur->depth;1762 break;1763 }17641765 /*1766 * We break cycles before looping, so an ACTIVE state (or any1767 * other cruft which made its way into the state variable)1768 * is a bug.1769 */1770 if (cur->dfs_state != DFS_NONE)1771 BUG("confusing delta dfs state in first pass: %d",1772 cur->dfs_state);17731774 /*1775 * Now we know this is the first time we've seen the object. If1776 * it's not a delta, we're done traversing, but we'll mark it1777 * done to save time on future traversals.1778 */1779 if (!DELTA(cur)) {1780 cur->dfs_state = DFS_DONE;1781 break;1782 }17831784 /*1785 * Mark ourselves as active and see if the next step causes1786 * us to cycle to another active object. It's important to do1787 * this _before_ we loop, because it impacts where we make the1788 * cut, and thus how our total_depth counter works.1789 * E.g., We may see a partial loop like:1790 *1791 * A -> B -> C -> D -> B1792 *1793 * Cutting B->C breaks the cycle. But now the depth of A is1794 * only 1, and our total_depth counter is at 3. The size of the1795 * error is always one less than the size of the cycle we1796 * broke. Commits C and D were "lost" from A's chain.1797 *1798 * If we instead cut D->B, then the depth of A is correct at 3.1799 * We keep all commits in the chain that we examined.1800 */1801 cur->dfs_state = DFS_ACTIVE;1802 if (DELTA(cur)->dfs_state == DFS_ACTIVE) {1803 drop_reused_delta(cur);1804 cur->dfs_state = DFS_DONE;1805 break;1806 }1807 }18081809 /*1810 * And now that we've gone all the way to the bottom of the chain, we1811 * need to clear the active flags and set the depth fields as1812 * appropriate. Unlike the loop above, which can quit when it drops a1813 * delta, we need to keep going to look for more depth cuts. So we need1814 * an extra "next" pointer to keep going after we reset cur->delta.1815 */1816 for (cur = entry; cur; cur = next) {1817 next = DELTA(cur);18181819 /*1820 * We should have a chain of zero or more ACTIVE states down to1821 * a final DONE. We can quit after the DONE, because either it1822 * has no bases, or we've already handled them in a previous1823 * call.1824 */1825 if (cur->dfs_state == DFS_DONE)1826 break;1827 else if (cur->dfs_state != DFS_ACTIVE)1828 BUG("confusing delta dfs state in second pass: %d",1829 cur->dfs_state);18301831 /*1832 * If the total_depth is more than depth, then we need to snip1833 * the chain into two or more smaller chains that don't exceed1834 * the maximum depth. Most of the resulting chains will contain1835 * (depth + 1) entries (i.e., depth deltas plus one base), and1836 * the last chain (i.e., the one containing entry) will contain1837 * whatever entries are left over, namely1838 * (total_depth % (depth + 1)) of them.1839 *1840 * Since we are iterating towards decreasing depth, we need to1841 * decrement total_depth as we go, and we need to write to the1842 * entry what its final depth will be after all of the1843 * snipping. Since we're snipping into chains of length (depth1844 * + 1) entries, the final depth of an entry will be its1845 * original depth modulo (depth + 1). Any time we encounter an1846 * entry whose final depth is supposed to be zero, we snip it1847 * from its delta base, thereby making it so.1848 */1849 cur->depth = (total_depth--) % (depth + 1);1850 if (!cur->depth)1851 drop_reused_delta(cur);18521853 cur->dfs_state = DFS_DONE;1854 }1855}18561857static void get_object_details(void)1858{1859 uint32_t i;1860 struct object_entry **sorted_by_offset;18611862 if (progress)1863 progress_state = start_progress(_("Counting objects"),1864 to_pack.nr_objects);18651866 sorted_by_offset = xcalloc(to_pack.nr_objects, sizeof(struct object_entry *));1867 for (i = 0; i < to_pack.nr_objects; i++)1868 sorted_by_offset[i] = to_pack.objects + i;1869 QSORT(sorted_by_offset, to_pack.nr_objects, pack_offset_sort);18701871 for (i = 0; i < to_pack.nr_objects; i++) {1872 struct object_entry *entry = sorted_by_offset[i];1873 check_object(entry);1874 if (entry->type_valid &&1875 oe_size_greater_than(&to_pack, entry, big_file_threshold))1876 entry->no_try_delta = 1;1877 display_progress(progress_state, i + 1);1878 }1879 stop_progress(&progress_state);18801881 /*1882 * This must happen in a second pass, since we rely on the delta1883 * information for the whole list being completed.1884 */1885 for (i = 0; i < to_pack.nr_objects; i++)1886 break_delta_chains(&to_pack.objects[i]);18871888 free(sorted_by_offset);1889}18901891/*1892 * We search for deltas in a list sorted by type, by filename hash, and then1893 * by size, so that we see progressively smaller and smaller files.1894 * That's because we prefer deltas to be from the bigger file1895 * to the smaller -- deletes are potentially cheaper, but perhaps1896 * more importantly, the bigger file is likely the more recent1897 * one. The deepest deltas are therefore the oldest objects which are1898 * less susceptible to be accessed often.1899 */1900static int type_size_sort(const void *_a, const void *_b)1901{1902 const struct object_entry *a = *(struct object_entry **)_a;1903 const struct object_entry *b = *(struct object_entry **)_b;1904 enum object_type a_type = oe_type(a);1905 enum object_type b_type = oe_type(b);1906 unsigned long a_size = SIZE(a);1907 unsigned long b_size = SIZE(b);19081909 if (a_type > b_type)1910 return -1;1911 if (a_type < b_type)1912 return 1;1913 if (a->hash > b->hash)1914 return -1;1915 if (a->hash < b->hash)1916 return 1;1917 if (a->preferred_base > b->preferred_base)1918 return -1;1919 if (a->preferred_base < b->preferred_base)1920 return 1;1921 if (use_delta_islands) {1922 int island_cmp = island_delta_cmp(&a->idx.oid, &b->idx.oid);1923 if (island_cmp)1924 return island_cmp;1925 }1926 if (a_size > b_size)1927 return -1;1928 if (a_size < b_size)1929 return 1;1930 return a < b ? -1 : (a > b); /* newest first */1931}19321933struct unpacked {1934 struct object_entry *entry;1935 void *data;1936 struct delta_index *index;1937 unsigned depth;1938};19391940static int delta_cacheable(unsigned long src_size, unsigned long trg_size,1941 unsigned long delta_size)1942{1943 if (max_delta_cache_size && delta_cache_size + delta_size > max_delta_cache_size)1944 return 0;19451946 if (delta_size < cache_max_small_delta_size)1947 return 1;19481949 /* cache delta, if objects are large enough compared to delta size */1950 if ((src_size >> 20) + (trg_size >> 21) > (delta_size >> 10))1951 return 1;19521953 return 0;1954}19551956/* Protect access to object database */1957static pthread_mutex_t read_mutex;1958#define read_lock() pthread_mutex_lock(&read_mutex)1959#define read_unlock() pthread_mutex_unlock(&read_mutex)19601961/* Protect delta_cache_size */1962static pthread_mutex_t cache_mutex;1963#define cache_lock() pthread_mutex_lock(&cache_mutex)1964#define cache_unlock() pthread_mutex_unlock(&cache_mutex)19651966/*1967 * Protect object list partitioning (e.g. struct thread_param) and1968 * progress_state1969 */1970static pthread_mutex_t progress_mutex;1971#define progress_lock() pthread_mutex_lock(&progress_mutex)1972#define progress_unlock() pthread_mutex_unlock(&progress_mutex)19731974/*1975 * Access to struct object_entry is unprotected since each thread owns1976 * a portion of the main object list. Just don't access object entries1977 * ahead in the list because they can be stolen and would need1978 * progress_mutex for protection.1979 */19801981/*1982 * Return the size of the object without doing any delta1983 * reconstruction (so non-deltas are true object sizes, but deltas1984 * return the size of the delta data).1985 */1986unsigned long oe_get_size_slow(struct packing_data *pack,1987 const struct object_entry *e)1988{1989 struct packed_git *p;1990 struct pack_window *w_curs;1991 unsigned char *buf;1992 enum object_type type;1993 unsigned long used, avail, size;19941995 if (e->type_ != OBJ_OFS_DELTA && e->type_ != OBJ_REF_DELTA) {1996 read_lock();1997 if (oid_object_info(the_repository, &e->idx.oid, &size) < 0)1998 die(_("unable to get size of %s"),1999 oid_to_hex(&e->idx.oid));2000 read_unlock();2001 return size;2002 }20032004 p = oe_in_pack(pack, e);2005 if (!p)2006 BUG("when e->type is a delta, it must belong to a pack");20072008 read_lock();2009 w_curs = NULL;2010 buf = use_pack(p, &w_curs, e->in_pack_offset, &avail);2011 used = unpack_object_header_buffer(buf, avail, &type, &size);2012 if (used == 0)2013 die(_("unable to parse object header of %s"),2014 oid_to_hex(&e->idx.oid));20152016 unuse_pack(&w_curs);2017 read_unlock();2018 return size;2019}20202021static int try_delta(struct unpacked *trg, struct unpacked *src,2022 unsigned max_depth, unsigned long *mem_usage)2023{2024 struct object_entry *trg_entry = trg->entry;2025 struct object_entry *src_entry = src->entry;2026 unsigned long trg_size, src_size, delta_size, sizediff, max_size, sz;2027 unsigned ref_depth;2028 enum object_type type;2029 void *delta_buf;20302031 /* Don't bother doing diffs between different types */2032 if (oe_type(trg_entry) != oe_type(src_entry))2033 return -1;20342035 /*2036 * We do not bother to try a delta that we discarded on an2037 * earlier try, but only when reusing delta data. Note that2038 * src_entry that is marked as the preferred_base should always2039 * be considered, as even if we produce a suboptimal delta against2040 * it, we will still save the transfer cost, as we already know2041 * the other side has it and we won't send src_entry at all.2042 */2043 if (reuse_delta && IN_PACK(trg_entry) &&2044 IN_PACK(trg_entry) == IN_PACK(src_entry) &&2045 !src_entry->preferred_base &&2046 trg_entry->in_pack_type != OBJ_REF_DELTA &&2047 trg_entry->in_pack_type != OBJ_OFS_DELTA)2048 return 0;20492050 /* Let's not bust the allowed depth. */2051 if (src->depth >= max_depth)2052 return 0;20532054 /* Now some size filtering heuristics. */2055 trg_size = SIZE(trg_entry);2056 if (!DELTA(trg_entry)) {2057 max_size = trg_size/2 - the_hash_algo->rawsz;2058 ref_depth = 1;2059 } else {2060 max_size = DELTA_SIZE(trg_entry);2061 ref_depth = trg->depth;2062 }2063 max_size = (uint64_t)max_size * (max_depth - src->depth) /2064 (max_depth - ref_depth + 1);2065 if (max_size == 0)2066 return 0;2067 src_size = SIZE(src_entry);2068 sizediff = src_size < trg_size ? trg_size - src_size : 0;2069 if (sizediff >= max_size)2070 return 0;2071 if (trg_size < src_size / 32)2072 return 0;20732074 if (!in_same_island(&trg->entry->idx.oid, &src->entry->idx.oid))2075 return 0;20762077 /* Load data if not already done */2078 if (!trg->data) {2079 read_lock();2080 trg->data = read_object_file(&trg_entry->idx.oid, &type, &sz);2081 read_unlock();2082 if (!trg->data)2083 die(_("object %s cannot be read"),2084 oid_to_hex(&trg_entry->idx.oid));2085 if (sz != trg_size)2086 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2087 oid_to_hex(&trg_entry->idx.oid), (uintmax_t)sz,2088 (uintmax_t)trg_size);2089 *mem_usage += sz;2090 }2091 if (!src->data) {2092 read_lock();2093 src->data = read_object_file(&src_entry->idx.oid, &type, &sz);2094 read_unlock();2095 if (!src->data) {2096 if (src_entry->preferred_base) {2097 static int warned = 0;2098 if (!warned++)2099 warning(_("object %s cannot be read"),2100 oid_to_hex(&src_entry->idx.oid));2101 /*2102 * Those objects are not included in the2103 * resulting pack. Be resilient and ignore2104 * them if they can't be read, in case the2105 * pack could be created nevertheless.2106 */2107 return 0;2108 }2109 die(_("object %s cannot be read"),2110 oid_to_hex(&src_entry->idx.oid));2111 }2112 if (sz != src_size)2113 die(_("object %s inconsistent object length (%"PRIuMAX" vs %"PRIuMAX")"),2114 oid_to_hex(&src_entry->idx.oid), (uintmax_t)sz,2115 (uintmax_t)src_size);2116 *mem_usage += sz;2117 }2118 if (!src->index) {2119 src->index = create_delta_index(src->data, src_size);2120 if (!src->index) {2121 static int warned = 0;2122 if (!warned++)2123 warning(_("suboptimal pack - out of memory"));2124 return 0;2125 }2126 *mem_usage += sizeof_delta_index(src->index);2127 }21282129 delta_buf = create_delta(src->index, trg->data, trg_size, &delta_size, max_size);2130 if (!delta_buf)2131 return 0;21322133 if (DELTA(trg_entry)) {2134 /* Prefer only shallower same-sized deltas. */2135 if (delta_size == DELTA_SIZE(trg_entry) &&2136 src->depth + 1 >= trg->depth) {2137 free(delta_buf);2138 return 0;2139 }2140 }21412142 /*2143 * Handle memory allocation outside of the cache2144 * accounting lock. Compiler will optimize the strangeness2145 * away when NO_PTHREADS is defined.2146 */2147 free(trg_entry->delta_data);2148 cache_lock();2149 if (trg_entry->delta_data) {2150 delta_cache_size -= DELTA_SIZE(trg_entry);2151 trg_entry->delta_data = NULL;2152 }2153 if (delta_cacheable(src_size, trg_size, delta_size)) {2154 delta_cache_size += delta_size;2155 cache_unlock();2156 trg_entry->delta_data = xrealloc(delta_buf, delta_size);2157 } else {2158 cache_unlock();2159 free(delta_buf);2160 }21612162 SET_DELTA(trg_entry, src_entry);2163 SET_DELTA_SIZE(trg_entry, delta_size);2164 trg->depth = src->depth + 1;21652166 return 1;2167}21682169static unsigned int check_delta_limit(struct object_entry *me, unsigned int n)2170{2171 struct object_entry *child = DELTA_CHILD(me);2172 unsigned int m = n;2173 while (child) {2174 unsigned int c = check_delta_limit(child, n + 1);2175 if (m < c)2176 m = c;2177 child = DELTA_SIBLING(child);2178 }2179 return m;2180}21812182static unsigned long free_unpacked(struct unpacked *n)2183{2184 unsigned long freed_mem = sizeof_delta_index(n->index);2185 free_delta_index(n->index);2186 n->index = NULL;2187 if (n->data) {2188 freed_mem += SIZE(n->entry);2189 FREE_AND_NULL(n->data);2190 }2191 n->entry = NULL;2192 n->depth = 0;2193 return freed_mem;2194}21952196static void find_deltas(struct object_entry **list, unsigned *list_size,2197 int window, int depth, unsigned *processed)2198{2199 uint32_t i, idx = 0, count = 0;2200 struct unpacked *array;2201 unsigned long mem_usage = 0;22022203 array = xcalloc(window, sizeof(struct unpacked));22042205 for (;;) {2206 struct object_entry *entry;2207 struct unpacked *n = array + idx;2208 int j, max_depth, best_base = -1;22092210 progress_lock();2211 if (!*list_size) {2212 progress_unlock();2213 break;2214 }2215 entry = *list++;2216 (*list_size)--;2217 if (!entry->preferred_base) {2218 (*processed)++;2219 display_progress(progress_state, *processed);2220 }2221 progress_unlock();22222223 mem_usage -= free_unpacked(n);2224 n->entry = entry;22252226 while (window_memory_limit &&2227 mem_usage > window_memory_limit &&2228 count > 1) {2229 uint32_t tail = (idx + window - count) % window;2230 mem_usage -= free_unpacked(array + tail);2231 count--;2232 }22332234 /* We do not compute delta to *create* objects we are not2235 * going to pack.2236 */2237 if (entry->preferred_base)2238 goto next;22392240 /*2241 * If the current object is at pack edge, take the depth the2242 * objects that depend on the current object into account2243 * otherwise they would become too deep.2244 */2245 max_depth = depth;2246 if (DELTA_CHILD(entry)) {2247 max_depth -= check_delta_limit(entry, 0);2248 if (max_depth <= 0)2249 goto next;2250 }22512252 j = window;2253 while (--j > 0) {2254 int ret;2255 uint32_t other_idx = idx + j;2256 struct unpacked *m;2257 if (other_idx >= window)2258 other_idx -= window;2259 m = array + other_idx;2260 if (!m->entry)2261 break;2262 ret = try_delta(n, m, max_depth, &mem_usage);2263 if (ret < 0)2264 break;2265 else if (ret > 0)2266 best_base = other_idx;2267 }22682269 /*2270 * If we decided to cache the delta data, then it is best2271 * to compress it right away. First because we have to do2272 * it anyway, and doing it here while we're threaded will2273 * save a lot of time in the non threaded write phase,2274 * as well as allow for caching more deltas within2275 * the same cache size limit.2276 * ...2277 * But only if not writing to stdout, since in that case2278 * the network is most likely throttling writes anyway,2279 * and therefore it is best to go to the write phase ASAP2280 * instead, as we can afford spending more time compressing2281 * between writes at that moment.2282 */2283 if (entry->delta_data && !pack_to_stdout) {2284 unsigned long size;22852286 size = do_compress(&entry->delta_data, DELTA_SIZE(entry));2287 if (size < (1U << OE_Z_DELTA_BITS)) {2288 entry->z_delta_size = size;2289 cache_lock();2290 delta_cache_size -= DELTA_SIZE(entry);2291 delta_cache_size += entry->z_delta_size;2292 cache_unlock();2293 } else {2294 FREE_AND_NULL(entry->delta_data);2295 entry->z_delta_size = 0;2296 }2297 }22982299 /* if we made n a delta, and if n is already at max2300 * depth, leaving it in the window is pointless. we2301 * should evict it first.2302 */2303 if (DELTA(entry) && max_depth <= n->depth)2304 continue;23052306 /*2307 * Move the best delta base up in the window, after the2308 * currently deltified object, to keep it longer. It will2309 * be the first base object to be attempted next.2310 */2311 if (DELTA(entry)) {2312 struct unpacked swap = array[best_base];2313 int dist = (window + idx - best_base) % window;2314 int dst = best_base;2315 while (dist--) {2316 int src = (dst + 1) % window;2317 array[dst] = array[src];2318 dst = src;2319 }2320 array[dst] = swap;2321 }23222323 next:2324 idx++;2325 if (count + 1 < window)2326 count++;2327 if (idx >= window)2328 idx = 0;2329 }23302331 for (i = 0; i < window; ++i) {2332 free_delta_index(array[i].index);2333 free(array[i].data);2334 }2335 free(array);2336}23372338static void try_to_free_from_threads(size_t size)2339{2340 read_lock();2341 release_pack_memory(size);2342 read_unlock();2343}23442345static try_to_free_t old_try_to_free_routine;23462347/*2348 * The main object list is split into smaller lists, each is handed to2349 * one worker.2350 *2351 * The main thread waits on the condition that (at least) one of the workers2352 * has stopped working (which is indicated in the .working member of2353 * struct thread_params).2354 *2355 * When a work thread has completed its work, it sets .working to 0 and2356 * signals the main thread and waits on the condition that .data_ready2357 * becomes 1.2358 *2359 * The main thread steals half of the work from the worker that has2360 * most work left to hand it to the idle worker.2361 */23622363struct thread_params {2364 pthread_t thread;2365 struct object_entry **list;2366 unsigned list_size;2367 unsigned remaining;2368 int window;2369 int depth;2370 int working;2371 int data_ready;2372 pthread_mutex_t mutex;2373 pthread_cond_t cond;2374 unsigned *processed;2375};23762377static pthread_cond_t progress_cond;23782379/*2380 * Mutex and conditional variable can't be statically-initialized on Windows.2381 */2382static void init_threaded_search(void)2383{2384 init_recursive_mutex(&read_mutex);2385 pthread_mutex_init(&cache_mutex, NULL);2386 pthread_mutex_init(&progress_mutex, NULL);2387 pthread_cond_init(&progress_cond, NULL);2388 old_try_to_free_routine = set_try_to_free_routine(try_to_free_from_threads);2389}23902391static void cleanup_threaded_search(void)2392{2393 set_try_to_free_routine(old_try_to_free_routine);2394 pthread_cond_destroy(&progress_cond);2395 pthread_mutex_destroy(&read_mutex);2396 pthread_mutex_destroy(&cache_mutex);2397 pthread_mutex_destroy(&progress_mutex);2398}23992400static void *threaded_find_deltas(void *arg)2401{2402 struct thread_params *me = arg;24032404 progress_lock();2405 while (me->remaining) {2406 progress_unlock();24072408 find_deltas(me->list, &me->remaining,2409 me->window, me->depth, me->processed);24102411 progress_lock();2412 me->working = 0;2413 pthread_cond_signal(&progress_cond);2414 progress_unlock();24152416 /*2417 * We must not set ->data_ready before we wait on the2418 * condition because the main thread may have set it to 12419 * before we get here. In order to be sure that new2420 * work is available if we see 1 in ->data_ready, it2421 * was initialized to 0 before this thread was spawned2422 * and we reset it to 0 right away.2423 */2424 pthread_mutex_lock(&me->mutex);2425 while (!me->data_ready)2426 pthread_cond_wait(&me->cond, &me->mutex);2427 me->data_ready = 0;2428 pthread_mutex_unlock(&me->mutex);24292430 progress_lock();2431 }2432 progress_unlock();2433 /* leave ->working 1 so that this doesn't get more work assigned */2434 return NULL;2435}24362437static void ll_find_deltas(struct object_entry **list, unsigned list_size,2438 int window, int depth, unsigned *processed)2439{2440 struct thread_params *p;2441 int i, ret, active_threads = 0;24422443 init_threaded_search();24442445 if (delta_search_threads <= 1) {2446 find_deltas(list, &list_size, window, depth, processed);2447 cleanup_threaded_search();2448 return;2449 }2450 if (progress > pack_to_stdout)2451 fprintf_ln(stderr, _("Delta compression using up to %d threads"),2452 delta_search_threads);2453 p = xcalloc(delta_search_threads, sizeof(*p));24542455 /* Partition the work amongst work threads. */2456 for (i = 0; i < delta_search_threads; i++) {2457 unsigned sub_size = list_size / (delta_search_threads - i);24582459 /* don't use too small segments or no deltas will be found */2460 if (sub_size < 2*window && i+1 < delta_search_threads)2461 sub_size = 0;24622463 p[i].window = window;2464 p[i].depth = depth;2465 p[i].processed = processed;2466 p[i].working = 1;2467 p[i].data_ready = 0;24682469 /* try to split chunks on "path" boundaries */2470 while (sub_size && sub_size < list_size &&2471 list[sub_size]->hash &&2472 list[sub_size]->hash == list[sub_size-1]->hash)2473 sub_size++;24742475 p[i].list = list;2476 p[i].list_size = sub_size;2477 p[i].remaining = sub_size;24782479 list += sub_size;2480 list_size -= sub_size;2481 }24822483 /* Start work threads. */2484 for (i = 0; i < delta_search_threads; i++) {2485 if (!p[i].list_size)2486 continue;2487 pthread_mutex_init(&p[i].mutex, NULL);2488 pthread_cond_init(&p[i].cond, NULL);2489 ret = pthread_create(&p[i].thread, NULL,2490 threaded_find_deltas, &p[i]);2491 if (ret)2492 die(_("unable to create thread: %s"), strerror(ret));2493 active_threads++;2494 }24952496 /*2497 * Now let's wait for work completion. Each time a thread is done2498 * with its work, we steal half of the remaining work from the2499 * thread with the largest number of unprocessed objects and give2500 * it to that newly idle thread. This ensure good load balancing2501 * until the remaining object list segments are simply too short2502 * to be worth splitting anymore.2503 */2504 while (active_threads) {2505 struct thread_params *target = NULL;2506 struct thread_params *victim = NULL;2507 unsigned sub_size = 0;25082509 progress_lock();2510 for (;;) {2511 for (i = 0; !target && i < delta_search_threads; i++)2512 if (!p[i].working)2513 target = &p[i];2514 if (target)2515 break;2516 pthread_cond_wait(&progress_cond, &progress_mutex);2517 }25182519 for (i = 0; i < delta_search_threads; i++)2520 if (p[i].remaining > 2*window &&2521 (!victim || victim->remaining < p[i].remaining))2522 victim = &p[i];2523 if (victim) {2524 sub_size = victim->remaining / 2;2525 list = victim->list + victim->list_size - sub_size;2526 while (sub_size && list[0]->hash &&2527 list[0]->hash == list[-1]->hash) {2528 list++;2529 sub_size--;2530 }2531 if (!sub_size) {2532 /*2533 * It is possible for some "paths" to have2534 * so many objects that no hash boundary2535 * might be found. Let's just steal the2536 * exact half in that case.2537 */2538 sub_size = victim->remaining / 2;2539 list -= sub_size;2540 }2541 target->list = list;2542 victim->list_size -= sub_size;2543 victim->remaining -= sub_size;2544 }2545 target->list_size = sub_size;2546 target->remaining = sub_size;2547 target->working = 1;2548 progress_unlock();25492550 pthread_mutex_lock(&target->mutex);2551 target->data_ready = 1;2552 pthread_cond_signal(&target->cond);2553 pthread_mutex_unlock(&target->mutex);25542555 if (!sub_size) {2556 pthread_join(target->thread, NULL);2557 pthread_cond_destroy(&target->cond);2558 pthread_mutex_destroy(&target->mutex);2559 active_threads--;2560 }2561 }2562 cleanup_threaded_search();2563 free(p);2564}25652566static void add_tag_chain(const struct object_id *oid)2567{2568 struct tag *tag;25692570 /*2571 * We catch duplicates already in add_object_entry(), but we'd2572 * prefer to do this extra check to avoid having to parse the2573 * tag at all if we already know that it's being packed (e.g., if2574 * it was included via bitmaps, we would not have parsed it2575 * previously).2576 */2577 if (packlist_find(&to_pack, oid->hash, NULL))2578 return;25792580 tag = lookup_tag(the_repository, oid);2581 while (1) {2582 if (!tag || parse_tag(tag) || !tag->tagged)2583 die(_("unable to pack objects reachable from tag %s"),2584 oid_to_hex(oid));25852586 add_object_entry(&tag->object.oid, OBJ_TAG, NULL, 0);25872588 if (tag->tagged->type != OBJ_TAG)2589 return;25902591 tag = (struct tag *)tag->tagged;2592 }2593}25942595static int add_ref_tag(const char *path, const struct object_id *oid, int flag, void *cb_data)2596{2597 struct object_id peeled;25982599 if (starts_with(path, "refs/tags/") && /* is a tag? */2600 !peel_ref(path, &peeled) && /* peelable? */2601 packlist_find(&to_pack, peeled.hash, NULL)) /* object packed? */2602 add_tag_chain(oid);2603 return 0;2604}26052606static void prepare_pack(int window, int depth)2607{2608 struct object_entry **delta_list;2609 uint32_t i, nr_deltas;2610 unsigned n;26112612 if (use_delta_islands)2613 resolve_tree_islands(progress, &to_pack);26142615 get_object_details();26162617 /*2618 * If we're locally repacking then we need to be doubly careful2619 * from now on in order to make sure no stealth corruption gets2620 * propagated to the new pack. Clients receiving streamed packs2621 * should validate everything they get anyway so no need to incur2622 * the additional cost here in that case.2623 */2624 if (!pack_to_stdout)2625 do_check_packed_object_crc = 1;26262627 if (!to_pack.nr_objects || !window || !depth)2628 return;26292630 ALLOC_ARRAY(delta_list, to_pack.nr_objects);2631 nr_deltas = n = 0;26322633 for (i = 0; i < to_pack.nr_objects; i++) {2634 struct object_entry *entry = to_pack.objects + i;26352636 if (DELTA(entry))2637 /* This happens if we decided to reuse existing2638 * delta from a pack. "reuse_delta &&" is implied.2639 */2640 continue;26412642 if (!entry->type_valid ||2643 oe_size_less_than(&to_pack, entry, 50))2644 continue;26452646 if (entry->no_try_delta)2647 continue;26482649 if (!entry->preferred_base) {2650 nr_deltas++;2651 if (oe_type(entry) < 0)2652 die(_("unable to get type of object %s"),2653 oid_to_hex(&entry->idx.oid));2654 } else {2655 if (oe_type(entry) < 0) {2656 /*2657 * This object is not found, but we2658 * don't have to include it anyway.2659 */2660 continue;2661 }2662 }26632664 delta_list[n++] = entry;2665 }26662667 if (nr_deltas && n > 1) {2668 unsigned nr_done = 0;2669 if (progress)2670 progress_state = start_progress(_("Compressing objects"),2671 nr_deltas);2672 QSORT(delta_list, n, type_size_sort);2673 ll_find_deltas(delta_list, n, window+1, depth, &nr_done);2674 stop_progress(&progress_state);2675 if (nr_done != nr_deltas)2676 die(_("inconsistency with delta count"));2677 }2678 free(delta_list);2679}26802681static int git_pack_config(const char *k, const char *v, void *cb)2682{2683 if (!strcmp(k, "pack.window")) {2684 window = git_config_int(k, v);2685 return 0;2686 }2687 if (!strcmp(k, "pack.windowmemory")) {2688 window_memory_limit = git_config_ulong(k, v);2689 return 0;2690 }2691 if (!strcmp(k, "pack.depth")) {2692 depth = git_config_int(k, v);2693 return 0;2694 }2695 if (!strcmp(k, "pack.deltacachesize")) {2696 max_delta_cache_size = git_config_int(k, v);2697 return 0;2698 }2699 if (!strcmp(k, "pack.deltacachelimit")) {2700 cache_max_small_delta_size = git_config_int(k, v);2701 return 0;2702 }2703 if (!strcmp(k, "pack.writebitmaphashcache")) {2704 if (git_config_bool(k, v))2705 write_bitmap_options |= BITMAP_OPT_HASH_CACHE;2706 else2707 write_bitmap_options &= ~BITMAP_OPT_HASH_CACHE;2708 }2709 if (!strcmp(k, "pack.usebitmaps")) {2710 use_bitmap_index_default = git_config_bool(k, v);2711 return 0;2712 }2713 if (!strcmp(k, "pack.threads")) {2714 delta_search_threads = git_config_int(k, v);2715 if (delta_search_threads < 0)2716 die(_("invalid number of threads specified (%d)"),2717 delta_search_threads);2718 if (!HAVE_THREADS && delta_search_threads != 1) {2719 warning(_("no threads support, ignoring %s"), k);2720 delta_search_threads = 0;2721 }2722 return 0;2723 }2724 if (!strcmp(k, "pack.indexversion")) {2725 pack_idx_opts.version = git_config_int(k, v);2726 if (pack_idx_opts.version > 2)2727 die(_("bad pack.indexversion=%"PRIu32),2728 pack_idx_opts.version);2729 return 0;2730 }2731 return git_default_config(k, v, cb);2732}27332734static void read_object_list_from_stdin(void)2735{2736 char line[GIT_MAX_HEXSZ + 1 + PATH_MAX + 2];2737 struct object_id oid;2738 const char *p;27392740 for (;;) {2741 if (!fgets(line, sizeof(line), stdin)) {2742 if (feof(stdin))2743 break;2744 if (!ferror(stdin))2745 die("BUG: fgets returned NULL, not EOF, not error!");2746 if (errno != EINTR)2747 die_errno("fgets");2748 clearerr(stdin);2749 continue;2750 }2751 if (line[0] == '-') {2752 if (get_oid_hex(line+1, &oid))2753 die(_("expected edge object ID, got garbage:\n %s"),2754 line);2755 add_preferred_base(&oid);2756 continue;2757 }2758 if (parse_oid_hex(line, &oid, &p))2759 die(_("expected object ID, got garbage:\n %s"), line);27602761 add_preferred_base_object(p + 1);2762 add_object_entry(&oid, OBJ_NONE, p + 1, 0);2763 }2764}27652766/* Remember to update object flag allocation in object.h */2767#define OBJECT_ADDED (1u<<20)27682769static void show_commit(struct commit *commit, void *data)2770{2771 add_object_entry(&commit->object.oid, OBJ_COMMIT, NULL, 0);2772 commit->object.flags |= OBJECT_ADDED;27732774 if (write_bitmap_index)2775 index_commit_for_bitmap(commit);27762777 if (use_delta_islands)2778 propagate_island_marks(commit);2779}27802781static void show_object(struct object *obj, const char *name, void *data)2782{2783 add_preferred_base_object(name);2784 add_object_entry(&obj->oid, obj->type, name, 0);2785 obj->flags |= OBJECT_ADDED;27862787 if (use_delta_islands) {2788 const char *p;2789 unsigned depth;2790 struct object_entry *ent;27912792 /* the empty string is a root tree, which is depth 0 */2793 depth = *name ? 1 : 0;2794 for (p = strchr(name, '/'); p; p = strchr(p + 1, '/'))2795 depth++;27962797 ent = packlist_find(&to_pack, obj->oid.hash, NULL);2798 if (ent && depth > oe_tree_depth(&to_pack, ent))2799 oe_set_tree_depth(&to_pack, ent, depth);2800 }2801}28022803static void show_object__ma_allow_any(struct object *obj, const char *name, void *data)2804{2805 assert(arg_missing_action == MA_ALLOW_ANY);28062807 /*2808 * Quietly ignore ALL missing objects. This avoids problems with2809 * staging them now and getting an odd error later.2810 */2811 if (!has_object_file(&obj->oid))2812 return;28132814 show_object(obj, name, data);2815}28162817static void show_object__ma_allow_promisor(struct object *obj, const char *name, void *data)2818{2819 assert(arg_missing_action == MA_ALLOW_PROMISOR);28202821 /*2822 * Quietly ignore EXPECTED missing objects. This avoids problems with2823 * staging them now and getting an odd error later.2824 */2825 if (!has_object_file(&obj->oid) && is_promisor_object(&obj->oid))2826 return;28272828 show_object(obj, name, data);2829}28302831static int option_parse_missing_action(const struct option *opt,2832 const char *arg, int unset)2833{2834 assert(arg);2835 assert(!unset);28362837 if (!strcmp(arg, "error")) {2838 arg_missing_action = MA_ERROR;2839 fn_show_object = show_object;2840 return 0;2841 }28422843 if (!strcmp(arg, "allow-any")) {2844 arg_missing_action = MA_ALLOW_ANY;2845 fetch_if_missing = 0;2846 fn_show_object = show_object__ma_allow_any;2847 return 0;2848 }28492850 if (!strcmp(arg, "allow-promisor")) {2851 arg_missing_action = MA_ALLOW_PROMISOR;2852 fetch_if_missing = 0;2853 fn_show_object = show_object__ma_allow_promisor;2854 return 0;2855 }28562857 die(_("invalid value for --missing"));2858 return 0;2859}28602861static void show_edge(struct commit *commit)2862{2863 add_preferred_base(&commit->object.oid);2864}28652866struct in_pack_object {2867 off_t offset;2868 struct object *object;2869};28702871struct in_pack {2872 unsigned int alloc;2873 unsigned int nr;2874 struct in_pack_object *array;2875};28762877static void mark_in_pack_object(struct object *object, struct packed_git *p, struct in_pack *in_pack)2878{2879 in_pack->array[in_pack->nr].offset = find_pack_entry_one(object->oid.hash, p);2880 in_pack->array[in_pack->nr].object = object;2881 in_pack->nr++;2882}28832884/*2885 * Compare the objects in the offset order, in order to emulate the2886 * "git rev-list --objects" output that produced the pack originally.2887 */2888static int ofscmp(const void *a_, const void *b_)2889{2890 struct in_pack_object *a = (struct in_pack_object *)a_;2891 struct in_pack_object *b = (struct in_pack_object *)b_;28922893 if (a->offset < b->offset)2894 return -1;2895 else if (a->offset > b->offset)2896 return 1;2897 else2898 return oidcmp(&a->object->oid, &b->object->oid);2899}29002901static void add_objects_in_unpacked_packs(struct rev_info *revs)2902{2903 struct packed_git *p;2904 struct in_pack in_pack;2905 uint32_t i;29062907 memset(&in_pack, 0, sizeof(in_pack));29082909 for (p = get_all_packs(the_repository); p; p = p->next) {2910 struct object_id oid;2911 struct object *o;29122913 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)2914 continue;2915 if (open_pack_index(p))2916 die(_("cannot open pack index"));29172918 ALLOC_GROW(in_pack.array,2919 in_pack.nr + p->num_objects,2920 in_pack.alloc);29212922 for (i = 0; i < p->num_objects; i++) {2923 nth_packed_object_oid(&oid, p, i);2924 o = lookup_unknown_object(oid.hash);2925 if (!(o->flags & OBJECT_ADDED))2926 mark_in_pack_object(o, p, &in_pack);2927 o->flags |= OBJECT_ADDED;2928 }2929 }29302931 if (in_pack.nr) {2932 QSORT(in_pack.array, in_pack.nr, ofscmp);2933 for (i = 0; i < in_pack.nr; i++) {2934 struct object *o = in_pack.array[i].object;2935 add_object_entry(&o->oid, o->type, "", 0);2936 }2937 }2938 free(in_pack.array);2939}29402941static int add_loose_object(const struct object_id *oid, const char *path,2942 void *data)2943{2944 enum object_type type = oid_object_info(the_repository, oid, NULL);29452946 if (type < 0) {2947 warning(_("loose object at %s could not be examined"), path);2948 return 0;2949 }29502951 add_object_entry(oid, type, "", 0);2952 return 0;2953}29542955/*2956 * We actually don't even have to worry about reachability here.2957 * add_object_entry will weed out duplicates, so we just add every2958 * loose object we find.2959 */2960static void add_unreachable_loose_objects(void)2961{2962 for_each_loose_file_in_objdir(get_object_directory(),2963 add_loose_object,2964 NULL, NULL, NULL);2965}29662967static int has_sha1_pack_kept_or_nonlocal(const struct object_id *oid)2968{2969 static struct packed_git *last_found = (void *)1;2970 struct packed_git *p;29712972 p = (last_found != (void *)1) ? last_found :2973 get_all_packs(the_repository);29742975 while (p) {2976 if ((!p->pack_local || p->pack_keep ||2977 p->pack_keep_in_core) &&2978 find_pack_entry_one(oid->hash, p)) {2979 last_found = p;2980 return 1;2981 }2982 if (p == last_found)2983 p = get_all_packs(the_repository);2984 else2985 p = p->next;2986 if (p == last_found)2987 p = p->next;2988 }2989 return 0;2990}29912992/*2993 * Store a list of sha1s that are should not be discarded2994 * because they are either written too recently, or are2995 * reachable from another object that was.2996 *2997 * This is filled by get_object_list.2998 */2999static struct oid_array recent_objects;30003001static int loosened_object_can_be_discarded(const struct object_id *oid,3002 timestamp_t mtime)3003{3004 if (!unpack_unreachable_expiration)3005 return 0;3006 if (mtime > unpack_unreachable_expiration)3007 return 0;3008 if (oid_array_lookup(&recent_objects, oid) >= 0)3009 return 0;3010 return 1;3011}30123013static void loosen_unused_packed_objects(struct rev_info *revs)3014{3015 struct packed_git *p;3016 uint32_t i;3017 struct object_id oid;30183019 for (p = get_all_packs(the_repository); p; p = p->next) {3020 if (!p->pack_local || p->pack_keep || p->pack_keep_in_core)3021 continue;30223023 if (open_pack_index(p))3024 die(_("cannot open pack index"));30253026 for (i = 0; i < p->num_objects; i++) {3027 nth_packed_object_oid(&oid, p, i);3028 if (!packlist_find(&to_pack, oid.hash, NULL) &&3029 !has_sha1_pack_kept_or_nonlocal(&oid) &&3030 !loosened_object_can_be_discarded(&oid, p->mtime))3031 if (force_object_loose(&oid, p->mtime))3032 die(_("unable to force loose object"));3033 }3034 }3035}30363037/*3038 * This tracks any options which pack-reuse code expects to be on, or which a3039 * reader of the pack might not understand, and which would therefore prevent3040 * blind reuse of what we have on disk.3041 */3042static int pack_options_allow_reuse(void)3043{3044 return pack_to_stdout &&3045 allow_ofs_delta &&3046 !ignore_packed_keep_on_disk &&3047 !ignore_packed_keep_in_core &&3048 (!local || !have_non_local_packs) &&3049 !incremental;3050}30513052static int get_object_list_from_bitmap(struct rev_info *revs)3053{3054 if (!(bitmap_git = prepare_bitmap_walk(revs)))3055 return -1;30563057 if (pack_options_allow_reuse() &&3058 !reuse_partial_packfile_from_bitmap(3059 bitmap_git,3060 &reuse_packfile,3061 &reuse_packfile_objects,3062 &reuse_packfile_offset)) {3063 assert(reuse_packfile_objects);3064 nr_result += reuse_packfile_objects;3065 display_progress(progress_state, nr_result);3066 }30673068 traverse_bitmap_commit_list(bitmap_git, &add_object_entry_from_bitmap);3069 return 0;3070}30713072static void record_recent_object(struct object *obj,3073 const char *name,3074 void *data)3075{3076 oid_array_append(&recent_objects, &obj->oid);3077}30783079static void record_recent_commit(struct commit *commit, void *data)3080{3081 oid_array_append(&recent_objects, &commit->object.oid);3082}30833084static void get_object_list(int ac, const char **av)3085{3086 struct rev_info revs;3087 char line[1000];3088 int flags = 0;3089 int save_warning;30903091 repo_init_revisions(the_repository, &revs, NULL);3092 save_commit_buffer = 0;3093 revs.allow_exclude_promisor_objects_opt = 1;3094 setup_revisions(ac, av, &revs, NULL);30953096 /* make sure shallows are read */3097 is_repository_shallow(the_repository);30983099 save_warning = warn_on_object_refname_ambiguity;3100 warn_on_object_refname_ambiguity = 0;31013102 while (fgets(line, sizeof(line), stdin) != NULL) {3103 int len = strlen(line);3104 if (len && line[len - 1] == '\n')3105 line[--len] = 0;3106 if (!len)3107 break;3108 if (*line == '-') {3109 if (!strcmp(line, "--not")) {3110 flags ^= UNINTERESTING;3111 write_bitmap_index = 0;3112 continue;3113 }3114 if (starts_with(line, "--shallow ")) {3115 struct object_id oid;3116 if (get_oid_hex(line + 10, &oid))3117 die("not an SHA-1 '%s'", line + 10);3118 register_shallow(the_repository, &oid);3119 use_bitmap_index = 0;3120 continue;3121 }3122 die(_("not a rev '%s'"), line);3123 }3124 if (handle_revision_arg(line, &revs, flags, REVARG_CANNOT_BE_FILENAME))3125 die(_("bad revision '%s'"), line);3126 }31273128 warn_on_object_refname_ambiguity = save_warning;31293130 if (use_bitmap_index && !get_object_list_from_bitmap(&revs))3131 return;31323133 if (use_delta_islands)3134 load_delta_islands();31353136 if (prepare_revision_walk(&revs))3137 die(_("revision walk setup failed"));3138 mark_edges_uninteresting(&revs, show_edge);31393140 if (!fn_show_object)3141 fn_show_object = show_object;3142 traverse_commit_list_filtered(&filter_options, &revs,3143 show_commit, fn_show_object, NULL,3144 NULL);31453146 if (unpack_unreachable_expiration) {3147 revs.ignore_missing_links = 1;3148 if (add_unseen_recent_objects_to_traversal(&revs,3149 unpack_unreachable_expiration))3150 die(_("unable to add recent objects"));3151 if (prepare_revision_walk(&revs))3152 die(_("revision walk setup failed"));3153 traverse_commit_list(&revs, record_recent_commit,3154 record_recent_object, NULL);3155 }31563157 if (keep_unreachable)3158 add_objects_in_unpacked_packs(&revs);3159 if (pack_loose_unreachable)3160 add_unreachable_loose_objects();3161 if (unpack_unreachable)3162 loosen_unused_packed_objects(&revs);31633164 oid_array_clear(&recent_objects);3165}31663167static void add_extra_kept_packs(const struct string_list *names)3168{3169 struct packed_git *p;31703171 if (!names->nr)3172 return;31733174 for (p = get_all_packs(the_repository); p; p = p->next) {3175 const char *name = basename(p->pack_name);3176 int i;31773178 if (!p->pack_local)3179 continue;31803181 for (i = 0; i < names->nr; i++)3182 if (!fspathcmp(name, names->items[i].string))3183 break;31843185 if (i < names->nr) {3186 p->pack_keep_in_core = 1;3187 ignore_packed_keep_in_core = 1;3188 continue;3189 }3190 }3191}31923193static int option_parse_index_version(const struct option *opt,3194 const char *arg, int unset)3195{3196 char *c;3197 const char *val = arg;31983199 BUG_ON_OPT_NEG(unset);32003201 pack_idx_opts.version = strtoul(val, &c, 10);3202 if (pack_idx_opts.version > 2)3203 die(_("unsupported index version %s"), val);3204 if (*c == ',' && c[1])3205 pack_idx_opts.off32_limit = strtoul(c+1, &c, 0);3206 if (*c || pack_idx_opts.off32_limit & 0x80000000)3207 die(_("bad index version '%s'"), val);3208 return 0;3209}32103211static int option_parse_unpack_unreachable(const struct option *opt,3212 const char *arg, int unset)3213{3214 if (unset) {3215 unpack_unreachable = 0;3216 unpack_unreachable_expiration = 0;3217 }3218 else {3219 unpack_unreachable = 1;3220 if (arg)3221 unpack_unreachable_expiration = approxidate(arg);3222 }3223 return 0;3224}32253226int cmd_pack_objects(int argc, const char **argv, const char *prefix)3227{3228 int use_internal_rev_list = 0;3229 int shallow = 0;3230 int all_progress_implied = 0;3231 struct argv_array rp = ARGV_ARRAY_INIT;3232 int rev_list_unpacked = 0, rev_list_all = 0, rev_list_reflog = 0;3233 int rev_list_index = 0;3234 struct string_list keep_pack_list = STRING_LIST_INIT_NODUP;3235 struct option pack_objects_options[] = {3236 OPT_SET_INT('q', "quiet", &progress,3237 N_("do not show progress meter"), 0),3238 OPT_SET_INT(0, "progress", &progress,3239 N_("show progress meter"), 1),3240 OPT_SET_INT(0, "all-progress", &progress,3241 N_("show progress meter during object writing phase"), 2),3242 OPT_BOOL(0, "all-progress-implied",3243 &all_progress_implied,3244 N_("similar to --all-progress when progress meter is shown")),3245 { OPTION_CALLBACK, 0, "index-version", NULL, N_("<version>[,<offset>]"),3246 N_("write the pack index file in the specified idx format version"),3247 PARSE_OPT_NONEG, option_parse_index_version },3248 OPT_MAGNITUDE(0, "max-pack-size", &pack_size_limit,3249 N_("maximum size of each output pack file")),3250 OPT_BOOL(0, "local", &local,3251 N_("ignore borrowed objects from alternate object store")),3252 OPT_BOOL(0, "incremental", &incremental,3253 N_("ignore packed objects")),3254 OPT_INTEGER(0, "window", &window,3255 N_("limit pack window by objects")),3256 OPT_MAGNITUDE(0, "window-memory", &window_memory_limit,3257 N_("limit pack window by memory in addition to object limit")),3258 OPT_INTEGER(0, "depth", &depth,3259 N_("maximum length of delta chain allowed in the resulting pack")),3260 OPT_BOOL(0, "reuse-delta", &reuse_delta,3261 N_("reuse existing deltas")),3262 OPT_BOOL(0, "reuse-object", &reuse_object,3263 N_("reuse existing objects")),3264 OPT_BOOL(0, "delta-base-offset", &allow_ofs_delta,3265 N_("use OFS_DELTA objects")),3266 OPT_INTEGER(0, "threads", &delta_search_threads,3267 N_("use threads when searching for best delta matches")),3268 OPT_BOOL(0, "non-empty", &non_empty,3269 N_("do not create an empty pack output")),3270 OPT_BOOL(0, "revs", &use_internal_rev_list,3271 N_("read revision arguments from standard input")),3272 OPT_SET_INT_F(0, "unpacked", &rev_list_unpacked,3273 N_("limit the objects to those that are not yet packed"),3274 1, PARSE_OPT_NONEG),3275 OPT_SET_INT_F(0, "all", &rev_list_all,3276 N_("include objects reachable from any reference"),3277 1, PARSE_OPT_NONEG),3278 OPT_SET_INT_F(0, "reflog", &rev_list_reflog,3279 N_("include objects referred by reflog entries"),3280 1, PARSE_OPT_NONEG),3281 OPT_SET_INT_F(0, "indexed-objects", &rev_list_index,3282 N_("include objects referred to by the index"),3283 1, PARSE_OPT_NONEG),3284 OPT_BOOL(0, "stdout", &pack_to_stdout,3285 N_("output pack to stdout")),3286 OPT_BOOL(0, "include-tag", &include_tag,3287 N_("include tag objects that refer to objects to be packed")),3288 OPT_BOOL(0, "keep-unreachable", &keep_unreachable,3289 N_("keep unreachable objects")),3290 OPT_BOOL(0, "pack-loose-unreachable", &pack_loose_unreachable,3291 N_("pack loose unreachable objects")),3292 { OPTION_CALLBACK, 0, "unpack-unreachable", NULL, N_("time"),3293 N_("unpack unreachable objects newer than <time>"),3294 PARSE_OPT_OPTARG, option_parse_unpack_unreachable },3295 OPT_BOOL(0, "thin", &thin,3296 N_("create thin packs")),3297 OPT_BOOL(0, "shallow", &shallow,3298 N_("create packs suitable for shallow fetches")),3299 OPT_BOOL(0, "honor-pack-keep", &ignore_packed_keep_on_disk,3300 N_("ignore packs that have companion .keep file")),3301 OPT_STRING_LIST(0, "keep-pack", &keep_pack_list, N_("name"),3302 N_("ignore this pack")),3303 OPT_INTEGER(0, "compression", &pack_compression_level,3304 N_("pack compression level")),3305 OPT_SET_INT(0, "keep-true-parents", &grafts_replace_parents,3306 N_("do not hide commits by grafts"), 0),3307 OPT_BOOL(0, "use-bitmap-index", &use_bitmap_index,3308 N_("use a bitmap index if available to speed up counting objects")),3309 OPT_BOOL(0, "write-bitmap-index", &write_bitmap_index,3310 N_("write a bitmap index together with the pack index")),3311 OPT_PARSE_LIST_OBJECTS_FILTER(&filter_options),3312 { OPTION_CALLBACK, 0, "missing", NULL, N_("action"),3313 N_("handling for missing objects"), PARSE_OPT_NONEG,3314 option_parse_missing_action },3315 OPT_BOOL(0, "exclude-promisor-objects", &exclude_promisor_objects,3316 N_("do not pack objects in promisor packfiles")),3317 OPT_BOOL(0, "delta-islands", &use_delta_islands,3318 N_("respect islands during delta compression")),3319 OPT_END(),3320 };33213322 if (DFS_NUM_STATES > (1 << OE_DFS_STATE_BITS))3323 BUG("too many dfs states, increase OE_DFS_STATE_BITS");33243325 read_replace_refs = 0;33263327 reset_pack_idx_option(&pack_idx_opts);3328 git_config(git_pack_config, NULL);33293330 progress = isatty(2);3331 argc = parse_options(argc, argv, prefix, pack_objects_options,3332 pack_usage, 0);33333334 if (argc) {3335 base_name = argv[0];3336 argc--;3337 }3338 if (pack_to_stdout != !base_name || argc)3339 usage_with_options(pack_usage, pack_objects_options);33403341 if (depth >= (1 << OE_DEPTH_BITS)) {3342 warning(_("delta chain depth %d is too deep, forcing %d"),3343 depth, (1 << OE_DEPTH_BITS) - 1);3344 depth = (1 << OE_DEPTH_BITS) - 1;3345 }3346 if (cache_max_small_delta_size >= (1U << OE_Z_DELTA_BITS)) {3347 warning(_("pack.deltaCacheLimit is too high, forcing %d"),3348 (1U << OE_Z_DELTA_BITS) - 1);3349 cache_max_small_delta_size = (1U << OE_Z_DELTA_BITS) - 1;3350 }33513352 argv_array_push(&rp, "pack-objects");3353 if (thin) {3354 use_internal_rev_list = 1;3355 argv_array_push(&rp, shallow3356 ? "--objects-edge-aggressive"3357 : "--objects-edge");3358 } else3359 argv_array_push(&rp, "--objects");33603361 if (rev_list_all) {3362 use_internal_rev_list = 1;3363 argv_array_push(&rp, "--all");3364 }3365 if (rev_list_reflog) {3366 use_internal_rev_list = 1;3367 argv_array_push(&rp, "--reflog");3368 }3369 if (rev_list_index) {3370 use_internal_rev_list = 1;3371 argv_array_push(&rp, "--indexed-objects");3372 }3373 if (rev_list_unpacked) {3374 use_internal_rev_list = 1;3375 argv_array_push(&rp, "--unpacked");3376 }33773378 if (exclude_promisor_objects) {3379 use_internal_rev_list = 1;3380 fetch_if_missing = 0;3381 argv_array_push(&rp, "--exclude-promisor-objects");3382 }3383 if (unpack_unreachable || keep_unreachable || pack_loose_unreachable)3384 use_internal_rev_list = 1;33853386 if (!reuse_object)3387 reuse_delta = 0;3388 if (pack_compression_level == -1)3389 pack_compression_level = Z_DEFAULT_COMPRESSION;3390 else if (pack_compression_level < 0 || pack_compression_level > Z_BEST_COMPRESSION)3391 die(_("bad pack compression level %d"), pack_compression_level);33923393 if (!delta_search_threads) /* --threads=0 means autodetect */3394 delta_search_threads = online_cpus();33953396 if (!HAVE_THREADS && delta_search_threads != 1)3397 warning(_("no threads support, ignoring --threads"));3398 if (!pack_to_stdout && !pack_size_limit)3399 pack_size_limit = pack_size_limit_cfg;3400 if (pack_to_stdout && pack_size_limit)3401 die(_("--max-pack-size cannot be used to build a pack for transfer"));3402 if (pack_size_limit && pack_size_limit < 1024*1024) {3403 warning(_("minimum pack size limit is 1 MiB"));3404 pack_size_limit = 1024*1024;3405 }34063407 if (!pack_to_stdout && thin)3408 die(_("--thin cannot be used to build an indexable pack"));34093410 if (keep_unreachable && unpack_unreachable)3411 die(_("--keep-unreachable and --unpack-unreachable are incompatible"));3412 if (!rev_list_all || !rev_list_reflog || !rev_list_index)3413 unpack_unreachable_expiration = 0;34143415 if (filter_options.choice) {3416 if (!pack_to_stdout)3417 die(_("cannot use --filter without --stdout"));3418 use_bitmap_index = 0;3419 }34203421 /*3422 * "soft" reasons not to use bitmaps - for on-disk repack by default we want3423 *3424 * - to produce good pack (with bitmap index not-yet-packed objects are3425 * packed in suboptimal order).3426 *3427 * - to use more robust pack-generation codepath (avoiding possible3428 * bugs in bitmap code and possible bitmap index corruption).3429 */3430 if (!pack_to_stdout)3431 use_bitmap_index_default = 0;34323433 if (use_bitmap_index < 0)3434 use_bitmap_index = use_bitmap_index_default;34353436 /* "hard" reasons not to use bitmaps; these just won't work at all */3437 if (!use_internal_rev_list || (!pack_to_stdout && write_bitmap_index) || is_repository_shallow(the_repository))3438 use_bitmap_index = 0;34393440 if (pack_to_stdout || !rev_list_all)3441 write_bitmap_index = 0;34423443 if (use_delta_islands)3444 argv_array_push(&rp, "--topo-order");34453446 if (progress && all_progress_implied)3447 progress = 2;34483449 add_extra_kept_packs(&keep_pack_list);3450 if (ignore_packed_keep_on_disk) {3451 struct packed_git *p;3452 for (p = get_all_packs(the_repository); p; p = p->next)3453 if (p->pack_local && p->pack_keep)3454 break;3455 if (!p) /* no keep-able packs found */3456 ignore_packed_keep_on_disk = 0;3457 }3458 if (local) {3459 /*3460 * unlike ignore_packed_keep_on_disk above, we do not3461 * want to unset "local" based on looking at packs, as3462 * it also covers non-local objects3463 */3464 struct packed_git *p;3465 for (p = get_all_packs(the_repository); p; p = p->next) {3466 if (!p->pack_local) {3467 have_non_local_packs = 1;3468 break;3469 }3470 }3471 }34723473 prepare_packing_data(&to_pack);34743475 if (progress)3476 progress_state = start_progress(_("Enumerating objects"), 0);3477 if (!use_internal_rev_list)3478 read_object_list_from_stdin();3479 else {3480 get_object_list(rp.argc, rp.argv);3481 argv_array_clear(&rp);3482 }3483 cleanup_preferred_base();3484 if (include_tag && nr_result)3485 for_each_ref(add_ref_tag, NULL);3486 stop_progress(&progress_state);34873488 if (non_empty && !nr_result)3489 return 0;3490 if (nr_result)3491 prepare_pack(window, depth);3492 write_pack_file();3493 if (progress)3494 fprintf_ln(stderr,3495 _("Total %"PRIu32" (delta %"PRIu32"),"3496 " reused %"PRIu32" (delta %"PRIu32")"),3497 written, written_delta, reused, reused_delta);3498 return 0;3499}